TheRock/rocm-systems/.github/workflows/amdsmi-build.yml

# ---------------------------------------------------------------------------
# AMD-SMI CI.
#
# Builds, installs, and tests AMD-SMI against AMD GPUs on self-hosted
# runners BM/VM, using per-distro containers to cover a matrix of
# Debian/RPM distros. TheRock CI only runs in containers so it doesn't cover
# every scenario
#
# Aims to follow TheRock CI conventions as closely as possible.
# ---------------------------------------------------------------------------

name: AMDSMI CI

on:
  pull_request:
    branches: [develop]
    paths:
      - 'projects/amdsmi/**'
      - '.github/workflows/amdsmi-build.yml'
  push:
    branches: [develop]
    paths:
      - 'projects/amdsmi/**'
      - '.github/workflows/amdsmi-build.yml'
  workflow_dispatch:

permissions:
  contents: read

env:
  DEBIAN_FRONTEND: noninteractive
  DEBCONF_NONINTERACTIVE_SEEN: true
  BUILD_TYPE: Release
  ROCM_DIR: /opt/rocm

defaults:
  run:
    shell: bash

jobs:
  # Single job covering every distro + package format. The build-time logic
  # (package manager, distro quirks) is autodetected by run_amdsmi_build.py
  # from /etc/os-release, so the matrix only needs the OS label plus the
  # artifact format used for upload.
  build-and-test:
    name: ${{ matrix.os }}
    # Privileged self-hosted runner (SYS_MODULE + host /lib/modules): never
    # auto-run untrusted fork-PR code on it. Mirrors the guard other CI jobs use.
    if: ${{ github.event_name != 'pull_request' || !github.event.pull_request.head.repo.fork }}
    runs-on:
      - self-hosted
      - ${{ vars.RUNNER_TYPE }}
    continue-on-error: true
    strategy:
      max-parallel: 10
      fail-fast: false
      matrix:
        include:
          - { os: Ubuntu20,    package_format: deb, arch_suffix: _amd64.deb }
          - { os: Ubuntu22,    package_format: deb, arch_suffix: _amd64.deb }
          - { os: Debian10,    package_format: deb, arch_suffix: _amd64.deb }
          - { os: SLES,        package_format: rpm, arch_suffix: .rpm }
          - { os: RHEL8,       package_format: rpm, arch_suffix: .rpm }
          - { os: RHEL9,       package_format: rpm, arch_suffix: .rpm }
          - { os: RHEL10,      package_format: rpm, arch_suffix: .rpm }
          - { os: AlmaLinux8,  package_format: rpm, arch_suffix: .rpm }
          - { os: AzureLinux3, package_format: rpm, arch_suffix: .rpm }
    container:
      image: ${{ vars[format('{0}_DOCKER_IMAGE', matrix.os)] }}
      options: --rm --privileged --device=/dev/kfd --device=/dev/dri --group-add video --cap-add=SYS_PTRACE --security-opt seccomp=unconfined --shm-size=64G --cap-add=SYS_MODULE -v /lib/modules:/lib/modules -u root

    steps:
      - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd

      - name: Mark workspace safe for git
        run: git config --global --add safe.directory "$GITHUB_WORKSPACE"

      - name: Set Artifact Metadata
        if: github.event_name == 'pull_request'
        run: |
          set -euo pipefail
          echo "PR_NUMBER=PR${{ github.event.pull_request.number }}" >> $GITHUB_ENV
          echo "BUILD_DATE=$(date -u +%m%d%y-%H%M)" >> $GITHUB_ENV

      - name: Set Project Directory
        run: |
          set -euo pipefail
          TARGET_DIR=$(find $GITHUB_WORKSPACE -path "*/projects/amdsmi/CMakeLists.txt" -exec dirname {} \;)
          if [ -z "$TARGET_DIR" ]; then
            TARGET_DIR=$(find $GITHUB_WORKSPACE -maxdepth 2 -name "CMakeLists.txt" -exec dirname {} \; | head -n 1)
          fi
          echo "PROJECT_DIR=$TARGET_DIR" >> $GITHUB_ENV

      - name: Run AMDSMI Build and Install
        run: |
          set -euo pipefail
          SCRIPT=$(find $GITHUB_WORKSPACE -path "*/projects/amdsmi/tests/amdsmi_build/run_amdsmi_build.py" | head -1)
          if [ -z "$SCRIPT" ]; then
            echo "ERROR: run_amdsmi_build.py not found"
            exit 1
          fi
          echo "Using runner script: $SCRIPT"

          # The script autodetects package manager, format, and distro
          # quirks (Debian10 archived repos, RHEL10/AlmaLinux8 QA_RPATHS,
          # AzureLinux3 more_itertools, etc.) from /etc/os-release.
          python3 "$SCRIPT" \
            --project-dir "${{ env.PROJECT_DIR }}" \
            --os-label "${{ matrix.os }}" \
            --test-results-dir "/tmp/test-results-${{ matrix.os }}"

      # The previous step exits non-zero if any build/install/wheel-verify
      # stage failed; on failure we dump the per-stage result files for
      # quick triage in the GitHub Actions log. Full logs are still
      # uploaded via the test-results artifact below.
      - name: Dump build/install/wheel result files on failure
        if: failure()
        run: |
          set +e
          for f in build_result.txt install_result.txt verify_wheel_result.txt; do
            path="/tmp/test-results-${{ matrix.os }}/$f"
            if [ -f "$path" ]; then
              echo "===== $f ====="
              cat "$path"
            fi
          done

      - name: Upload Package Artifact
        if: github.event_name == 'pull_request'
        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
        with:
          name: amd-smi-lib-${{ matrix.package_format }}-${{ matrix.os }}-${{ env.PR_NUMBER }}-${{ env.BUILD_DATE }}
          path: |
            ${{ env.PROJECT_DIR }}/build/amd-smi-lib*99999-local${{ matrix.arch_suffix }}
            ${{ env.PROJECT_DIR }}/build/amd-smi-lib-*99999-local*.rpm
          if-no-files-found: warn
          retention-days: 7

      - name: AMDSMI Command Tests
        run: |
          set -euo pipefail
          mkdir -p /tmp/test-results-${{ matrix.os }}
          commands=(
            "amd-smi version"
            "amd-smi list"
            "amd-smi static"
            "amd-smi firmware"
            "amd-smi ucode"
            "amd-smi bad-pages"
            "amd-smi metric"
            "amd-smi process"
            "amd-smi topology"
            "amd-smi monitor"
            "amd-smi dmon"
            "amd-smi xgmi"
            "amd-smi partition"
          )
          # Command failures are not hard fails here; the AMDSMI CI
          # Summary step scans amd-smi_*.log for 'Error code:' / 'Traceback'
          # / 'AmdSmiException' and surfaces any failures in the summary.
          for cmd in "${commands[@]}"; do
            debug_cmd="$cmd --loglevel debug"
            echo "Running: $debug_cmd"
            log="/tmp/test-results-${{ matrix.os }}/$(echo $cmd | tr ' ' '_').log"
            rc=0
            eval "$debug_cmd" > "$log" 2>&1 || rc=$?
            if [ "$rc" -ne 0 ]; then
              echo "Command '$debug_cmd' failed (exit $rc)."
              cat "$log"
              echo "Error code: $rc" >> "$log"
            fi
          done

      - name: Upload AMDSMI Command Test Results
        if: always()
        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
        with:
          name: amdsmi-command-tests-${{ matrix.os }}
          path: /tmp/test-results-${{ matrix.os }}

      - name: Run AMDSMI, Python, and Example Tests
        run: |
          set -euo pipefail
          RESULTS_DIR=/tmp/test-results-${{ matrix.os }}

          cd /opt/rocm/share/amd_smi/tests
          source amdsmitst.exclude
          source detect_asic_filter.sh
          echo "GTEST_EXCLUDE=${GTEST_EXCLUDE}"
          ./amdsmitst --gtest_filter="-${GTEST_EXCLUDE}" > "$RESULTS_DIR/amdsmi_tests.log" 2>&1

          cd /opt/rocm/share/amd_smi/tests/python_unittest
          ./integration_test.py -v > "$RESULTS_DIR/integration_test_output.txt" 2>&1
          ./unit_tests.py -v > "$RESULTS_DIR/unit_test_output.txt" 2>&1
          if [ -f ./perf_tests.py ]; then
            ./perf_tests.py -v > "$RESULTS_DIR/perf_test_output.txt" 2>&1 || echo 'perf_tests.py failed'
          else
            echo "perf_tests.py not present; skipping" > "$RESULTS_DIR/perf_test_output.txt"
          fi

          cd ${{ env.PROJECT_DIR }}/example
          rm -rf build
          cmake -B build -DENABLE_ESMI_LIB=OFF
          make -C build -j $(nproc)
          cd build
          ./amd_smi_drm_ex > "$RESULTS_DIR/amd_smi_drm_ex.log" 2>&1 || echo 'amd_smi_drm_ex failed'
          ./amd_smi_nodrm_ex > "$RESULTS_DIR/amd_smi_nodrm_ex.log" 2>&1 || echo 'amd_smi_nodrm_ex failed'

      - name: AMDSMI Test Results
        if: always()
        run: cat /tmp/test-results-${{ matrix.os }}/amdsmi_tests.log || echo "No AMDSMI test results"

      - name: Integration Test Results
        if: always()
        run: cat /tmp/test-results-${{ matrix.os }}/integration_test_output.txt || echo "No integration test results"

      - name: Unit Test Results
        if: always()
        run: cat /tmp/test-results-${{ matrix.os }}/unit_test_output.txt || echo "No unit test results"

      - name: Perf Test Results
        if: always()
        run: cat /tmp/test-results-${{ matrix.os }}/perf_test_output.txt || echo "No perf test results"

      - name: Example DRM Test Results
        if: always()
        run: cat /tmp/test-results-${{ matrix.os }}/amd_smi_drm_ex.log || echo "No DRM example results"

      - name: Example NoDRM Test Results
        if: always()
        run: cat /tmp/test-results-${{ matrix.os }}/amd_smi_nodrm_ex.log || echo "No NoDRM example results"

      - name: AMDSMI CI Summary
        if: always()
        run: |
          set -euo pipefail
          SCRIPT=$(find $GITHUB_WORKSPACE -path "*/projects/amdsmi/tests/amdsmi_build/run_amdsmi_build.py" | head -1)
          python3 "$SCRIPT" summarize \
            --results-dir "/tmp/test-results-${{ matrix.os }}" \
            --os-label "${{ matrix.os }}" \
            --summary-file "$GITHUB_STEP_SUMMARY"

      - name: Uninstall
        if: always()
        run: |
          set -euo pipefail
          case "${{ matrix.package_format }}" in
            deb) apt remove -y amd-smi-lib || true ;;
            rpm)
              case "${{ matrix.os }}" in
                SLES) zypper remove -y amd-smi-lib || true ;;
                *) dnf remove -y amd-smi-lib || true ;;
              esac
              ;;
          esac
          rm -f /usr/local/bin/amd-smi
          if [ -d /opt/rocm/share/amd_smi ]; then
            rm -rf /opt/rocm/share/amd_smi
          fi