TheRock/rocm-libraries/.github/workflows/therock-ci-nightly.yml

name: TheRock CI Nightly

on:
  schedule:
    - cron: "0 7 * * *" # Runs nightly at 7 AM UTC
  workflow_dispatch:
    inputs:
      projects:
        type: string
        description: "Insert space-separated list of projects to test or 'all' to test all projects. ex: 'projects/rocprim projects/hipcub'"
      therock_ref_override:
        type: string
        description: "Optional explicit TheRock ref (branch/tag/SHA) to build against. Empty resolves the live tip of ROCm/TheRock@main."
        default: ""

permissions:
  contents: read

jobs:
  setup:
    name: "Setup"
    runs-on: ubuntu-24.04
    outputs:
      therock_ref: ${{ steps.checkout_therock.outputs.commit }}
      docker_image: ${{ steps.ci-env.outputs.docker-image }}
      rocm_package_version: ${{ steps.rocm_package_version.outputs.rocm_package_version }}
      linux_projects: ${{ steps.linux_projects.outputs.linux_projects }}
      linux_build_runs_on: ${{ steps.linux_projects.outputs.build_runs_on }}
      windows_projects: ${{ steps.windows_projects.outputs.windows_projects }}
      windows_build_runs_on: ${{ steps.windows_projects.outputs.build_runs_on }}
      test_type: ${{ steps.linux_projects.outputs.test_type }}
      linux_package_targets: ${{ steps.configure_linux.outputs.package_targets }}
      windows_package_targets: ${{ steps.configure_windows.outputs.package_targets }}
    steps:
      - name: Checkout code
        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
        with:
          sparse-checkout: .github
          sparse-checkout-cone-mode: true
          fetch-depth: 2

      - name: Load CI environment
        id: ci-env
        uses: ./.github/actions/ci-env

      # Resolve the TheRock ref to build against exactly once here. By default
      # this is the live tip of ROCm/TheRock@main; the checkout records the
      # concrete SHA as steps.checkout_therock.outputs.commit, which is what
      # every downstream job consumes (via needs.setup.outputs.therock_ref) so
      # no other job re-resolves main. Set the therock_ref_override input to
      # pin an explicit ref instead.
      - name: Checkout TheRock repository
        id: checkout_therock
        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
        with:
          repository: "ROCm/TheRock"
          path: TheRock
          ref: ${{ inputs.therock_ref_override || 'main' }}

      - name: Log locked TheRock ref
        run: |
          RESOLVED_REF="${{ steps.checkout_therock.outputs.commit }}"
          COMMIT_URL="https://github.com/ROCm/TheRock/commit/${RESOLVED_REF}"
          echo "Locked TheRock ref for this run: ${RESOLVED_REF} (${COMMIT_URL})"
          {
            echo "### Locked TheRock ref"
            echo ""
            echo "This nightly builds every job against ROCm/TheRock commit [\`${RESOLVED_REF}\`](${COMMIT_URL})."
          } >> "$GITHUB_STEP_SUMMARY"

      - 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: Set up Python
        uses: actions/setup-python@a309ff8b426b58ec0e2a45f0f869d46889d02405 # v6.2.0
        with:
          python-version: "3.12"

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

      - name: Compute package version
        id: rocm_package_version
        run: |
          python TheRock/build_tools/compute_rocm_package_version.py \
            --release-type=dev \
            --override-git-sha=${{ steps.checkout_therock.outputs.commit }}

      - name: Determine Linux projects to run
        id: linux_projects
        env:
          PROJECTS: ${{ inputs.projects }}
          PLATFORM: "linux"
          CI_CONFIG_PATH: ci-config
        run: |
          python .github/scripts/therock_configure_ci.py

      - name: Determine Windows projects to run
        id: windows_projects
        env:
          PROJECTS: ${{ inputs.projects }}
          PLATFORM: "windows"
          CI_CONFIG_PATH: ci-config
        run: |
          python .github/scripts/therock_configure_ci.py

      - name: Fetch Linux targets for build and test
        env:
          THEROCK_PACKAGE_PLATFORM: "linux"
          # TODO(geomin12): Allow dynamic values of AMDGPU_FAMILIES, with opt-in options
          AMDGPU_FAMILIES: "gfx94X, gfx950"
          # Variable comes from ROCm organization variable 'ROCM_THEROCK_TEST_RUNNERS'
          ROCM_THEROCK_TEST_RUNNERS: ${{ vars.ROCM_THEROCK_TEST_RUNNERS }}
          LOAD_TEST_RUNNERS_FROM_VAR: true
        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: "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:
      therock_ref: ${{ needs.setup.outputs.therock_ref }}
      docker_image: ${{ needs.setup.outputs.docker_image }}
      rocm_package_version: ${{ needs.setup.outputs.rocm_package_version }}
      cmake_options: ${{ matrix.projects.cmake_options }}
      projects_to_test: ${{ matrix.projects.projects_to_test }}
      test_type: ${{ needs.setup.outputs.test_type }}
      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 }}

  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:
      therock_ref: ${{ needs.setup.outputs.therock_ref }}
      rocm_package_version: ${{ needs.setup.outputs.rocm_package_version }}
      cmake_options: ${{ matrix.projects.cmake_options }}
      projects_to_test: ${{ matrix.projects.projects_to_test }}
      test_type: ${{ needs.setup.outputs.test_type }}
      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_nightly_summary:
    name: TheRock CI Nightly Summary
    if: always()
    needs:
      - setup
      - therock-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