Python ncnn with vulkan

.github/workflows/release-python.yml

@@ -5,6 +5,14 @@ on:
     tags:
       - '*'
 
+env:
+  DEVELOPER_DIR: /Applications/Xcode_13.4.1.app/Contents/Developer
+  MAC_DEPLOYMENT_TARGET: '10.9'
+  MAC_ARM64_DEPLOYMENT_TARGET: '11.0'
+  ENABLE_BITCODE: OFF
+  ENABLE_ARC: OFF
+  ENABLE_VISIBILITY: OFF
+
 jobs:
   build_sdist:
     name: Build SDist
@@ -62,20 +70,224 @@ jobs:
       with:
         python-version: '3.x'
 
-    - name: brew uninstall libomp
+    # build wheels for ubuntu-20.04
+    - name: Build wheels for ubuntu manylinux
+      if: matrix.os == 'ubuntu-20.04'  &&  matrix.build != 'cp*-musllinux*'
+      uses: pypa/[email protected]
+      env:
+        CIBW_ARCHS_LINUX: ${{ matrix.arch }}
+        CIBW_BUILD: ${{ matrix.build }}
+        CIBW_BUILD_VERBOSITY: 1
+        CIBW_ENVIRONMENT: CMAKE_BUILD_PARALLEL_LEVEL=2 NCNN_VULKAN=ON 
+          VULKAN_SDK=/project/Vulkan-Loader/build/Vulkan-Headers
+          LD_LIBRARY_PATH=/project/Vulkan-Loader/build/loader
+        CIBW_BEFORE_ALL: yum -y install libXrandr-devel &&
+          git clone https://github.com/KhronosGroup/Vulkan-Loader.git &&
+          cd Vulkan-Loader && mkdir build && cd build &&
+          ../scripts/update_deps.py &&
+          cmake -DCMAKE_BUILD_TYPE=Release -DVULKAN_HEADERS_INSTALL_DIR=$(pwd)/Vulkan-Headers/build/install .. &&
+          make -j$(nproc) &&
+          cd Vulkan-Headers &&
+          ln -s ../loader lib
+      with:
+        output-dir: wheelhouse
+
+    - name: Build wheels for ubuntu musllinux
+      if: matrix.os == 'ubuntu-20.04'  &&  matrix.build == 'cp*-musllinux*'
+      uses: pypa/[email protected]
+      env:
+        CIBW_ARCHS_LINUX: ${{ matrix.arch }}
+        CIBW_BUILD: ${{ matrix.build }}
+        CIBW_BUILD_VERBOSITY: 1
+        CIBW_ENVIRONMENT: CMAKE_BUILD_PARALLEL_LEVEL=2 NCNN_VULKAN=ON 
+          VULKAN_SDK=/project/Vulkan-Loader/build/Vulkan-Headers
+          LD_LIBRARY_PATH=/project/Vulkan-Loader/build/loader
+        CIBW_BEFORE_ALL: apk add libxrandr-dev &&
+          git clone https://github.com/KhronosGroup/Vulkan-Loader.git &&
+          cd Vulkan-Loader && mkdir build && cd build &&
+          ../scripts/update_deps.py &&
+          cmake -DCMAKE_BUILD_TYPE=Release -DVULKAN_HEADERS_INSTALL_DIR=$(pwd)/Vulkan-Headers/build/install .. &&
+          make -j$(nproc) &&
+          cd Vulkan-Headers &&
+          ln -s ../loader lib
+      with:
+        output-dir: wheelhouse
+
+    # build wheels for windows-2019
+    - name: Build wheels for windows amd64
+      if: matrix.os == 'windows-2019' && matrix.arch == 'AMD64'
+      uses: pypa/[email protected]
+      env:
+        CIBW_ARCHS_WINDOWS: ${{ matrix.arch }}
+        CIBW_BUILD: ${{ matrix.build }}
+        CIBW_BUILD_VERBOSITY: 1
+        CIBW_ENVIRONMENT_WINDOWS: >
+          PATH="D:\\a\\ncnn\\ncnn\\Vulkan-Loader\\build\\loader\\Release;$PATH"
+          CMAKE_BUILD_PARALLEL_LEVEL=2 NCNN_VULKAN=ON 
+          VULKAN_SDK=D:/a/ncnn/ncnn/Vulkan-Loader/external/Vulkan-Headers
+        CIBW_BEFORE_ALL: git clone https://github.com/KhronosGroup/Vulkan-Loader.git &&
+          cd Vulkan-Loader && mkdir build && cd build &&
+          python3 ../scripts/update_deps.py --dir ../external --config release &&
+          cmake -C ../external/helper.cmake -G "Visual Studio 16 2019" -A x64 -DCMAKE_BUILD_TYPE=Release .. &&
+          cmake --build . --config Release &&
+          mklink /d "D:/a/ncnn/ncnn/Vulkan-Loader/external/Vulkan-Headers/build/install/lib"
+          "D:/a/ncnn/ncnn/Vulkan-Loader/build/loader/Release"
+        CIBW_BEFORE_BUILD: pip install delvewheel
+        CIBW_REPAIR_WHEEL_COMMAND: delvewheel repair -w {dest_dir} {wheel}
+      with:
+        output-dir: wheelhouse
+
+    - name: Build wheels for windows x86
+      if: matrix.os == 'windows-2019' && matrix.arch == 'x86'
+      uses: pypa/[email protected]
+      env:
+        CIBW_ARCHS_WINDOWS: ${{ matrix.arch }}
+        CIBW_BUILD: ${{ matrix.build }}
+        CIBW_BUILD_VERBOSITY: 1
+        CIBW_ENVIRONMENT_WINDOWS: >
+          PATH="D:\\a\\ncnn\\ncnn\\Vulkan-Loader\\build\\loader\\Release;$PATH"
+          CMAKE_BUILD_PARALLEL_LEVEL=2 NCNN_VULKAN=ON 
+          VULKAN_SDK=D:/a/ncnn/ncnn/Vulkan-Loader/external/Vulkan-Headers
+        CIBW_BEFORE_ALL: git clone https://github.com/KhronosGroup/Vulkan-Loader.git &&
+          cd Vulkan-Loader && mkdir build && cd build &&
+          python3 ../scripts/update_deps.py --dir ../external --arch ${{ matrix.arch }} --config release &&
+          cmake -C ../external/helper.cmake -G "Visual Studio 16 2019" -A Win32 -DCMAKE_BUILD_TYPE=Release .. &&
+          cmake --build . --config Release &&
+          mklink /d "D:/a/ncnn/ncnn/Vulkan-Loader/external/Vulkan-Headers/build/install/lib"
+          "D:/a/ncnn/ncnn/Vulkan-Loader/build/loader/Release"
+        CIBW_BEFORE_BUILD: pip install delvewheel
+        CIBW_REPAIR_WHEEL_COMMAND: delvewheel repair -w {dest_dir} {wheel}
+      with:
+        output-dir: wheelhouse
+
+    - name: Build wheels for windows ARM64
+      if: matrix.os == 'windows-2019' && matrix.arch == 'ARM64'
+      uses: pypa/[email protected]
+      env:
+        CIBW_ARCHS_WINDOWS: ${{ matrix.arch }}
+        CIBW_BUILD: ${{ matrix.build }}
+        CIBW_BUILD_VERBOSITY: 1
+        CIBW_ENVIRONMENT_WINDOWS: >
+          PATH="D:\\a\\ncnn\\ncnn\\Vulkan-Loader\\build\\loader\\Release;$PATH"
+          CMAKE_BUILD_PARALLEL_LEVEL=2 NCNN_VULKAN=ON 
+          VULKAN_SDK=D:/a/ncnn/ncnn/Vulkan-Loader/external/Vulkan-Headers
+        CIBW_BEFORE_ALL: git clone https://github.com/KhronosGroup/Vulkan-Loader.git &&
+          cd Vulkan-Loader && mkdir build && cd build &&
+          python3 ../scripts/update_deps.py --dir ../external --config release &&
+          cmake -C ../external/helper.cmake -G "Visual Studio 16 2019" -A ARM64 -DCMAKE_BUILD_TYPE=Release -DUSE_MASM=OFF ..  &&
+          cmake --build . --config Release &&
+          mklink /d "D:/a/ncnn/ncnn/Vulkan-Loader/external/Vulkan-Headers/build/install/lib"
+          "D:/a/ncnn/ncnn/Vulkan-Loader/build/loader/Release"
+        CIBW_BEFORE_BUILD: pip install delvewheel
+        CIBW_REPAIR_WHEEL_COMMAND: delvewheel repair -w {dest_dir} {wheel} --no-dll "msvcp140.dll;vcomp140.dll"
+      with:
+        output-dir: wheelhouse
+
+    # build wheels for macos-latest
+    - name: cache-openmp for macos
       if: matrix.os == 'macos-latest'
+      id: cache-openmp
+      uses: actions/cache@v3
+      with:
+        path: openmp-install
+        key: openmp-macos-install-20230504
+
+    - name: openmp for macos
+      if: matrix.os == 'macos-latest' && steps.cache-openmp.outputs.cache-hit != 'true'
       run: |
-        brew uninstall --ignore-dependencies libomp
+        wget https://github.com/llvm/llvm-project/releases/download/llvmorg-11.0.0/openmp-11.0.0.src.tar.xz
+        tar -xf openmp-11.0.0.src.tar.xz
+        cd openmp-11.0.0.src
+        sed -i'' -e '/.size __kmp_unnamed_critical_addr/d' runtime/src/z_Linux_asm.S
+        sed -i'' -e 's/__kmp_unnamed_critical_addr/___kmp_unnamed_critical_addr/g' runtime/src/z_Linux_asm.S
 
-    - name: Build wheels
+    - name: openmp-build-x86_64 for macos
+      if: matrix.os == 'macos-latest' && steps.cache-openmp.outputs.cache-hit != 'true'
+      run: |
+        cd openmp-11.0.0.src
+        mkdir -p build-x86_64 && cd build-x86_64
+        cmake -DCMAKE_TOOLCHAIN_FILE=$GITHUB_WORKSPACE/toolchains/ios.toolchain.cmake -DPLATFORM=MAC -DARCHS="x86_64" \
+            -DDEPLOYMENT_TARGET=$MAC_DEPLOYMENT_TARGET -DENABLE_BITCODE=$ENABLE_BITCODE -DENABLE_ARC=$ENABLE_ARC -DENABLE_VISIBILITY=$ENABLE_VISIBILITY \
+            -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=install \
+            -DPERL_EXECUTABLE=/usr/local/bin/perl \
+            -DLIBOMP_ENABLE_SHARED=OFF -DLIBOMP_OMPT_SUPPORT=OFF -DLIBOMP_USE_HWLOC=OFF ..
+        cmake --build . -j 3
+        cmake --build . --target install
+
+    - name: openmp-build-arm64 for macos
+      if: matrix.os == 'macos-latest' && steps.cache-openmp.outputs.cache-hit != 'true'
+      run: |
+        cd openmp-11.0.0.src
+        mkdir -p build-arm64 && cd build-arm64
+        cmake -DCMAKE_TOOLCHAIN_FILE=$GITHUB_WORKSPACE/toolchains/ios.toolchain.cmake -DPLATFORM=MAC_ARM64 -DARCHS="arm64" \
+            -DDEPLOYMENT_TARGET=$MAC_ARM64_DEPLOYMENT_TARGET -DENABLE_BITCODE=$ENABLE_BITCODE -DENABLE_ARC=$ENABLE_ARC -DENABLE_VISIBILITY=$ENABLE_VISIBILITY \
+            -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=install \
+            -DPERL_EXECUTABLE=/usr/local/bin/perl \
+            -DLIBOMP_ENABLE_SHARED=OFF -DLIBOMP_OMPT_SUPPORT=OFF -DLIBOMP_USE_HWLOC=OFF ..
+        cmake --build . -j 3
+        cmake --build . --target install
+
+    - name: openmp-merge-fat-library for macos
+      if: matrix.os == 'macos-latest' && steps.cache-openmp.outputs.cache-hit != 'true'
+      run: |
+        mkdir -p $GITHUB_WORKSPACE/openmp-install
+        cp -a openmp-11.0.0.src/build-x86_64/install/include $GITHUB_WORKSPACE/openmp-install
+        mkdir -p $GITHUB_WORKSPACE/openmp-install/lib
+        lipo -create \
+            openmp-11.0.0.src/build-x86_64/install/lib/libomp.a \
+            openmp-11.0.0.src/build-arm64/install/lib/libomp.a \
+            -o $GITHUB_WORKSPACE/openmp-install/lib/libomp.a
+
+    - name: install-openmp for macos
+      if: matrix.os == 'macos-latest'
+      run: |
+        sudo cp $GITHUB_WORKSPACE/openmp-install/include/* $DEVELOPER_DIR/Platforms/MacOSX.platform/Developer/SDKs/MacOSX.sdk/usr/include
+        sudo cp $GITHUB_WORKSPACE/openmp-install/lib/libomp.a $DEVELOPER_DIR/Platforms/MacOSX.platform/Developer/SDKs/MacOSX.sdk/usr/lib
+
+    - name: vulkansdk for macos
+      if: matrix.os == 'macos-latest'
+      run: |
+        wget https://sdk.lunarg.com/sdk/download/1.3.236.0/mac/vulkansdk-macos-1.3.236.0.dmg?Human=true -O vulkansdk-macos-1.3.236.0.dmg
+        hdiutil attach vulkansdk-macos-1.3.236.0.dmg
+        sudo /Volumes/vulkansdk-macos-1.3.236.0/InstallVulkan.app/Contents/MacOS/InstallVulkan --root $GITHUB_WORKSPACE/vulkansdk-macos-1.3.236.0 --accept-licenses --default-answer --confirm-command install
+        hdiutil detach /Volumes/vulkansdk-macos-1.3.236.0
+
+    - name: Build wheels for macos x86_64
+      if: matrix.os == 'macos-latest' && matrix.arch == 'x86_64'
       uses: pypa/[email protected]
       env:
         CIBW_ARCHS_MACOS: ${{ matrix.arch }}
-        CIBW_ARCHS_LINUX: ${{ matrix.arch }}
-        CIBW_ARCHS_WINDOWS: ${{ matrix.arch }}
         CIBW_BUILD: ${{ matrix.build }}
         CIBW_BUILD_VERBOSITY: 1
-        CIBW_ENVIRONMENT: CMAKE_BUILD_PARALLEL_LEVEL=2
+        CIBW_ENVIRONMENT: CMAKE_BUILD_PARALLEL_LEVEL=2 NCNN_VULKAN=ON VULKAN_SDK=$GITHUB_WORKSPACE/vulkansdk-macos-1.3.236.0/macOS
+          CMAKE_TOOLCHAIN_FILE=$GITHUB_WORKSPACE/toolchains/ios.toolchain.cmake PLATFORM=MAC ARCHS="x86_64"
+          DEPLOYMENT_TARGET="10.9" ENABLE_BITCODE=OFF ENABLE_ARC=OFF ENABLE_VISIBILITY=OFF
+          OpenMP_C_FLAGS="-Xclang -fopenmp" OpenMP_CXX_FLAGS="-Xclang -fopenmp"
+          OpenMP_C_LIB_NAMES="libomp" OpenMP_CXX_LIB_NAMES="libomp"
+          OpenMP_libomp_LIBRARY="$DEVELOPER_DIR/Platforms/MacOSX.platform/Developer/SDKs/MacOSX.sdk/usr/lib/libomp.a"
+          Vulkan_INCLUDE_DIR=$GITHUB_WORKSPACE/vulkansdk-macos-1.3.236.0/MoltenVK/include
+          Vulkan_LIBRARY=$GITHUB_WORKSPACE/vulkansdk-macos-1.3.236.0/MoltenVK/dylib/macOS/libMoltenVK.dylib
+      with:
+        output-dir: wheelhouse
+
+    - name: Build wheels for macos arm64 and macos universal2
+      if: (matrix.os == 'macos-latest' && matrix.arch == 'arm64') ||
+        (matrix.os == 'macos-latest' && matrix.arch == 'universal2')
+      uses: pypa/[email protected]
+      env:
+        CIBW_ARCHS_MACOS: ${{ matrix.arch }}
+        CIBW_BUILD: ${{ matrix.build }}
+        CIBW_BUILD_VERBOSITY: 1
+        CIBW_ENVIRONMENT: CMAKE_BUILD_PARALLEL_LEVEL=2 NCNN_VULKAN=ON VULKAN_SDK=$GITHUB_WORKSPACE/vulkansdk-macos-1.3.236.0/macOS
+          CMAKE_TOOLCHAIN_FILE=$GITHUB_WORKSPACE/toolchains/ios.toolchain.cmake PLATFORM=MAC_ARM64 ARCHS="arm64"
+          DEPLOYMENT_TARGET="11.0" ENABLE_BITCODE=OFF ENABLE_ARC=OFF ENABLE_VISIBILITY=OFF
+          OpenMP_C_FLAGS="-Xclang -fopenmp" OpenMP_CXX_FLAGS="-Xclang -fopenmp"
+          OpenMP_C_LIB_NAMES="libomp" OpenMP_CXX_LIB_NAMES="libomp"
+          OpenMP_libomp_LIBRARY="$DEVELOPER_DIR/Platforms/MacOSX.platform/Developer/SDKs/MacOSX.sdk/usr/lib/libomp.a"
+          Vulkan_INCLUDE_DIR=$GITHUB_WORKSPACE/vulkansdk-macos-1.3.236.0/MoltenVK/include
+          Vulkan_LIBRARY=$GITHUB_WORKSPACE/vulkansdk-macos-1.3.236.0/MoltenVK/dylib/macOS/libMoltenVK.dylib
+      with:
+        output-dir: wheelhouse
 
     - name: Show files
       run: ls -lh wheelhouse
@@ -98,7 +310,11 @@ jobs:
       fail-fast: false
       matrix:
         arch: [aarch64, ppc64le, s390x]
-        build: ['cp36-*', 'cp37-*', 'cp38-*', 'cp39-*', 'cp310-*', 'cp311-*', 'cp312-*']
+        build: [ 'cp36-manylinux*',  'cp37-manylinux*',  'cp38-manylinux*',
+                 'cp39-manylinux*',  'cp310-manylinux*', 'cp311-manylinux*',
+                 'cp312-manylinux*', 'cp36-musllinux*',  'cp37-musllinux*',
+                 'cp38-musllinux*',  'cp39-musllinux*',  'cp310-musllinux*',
+                 'cp311-musllinux*', 'cp312-musllinux*' ]
         include:
           - arch: aarch64
             build: 'pp37-*'
@@ -123,13 +339,50 @@ jobs:
       with:
         platforms: all
 
-    - name: Build wheels
+    - name: Build wheels for manylinux with qemu
+      if: (matrix.build != 'cp36-musllinux*') && (matrix.build != 'cp37-musllinux*') &&
+          (matrix.build != 'cp38-musllinux*') && (matrix.build != 'cp39-musllinux*') &&
+          (matrix.build != 'cp310-musllinux*') && (matrix.build != 'cp311-musllinux*') &&
+          (matrix.build != 'cp312-musllinux*')
       uses: pypa/[email protected]
       env:
         CIBW_ARCHS_LINUX: ${{ matrix.arch }}
         CIBW_BUILD: ${{ matrix.build }}
         CIBW_BUILD_VERBOSITY: 1
-        CIBW_ENVIRONMENT: CMAKE_BUILD_PARALLEL_LEVEL=2
+        CIBW_ENVIRONMENT: CMAKE_BUILD_PARALLEL_LEVEL=2 NCNN_VULKAN=ON VULKAN_SDK=/project/Vulkan-Loader/build/Vulkan-Headers
+          LD_LIBRARY_PATH=/project/Vulkan-Loader/build/loader
+        CIBW_BEFORE_ALL: yum -y install libXrandr-devel &&
+          git clone https://github.com/KhronosGroup/Vulkan-Loader.git &&
+          cd Vulkan-Loader && mkdir build && cd build &&
+          ../scripts/update_deps.py &&
+          cmake -DCMAKE_BUILD_TYPE=Release -DVULKAN_HEADERS_INSTALL_DIR=$(pwd)/Vulkan-Headers/build/install .. &&
+          make -j$(nproc) &&
+          cd Vulkan-Headers &&
+          ln -s ../loader lib
+      with:
+        output-dir: wheelhouse
+
+    - name: Build wheels for musllinux with qemu
+      if: (matrix.build == 'cp36-musllinux*') || (matrix.build == 'cp37-musllinux*') ||
+          (matrix.build == 'cp38-musllinux*') || (matrix.build == 'cp39-musllinux*') ||
+          (matrix.build == 'cp310-musllinux*') || (matrix.build == 'cp311-musllinux*') ||
+          (matrix.build == 'cp312-musllinux*')
+      uses: pypa/[email protected]
+      env:
+        CIBW_ARCHS_LINUX: ${{ matrix.arch }}
+        CIBW_BUILD: ${{ matrix.build }}
+        CIBW_BUILD_VERBOSITY: 1
+        CIBW_ENVIRONMENT: CMAKE_BUILD_PARALLEL_LEVEL=2 NCNN_VULKAN=ON VULKAN_SDK=/project/Vulkan-Loader/build/Vulkan-Headers LD_LIBRARY_PATH=/project/Vulkan-Loader/build/loader
+        CIBW_BEFORE_ALL: apk add libxrandr-dev &&
+          git clone https://github.com/KhronosGroup/Vulkan-Loader.git &&
+          cd Vulkan-Loader && mkdir build && cd build &&
+          ../scripts/update_deps.py &&
+          cmake -DCMAKE_BUILD_TYPE=Release -DVULKAN_HEADERS_INSTALL_DIR=$(pwd)/Vulkan-Headers/build/install .. &&
+          make -j$(nproc) &&
+          cd Vulkan-Headers &&
+          ln -s ../loader lib
+      with:
+        output-dir: wheelhouse
 
     - name: Show files
       run: ls -lh wheelhouse

python/README.md

@@ -33,7 +33,41 @@ If you want to build ncnn with some options not as default, or just like to buil
 * Visual Studio 2015 or higher
 * CMake >= 3.4
 
-## Build
+##  Build & Install
+
+1. clone ncnn and init submodule.
+
+```bash
+cd /pathto/ncnn
+git submodule init && git submodule update
+```
+
+2. build and install.
+
+```
+python setup.py install
+```
+
+if you want to enable the usage of vulkan, you can install as following:
+
+```
+python setup.py install --vulkan=on
+```
+
+> **Attention:**
+>
+> To enable Vulkan support, you must first install the Vulkan SDK.
+>
+> **For Windows or Linux Users:**
+>
+> Ensure that the `VULKAN_SDK` environment variable is set to the path of the Vulkan SDK.
+>
+> **For MacOS Users:**
+>
+> On MacOS, you will need to specify additional environment variables. For guidance on setting these variables, please refer to lines 279-286 in the following file: [ncnn/.github/workflows/release-python.yml at master · Tencent/ncnn](https://github.com/Tencent/ncnn/blob/master/.github/workflows/release-python.yml).
+
+## Custom-build & Install
+
 1. clone ncnn and init submodule.
 ```bash
 cd /pathto/ncnn
@@ -47,7 +81,8 @@ cmake -DNCNN_PYTHON=ON ..
 make
 ```
 
-## Install
+3. install
+
 ```bash
 cd /pathto/ncnn
 pip install .
@@ -60,6 +95,7 @@ python3 setup.py install
 ```
 
 ## Tests
+
 **test**
 ```bash
 cd /pathto/ncnn/python

python/src/main.cpp

@@ -185,6 +185,9 @@ PYBIND11_MODULE(ncnn, m)
 #endif // NCNN_VULKAN
     .def_readwrite("openmp_blocktime", &Option::openmp_blocktime)
     .def_readwrite("use_winograd_convolution", &Option::use_winograd_convolution)
+    .def_readwrite("use_winograd23_convolution", &Option::use_winograd23_convolution)
+    .def_readwrite("use_winograd43_convolution", &Option::use_winograd43_convolution)
+    .def_readwrite("use_winograd63_convolution", &Option::use_winograd63_convolution)
     .def_readwrite("use_sgemm_convolution", &Option::use_sgemm_convolution)
     .def_readwrite("use_int8_inference", &Option::use_int8_inference)
     .def_readwrite("use_vulkan_compute", &Option::use_vulkan_compute)