53 #ifndef INCLUDED_volk_16i_max_star_horizontal_16i_a_H 54 #define INCLUDED_volk_16i_max_star_horizontal_16i_a_H 64 #include <emmintrin.h> 65 #include <tmmintrin.h> 66 #include <xmmintrin.h> 70 unsigned int num_points)
72 const unsigned int num_bytes = num_points * 2;
74 static const uint8_t shufmask0[16] = {
75 0x00, 0x01, 0x04, 0x05, 0x08, 0x09, 0x0c, 0x0d,
76 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff
78 static const uint8_t shufmask1[16] = {
79 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
80 0x00, 0x01, 0x04, 0x05, 0x08, 0x09, 0x0c, 0x0d
82 static const uint8_t andmask0[16] = {
83 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02,
84 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
86 static const uint8_t andmask1[16] = {
87 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
88 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02
91 __m128i xmm0, xmm1, xmm2, xmm3, xmm4;
92 __m128i xmm5, xmm6, xmm7, xmm8;
94 xmm4 = _mm_load_si128((__m128i*)shufmask0);
95 xmm5 = _mm_load_si128((__m128i*)shufmask1);
96 xmm6 = _mm_load_si128((__m128i*)andmask0);
97 xmm7 = _mm_load_si128((__m128i*)andmask1);
99 __m128i *p_target, *p_src0;
101 p_target = (__m128i*)target;
102 p_src0 = (__m128i*)src0;
104 int bound = num_bytes >> 5;
105 int intermediate = (num_bytes >> 4) & 1;
106 int leftovers = (num_bytes >> 1) & 7;
110 for (
i = 0;
i < bound; ++
i) {
111 xmm0 = _mm_load_si128(p_src0);
112 xmm1 = _mm_load_si128(&p_src0[1]);
114 xmm2 = _mm_xor_si128(xmm2, xmm2);
117 xmm3 = _mm_hsub_epi16(xmm0, xmm1);
119 xmm2 = _mm_cmpgt_epi16(xmm2, xmm3);
121 xmm8 = _mm_and_si128(xmm2, xmm6);
122 xmm3 = _mm_and_si128(xmm2, xmm7);
125 xmm8 = _mm_add_epi8(xmm8, xmm4);
126 xmm3 = _mm_add_epi8(xmm3, xmm5);
128 xmm0 = _mm_shuffle_epi8(xmm0, xmm8);
129 xmm1 = _mm_shuffle_epi8(xmm1, xmm3);
132 xmm3 = _mm_add_epi16(xmm0, xmm1);
135 _mm_store_si128(p_target, xmm3);
141 xmm0 = _mm_load_si128(p_src0);
143 xmm2 = _mm_xor_si128(xmm2, xmm2);
146 xmm3 = _mm_hsub_epi16(xmm0, xmm1);
147 xmm2 = _mm_cmpgt_epi16(xmm2, xmm3);
149 xmm8 = _mm_and_si128(xmm2, xmm6);
151 xmm3 = _mm_add_epi8(xmm8, xmm4);
153 xmm0 = _mm_shuffle_epi8(xmm0, xmm3);
155 _mm_storel_pd((
double*)p_target,
bit128_p(&xmm0)->double_vec);
157 p_target = (__m128i*)((int8_t*)p_target + 8);
160 for (
i = (bound << 4) + (intermediate << 3);
161 i < (bound << 4) + (intermediate << 3) + leftovers;
163 target[
i >> 1] = ((int16_t)(src0[
i] - src0[
i + 1]) > 0) ? src0[
i] : src0[
i + 1];
171 #include <arm_neon.h> 174 unsigned int num_points)
176 const unsigned int eighth_points = num_points / 16;
178 int16x8x2_t input_vec;
179 int16x8_t diff, max_vec, zeros;
180 uint16x8_t comp1, comp2;
181 zeros = vdupq_n_s16(0);
182 for (number = 0; number < eighth_points; ++number) {
183 input_vec = vld2q_s16(src0);
185 diff = vsubq_s16(input_vec.val[0], input_vec.val[1]);
186 comp1 = vcgeq_s16(diff, zeros);
187 comp2 = vcltq_s16(diff, zeros);
189 input_vec.val[0] = vandq_s16(input_vec.val[0], (int16x8_t)comp1);
190 input_vec.val[1] = vandq_s16(input_vec.val[1], (int16x8_t)comp2);
192 max_vec = vaddq_s16(input_vec.val[0], input_vec.val[1]);
193 vst1q_s16(target, max_vec);
197 for (number = 0; number < num_points % 16; number += 2) {
198 target[number >> 1] = ((int16_t)(src0[number] - src0[number + 1]) > 0)
205 #ifdef LV_HAVE_NEONV7 206 extern void volk_16i_max_star_horizontal_16i_a_neonasm(int16_t* target,
208 unsigned int num_points);
211 #ifdef LV_HAVE_GENERIC 214 unsigned int num_points)
216 const unsigned int num_bytes = num_points * 2;
220 int bound = num_bytes >> 1;
222 for (
i = 0;
i < bound;
i += 2) {
223 target[
i >> 1] = ((int16_t)(src0[
i] - src0[
i + 1]) > 0) ? src0[
i] : src0[
i + 1];
#define bit128_p(x)
Definition: volk_common.h:142
static void volk_16i_max_star_horizontal_16i_generic(int16_t *target, int16_t *src0, unsigned int num_points)
Definition: volk_16i_max_star_horizontal_16i.h:212
for i
Definition: volk_config_fixed.tmpl.h:25
static void volk_16i_max_star_horizontal_16i_neon(int16_t *target, int16_t *src0, unsigned int num_points)
Definition: volk_16i_max_star_horizontal_16i.h:172
static void volk_16i_max_star_horizontal_16i_a_ssse3(int16_t *target, int16_t *src0, unsigned int num_points)
Definition: volk_16i_max_star_horizontal_16i.h:68