Skip to content
tvos

arm neon optimization for layernorm fp32/bf16s/fp16s #349

Sign in to view logs

arm neon optimization for layernorm fp32/bf16s/fp16s

arm neon optimization for layernorm fp32/bf16s/fp16s #349

Workflow file for this run

.github/workflows/tvos.yml at 3f285fb
name: tvos
on:
push:
branches: [master]
paths:
- '.github/workflows/tvos.yml'
- 'toolchains/ios.toolchain.cmake'
- 'CMakeLists.txt'
- 'cmake/**'
- 'src/*'
- 'src/layer/*'
- 'src/layer/arm/**'
- 'src/layer/x86/**'
- 'src/layer/vulkan/**'
pull_request:
branches: [master]
paths:
- '.github/workflows/tvos.yml'
- 'toolchains/ios.toolchain.cmake'
- 'CMakeLists.txt'
- 'cmake/**'
- 'src/*'
- 'src/layer/*'
- 'src/layer/arm/**'
- 'src/layer/x86/**'
- 'src/layer/vulkan/**'
concurrency:
group: tvos-${{ github.ref }}
cancel-in-progress: true
env:
DEVELOPER_DIR: /Applications/Xcode_15.2.app/Contents/Developer
TVOS_DEPLOYMENT_TARGET: '11.0'
ENABLE_BITCODE: OFF
ENABLE_ARC: OFF
ENABLE_VISIBILITY: OFF
permissions:
contents: read
jobs:
build:
runs-on: macos-13
env:
OPENMP_VERSION: '18.1.2'
OPENMP_CMAKE_OPTIONS: |
-DCMAKE_TOOLCHAIN_FILE=../../toolchains/ios.toolchain.cmake \
-DDEPLOYMENT_TARGET=$TVOS_DEPLOYMENT_TARGET \
-DENABLE_BITCODE=$ENABLE_BITCODE \
-DENABLE_ARC=$ENABLE_ARC \
-DENABLE_VISIBILITY=$ENABLE_VISIBILITY \
-DCMAKE_INSTALL_PREFIX=install \
-DCMAKE_BUILD_TYPE=Release \
-DPERL_EXECUTABLE=/usr/local/bin/perl \
-DLIBOMP_ENABLE_SHARED=OFF \
-DLIBOMP_OMPT_SUPPORT=OFF \
-DLIBOMP_USE_HWLOC=OFF \
MOLTENVK_VERSION: 'v1.2.8'
NCNN_CMAKE_OPTIONS: |
-DCMAKE_TOOLCHAIN_FILE=../toolchains/ios.toolchain.cmake \
-DDEPLOYMENT_TARGET=$TVOS_DEPLOYMENT_TARGET \
-DENABLE_BITCODE=$ENABLE_BITCODE \
-DENABLE_ARC=$ENABLE_ARC \
-DENABLE_VISIBILITY=$ENABLE_VISIBILITY \
-DCMAKE_INSTALL_PREFIX=install \
-DCMAKE_BUILD_TYPE=Release \
-DOpenMP_C_FLAGS="-Xclang -fopenmp" -DOpenMP_CXX_FLAGS="-Xclang -fopenmp" \
-DOpenMP_C_LIB_NAMES="libomp" -DOpenMP_CXX_LIB_NAMES="libomp" \
-DOpenMP_libomp_LIBRARY="libomp.a" \
-DNCNN_VULKAN=ON \
steps:
- uses: actions/checkout@v4
with:
submodules: true
- name: cache-openmp
id: cache-openmp
uses: actions/cache@v4
with:
path: openmp-install
key: openmp-tvos-install-20240403
- name: openmp
if: steps.cache-openmp.outputs.cache-hit != 'true'
run: |
wget https://github.com/llvm/llvm-project/releases/download/llvmorg-${{ env.OPENMP_VERSION }}/cmake-${{ env.OPENMP_VERSION }}.src.tar.xz
tar -xf cmake-${{ env.OPENMP_VERSION }}.src.tar.xz
wget https://github.com/llvm/llvm-project/releases/download/llvmorg-${{ env.OPENMP_VERSION }}/openmp-${{ env.OPENMP_VERSION }}.src.tar.xz
tar -xf openmp-${{ env.OPENMP_VERSION }}.src.tar.xz
mv cmake-${{ env.OPENMP_VERSION }}.src/Modules/* openmp-${{ env.OPENMP_VERSION }}.src/cmake/
cd openmp-${{ env.OPENMP_VERSION }}.src
wget https://github.com/nihui/llvm-project/commit/ef8c35bcf5d9cfdb0764ffde6a63c04ec715bc37.patch
patch -p2 -i ef8c35bcf5d9cfdb0764ffde6a63c04ec715bc37.patch
wget https://github.com/nihui/llvm-project/commit/5c12711f9a21f41bea70566bf15a4026804d6b20.patch
patch -p2 -i 5c12711f9a21f41bea70566bf15a4026804d6b20.patch
- name: openmp-arm64
if: steps.cache-openmp.outputs.cache-hit != 'true'
run: |
cd openmp-${{ env.OPENMP_VERSION }}.src
mkdir -p build-arm64 && cd build-arm64
cmake ${{ env.OPENMP_CMAKE_OPTIONS }} -DPLATFORM=TVOS -DARCHS="arm64" ..
cmake --build . -j 4
cmake --build . --target install
- name: openmp-arm64e
if: steps.cache-openmp.outputs.cache-hit != 'true'
run: |
cd openmp-${{ env.OPENMP_VERSION }}.src
mkdir -p build-arm64e && cd build-arm64e
cmake ${{ env.OPENMP_CMAKE_OPTIONS }} -DPLATFORM=TVOS -DARCHS="arm64e" ..
cmake --build . -j 4
cmake --build . --target install
- name: openmp-simulator-x86_64
if: steps.cache-openmp.outputs.cache-hit != 'true'
run: |
cd openmp-${{ env.OPENMP_VERSION }}.src
mkdir -p build-simulator-x86_64 && cd build-simulator-x86_64
cmake ${{ env.OPENMP_CMAKE_OPTIONS }} -DPLATFORM=SIMULATOR_TVOS -DARCHS="x86_64" ..
cmake --build . -j 4
cmake --build . --target install
- name: openmp-simulator-arm64
if: steps.cache-openmp.outputs.cache-hit != 'true'
run: |
cd openmp-${{ env.OPENMP_VERSION }}.src
mkdir -p build-simulator-arm64 && cd build-simulator-arm64
cmake ${{ env.OPENMP_CMAKE_OPTIONS }} -DPLATFORM=SIMULATOR_TVOS -DARCHS="arm64" ..
cmake --build . -j 4
cmake --build . --target install
- name: openmp-merge-fat-library
if: steps.cache-openmp.outputs.cache-hit != 'true'
run: |
mkdir -p $GITHUB_WORKSPACE/openmp-install
mkdir -p $GITHUB_WORKSPACE/openmp-install/tvos
mkdir -p $GITHUB_WORKSPACE/openmp-install/tvos-simulator
cp -a openmp-${{ env.OPENMP_VERSION }}.src/build-arm64/install/include $GITHUB_WORKSPACE/openmp-install/tvos
mkdir -p $GITHUB_WORKSPACE/openmp-install/tvos/lib
lipo -create \
openmp-${{ env.OPENMP_VERSION }}.src/build-arm64/install/lib/libomp.a \
openmp-${{ env.OPENMP_VERSION }}.src/build-arm64e/install/lib/libomp.a \
-o $GITHUB_WORKSPACE/openmp-install/tvos/lib/libomp.a
cp -a openmp-${{ env.OPENMP_VERSION }}.src/build-simulator-x86_64/install/include $GITHUB_WORKSPACE/openmp-install/tvos-simulator
mkdir -p $GITHUB_WORKSPACE/openmp-install/tvos-simulator/lib
lipo -create \
openmp-${{ env.OPENMP_VERSION }}.src/build-simulator-x86_64/install/lib/libomp.a \
openmp-${{ env.OPENMP_VERSION }}.src/build-simulator-arm64/install/lib/libomp.a \
-o $GITHUB_WORKSPACE/openmp-install/tvos-simulator/lib/libomp.a
- name: install-openmp
run: |
sudo cp $GITHUB_WORKSPACE/openmp-install/tvos/include/* $DEVELOPER_DIR/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/usr/include
sudo cp $GITHUB_WORKSPACE/openmp-install/tvos/lib/libomp.a $DEVELOPER_DIR/Platforms/AppleTVOS.platform/Developer/SDKs/AppleTVOS.sdk/usr/lib
sudo cp $GITHUB_WORKSPACE/openmp-install/tvos-simulator/include/* $DEVELOPER_DIR/Platforms/AppleTVSimulator.platform/Developer/SDKs/AppleTVSimulator.sdk/usr/include
sudo cp $GITHUB_WORKSPACE/openmp-install/tvos-simulator/lib/libomp.a $DEVELOPER_DIR/Platforms/AppleTVSimulator.platform/Developer/SDKs/AppleTVSimulator.sdk/usr/lib
- name: moltenvk
run: |
wget -q https://github.com/KhronosGroup/MoltenVK/releases/download/${{ env.MOLTENVK_VERSION }}/MoltenVK-all.tar
tar -xf MoltenVK-all.tar
- name: arm64
run: |
mkdir build-arm64 && cd build-arm64
cmake ${{ env.NCNN_CMAKE_OPTIONS }} -DPLATFORM=TVOS -DARCHS="arm64" \
-DVulkan_LIBRARY=$GITHUB_WORKSPACE/MoltenVK/MoltenVK/static/MoltenVK.xcframework/tvos-arm64_arm64e/libMoltenVK.a ..
cmake --build . -j 4
- name: arm64e
run: |
mkdir build-arm64e && cd build-arm64e
cmake ${{ env.NCNN_CMAKE_OPTIONS }} -DPLATFORM=TVOS -DARCHS="arm64e" \
-DVulkan_LIBRARY=$GITHUB_WORKSPACE/MoltenVK/MoltenVK/static/MoltenVK.xcframework/tvos-arm64_arm64e/libMoltenVK.a ..
cmake --build . -j 4
- name: simulator-x86_64
run: |
mkdir build-simulator-x86_64 && cd build-simulator-x86_64
cmake ${{ env.NCNN_CMAKE_OPTIONS }} -DPLATFORM=SIMULATOR_TVOS -DARCHS="x86_64" \
-DVulkan_LIBRARY=$GITHUB_WORKSPACE/MoltenVK/MoltenVK/static/MoltenVK.xcframework/tvos-arm64_x86_64-simulator/libMoltenVK.a ..
cmake --build . -j 4
- name: simulator-arm64
run: |
mkdir build-simulator-arm64 && cd build-simulator-arm64
cmake ${{ env.NCNN_CMAKE_OPTIONS }} -DPLATFORM=SIMULATORARM64_TVOS -DARCHS="arm64" \
-DVulkan_LIBRARY=$GITHUB_WORKSPACE/MoltenVK/MoltenVK/static/MoltenVK.xcframework/tvos-arm64_x86_64-simulator/libMoltenVK.a ..
cmake --build . -j 4