TheRock/rocm-systems/.github/workflows/therock-rccl-test-rocprof.yml

name: TheRock RCCL rocprofv3 smoke test

on:
  workflow_call:
    inputs:
      amdgpu_families:
        type: string
      artifact_group:
        type: string
      test_runs_on:
        type: string
      artifact_run_id:
        type: string
      notify_email:
        type: string
        default: ""
      notify_webhook:
        type: string
        default: ""
  workflow_dispatch:
    inputs:
      amdgpu_families:
        type: string
        default: "gfx94X-dcgpu"
      artifact_group:
        type: string
        default: "gfx94X-dcgpu"
      test_runs_on:
        type: string
        default: "linux-gfx942-8gpu-ossci-rocm"
      artifact_run_id:
        type: string
        required: true
        description: "Run ID of a completed RCCL build to fetch artifacts from"
      notify_email:
        type: string
        default: ""
        description: "Email address for result notification"
      notify_webhook:
        description: 'Teams webhook URL for test summary notifications'
        type: string
        default: ''

permissions:
  contents: read

jobs:
  test_rccl_rocprof:
    name: 'rocprofv3 smoke test'
    runs-on: ${{ inputs.test_runs_on }}
    container:
      image: ghcr.io/rocm/no_rocm_image_ubuntu24_04@sha256:4150afe4759d14822f0e3f8930e1124f26e11f68b5c7b91ec9a02b20b1ebbb98
      options: --ipc host
        --group-add video
        --device /dev/kfd
        --device /dev/dri
        --group-add 110
        --group-add 993
        --group-add 992
        --ulimit memlock=-1:-1
        --security-opt seccomp=unconfined
        --env-file /etc/podinfo/gha-gpu-isolation-settings
        --user 0:0
    defaults:
      run:
        shell: bash
    env:
      VENV_DIR: ${{ github.workspace }}/.venv
      ARTIFACT_RUN_ID: "${{ inputs.artifact_run_id }}"
      OUTPUT_ARTIFACTS_DIR: "./build"
      THEROCK_BIN_DIR: "./build/bin"
    steps:
      - name: Checkout TheRock
        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
        with:
          repository: "ROCm/TheRock"
          ref: release/therock-10.0

      - name: Checkout rocm-systems scripts
        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
        with:
          sparse-checkout: .github/scripts
          path: "rocm-systems"

      - name: Run setup test environment
        uses: './.github/actions/setup_test_environment'
        with:
          ARTIFACT_RUN_ID: ${{ env.ARTIFACT_RUN_ID }}
          ARTIFACT_GROUP: ${{ inputs.artifact_group }}
          OUTPUT_ARTIFACTS_DIR: ${{ env.OUTPUT_ARTIFACTS_DIR }}
          VENV_DIR: ${{ env.VENV_DIR }}
          FETCH_ARTIFACT_ARGS: "--rccl --tests --rocprofiler-sdk"

      - name: Install pytest
        run: |
          pip install pytest

      - name: Run rocprofv3 RCCL smoke tests
        timeout-minutes: 15
        env:
          THEROCK_BIN_DIR: ${{ env.THEROCK_BIN_DIR }}
          NOTIFY_EMAIL: ${{ inputs.notify_email }}
          NOTIFY_WEBHOOK: ${{ inputs.notify_webhook }}
        run: |
          NOTIFY_ARGS=()
          [ -n "$NOTIFY_EMAIL" ] && NOTIFY_ARGS+=(--notify-email "$NOTIFY_EMAIL")
          [ -n "$NOTIFY_WEBHOOK" ] && NOTIFY_ARGS+=(--notify-webhook "$NOTIFY_WEBHOOK")
          set -o pipefail
          python rocm-systems/.github/scripts/test_rccl_rocprof.py \
            "${NOTIFY_ARGS[@]}" \
            2>&1 | tee rocprof_test_results.txt

      - name: Upload test results
        if: always()
        uses: actions/upload-artifact@v4
        with:
          name: rocprof-smoke-test-results-${{ inputs.amdgpu_families }}
          path: rocprof_test_results.txt