cuda.yaml

name: CUDA

# Concurrency control:
# - PRs: new commits on a feature branch will cancel in-progress (outdated) runs.
# - Push to develop: runs queue sequentially, never cancelled.
# - `workflow_dispatch`: groups by branch and queues if run on develop.
concurrency:
  group: ${{ github.workflow }}-${{ github.ref }}
  cancel-in-progress: ${{ github.ref != 'refs/heads/develop' }}
on:
  push:
    branches: [develop]
  pull_request: { }
  workflow_dispatch: { }

permissions:
  contents: read

env:
  CARGO_TERM_COLOR: always
  RUST_BACKTRACE: 1

jobs:
  changes:
    name: "Detect CUDA changes"
    runs-on: ubuntu-latest
    timeout-minutes: 10
    permissions:
      pull-requests: read
    outputs:
      run-cuda-san: ${{ github.event_name != 'pull_request' || steps.filter.outputs.cuda == 'true' }}
    steps:
      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0  # v7
      - uses: dorny/paths-filter@fbd0ab8f3e69293af611ebaee6363fc25e6d187d  # v4
        id: filter
        if: github.event_name == 'pull_request'
        with:
          filters: |
            cuda:
              - "vortex-cuda/**"
              - "vortex-python/python/vortex/__init__.py"
              - "vortex-python/pyproject.toml"
              - "vortex-python/test/test_cuda.py"
              - "vortex-python-cuda/**"
              - "vortex-test/**"
              - "pyproject.toml"
              - "uv.lock"
              - ".github/workflows/**"

  cuda-build-lint:
    needs: [changes]
    if: >-
      always() && github.repository == 'vortex-data/vortex' &&
      needs.changes.outputs.run-cuda-san == 'true'
    name: "CUDA build & lint"
    timeout-minutes: 30
    runs-on: runs-on=${{ github.run_id }}/runner=gpu/tag=cuda-build
    steps:
      - uses: runs-on/action@v2
        with:
          sccache: s3
      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0  # v7
      - uses: ./.github/actions/setup-rust
        with:
          repo-token: ${{ secrets.GITHUB_TOKEN }}
      - uses: ./.github/actions/check-rebuild
        with:
          command: >-
            cargo build --profile ci --locked --all-features --all-targets
            -p vortex-cuda -p vortex-cuda-ffi -p vortex-cub -p vortex-nvcomp
            -p gpu-scan-cli -p vortex-test-e2e-cuda -p vortex-python-cuda
      - name: Clippy CUDA crates
        run: |
          cargo clippy --profile ci --locked --all-features --all-targets \
            -p vortex-cuda \
            -p vortex-cuda-ffi \
            -p vortex-cub \
            -p vortex-nvcomp \
            -p gpu-scan-cli \
            -p vortex-test-e2e-cuda \
            -p vortex-python-cuda \
            -- -D warnings

  cuda-test:
    needs: [changes]
    if: >-
      always() && github.repository == 'vortex-data/vortex' &&
      needs.changes.outputs.run-cuda-san == 'true'
    name: "CUDA tests"
    timeout-minutes: 30
    runs-on: runs-on=${{ github.run_id }}/runner=gpu/tag=cuda-tests
    steps:
      - uses: runs-on/action@v2
        with:
          sccache: s3
      - name: Display NVIDIA SMI details
        run: |
          nvidia-smi
          nvidia-smi -L
          nvidia-smi -q -d Memory
      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0  # v7
      - uses: ./.github/actions/setup-rust
        with:
          repo-token: ${{ secrets.GITHUB_TOKEN }}
      - name: Install uv
        uses: spiraldb/actions/.github/actions/setup-uv@a746510eafaa926484c354541cfc49b2ec06cc63  # 0.18.6
        with:
          sync: false
      - name: Install nextest
        uses: taiki-e/install-action@9e1e5806d4a4822de933115878265be9aaa786d9  # v2
        with:
          tool: nextest
      - name: Rust Tests
        env:
          FLAT_LAYOUT_INLINE_ARRAY_NODE: true
        run: |
          cargo nextest run \
            --cargo-profile ci \
            --locked \
            -p vortex-file \
            -p vortex-cuda \
            -p vortex-cuda-ffi \
            -p vortex-cub \
            -p vortex-nvcomp \
            -p vortex-test-e2e-cuda \
            -p vortex-python-cuda \
            --all-features \
            --no-fail-fast \
            --target x86_64-unknown-linux-gnu \
            --verbose
      - name: Python CUDA extension checks
        env:
          MATURIN_PEP517_ARGS: "--profile ci"
        run: |
          # --all-packages installs the shared dev tooling (basedpyright) from the
          # root `dev` group; --extra cuda adds the vortex-data-cuda extension.
          uv run --all-packages --extra cuda basedpyright vortex-python vortex-python-cuda
          uv run --all-packages --extra cuda pytest --benchmark-disable vortex-python/test/test_cuda.py vortex-python-cuda/test

  cuda-test-sanitizer:
    needs: [changes]
    if: >-
      always() && github.repository == 'vortex-data/vortex' &&
      needs.changes.outputs.run-cuda-san == 'true'
    name: "CUDA tests (${{ matrix.sanitizer }})"
    timeout-minutes: 30
    runs-on: runs-on=${{ github.run_id }}/runner=gpu/tag=cuda-test-sanitizer
    strategy:
      fail-fast: false
      matrix:
        include:
          - sanitizer: memcheck
            runner_flags: "--tool memcheck --leak-check=full --error-exitcode 1"
          # TODO(joe): try to re-enable racecheck, it is hanging in CI.
          # - sanitizer: racecheck
          #   runner_flags: "--tool racecheck --error-exitcode 1"
          - sanitizer: synccheck
            runner_flags: "--tool synccheck --error-exitcode 1"
          - sanitizer: initcheck
            runner_flags: "--tool initcheck --error-exitcode 1"
    steps:
      - uses: runs-on/action@v2
        with:
          sccache: s3
      - name: Display NVIDIA SMI details
        run: |
          nvidia-smi
          nvidia-smi -L
          nvidia-smi -q -d Memory
      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0  # v7
      - uses: ./.github/actions/setup-rust
        with:
          repo-token: ${{ secrets.GITHUB_TOKEN }}
      - name: Build tests
        run: cargo test --profile ci --locked -p vortex-cuda --all-features --target x86_64-unknown-linux-gnu --no-run
      - name: "CUDA - ${{ matrix.sanitizer }}"
        env:
          CARGO_TARGET_X86_64_UNKNOWN_LINUX_GNU_RUNNER: "compute-sanitizer ${{ matrix.runner_flags }}"
        run: cargo test --profile ci --locked -p vortex-cuda --all-features --target x86_64-unknown-linux-gnu

  cuda-test-cudf:
    needs: [changes]
    if: >-
      always() && github.repository == 'vortex-data/vortex' &&
      needs.changes.outputs.run-cuda-san == 'true'
    name: "CUDA tests (cudf)"
    timeout-minutes: 30
    runs-on: runs-on=${{ github.run_id }}/runner=gpu/tag=cuda-test-cudf
    steps:
      - uses: runs-on/action@v2
        with:
          sccache: s3
      - name: Display NVIDIA SMI details
        run: |
          nvidia-smi
          nvidia-smi -L
          nvidia-smi -q -d Memory
      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0  # v7
      - uses: ./.github/actions/setup-rust
        with:
          repo-token: ${{ secrets.GITHUB_TOKEN }}
      - name: Build cudf test library
        run: cargo build --profile ci --locked -p vortex-test-e2e-cuda --target x86_64-unknown-linux-gnu
      - name: Download and run cudf-test-harness
        run: |
          curl -fsSL https://github.com/vortex-data/cudf-test-harness/releases/latest/download/cudf-test-harness-x86_64.tar.gz | tar -xz
          $GITHUB_WORKSPACE/target/x86_64-unknown-linux-gnu/ci/cudf_harness_runner \
            ./cudf-test-harness-x86_64/cudf-test-harness \
            $GITHUB_WORKSPACE/target/x86_64-unknown-linux-gnu/ci/libvortex_test_e2e_cuda.so