conda_build_and_test.yml

name: Conda build and test

on:
  workflow_dispatch:
  schedule:
    - cron: "5 2 * * *"
  push:

permissions: read-all

env:
  BACKEND: XPU
  TRITON_DISABLE_LINE_INFO: 1

jobs:

  integration-tests:
    name: Integration tests
    runs-on:
      - glados
      - spr
      - conda-0.0.1
    strategy:
      matrix:
        python:
          - "3.9"
          - "3.10"
    defaults:
      run:
        shell: bash -noprofile --norc -eo pipefail -c "source /home/runner/intel/oneapi/setvars.sh > /dev/null && source /home/runner/miniconda3/etc/profile.d/conda.sh && source {0}"
    steps:
      - name: Checkout repository
        uses: actions/checkout@v4

      - name: Load pip cache
        id: pip-cache
        uses: ./.github/actions/load
        env:
          # Increase this value to reset cache
          CACHE_NUMBER: 2
        with:
          path: $HOME/.cache/pip
          key: pip-${{ matrix.python }}-${{ hashFiles('python/pyproject.toml', 'python/setup.py') }}-${{ env.CACHE_NUMBER }}

      - name: Get LLVM commit id
        uses: ./.github/actions/get-commit-id
        with:
          repository: intel/llvm.git
          branch: genx
          variable: LLVM_COMMIT_ID

      - name: Calculate env
        run: |
          PACKAGES_CACHE_KEY=$(echo $LLVM_COMMIT_ID ${{ hashFiles('scripts/compile-triton.sh') }} | sha256sum - | cut -d\  -f1)
          echo "PACKAGES_CACHE_KEY=$PACKAGES_CACHE_KEY" >>$GITHUB_ENV
          echo $HOME/miniconda3/bin >>$GITHUB_PATH

      - name: Load packages cache
        id: packages-cache
        uses: ./.github/actions/load
        env:
          # Increase this value to reset cache
          CACHE_NUMBER: 3
        with:
          path: $HOME/packages
          key: conda-packages-${{ env.PACKAGES_CACHE_KEY }}-${{ env.CACHE_NUMBER }}

      - name: Load conda cache
        id: conda-cache
        uses: ./.github/actions/load
        env:
          CACHE_NUMBER: 5
        with:
          path: $HOME/miniconda3/envs/triton
          key: conda-py${{ matrix.python }}-${{ hashFiles('scripts/triton.yml') }}-${{ env.CACHE_NUMBER }}

      - name: Update conda env
        if: ${{ steps.conda-cache.outputs.status == 'miss' }}
        run: |
          conda create -n triton --override-channels -c conda-forge python=${{ matrix.python }}.*
          conda env update -f scripts/triton.yml
          find /home/runner/intel/oneapi/ \( -name '*.so' -or -name '*.so.*' \) -exec cp -n {} $HOME/miniconda3/envs/triton/lib \;
          ln -snf /usr/include/level_zero $HOME/miniconda3/envs/triton/bin/../x86_64-conda-linux-gnu/sysroot/usr/include/level_zero
          find /usr -name libze_\* -exec cp -n {} $HOME/miniconda3/envs/triton/lib \;

      - name: Add conda info to log
        run: |
          conda info
          conda list -n triton

      - name: Build packages
        if: ${{ steps.packages-cache.outputs.status == 'miss' }}
        run: |
          conda run --no-capture-output -n triton ./scripts/compile-triton.sh --llvm

      - name: Save packages cache
        if: ${{ steps.packages-cache.outputs.status == 'miss' }}
        uses: ./.github/actions/save
        with:
          path: ${{ steps.packages-cache.outputs.path }}
          dest: ${{ steps.packages-cache.outputs.dest }}

      - name: Build Triton
        run: |
          conda run -n triton scripts/compile-triton.sh --triton

      - name: Run core tests
        run: |
          conda run -n triton scripts/test-triton.sh

      - name: Save conda cache
        if: ${{ steps.conda-cache.outputs.status == 'miss' }}
        uses: ./.github/actions/save
        with:
          path: ${{ steps.conda-cache.outputs.path }}
          dest: ${{ steps.conda-cache.outputs.dest }}

      - name: Run E2E test
        run: |
          conda activate triton
          # Set WORKSPACE for inductor_xpu_test.sh to make sure it creates "inductor_log" outside of pytorch cloned directory
          export WORKSPACE=$GITHUB_WORKSPACE
          cd ../pytorch
          TRANSFORMERS_VERSION="$(<.ci/docker/ci_commit_pins/huggingface.txt)"
          pip install pyyaml pandas scipy numpy psutil pyre_extensions torchrec transformers==$TRANSFORMERS_VERSION
          # TODO: Find the fastest Hugging Face model
          $GITHUB_WORKSPACE/scripts/inductor_xpu_test.sh huggingface float32 inference accuracy xpu 0 static 1 0 AlbertForMaskedLM
          # The script above always returns 0, so we need an additional check to see if the accuracy test passed
          cat $WORKSPACE/inductor_log/*/*/*.csv
          grep AlbertForMaskedLM $WORKSPACE/inductor_log/*/*/*.csv | grep -q ,pass,

      - name: Save pip cache
        if: ${{ steps.pip-cache.outputs.status == 'miss' }}
        uses: ./.github/actions/save
        with:
          path: ${{ steps.pip-cache.outputs.path }}
          dest: ${{ steps.pip-cache.outputs.dest }}