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

name: TheRock CI for rccl

on:
  schedule:
    - cron: '17 6 * * 0'  # Weekly, Sunday 06:17 UTC
  workflow_dispatch:
    inputs:
      artifact_run_id:
        description: 'Reuse artifacts from a previous run ID (skips build)'
        type: string
        default: ''
      test_scope:
        description: 'smoke (22 curated tests, ~3min) or all'
        type: string
        default: 'smoke'
      notify_email:
        description: 'Email address to send test summary report'
        type: string
        default: 'collectives-ci@amd.com'
      notify_webhook:
        description: 'Teams webhook URL for test summary notifications'
        type: string
        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:
    runs-on: ubuntu-24.04
    outputs:
      run_linux_ci: ${{ steps.rccl_check.outputs.run_linux_ci }}

    steps:
      - name: "Checking out repository"
        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
        with:
          # We need the parent commit to do a diff
          fetch-depth: 2

      - name: "Detect RCCL file changes"
        id: rccl_check
        shell: bash
        run: |
          echo "Changed files:"
          git diff --name-only HEAD^

          # Scheduled and manual runs always execute the full CI pipeline.
          # PR/push runs only trigger CI when files under projects/rccl/
          # have changed, to avoid burning GPU runner time on unrelated commits.
          if [ "${{ github.event_name }}" = "schedule" ] || [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
            echo "run_linux_ci=true" >> "$GITHUB_OUTPUT"
            echo "Scheduled/manual run — always run CI"
          elif git diff --name-only HEAD^ | grep -q '^projects/rccl/'; then
            echo "run_linux_ci=true" >> "$GITHUB_OUTPUT"
            echo "RCCL changes detected"
          else
            echo "run_linux_ci=false" >> "$GITHUB_OUTPUT"
            echo "No RCCL changes detected"
          fi

  therock-ci-linux:
    name: TheRock CI Linux
    needs: setup
    if: ${{ needs.setup.outputs.run_linux_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 }}
      event_name: ${{ github.event_name }}
      package_version: ADHOCBUILD
      artifact_run_id: ${{ inputs.artifact_run_id }}
      test_scope: ${{ inputs.test_scope || 'smoke' }}
      notify_email: ${{ inputs.notify_email || 'collectives-ci@amd.com' }}
      notify_webhook: ${{ inputs.notify_webhook }}
      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_summary:
    name: TheRock CI Summary
    if: always()
    needs:
      - setup
      - therock-ci-linux
    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