TheRock/rocm-systems/.github/workflows/therock-test-packages.yml

name: TheRock Test Packages

on:
  workflow_dispatch:
    inputs:
      projects_to_test:
        type: string
        description: "Comma-separated list of projects to test"
      test_type:
        type: string
        default: "standard"
        description: "Test type (quick, standard, comprehensive, full)"
      amdgpu_families:
        type: string
        description: "GPU family to test (e.g., gfx94X-dcgpu, gfx125X-dcgpu)"
      test_runs_on:
        type: string
        description: "Runner label to use for tests"
      platform:
        type: string
        default: "linux"
        description: "Platform (linux or windows)"
      artifact_run_id:
        type: string
        default: ""
        description: "Run ID to fetch artifacts from (leave empty for current run)"
      sanity_check_only_for_family:
        type: boolean
        default: false
        description: "Run only sanity check for this family"
  workflow_call:
    inputs:
      projects_to_test:
        type: string
      test_type:
        type: string
        default: "standard"
      amdgpu_families:
        type: string
      test_runs_on:
        type: string
      platform:
        type: string
      artifact_run_id:
        type: string
        default: ""
      sanity_check_only_for_family:
        type: boolean
        default: false

permissions:
  contents: read

jobs:
  configure_test_matrix:
    name: "Configure test matrix"
    # Use standard GitHub runner for config - no GPU needed
    runs-on: ubuntu-24.04
    if: ${{ inputs.test_runs_on != '' }}
    outputs:
      components: ${{ steps.configure.outputs.components }}
      platform: ${{ steps.configure.outputs.platform }}
      sanity_component: ${{ steps.configure.outputs.sanity_component }}
    steps:
      - name: "Fetch 'build_tools' from repository"
        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
        with:
          sparse-checkout: build_tools
          path: "prejob"
          repository: "ROCm/TheRock"

      # Checkout failure is possible on Windows, as it's the first job on a GPU test runner.
      # Post-job cleanup isn't necessary since no executables are launched in this job.
      - name: Pre-job cleanup processes on Windows
        if: ${{ runner.os == 'Windows' }}
        shell: powershell
        run: . '${{ github.workspace }}\prejob\build_tools\github_actions\cleanup_processes.ps1'

      - name: "Checking out repository"
        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
        with:
          repository: "ROCm/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: Setting up Python
        uses: actions/setup-python@e797f83bcb11b83ae66e0230d6156d7c80228e7c # v6.0.0
        with:
          python-version: 3.12

      - name: "Configuring CI options"
        env:
          PROJECTS_TO_TEST: ${{ inputs.projects_to_test }}
          AMDGPU_FAMILIES: ${{ inputs.amdgpu_families }}
          WINDOWS_HIP_ROCR_TESTS: ${{ contains(inputs.projects_to_test, 'hip-tests') }}
          CI_CONFIG_PATH: ci-config
        id: configure
        run: python ./build_tools/github_actions/fetch_test_configurations.py --platform ${{ inputs.platform }}

  test_sanity_check:
    name: 'Test Sanity Check'
    needs: configure_test_matrix
    uses: './.github/workflows/therock-test-component.yml'
    with:
      artifact_run_id: ${{ inputs.artifact_run_id }}
      amdgpu_families: ${{ inputs.amdgpu_families }}
      # Use the runner selected by fetch_test_configurations.py
      # TEMPORARY: For gfx125X-dcgpu, use inputs.test_runs_on since test_runner is not
      # configured in therock-ci-config yet (limited MI455 machines)
      test_runs_on: >-
        ${{
          (contains(inputs.amdgpu_families, 'gfx125X') && inputs.test_runs_on) ||
          fromJSON(needs.configure_test_matrix.outputs.sanity_component).test_runner
        }}
      platform: ${{ needs.configure_test_matrix.outputs.platform }}
      component: ${{ needs.configure_test_matrix.outputs.sanity_component }}

  test_components:
    name: 'Test ${{ matrix.components.job_name }}'
    needs: [test_sanity_check, configure_test_matrix]
    # skip tests if no test matrix to run and sanity check only requested
    if: ${{ needs.configure_test_matrix.outputs.components != '[]' && !inputs.sanity_check_only_for_family }}
    strategy:
      fail-fast: false
      matrix:
        components: ${{ fromJSON(needs.configure_test_matrix.outputs.components) }}
    uses: './.github/workflows/therock-test-component.yml'
    with:
      artifact_run_id: ${{ inputs.artifact_run_id }}
      amdgpu_families: ${{ inputs.amdgpu_families }}
      # Route to the appropriate runner:
      #   1. Multi-GPU components use the multi-GPU runner
      #   2. CPU-only components use standard GitHub runners (azure-linux-scale-rocm)
      #   3. Per-component runner (set by fetch_test_configurations.py)
      # TEMPORARY: For gfx125X-dcgpu, use inputs.test_runs_on since test_runner is not
      # configured in therock-ci-config yet (limited MI455 machines)
      test_runs_on: >-
        ${{
          (contains(inputs.amdgpu_families, 'gfx125X') && inputs.test_runs_on) ||
          (matrix.components.multi_gpu_runner != '' &&
            matrix.components.multi_gpu_runner) ||
          (matrix.components.linux_cpu_runner == true &&
            'azure-linux-scale-rocm') ||
          matrix.components.test_runner
        }}
      platform: ${{ needs.configure_test_matrix.outputs.platform }}
      component: ${{ toJSON(matrix.components) }}