Skip to content

quantize_block C->C++, use std::thread everywhere (#1024) #85

quantize_block C->C++, use std::thread everywhere (#1024)

quantize_block C->C++, use std::thread everywhere (#1024) #85

Workflow file for this run

name: CMake on multiple platforms
on:
push:
branches: [ "main" ]
pull_request:
branches: [ "main" ]
concurrency:
group: cmake-${{ github.ref }}
cancel-in-progress: true
jobs:
build-shared-libs:
runs-on: ${{ matrix.os }}
strategy:
# Set fail-fast to false to ensure that feedback is delivered for all matrix combinations. Consider changing this to true when your workflow is stable.
fail-fast: false
matrix:
os: [ubuntu-latest, windows-latest]
arch: [x86_64, aarch64]
build_type: [Release]
exclude:
- os: windows-latest
arch: aarch64
steps:
- uses: actions/checkout@v4
- name: Set up MSVC
if: matrix.os == 'windows-latest'
uses: ilammy/[email protected]
with:
arch: amd64
- name: Set reusable strings
# Turn repeated input strings (such as the build output directory) into step outputs. These step outputs can be used throughout the workflow file.
id: strings
shell: bash
run: |
echo "build-output-dir=${{ github.workspace }}/build" >> "$GITHUB_OUTPUT"
- name: Prep build
run: python3 -m pip install cmake==3.27.9 ninja setuptools wheel
- name: Prep Compilers
shell: bash -el {0}
run: |
if [ "${{ matrix.os }}" = "windows-latest" ]; then
echo CXX_COMPILER=cl >> "$GITHUB_ENV"
echo C_COMPILER=cl >> "$GITHUB_ENV"
else
echo CXX_COMPILER=g++ >> "$GITHUB_ENV"
echo C_COMPILER=gcc >> "$GITHUB_ENV"
fi
- name: Configure CPU
run: >
cmake -B ${{ steps.strings.outputs.build-output-dir }}
-G Ninja
-DCMAKE_CXX_COMPILER=${{ env.CXX_COMPILER }}
-DCMAKE_C_COMPILER=${{ env.C_COMPILER }}
-DCMAKE_BUILD_TYPE=${{ matrix.build_type }}
-DBUILD_CUDA=OFF
-S ${{ github.workspace }}
- name: Build CPU
run: cmake --build ${{ steps.strings.outputs.build-output-dir }} --config ${{ matrix.build_type }}
- name: Copy libraries
shell: bash
run: |
mkdir -p output/${{ matrix.os }}/${{ matrix.arch }}
( shopt -s nullglob && cp -a bitsandbytes/*.{so,dylib,dll} output/${{ matrix.os }}/${{ matrix.arch }} )
- name: Upload Build Artifacts
uses: actions/upload-artifact@v4
with:
name: shared_library-${{ matrix.os }}-${{ matrix.arch }}
path: output/*
build-shared-libs-cuda:
runs-on: ${{ matrix.os }}
strategy:
# Set fail-fast to false to ensure that feedback is delivered for all matrix combinations. Consider changing this to true when your workflow is stable.
fail-fast: false
matrix:
os: [ubuntu-latest, windows-latest]
cuda-version: ['11.8', '12.1']
arch: [x86_64, aarch64]
build_type: [Release]
exclude:
- os: windows-latest
arch: aarch64
steps:
- uses: actions/checkout@v4
- name: Set up Python 3.10
uses: actions/setup-python@v5
with:
python-version: "3.10"
- name: Set up MSVC
if: matrix.os == 'windows-latest'
uses: ilammy/[email protected]
with:
arch: amd64
- name: Setup Mambaforge
uses: conda-incubator/[email protected]
with:
miniforge-variant: Mambaforge
miniforge-version: latest
activate-environment: bnb-env
use-mamba: true
- uses: conda-incubator/[email protected]
with:
auto-update-conda: true
activate-environment: bnb-env
environment-file: environment-bnb.yml
use-only-tar-bz2: false
auto-activate-base: true
python-version: "3.10"
mamba-version: "*"
- name: Set reusable strings
# Turn repeated input strings (such as the build output directory) into step outputs. These step outputs can be used throughout the workflow file.
id: strings
shell: bash
run: |
echo "build-output-dir=${{ github.workspace }}/build" >> "$GITHUB_OUTPUT"
- name: CUDA Toolkit
shell: bash -el {0}
run: |
if [ "${{ matrix.os }}" = "ubuntu-latest" ]; then
# to prepare space
sudo rm -rf /usr/share/dotnet
sudo rm -rf /opt/ghc
sudo rm -rf /usr/local/share/boost
fi
addon=""
cuda_version=${{ matrix.cuda-version }}
[ "$cuda_version" = "12.1" ] && [ "${{ matrix.os }}" = "ubuntu-latest" ] && addon="cuda-cudart-static cuda-nvrtc"
[ "$cuda_version" = "12.1" ] && [ "${{ matrix.os }}" = "windows-latest" ] && addon="cuda-nvrtc"
[ "$cuda_version" = "11.8" ] && cuda_version="11.8.0"
[ "$cuda_version" = "12.1" ] && cuda_version="12.1.1"
conda install pytorch-cuda=${{ matrix.cuda-version }} -c pytorch # it's dependency not correctly resolved sometime
conda install cuda-python=${{ matrix.cuda-version }} cuda-libraries-dev cuda-nvcc cuda-nvtx cuda-cupti cuda-cudart cuda-cudart-dev cuda-runtime cuda-libraries $addon -c "nvidia/label/cuda-$cuda_version"
[ "${{ matrix.os }}" = "windows-latest" ] && conda install "clang>=17.0.6" "clangxx>=17.0.6" -c conda-forge
CUDA_HOME="${{ env.CONDA }}/envs/bnb-env"
echo CUDA_HOME=$CUDA_HOME >> "$GITHUB_ENV"
echo CUDA_PATH=$CUDA_HOME >> "$GITHUB_ENV"
if [ "${{ matrix.os }}" = "windows-latest" ]; then
echo CXX_COMPILER=cl >> "$GITHUB_ENV"
echo C_COMPILER=cl >> "$GITHUB_ENV"
# without -DCMAKE_CUDA_COMPILER=nvcc, cmake config always fail for cuda-11.8
echo DCMAKE_CUDA_COMPILER=-DCMAKE_CUDA_COMPILER=nvcc >> "$GITHUB_ENV"
else
echo CXX_COMPILER=g++ >> "$GITHUB_ENV"
echo C_COMPILER=gcc >> "$GITHUB_ENV"
fi
nvcc --version
- name: Update environment
run: mamba env update -n bnb-env -f environment-bnb.yml
- name: Prep build
run: python -m pip install cmake==3.27.9 ninja setuptools wheel
# TODO: the following steps (CUDA, NOBLASLT, CPU) could be moved to the matrix, so they're built in parallel
- name: Configure CUDA
run: >
cmake -B ${{ steps.strings.outputs.build-output-dir }}
-G Ninja ${{ env.DCMAKE_CUDA_COMPILER }}
-DCMAKE_CXX_COMPILER=${{ env.CXX_COMPILER }}
-DCMAKE_C_COMPILER=${{ env.C_COMPILER }}
-DCMAKE_BUILD_TYPE=${{ matrix.build_type }}
-DCOMPUTE_CAPABILITY="50;52;60;61;62;70;72;75;80;86;87;89;90"
-S ${{ github.workspace }}
- name: Build CUDA
run: cmake --build ${{ steps.strings.outputs.build-output-dir }} --config ${{ matrix.build_type }}
- name: Configure NOBLASLT
run: >
cmake -B ${{ steps.strings.outputs.build-output-dir }}
-G Ninja ${{ env.DCMAKE_CUDA_COMPILER }}
-DCMAKE_CXX_COMPILER=${{ env.CXX_COMPILER }}
-DCMAKE_C_COMPILER=${{ env.C_COMPILER }}
-DCMAKE_BUILD_TYPE=${{ matrix.build_type }}
-DCOMPUTE_CAPABILITY="50;52;60;61;62;70;72;75;80;86;87;89;90"
-DNO_CUBLASLT=ON
-S ${{ github.workspace }}
- name: Build NOBLASLT
run: cmake --build ${{ steps.strings.outputs.build-output-dir }} --config ${{ matrix.build_type }}
- name: Copy libraries
shell: bash
run: |
mkdir -p output/${{ matrix.os }}/${{ matrix.arch }}
( shopt -s nullglob && cp -a bitsandbytes/*.{so,dylib,dll} output/${{ matrix.os }}/${{ matrix.arch }} )
- name: Upload Build Artifacts
uses: actions/upload-artifact@v4
with:
name: shared_library_cuda-${{ matrix.os }}-${{ matrix.cuda-version }}-${{ matrix.arch }}
path: output/*
build-wheels:
needs:
- build-shared-libs
- build-shared-libs-cuda
runs-on: ${{ matrix.os }}
strategy:
matrix:
os: [ubuntu-latest, windows-latest]
arch: [x86_64, aarch64]
exclude:
- os: windows-latest
arch: aarch64
steps:
# Check out code
- uses: actions/checkout@v4
# Download shared libraries
- name: Download build artifact
uses: actions/download-artifact@v4
with:
merge-multiple: true
path: output/
- name: Copy correct platform shared libraries
shell: bash
run: |
cp output/${{ matrix.os }}/${{ matrix.arch }}/* bitsandbytes/
# Set up the Python version needed
- name: Set up Python 3.10
uses: actions/setup-python@v5
with:
python-version: "3.10"
cache: pip
- name: Install build package
shell: bash
run: pip install build
- name: Build wheel
shell: bash
run: python -m build . --wheel
- name: Upload Build Artifacts
uses: actions/upload-artifact@v4
with:
name: bdist_wheel-${{ matrix.os }}-${{ matrix.arch }}
path: |
${{ github.workspace }}/dist/