arm neon optimization for layernorm fp32/bf16s/fp16s #363
Workflow file for this run
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
name: watchos | |
on: | |
push: | |
branches: [master] | |
paths: | |
- '.github/workflows/watchos.yml' | |
- 'toolchains/ios.toolchain.cmake' | |
- 'CMakeLists.txt' | |
- 'cmake/**' | |
- 'src/*' | |
- 'src/layer/*' | |
- 'src/layer/arm/**' | |
- 'src/layer/x86/**' | |
pull_request: | |
branches: [master] | |
paths: | |
- '.github/workflows/watchos.yml' | |
- 'toolchains/ios.toolchain.cmake' | |
- 'CMakeLists.txt' | |
- 'cmake/**' | |
- 'src/*' | |
- 'src/layer/*' | |
- 'src/layer/arm/**' | |
- 'src/layer/x86/**' | |
concurrency: | |
group: watchos-${{ github.ref }} | |
cancel-in-progress: true | |
env: | |
DEVELOPER_DIR: /Applications/Xcode_15.2.app/Contents/Developer | |
WATCHOS_DEPLOYMENT_TARGET: '6.0' | |
ENABLE_BITCODE: OFF | |
ENABLE_ARC: OFF | |
ENABLE_VISIBILITY: OFF | |
permissions: | |
contents: read | |
jobs: | |
build: | |
runs-on: macos-13 | |
env: | |
OPENMP_VERSION: '18.1.2' | |
OPENMP_CMAKE_OPTIONS: | | |
-DCMAKE_TOOLCHAIN_FILE=../../toolchains/ios.toolchain.cmake \ | |
-DDEPLOYMENT_TARGET=$WATCHOS_DEPLOYMENT_TARGET \ | |
-DENABLE_BITCODE=$ENABLE_BITCODE \ | |
-DENABLE_ARC=$ENABLE_ARC \ | |
-DENABLE_VISIBILITY=$ENABLE_VISIBILITY \ | |
-DCMAKE_INSTALL_PREFIX=install \ | |
-DCMAKE_BUILD_TYPE=Release \ | |
-DPERL_EXECUTABLE=/usr/local/bin/perl \ | |
-DLIBOMP_ENABLE_SHARED=OFF \ | |
-DLIBOMP_OMPT_SUPPORT=OFF \ | |
-DLIBOMP_USE_HWLOC=OFF \ | |
NCNN_CMAKE_OPTIONS: | | |
-DCMAKE_TOOLCHAIN_FILE=../toolchains/ios.toolchain.cmake \ | |
-DDEPLOYMENT_TARGET=$WATCHOS_DEPLOYMENT_TARGET \ | |
-DENABLE_BITCODE=$ENABLE_BITCODE \ | |
-DENABLE_ARC=$ENABLE_ARC \ | |
-DENABLE_VISIBILITY=$ENABLE_VISIBILITY \ | |
-DCMAKE_INSTALL_PREFIX=install \ | |
-DCMAKE_BUILD_TYPE=Release \ | |
-DOpenMP_C_FLAGS="-Xclang -fopenmp" -DOpenMP_CXX_FLAGS="-Xclang -fopenmp" \ | |
-DOpenMP_C_LIB_NAMES="libomp" -DOpenMP_CXX_LIB_NAMES="libomp" \ | |
-DOpenMP_libomp_LIBRARY="libomp.a" \ | |
steps: | |
- uses: actions/checkout@v4 | |
- name: cache-openmp | |
id: cache-openmp | |
uses: actions/cache@v4 | |
with: | |
path: openmp-install | |
key: openmp-watchos-install-20240402 | |
- name: openmp | |
if: steps.cache-openmp.outputs.cache-hit != 'true' | |
run: | | |
wget https://github.com/llvm/llvm-project/releases/download/llvmorg-${{ 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-armv7k | |
if: steps.cache-openmp.outputs.cache-hit != 'true' | |
run: | | |
cd openmp-${{ env.OPENMP_VERSION }}.src | |
mkdir -p build-armv7k && cd build-armv7k | |
cmake ${{ env.OPENMP_CMAKE_OPTIONS }} -DPLATFORM=WATCHOS -DARCHS="armv7k" .. | |
cmake --build . -j 4 | |
cmake --build . --target install | |
- name: openmp-arm64_32 | |
if: steps.cache-openmp.outputs.cache-hit != 'true' | |
run: | | |
cd openmp-${{ env.OPENMP_VERSION }}.src | |
mkdir -p build-arm64_32 && cd build-arm64_32 | |
cmake ${{ env.OPENMP_CMAKE_OPTIONS }} -DPLATFORM=WATCHOS -DARCHS="arm64_32" .. | |
cmake --build . -j 4 | |
cmake --build . --target install | |
- name: openmp-simulator-x86_64 | |
if: steps.cache-openmp.outputs.cache-hit != 'true' | |
run: | | |
cd openmp-${{ env.OPENMP_VERSION }}.src | |
mkdir -p build-simulator-x86_64 && cd build-simulator-x86_64 | |
cmake ${{ env.OPENMP_CMAKE_OPTIONS }} -DPLATFORM=SIMULATOR_WATCHOS -DARCHS="x86_64" .. | |
cmake --build . -j 4 | |
cmake --build . --target install | |
- name: openmp-simulator-arm64 | |
if: steps.cache-openmp.outputs.cache-hit != 'true' | |
run: | | |
cd openmp-${{ env.OPENMP_VERSION }}.src | |
mkdir -p build-simulator-arm64 && cd build-simulator-arm64 | |
cmake ${{ env.OPENMP_CMAKE_OPTIONS }} -DPLATFORM=SIMULATOR_WATCHOS -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/watchos | |
mkdir -p $GITHUB_WORKSPACE/openmp-install/watchos-simulator | |
cp -a openmp-${{ env.OPENMP_VERSION }}.src/build-arm64_32/install/include $GITHUB_WORKSPACE/openmp-install/watchos | |
mkdir -p $GITHUB_WORKSPACE/openmp-install/watchos/lib | |
lipo -create \ | |
openmp-${{ env.OPENMP_VERSION }}.src/build-armv7k/install/lib/libomp.a \ | |
openmp-${{ env.OPENMP_VERSION }}.src/build-arm64_32/install/lib/libomp.a \ | |
-o $GITHUB_WORKSPACE/openmp-install/watchos/lib/libomp.a | |
cp -a openmp-${{ env.OPENMP_VERSION }}.src/build-simulator-x86_64/install/include $GITHUB_WORKSPACE/openmp-install/watchos-simulator | |
mkdir -p $GITHUB_WORKSPACE/openmp-install/watchos-simulator/lib | |
lipo -create \ | |
openmp-${{ 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/watchos-simulator/lib/libomp.a | |
- name: install-openmp | |
run: | | |
sudo cp $GITHUB_WORKSPACE/openmp-install/watchos/include/* $DEVELOPER_DIR/Platforms/WatchOS.platform/Developer/SDKs/WatchOS.sdk/usr/include | |
sudo cp $GITHUB_WORKSPACE/openmp-install/watchos/lib/libomp.a $DEVELOPER_DIR/Platforms/WatchOS.platform/Developer/SDKs/WatchOS.sdk/usr/lib | |
sudo cp $GITHUB_WORKSPACE/openmp-install/watchos-simulator/include/* $DEVELOPER_DIR/Platforms/WatchSimulator.platform/Developer/SDKs/WatchSimulator.sdk/usr/include | |
sudo cp $GITHUB_WORKSPACE/openmp-install/watchos-simulator/lib/libomp.a $DEVELOPER_DIR/Platforms/WatchSimulator.platform/Developer/SDKs/WatchSimulator.sdk/usr/lib | |
- name: armv7k | |
run: | | |
mkdir build-armv7k && cd build-armv7k | |
cmake ${{ env.NCNN_CMAKE_OPTIONS }} -DPLATFORM=WATCHOS -DARCHS="armv7k" .. | |
cmake --build . -j 4 | |
- name: arm64_32 | |
run: | | |
mkdir build-arm64_32 && cd build-arm64_32 | |
cmake ${{ env.NCNN_CMAKE_OPTIONS }} -DPLATFORM=WATCHOS -DARCHS="arm64_32" .. | |
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_WATCHOS -DARCHS="x86_64" .. | |
cmake --build . -j 4 | |
- name: simulator-arm64 | |
run: | | |
mkdir build-simulator-arm64 && cd build-simulator-arm64 | |
cmake ${{ env.NCNN_CMAKE_OPTIONS }} -DPLATFORM=SIMULATOR_WATCHOS -DARCHS="arm64" .. | |
cmake --build . -j 4 |