TheRock/rocm-systems/.github/workflows/rocprofiler-compute-continuous-integration.yml

name: rocprofiler-compute Continuous Integration
run-name: rocprofiler-compute-ci

on:
  schedule:
    - cron: '0 6 * * *'
  workflow_dispatch:
    inputs:
      mode:
        description: 'Build mode'
        required: false
        default: 'continuous'
        type: choice
        options:
          - continuous
          - nightly
  push:
    branches: [ develop ]
    paths:
      - '.github/workflows/rocprofiler-compute-continuous-integration.yml'
      - 'projects/rocprofiler-compute/**'
      - '!**/*.md'
      - '!**/*.rtf'
      - '!**/*.rst'
      - '!**/.markdownlint-ci2.yaml'
      - '!**/.readthedocs.yaml'
      - '!**/.spellcheck.local.yaml'
      - '!**/.wordlist.txt'
      - '!projects/rocprofiler-compute/docs/**'
      - '!projects/rocprofiler-compute/source/docs/**'
      - '!projects/rocprofiler-compute/source/python/gui/**'
      - '!projects/rocprofiler-compute/docker/**'
      - '!projects/rocprofiler-compute/CMakePresets.json'
  pull_request:
    paths:
      - '.github/workflows/rocprofiler-compute-continuous-integration.yml'
      - 'projects/rocprofiler-compute/**'
      - '!**/*.md'
      - '!**/*.rtf'
      - '!**/*.rst'
      - '!**/.markdownlint-ci2.yaml'
      - '!**/.readthedocs.yaml'
      - '!**/.spellcheck.local.yaml'
      - '!**/.wordlist.txt'
      - '!projects/rocprofiler-compute/docs/**'
      - '!projects/rocprofiler-compute/source/docs/**'
      - '!projects/rocprofiler-compute/source/python/gui/**'
      - '!projects/rocprofiler-compute/docker/**'
      - '!projects/rocprofiler-compute/CMakePresets.json'

concurrency:
  group: ${{ github.workflow }}-${{ github.ref }}
  cancel-in-progress: true

env:
  ROCPROFSYS_CI: ON
  ROCM_PATH: "/opt/rocm"

jobs:
  prepare_matrix:
    runs-on: ubuntu-latest
    outputs:
      matrix: ${{ steps.generate_matrix.outputs.matrix }}

    steps:
      - uses: actions/checkout@v6
        with:
          sparse-checkout: projects/rocprofiler-compute/.github

      - name: Generate and output matrix
        id: generate_matrix
        working-directory: projects/rocprofiler-compute/.github
        run: |
          if [ '${{ github.event_name }}' = 'schedule' ] || [ '${{ inputs.mode }}' = 'nightly' ]; then
            MATRIX_CONTENT=$(cat ci-matrix.yml | yq '.matrix-ubuntu-nightly' -I=0 -o=json)
          else
            MATRIX_CONTENT=$(cat ci-matrix.yml | yq '.matrix-ubuntu-ci' -I=0 -o=json)
          fi
          echo "matrix=${MATRIX_CONTENT}" >> $GITHUB_OUTPUT

  ubuntu:
    name: Ubuntu ${{ matrix.system.os-release }} • ${{ matrix.system.arch }}
    needs: prepare_matrix
    strategy:
      fail-fast: false
      matrix:
        system: ${{ fromJSON(needs.prepare_matrix.outputs.matrix) }}
    runs-on: ${{ matrix.system.runner }}
    env:
      HIP_PLATFORM: "amd"
      OMPI_ALLOW_RUN_AS_ROOT: 1
      OMPI_ALLOW_RUN_AS_ROOT_CONFIRM: 1
    permissions:
      packages: read
    container:
      image: ghcr.io/rocm/rocprofiler-ubuntu:${{ matrix.system.os-release }}-compute-ci-multiarch
      # --env-file /etc/podinfo/gha-gpu-isolation-settings - Required for GPU isolation on OSSCI MIXXX runners
      options:
        --privileged
        --ipc host
        --group-add video
        --device /dev/kfd
        --device /dev/dri
        --cap-add CAP_SYS_ADMIN
        --env-file /etc/podinfo/gha-gpu-isolation-settings
        --pull always

    steps:
      - uses: actions/checkout@v6
        with:
          ref: ${{ github.event_name == 'schedule' && 'rocprofiler-compute-develop' || github.ref }}
          sparse-checkout: |
            projects/rocprofiler-compute
            shared/ctest

      - name: Setup Environment
        id: setup_env
        run: |
          if [ '${{ github.event_name }}' = 'schedule' ]; then
            MODE=Nightly
            ADD_COVERAGE="--coverage"
          else
            MODE=Continuous
            ADD_COVERAGE=""
          fi
          echo "mode=${MODE}" >> $GITHUB_OUTPUT
          echo "add_coverage=${ADD_COVERAGE}" >> $GITHUB_OUTPUT

          if [ '${{ matrix.system.os-release }}' = '24.04' ]; then
            CODE_NAME=noble
          else
            CODE_NAME=jammy
          fi
          echo "code_name=${CODE_NAME}" >> $GITHUB_OUTPUT

      - name: Install dependencies
        shell: bash
        run: |
          apt-get update
          apt install -y libdw-dev pkg-config
          echo "✅ ROCm Dependencies Installed!"

      - name: Install Latest Nightly ROCm
        run: |
          set -e
          mkdir ${{ env.ROCM_PATH }}
          TARBALL_ROCM_VERSION=$(ls /opt/*.tar.gz | grep -Eo '*([0-9]+\.[0-9]+\.[0-9]+)*')
          tar -xf ${{ env.ROCM_PATH }}-${TARBALL_ROCM_VERSION}-multiarch.tar.gz -C ${{ env.ROCM_PATH }}
          rm ${{ env.ROCM_PATH }}-${TARBALL_ROCM_VERSION}-multiarch.tar.gz
          echo "ROCm installed to: ${{ env.ROCM_PATH }}"
          echo "✅ ROCm Installation Complete!"

      - name: Output TheRock Manifest
        run: |
          cat ${{ env.ROCM_PATH }}/share/therock/therock_manifest.json

      - name: Install Python Requirements
        working-directory: projects/rocprofiler-compute
        run: |
          pip --version
          for i in 1 2 3; do
            pip install -r requirements.txt --break-system-packages --ignore-installed --timeout 60 && break
            echo "⚠️ pip install attempt $i failed, retrying..."
            sleep 10
          done
          for i in 1 2 3; do
            pip install -r requirements-test.txt --break-system-packages --ignore-installed --timeout 60 && break
            echo "⚠️ pip install attempt $i failed, retrying..."
            sleep 10
          done
          echo "✅ pip requirements installed!"

      - name: Configure, Build, and Test
        id: test
        continue-on-error: true
        timeout-minutes: ${{ steps.setup_env.outputs.mode == 'Nightly' && 120 || 90 }}
        working-directory: projects/rocprofiler-compute
        run: |
          set -e
          git config --global --add safe.directory ${GITHUB_WORKSPACE}
          git config --global --add safe.directory ${PWD}
          PATH=${{ env.ROCM_PATH }}/bin:${{ env.ROCM_PATH }}/llvm/bin:$PATH \
          LD_LIBRARY_PATH=${{ env.ROCM_PATH }}/lib:${{ env.ROCM_PATH }}/llvm/lib:${{ env.ROCM_PATH }}/lib/rocm_sysdeps/lib:$LD_LIBRARY_PATH \
          python3 ./tools/run-ci.py \
            --name "ROCm/rocprofiler-compute-${{ github.ref_name }}-ubuntu-${{ matrix.system.os-release }}-${{ matrix.system.gpu }}" \
            --actor "${{ github.actor }}" \
            --site ${{ matrix.system.runner }} \
            --mode ${{ steps.setup_env.outputs.mode }} \
            --build-jobs 16 \
            ${{ steps.setup_env.outputs.add_coverage }} \
            -- \
            -DCMAKE_BUILD_TYPE=Release \
            -DCMAKE_PREFIX_PATH=${{ env.ROCM_PATH }} \
            -DENABLE_TESTS=ON \
            -DINSTALL_TESTS=ON \
            -DPYTEST_NUMPROCS=8 \
            -- \
            -L full \
            -LE full_rocprofiler_compute_ci_exclude

      - name: Upload Test Logs (on failure)
        if: steps.test.outcome == 'failure'
        uses: actions/upload-artifact@v7
        with:
          name: failed-test-logs-ubuntu-${{ matrix.system.os-release }}-${{ matrix.system.arch }}
          path: |
            projects/rocprofiler-compute/build/ctest-results.xml
            projects/rocprofiler-compute/build/Testing/Temporary/**
          retention-days: 2

      - name: Check Test Results
        if: always()
        shell: bash
        run: |
          if [[ "${{ steps.test.outcome }}" == "failure" ]]; then
            echo "::error::Tests failed - see CDash for details: https://my.cdash.org/index.php?project=rocprofiler-compute"
            exit 1
          fi
          echo "✅ All tests passed!"