# # Copyright (C) 2026 Intel Corporation # # Please submit issues or comments https://github.com/openvinotoolkit/openvino/issues # # Fedora's openvino (f45, 2026.0.0) updated to 2026.4.1 with the Intel NPU # plugin enabled. The plugin compiles models with Intel's NPU compiler # (libopenvino_intel_npu_compiler*.so from intel-npu-compiler); the build-time # download of that compiler (ENABLE_INTEL_NPU_COMPILER) is disabled. %global so_ver 2641 %global onednn_commit 1289c3b65dd6a119a5ed12a816517d9c3a21d81b %global mlas_commit d1bc25ec4660cddd87804fcf03b2411b5dfb2e94 %global onednn_gpu_commit a3d459721b72c3a9d2685b46c03dafee7af0f25c %global npu_ext_commit b11dc98b48f9f0dbb8c3b44f73b42f295a32097f %global desc %{expand:OpenVINO is an open-source toolkit for optimizing and deploying deep learning models from cloud to edge. It accelerates deep learning inference across various use cases, such as generative AI, video, audio, and language with models from popular frameworks like PyTorch, TensorFlow, ONNX, and more.} Name: openvino Version: 2026.4.1 Release: 2%{?dist} Summary: Toolkit for optimizing and deploying AI inference # Most of the source code is Apache-2.0, with the following exceptions: # src/core/reference/include/openvino/reference/deformable_psroi_pooling.hpp : MIT # src/core/src/type/nf4.cpp : MIT # src/plugins/intel_cpu/src/hash_builder.hpp : BSL-1.0 # src/core/reference/include/openvino/reference/interpolate_pil.hpp : HPND # oneDNN: BSL-1.0, BSD-3-Clause, (GPL-2.0-only OR BSD-3-Clause), MIT License: Apache-2.0 AND MIT AND BSL-1.0 AND HPND AND BSD-3-Clause AND (GPL-2.0-only OR BSD-3-Clause) URL: https://github.com/openvinotoolkit/openvino Source0: %{url}/archive/%{version}/%{name}-%{version}.tar.gz Source1: https://github.com/openvinotoolkit/oneDNN/archive/%{onednn_commit}/onednn-%{sub %{onednn_commit} 1 7}.tar.gz Source2: https://github.com/openvinotoolkit/mlas/archive/%{mlas_commit}/mlas-%{sub %{mlas_commit} 1 7}.tar.gz Source3: https://github.com/oneapi-src/oneDNN/archive/%{onednn_gpu_commit}/onednn-gpu-%{sub %{onednn_gpu_commit} 1 7}.tar.gz Source4: https://github.com/intel/level-zero-npu-extensions/archive/%{npu_ext_commit}/level-zero-npu-extensions-%{sub %{npu_ext_commit} 1 7}.tar.gz Patch0: protobuf_version.patch Patch1: xbyak-gflags-system-modules.patch Patch2: samples-system-gflags-json.patch # Configure the NPU plugin against the system Level Zero Patch3: npu-system-level-zero.patch # The pybind11 3.x call_guard and OpenCL-CLHPP macro fixes Fedora carries # for 2026.0.0 are upstream in 2026.4. ExclusiveArch: x86_64 BuildRequires: cmake BuildRequires: flatbuffers-compiler BuildRequires: flatbuffers-devel BuildRequires: gcc BuildRequires: gcc-c++ BuildRequires: gflags-devel BuildRequires: glibc-devel BuildRequires: nlohmann-json-devel BuildRequires: oneapi-level-zero-devel BuildRequires: onnx-devel >= 1.22 BuildRequires: opencl-headers BuildRequires: pkgconfig(OpenCL) %if 0%{?fedora} > 44 BuildRequires: protobuf-devel >= 4 %else BuildRequires: protobuf-devel %endif BuildRequires: pugixml-devel BuildRequires: pybind11-devel BuildRequires: python3-devel BuildRequires: python3-numpy BuildRequires: python3-setuptools BuildRequires: python3-wheel BuildRequires: snappy-devel BuildRequires: tbb-devel BuildRequires: xbyak-devel BuildRequires: zlib-devel Requires: lib%{name}-ir-frontend = %{version}-%{release} Requires: lib%{name}-gguf-frontend = %{version}-%{release} Requires: lib%{name}-pytorch-frontend = %{version}-%{release} Requires: lib%{name}-onnx-frontend = %{version}-%{release} Requires: lib%{name}-paddle-frontend = %{version}-%{release} Requires: lib%{name}-tensorflow-frontend = %{version}-%{release} Requires: lib%{name}-tensorflow-lite-frontend = %{version}-%{release} Recommends: lib%{name}-auto-plugin = %{version}-%{release} Recommends: lib%{name}-auto-batch-plugin = %{version}-%{release} Recommends: lib%{name}-hetero-plugin = %{version}-%{release} Recommends: lib%{name}-intel-cpu-plugin = %{version}-%{release} Recommends: lib%{name}-intel-gpu-plugin = %{version}-%{release} Recommends: lib%{name}-intel-npu-plugin = %{version}-%{release} %description %{desc} %package -n lib%{name}-devel Summary: Development files for %{name} Provides: %{name}-devel = %{version}-%{release} Obsoletes: %{name}-devel < 2025.1.0-14 Requires: %{name}%{?_isa} = %{version}-%{release} Requires: tbb-devel %description -n lib%{name}-devel %{desc} . This package provides the headers and libraries for developing applications with OpenVINO. ## Plugins ## %package -n lib%{name}-auto-plugin Summary: Auto / Multi software plugin for OpenVINO %description -n lib%{name}-auto-plugin %{desc} . This package provides the Auto / Multi software plugin for OpenVINO. %package -n lib%{name}-auto-batch-plugin Summary: Automatic batch software plugin for OpenVINO %description -n lib%{name}-auto-batch-plugin %{desc} . This package provides the automatic batch software plugin for OpenVINO. %package -n lib%{name}-hetero-plugin Summary: Hetero plugin for OpenVINO %description -n lib%{name}-hetero-plugin %{desc} . This package provides the hetero plugin for OpenVINO. %package -n lib%{name}-intel-cpu-plugin Summary: Intel CPU plugin for OpenVINO # Forked version of OpenVINO oneDNN does not have a proper version Provides: bundled(onednn) # Intel MLAS Provides: bundled(mlas) %description -n lib%{name}-intel-cpu-plugin %{desc} . This package provides the Intel CPU plugin for OpenVINO. %package -n lib%{name}-intel-gpu-plugin Summary: Intel GPU plugin for OpenVINO # Forked version of oneapi-src/oneDNN used by the GPU plugin Provides: bundled(onednn-gpu) %description -n lib%{name}-intel-gpu-plugin %{desc} . This package provides the Intel GPU plugin for OpenVINO. %package -n lib%{name}-intel-npu-plugin Summary: Intel NPU plugin for OpenVINO Provides: bundled(level-zero-npu-extensions) # Model compilation for the NPU; also used by intel-npu-driver Recommends: intel-npu-compiler Recommends: intel-npu-driver %description -n lib%{name}-intel-npu-plugin %{desc} . This package provides the Intel NPU plugin for OpenVINO. ## Frontend shared libraries ## %package -n lib%{name}-ir-frontend Summary: OpenVINO IR Frontend Requires: %{name}%{?_isa} = %{version}-%{release} %description -n lib%{name}-ir-frontend %{desc} . This package provides the IR frontend for OpenVINO. %package -n lib%{name}-gguf-frontend Summary: OpenVINO GGUF Frontend Requires: %{name}%{?_isa} = %{version}-%{release} %description -n lib%{name}-gguf-frontend %{desc} . This package provides the GGUF (llama.cpp model format) frontend for OpenVINO. %package -n lib%{name}-onnx-frontend Summary: OpenVINO ONNX Frontend Requires: %{name}%{?_isa} = %{version}-%{release} %description -n lib%{name}-onnx-frontend %{desc} . This package provides the ONNX frontend for OpenVINO. %package -n lib%{name}-paddle-frontend Summary: OpenVINO Paddle Frontend Requires: %{name}%{?_isa} = %{version}-%{release} %description -n lib%{name}-paddle-frontend %{desc} . This package provides the Paddle frontend for OpenVINO. %package -n lib%{name}-pytorch-frontend Summary: OpenVINO PyTorch Frontend Requires: %{name}%{?_isa} = %{version}-%{release} %description -n lib%{name}-pytorch-frontend %{desc} . This package provides the PyTorch frontend for OpenVINO. %package -n lib%{name}-tensorflow-frontend Summary: OpenVINO TensorFlow Frontend Requires: %{name}%{?_isa} = %{version}-%{release} %description -n lib%{name}-tensorflow-frontend %{desc} . This package provides the TensorFlow frontend for OpenVINO. %package -n lib%{name}-tensorflow-lite-frontend Summary: OpenVINO TensorFlow Lite Frontend Requires: %{name}%{?_isa} = %{version}-%{release} %description -n lib%{name}-tensorflow-lite-frontend %{desc} . This package provides the TensorFlow Lite frontend for OpenVINO. ## Samples ## %package -n %{name}-samples Summary: C and C++ samples for OpenVINO BuildArch: noarch Requires: lib%{name}-devel = %{version}-%{release} Requires: cmake Requires: gcc-c++ Requires: gcc Requires: glibc-devel Requires: make Requires: pkgconf-pkg-config Suggests: opencv-devel >= 3.0 Suggests: ocl-icd-devel Suggests: opencl-headers %description -n %{name}-samples %{desc} . This package provides C and C++ source code samples demonstrating how to use OpenVINO runtime and APIs. %package -n python3-%{name} Summary: OpenVINO Python API Requires: %{name}%{?_isa} = %{version}-%{release} Requires: python3-numpy %description -n python3-%{name} OpenVINO Python API allowing users to use the OpenVINO library in their Python code. Python API provides bindings to basic and advanced APIs from OpenVINO runtime. %prep %autosetup -p1 # Fedora provides newer NumPy versions sed -i 's/numpy>=1\.16\.6,<2\.6\.0/numpy>=1.16.6/' pyproject.toml # Intel CPU plugin thirdparty deps tar xf %{SOURCE1} cp -r oneDNN-%{onednn_commit}/* src/plugins/intel_cpu/thirdparty/onednn tar xf %{SOURCE2} cp -r mlas-%{mlas_commit}/* src/plugins/intel_cpu/thirdparty/mlas # Intel GPU plugin thirdparty deps tar xf %{SOURCE3} cp -r oneDNN-%{onednn_gpu_commit}/* src/plugins/intel_gpu/thirdparty/onednn_gpu # Intel NPU plugin thirdparty deps (headers only) tar xf %{SOURCE4} cp -r level-zero-npu-extensions-%{npu_ext_commit}/* src/plugins/intel_npu/thirdparty/level-zero-ext # Python bindings: remove openvino-telemetry (not in Fedora) # It is present in both pyproject.toml (root) and src/bindings/python/requirements.txt sed -i '/openvino-telemetry/d' pyproject.toml sed -i '/openvino-telemetry/d' src/bindings/python/requirements.txt %build %cmake \ -DCMAKE_BUILD_TYPE=RelWithDebInfo \ -DCMAKE_POLICY_VERSION_MINIMUM="3.5.0" \ -DCMAKE_CXX_FLAGS="%{optflags} -Wformat -Wformat-security -Wno-free-nonheap-object -Wno-array-bounds -Wno-stringop-overflow" \ -DCMAKE_COMPILE_WARNING_AS_ERROR=OFF \ -DCMAKE_NO_SYSTEM_FROM_IMPORTED=ON \ -DENABLE_CLANG_FORMAT=OFF \ -DENABLE_PRECOMPILED_HEADERS=OFF \ -DBUILD_SHARED_LIBS=ON \ -DCPACK_GENERATOR=RPM \ -DENABLE_OV_ONNX_FRONTEND=ON \ -DENABLE_OV_PADDLE_FRONTEND=ON \ -DENABLE_OV_PYTORCH_FRONTEND=ON \ -DENABLE_OV_IR_FRONTEND=ON \ -DENABLE_OV_TF_FRONTEND=ON \ -DENABLE_OV_TF_LITE_FRONTEND=ON \ -DENABLE_OV_JAX_FRONTEND=OFF \ -DENABLE_INTEL_NPU=ON \ -DENABLE_INTEL_NPU_COMPILER=OFF \ -DENABLE_INTEL_NPU_INTERNAL=OFF \ -DENABLE_SYSTEM_LEVEL_ZERO=ON \ -DENABLE_TEMPLATE=OFF \ -DENABLE_PROXY=OFF \ -DENABLE_PYTHON=ON \ -DPython3_EXECUTABLE=%{python3} \ -DENABLE_WHEEL=OFF \ -DENABLE_JS=OFF \ -DENABLE_SAMPLES=ON \ -DENABLE_TESTS=OFF \ -DENABLE_PROFILING_ITT=OFF \ -DENABLE_PKGCONFIG_GEN=ON \ -DENABLE_SYSTEM_TBB=ON \ -DENABLE_SYSTEM_OPENCL=ON \ -DENABLE_SYSTEM_PUGIXML=ON \ -DENABLE_SYSTEM_SNAPPY=ON \ -DENABLE_SYSTEM_PROTOBUF=ON \ -DProtobuf_USE_STATIC_LIBS=OFF \ -DENABLE_SYSTEM_FLATBUFFERS=ON \ -DTHREADING=TBB %cmake_build %install %cmake_install # Remove unnecessary files that may be installed rm -rf %{buildroot}%{_prefix}/install_dependencies \ %{buildroot}%{_prefix}/setupvars.sh rm -rf %{buildroot}%{_datadir}/licenses/* rm -rf %{buildroot}%{_datadir}/doc # Install Python components directly (CPACK_GENERATOR=RPM marks them EXCLUDE_FROM_ALL # for the default cmake install, but explicit --component bypasses that) cmake --install %{__cmake_builddir} --prefix %{buildroot}%{_prefix} \ --component pyopenvino_python%{python3_version} cmake --install %{__cmake_builddir} --prefix %{buildroot}%{_prefix} \ --component ovc cmake --install %{__cmake_builddir} --prefix %{buildroot}%{_prefix} \ --component benchmark_app # Generate dist-info metadata (dist_info does not trigger cmake sub-builds) WHEEL_VERSION=%{version} \ %{python3} setup.py dist_info -o %{buildroot}%{python3_sitearch} # The NPU plugin looks for Intel's NPU compiler next to itself; point it at # the copy intel-npu-compiler installs (the driver loads the same library) ln -s ../libopenvino_intel_npu_compiler_loader.so \ %{buildroot}%{_libdir}/%{name}-%{version}/libopenvino_intel_npu_compiler_loader.so # The loader then opens the compiler from its own directory ln -s ../libopenvino_intel_npu_compiler.so \ %{buildroot}%{_libdir}/%{name}-%{version}/libopenvino_intel_npu_compiler.so # Remove openvino-telemetry (not packaged in Fedora) rm -rf %{buildroot}%{python3_sitearch}/openvino_telemetry* rm -vf %{buildroot}%{python3_sitearch}/requirements.txt rm -vf %{buildroot}%{python3_sitearch}/%{name}/preprocess/torchvision/requirements.txt %check LD_LIBRARY_PATH=%{buildroot}%{_libdir} PYTHONPATH=%{buildroot}%{python3_sitearch} \ %{python3} samples/python/hello_query_device/hello_query_device.py LD_LIBRARY_PATH=%{buildroot}%{_libdir} PYTHONPATH=%{buildroot}%{python3_sitearch} \ %{python3} samples/python/model_creation_sample/model_creation_sample.py \ samples/python/model_creation_sample/lenet.bin CPU %ldconfig_scriptlets %ldconfig_scriptlets -n lib%{name}-ir-frontend %ldconfig_scriptlets -n lib%{name}-gguf-frontend %ldconfig_scriptlets -n lib%{name}-onnx-frontend %ldconfig_scriptlets -n lib%{name}-paddle-frontend %ldconfig_scriptlets -n lib%{name}-pytorch-frontend %ldconfig_scriptlets -n lib%{name}-tensorflow-frontend %ldconfig_scriptlets -n lib%{name}-tensorflow-lite-frontend %files %license LICENSE %doc README.md %{_libdir}/lib%{name}.so.%{version} %{_libdir}/lib%{name}.so.%{so_ver} %{_libdir}/lib%{name}_c.so.%{version} %{_libdir}/lib%{name}_c.so.%{so_ver} %files -n lib%{name}-devel %{_includedir}/%{name}/ %{_libdir}/cmake/%{name}%{version}/ %{_libdir}/*.so %{_libdir}/pkgconfig/%{name}.pc ## Plugins %files -n lib%{name}-auto-plugin %dir %{_libdir}/%{name}-%{version} %{_libdir}/%{name}-%{version}/lib%{name}_auto_plugin.so %files -n lib%{name}-auto-batch-plugin %dir %{_libdir}/%{name}-%{version} %{_libdir}/%{name}-%{version}/lib%{name}_auto_batch_plugin.so %files -n lib%{name}-hetero-plugin %dir %{_libdir}/%{name}-%{version} %{_libdir}/%{name}-%{version}/lib%{name}_hetero_plugin.so %files -n lib%{name}-intel-cpu-plugin %dir %{_libdir}/%{name}-%{version} %{_libdir}/%{name}-%{version}/lib%{name}_intel_cpu_plugin.so %files -n lib%{name}-intel-gpu-plugin %dir %{_libdir}/%{name}-%{version} %{_libdir}/%{name}-%{version}/lib%{name}_intel_gpu_plugin.so %{_libdir}/%{name}-%{version}/cache.json %files -n lib%{name}-intel-npu-plugin %dir %{_libdir}/%{name}-%{version} %{_libdir}/%{name}-%{version}/lib%{name}_intel_npu_plugin.so %{_libdir}/%{name}-%{version}/libopenvino_intel_npu_compiler_loader.so %{_libdir}/%{name}-%{version}/libopenvino_intel_npu_compiler.so ## Frontends %files -n lib%{name}-ir-frontend %{_libdir}/lib%{name}_ir_frontend.so.* %files -n lib%{name}-gguf-frontend %{_libdir}/lib%{name}_gguf_frontend.so.* %files -n lib%{name}-onnx-frontend %{_libdir}/lib%{name}_onnx_frontend.so.* %files -n lib%{name}-paddle-frontend %{_libdir}/lib%{name}_paddle_frontend.so.* %files -n lib%{name}-pytorch-frontend %{_libdir}/lib%{name}_pytorch_frontend.so.* %files -n lib%{name}-tensorflow-frontend %{_libdir}/lib%{name}_tensorflow_frontend.so.* %files -n lib%{name}-tensorflow-lite-frontend %{_libdir}/lib%{name}_tensorflow_lite_frontend.so.* ## Samples %files -n %{name}-samples %{_datadir}/openvino/samples/ %files -n python3-%{name} %{python3_sitearch}/%{name} %{python3_sitearch}/%{name}-%{version}.dist-info %changelog * Thu Oct 08 2026 Onuralp SEZER - 2026.4.1-2 - Link the NPU compiler into the plugin directory too; the loader opens it from there, so NPU compilation failed with "Failed to load compiler library" * Wed Oct 07 2026 Onuralp SEZER - 2026.4.1-1 - Update to 2026.4.1 - Enable the Intel NPU plugin (libopenvino-intel-npu-plugin), using intel-npu-compiler for model compilation - Drop the pybind11 and OpenCL-CLHPP patches, fixed upstream - Build against onnx 1.22 - Fix configuring the NPU plugin with the system Level Zero - New libopenvino-gguf-frontend subpackage - Based on Fedora's openvino 2026.0.0 packaging