Skip to content

chore(gpu): rework ci to adapt to the shortage of h100 #719

chore(gpu): rework ci to adapt to the shortage of h100

chore(gpu): rework ci to adapt to the shortage of h100 #719

# Compile and test tfhe-cuda-backend on an H100 VM on hyperstack
name: TFHE Cuda Backend - Fast tests on H100
env:
CARGO_TERM_COLOR: always
ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
RUSTFLAGS: "-C target-cpu=native"
RUST_BACKTRACE: "full"
RUST_MIN_STACK: "8388608"
SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
IS_PULL_REQUEST: ${{ github.event_name == 'pull_request' }}
on:
# Allows you to run this workflow manually from the Actions tab as an alternative.
workflow_dispatch:
pull_request:
types: [ labeled ]
jobs:
should-run:
runs-on: ubuntu-latest
permissions:
pull-requests: write
outputs:
gpu_test: ${{ env.IS_PULL_REQUEST == 'false' || steps.changed-files.outputs.gpu_any_changed }}
steps:
- name: Checkout tfhe-rs
uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
with:
fetch-depth: 0
- name: Check for file changes
id: changed-files
uses: tj-actions/changed-files@c3a1bb2c992d77180ae65be6ae6c166cf40f857c
with:
since_last_remote_commit: true
files_yaml: |
gpu:
- tfhe/Cargo.toml
- tfhe/build.rs
- backends/tfhe-cuda-backend/**
- tfhe/src/core_crypto/gpu/**
- tfhe/src/integer/gpu/**
- tfhe/src/shortint/parameters/**
- tfhe/src/high_level_api/**
- tfhe/src/c_api/**
- 'tfhe/docs/**.md'
- '.github/workflows/gpu_fast_h100_tests.yml'
- scripts/integer-tests.sh
- ci/slab.toml
setup-instance:
name: Setup instance (cuda-h100-tests)
needs: should-run
if: github.event_name != 'pull_request' ||
(github.event.action != 'labeled' && needs.should-run.outputs.gpu_test == 'true') ||
(github.event.action == 'labeled' && github.event.label.name == 'approved' && needs.should-run.outputs.gpu_test == 'true')
runs-on: ubuntu-latest
outputs:
runner-name: ${{ steps.start-instance.outputs.label }}
steps:
- name: Start instance
id: start-instance
uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
with:
mode: start
github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
slab-url: ${{ secrets.SLAB_BASE_URL }}
job-secret: ${{ secrets.JOB_SECRET }}
backend: hyperstack
profile: single-h100
cuda-tests-linux:
name: CUDA H100 tests
needs: [ should-run, setup-instance ]
if: github.event_name != 'pull_request' ||
(github.event_name == 'pull_request' && needs.setup-instance.result != 'skipped')
concurrency:
group: ${{ github.workflow }}_${{ github.ref }}
cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
runs-on: ${{ needs.setup-instance.outputs.runner-name }}
strategy:
fail-fast: false
# explicit include-based build matrix, of known valid options
matrix:
include:
- os: ubuntu-22.04
cuda: "12.2"
gcc: 11
env:
CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
CMAKE_VERSION: 3.29.6
steps:
# Mandatory on hyperstack since a bootable volume is not re-usable yet.
- name: Install dependencies
run: |
sudo apt update
sudo apt install -y checkinstall zlib1g-dev libssl-dev libclang-dev
wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
cd cmake-${{ env.CMAKE_VERSION }}
./bootstrap
make -j"$(nproc)"
sudo make install
- name: Checkout tfhe-rs
uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
with:
persist-credentials: 'false'
token: ${{ secrets.FHE_ACTIONS_TOKEN }}
- name: Set up home
run: |
echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
- name: Install latest stable
uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
with:
toolchain: stable
- name: Export CUDA variables
if: ${{ !cancelled() }}
run: |
echo "CUDA_PATH=$CUDA_PATH" >> "${GITHUB_ENV}"
echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH" >> "${GITHUB_ENV}"
echo "CUDACXX=/usr/local/cuda-${{ matrix.cuda }}/bin/nvcc" >> "${GITHUB_ENV}"
# Specify the correct host compilers
- name: Export gcc and g++ variables
if: ${{ !cancelled() }}
run: |
{
echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
echo "HOME=/home/ubuntu";
} >> "${GITHUB_ENV}"
- name: Check device is detected
if: ${{ !cancelled() }}
run: nvidia-smi
- name: Run core crypto and internal CUDA backend tests
run: |
BIG_TESTS_INSTANCE=TRUE make test_core_crypto_gpu
BIG_TESTS_INSTANCE=TRUE make test_integer_compression_gpu
BIG_TESTS_INSTANCE=TRUE make test_cuda_backend
- name: Run user docs tests
run: |
BIG_TESTS_INSTANCE=TRUE make test_user_doc_gpu
- name: Test C API
run: |
BIG_TESTS_INSTANCE=TRUE make test_c_api_gpu
- name: Run High Level API Tests
run: |
BIG_TESTS_INSTANCE=TRUE make test_high_level_api_gpu
slack-notify:
name: Slack Notification
needs: [ setup-instance, cuda-tests-linux ]
runs-on: ubuntu-latest
if: ${{ always() && needs.cuda-tests-linux.result != 'skipped' && failure() }}
continue-on-error: true
steps:
- name: Send message
uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
env:
SLACK_COLOR: ${{ needs.cuda-tests-linux.result }}
SLACK_MESSAGE: "Fast H100 tests finished with status: ${{ needs.cuda-tests-linux.result }}. (${{ env.ACTION_RUN_URL }})"
teardown-instance:
name: Teardown instance (cuda-h100-tests)
if: ${{ always() && needs.setup-instance.result != 'skipped' }}
needs: [ setup-instance, cuda-tests-linux ]
runs-on: ubuntu-latest
steps:
- name: Stop instance
id: stop-instance
uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
with:
mode: stop
github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
slab-url: ${{ secrets.SLAB_BASE_URL }}
job-secret: ${{ secrets.JOB_SECRET }}
label: ${{ needs.setup-instance.outputs.runner-name }}
- name: Slack Notification
if: ${{ failure() }}
continue-on-error: true
uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
env:
SLACK_COLOR: ${{ job.status }}
SLACK_MESSAGE: "Instance teardown (cuda-h100-tests) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"