TheRock/rocm-systems/.github/workflows/therock-ci.yml

name: TheRock CI

on:
  push:
    branches:
      - develop
      - release/therock-*
  pull_request:
    types:
      - opened
      - synchronize
      - reopened
      - labeled
  workflow_dispatch:
    inputs:
      projects:
        type: string
        description: "Insert space-separated list of projects to test or 'all' to test all projects. ex: 'projects/clr projects/rocminfo'"
      linux_amdgpu_families:
        type: string
        description: "Insert comma-separated list of Linux GPU families to build and test. ex: gfx94X, gfx950"
        default: ""
      windows_amdgpu_families:
        type: string
        description: "Insert comma-separated list of Windows GPU families to build and test. ex: gfx1151"
        default: ""

permissions:
  contents: read

concurrency:
  # A PR number if a pull request and otherwise the commit hash. This cancels
  # queued and in-progress runs for the same PR (presubmit) or commit
  # (postsubmit). The workflow name is prepended to avoid conflicts between
  # different workflows.
  group: ${{ github.workflow }}-${{ github.event.number || github.sha }}
  cancel-in-progress: true

jobs:
  setup:
    name: "Setup"
    runs-on: ubuntu-24.04
    env:
      # The commit being checked out is the merge commit for a PR. Its first
      # parent will be the tip of the base branch.
      BASE_REF: HEAD^
    outputs:
      linux_projects: ${{ steps.linux_projects.outputs.projects }}
      windows_projects: ${{ steps.windows_projects.outputs.projects }}
      run_linux_rccl_ci: ${{ steps.linux_projects.outputs.run_linux_rccl_ci }}
      run_mi455_test: ${{ steps.linux_projects.outputs.run_mi455_test }}
      test_type: ${{ steps.linux_projects.outputs.test_type }}
      package_version: ${{ steps.rocm_package_version.outputs.rocm_package_version }}
      linux_package_targets: ${{ steps.configure_linux.outputs.package_targets }}
      windows_package_targets: ${{ steps.configure_windows.outputs.package_targets }}
      linux_build_runs_on: ${{ steps.linux_projects.outputs.build_runs_on }}
      windows_build_runs_on: ${{ steps.windows_projects.outputs.build_runs_on }}
    steps:
      - name: Checkout code
        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
        with:
          sparse-checkout: .github
          sparse-checkout-cone-mode: true
          fetch-depth: 2

      - name: Set up Python
        uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0
        with:
          python-version: "3.12"

      - name: Install python dependencies
        run: |
          python -m pip install --upgrade pip
          pip install pydantic pytest requests

      - name: Run TheRock CI unit tests
        run: python -m pytest .github/scripts/tests/therock_configure_ci_test.py

      - name: Checkout TheRock repository
        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
        with:
          repository: "ROCm/TheRock"
          path: TheRock
          ref: release/therock-10.0

      - name: Checkout CI config
        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
        with:
          repository: ROCm/therock-ci-config
          ref: main
          path: ci-config
        continue-on-error: true

      - name: Detect changed subtrees
        id: detect
        if: github.event_name == 'pull_request'
        env:
          GH_TOKEN: ${{ github.token }}
        run: |
          python .github/scripts/pr_detect_changed_subtrees.py \
            --repo "${{ github.repository }}" \
            --pr "${{ github.event.pull_request.number }}" \
            --config ".github/repos-config.json"

      - name: Set PR_LABELS variable with labels assigned to pull request
        if: ${{ github.event.pull_request }}
        env:
          GITHUB_TOKEN: ${{ github.token }}
          PR_NUMBER: ${{ github.event.number }}
        run: |
          echo "PR_LABELS=$(gh pr view ${PR_NUMBER} --json labels)" >> $GITHUB_ENV

      - name: Determine Linux projects to run
        id: linux_projects
        env:
          SUBTREES: ${{ steps.detect.outputs.subtrees }}
          PROJECTS: ${{ inputs.projects }}
          PLATFORM: "linux"
        run: |
          python .github/scripts/therock_configure_ci.py

      - name: Determine Windows projects to run
        id: windows_projects
        env:
          SUBTREES: ${{ steps.detect.outputs.subtrees }}
          PROJECTS: ${{ inputs.projects }}
          PLATFORM: "windows"
        run: |
          python .github/scripts/therock_configure_ci.py

      - name: Compute package version
        id: rocm_package_version
        run: |
          python TheRock/build_tools/compute_rocm_package_version.py --release-type=dev

      - name: Fetch Linux targets for build and test
        env:
          THEROCK_PACKAGE_PLATFORM: "linux"
          AMDGPU_FAMILIES: ${{ inputs.linux_amdgpu_families || 'gfx94X, gfx950' }}
          CI_CONFIG_PATH: ci-config
        id: configure_linux
        run: python ./TheRock/build_tools/github_actions/fetch_package_targets.py

      - name: Fetch Windows targets for build and test
        env:
          THEROCK_PACKAGE_PLATFORM: "windows"
          AMDGPU_FAMILIES: ${{ inputs.windows_amdgpu_families || 'gfx1151' }}
          CI_CONFIG_PATH: ci-config
        id: configure_windows
        run: python ./TheRock/build_tools/github_actions/fetch_package_targets.py

  therock-ci-linux:
    name: Linux (${{ matrix.projects.projects_to_test }} | ${{ matrix.target_bundle.amdgpu_family }})
    permissions:
      contents: read
      id-token: write
    needs: setup
    if: ${{ needs.setup.outputs.linux_projects != '[]' }}
    strategy:
      fail-fast: false
      matrix:
        projects: ${{ fromJSON(needs.setup.outputs.linux_projects) }}
        target_bundle: ${{ fromJSON(needs.setup.outputs.linux_package_targets) }}
    uses: ./.github/workflows/therock-ci-linux.yml
    secrets: inherit
    with:
      cmake_options: ${{ matrix.projects.cmake_options }}
      projects_to_test: ${{ matrix.projects.projects_to_test }}
      test_type: ${{ needs.setup.outputs.test_type }}
      package_version: ${{ needs.setup.outputs.package_version }}
      amdgpu_families: ${{ matrix.target_bundle.amdgpu_family }}
      test_runs_on: ${{ matrix.target_bundle.test_machine }}
      build_runs_on: ${{ needs.setup.outputs.linux_build_runs_on }}

  # TEMPORARY: MI455 (gfx125X) Build - Blocking job (included in summary)
  # Separated from therock-ci-linux matrix so MI455 tests can depend only on this
  # build, not on other families' tests which may take longer.
  # TODO: Remove this separate job once MI455 test infrastructure is stable and
  # we have sufficient MI455 machine capacity to run tests without blocking PRs.
  therock-build-linux-mi455:
    name: Linux MI455 Build (gfx125X-dcgpu)
    permissions:
      contents: read
      id-token: write
    needs: setup
    if: ${{ needs.setup.outputs.linux_projects != '[]' && needs.setup.outputs.run_mi455_test == 'true' }}
    uses: ./.github/workflows/therock-build-linux.yml
    secrets: inherit
    with:
      package_version: ${{ needs.setup.outputs.package_version }}
      amdgpu_families: gfx125X-dcgpu
      build_runs_on: ${{ needs.setup.outputs.linux_build_runs_on }}

  # TEMPORARY: MI455 (gfx125X) Test - Non-blocking job (excluded from summary)
  # This job is intentionally excluded from therock_ci_summary to avoid blocking PRs.
  # MI455 machines are limited, so jobs may queue for extended periods. By keeping
  # tests separate, PRs can still merge while MI455 tests run or wait in queue.
  # TODO: Remove this separate job once MI455 test infrastructure is stable and
  # we have sufficient MI455 machine capacity to run tests without blocking PRs.
  therock-test-linux-mi455:
    name: Linux MI455 Test
    needs: [setup, therock-build-linux-mi455]
    if: ${{ needs.setup.outputs.run_mi455_test == 'true' }}
    uses: ./.github/workflows/therock-test-packages.yml
    with:
      projects_to_test: "hip-tests, rocrtst"
      amdgpu_families: gfx125X-dcgpu
      # MI455 machine is down - disabled until back online (was: linux-mi455-gpu-rocm)
      test_runs_on: ""
      platform: linux
      artifact_run_id: ${{ github.run_id }}

  therock-rccl-ci-linux:
    name: TheRock RCCL CI Linux
    needs: setup
    if: ${{ needs.setup.outputs.run_linux_rccl_ci == 'true' }}
    permissions:
      contents: read
      id-token: write
    strategy:
      fail-fast: false
      matrix:
        amdgpu_family: [gfx94X-dcgpu, gfx950-dcgpu]
    uses: ./.github/workflows/therock-rccl-ci-linux.yml
    secrets: inherit
    with:
      amdgpu_families: ${{ matrix.amdgpu_family }}
      artifact_group: ${{ matrix.amdgpu_family }}
      package_version: ${{ needs.setup.outputs.package_version }}
      cmake_options: >
        -DTHEROCK_ENABLE_ALL=OFF
        -DTHEROCK_BUILD_TESTING=ON
        -DTHEROCK_BUNDLE_SYSDEPS=ON
        -DTHEROCK_ENABLE_COMM_LIBS=ON
        -DTHEROCK_ENABLE_ROCPROFV3=ON
        -DTHEROCK_ENABLE_MPI=ON

  therock-ci-windows:
    name: Windows (${{ matrix.projects.projects_to_test }} | ${{ matrix.target_bundle.amdgpu_family }})
    permissions:
      contents: read
      id-token: write
    needs: setup
    if: ${{ needs.setup.outputs.windows_projects != '[]' }}
    strategy:
      fail-fast: false
      matrix:
        projects: ${{ fromJSON(needs.setup.outputs.windows_projects) }}
        target_bundle: ${{ fromJSON(needs.setup.outputs.windows_package_targets) }}
    uses: ./.github/workflows/therock-ci-windows.yml
    secrets: inherit
    with:
      cmake_options: ${{ matrix.projects.cmake_options }}
      projects_to_test: ${{ matrix.projects.projects_to_test }}
      test_type: ${{ needs.setup.outputs.test_type }}
      package_version: ${{ needs.setup.outputs.package_version }}
      amdgpu_families: ${{ matrix.target_bundle.amdgpu_family }}
      test_runs_on: ${{ matrix.target_bundle.test_machine }}
      build_runs_on: ${{ needs.setup.outputs.windows_build_runs_on }}

  therock_ci_summary:
    name: TheRock CI Summary
    if: always()
    needs:
      - setup
      - therock-ci-linux
      - therock-build-linux-mi455
      - therock-rccl-ci-linux
      - therock-ci-windows
    runs-on: ubuntu-24.04
    steps:
      - name: Output failed jobs
        run: |
          echo '${{ toJson(needs) }}'
          FAILED_JOBS="$(echo '${{ toJson(needs) }}' \
            | jq --raw-output \
            'map_values(select(.result!="success" and .result!="skipped")) | keys | join(",")' \
          )"
          if [[ "${FAILED_JOBS}" != "" ]]; then
            echo "The following jobs failed: ${FAILED_JOBS}"
            exit 1
          fi