name: TheRock Sanity Check

on:
  workflow_dispatch:
    inputs:
      artifact_group:
        type: string
      artifact_run_id:
        type: string
        default: ""
      amdgpu_families:
        type: string
        default: ""
      amdgpu_targets:
        type: string
        default: ""
      test_runs_on:
        type: string
      platform:
        type: string
  workflow_call:
    inputs:
      artifact_group:
        type: string
      artifact_run_id:
        type: string
        default: ""
      amdgpu_families:
        type: string
        default: ""
      amdgpu_targets:
        type: string
        default: ""
      test_runs_on:
        type: string
      platform:
        type: string
  push:
    branches:
      - ADHOCBUILD

permissions:
  contents: read

jobs:
  test_sanity_check:
    name: "Sanity ROCM Test (${{ inputs.amdgpu_families }})"
    runs-on: ${{ inputs.test_runs_on }}
    # Running docker with cap-add and -v /lib/modiles, by recommendation of Github: https://rocm.docs.amd.com/projects/amdsmi/en/amd-staging/how-to/setup-docker-container.html
    container:
      image: ${{ inputs.platform == 'linux' && 'ghcr.io/rocm/no_rocm_image_ubuntu24_04@sha256:4150afe4759d14822f0e3f8930e1124f26e11f68b5c7b91ec9a02b20b1ebbb98' || null }}
      # --ulimit memlock=-1:-1 - Prevents memory allocation issues with ROCm inside container
      # --security-opt seccomp=unconfined - enables memory mapping, and is recommended for containers running in HPC environments
      # --env-file /etc/podinfo/gha-gpu-isolation-settings - Required for GPU isolation on OSSCI MIXXX runners
      # --user 0:0 - Running as root, by recommendation of GitHub: https://docs.github.com/en/actions/reference/workflows-and-actions/dockerfile-support#user
      options: --ipc host
        --group-add video
        --device /dev/kfd
        --device /dev/dri
        --group-add 992
        --group-add 110
        --cap-add SYS_MODULE
        -v /lib/modules:/lib/modules
        --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 != '' && inputs.artifact_run_id || github.run_id }}"
      OUTPUT_ARTIFACTS_DIR: ${{ github.workspace }}/build
      THEROCK_BIN_DIR: ${{ github.workspace }}/build/bin
      AMDGPU_FAMILIES: ${{ inputs.amdgpu_families }}
      AMDGPU_TARGETS: ${{ inputs.amdgpu_targets }}
      ARTIFACT_GROUP: ${{ inputs.artifact_group }}
    steps:
      - name: "Fetch 'build_tools' from repository"
        if: ${{ runner.os == 'Windows' }}
        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
        with:
          sparse-checkout: build_tools
          path: prejob

      - name: Pre-job cleanup processes on Windows
        if: ${{ runner.os == 'Windows' }}
        timeout-minutes: 5
        shell: powershell
        run: . '${{ github.workspace }}\prejob\build_tools\github_actions\cleanup_processes.ps1'

      - name: Checkout Repository
        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
        with:
          repository: "ROCm/TheRock"
          ref: 'compiler/amd-staging'

      - name: Pre-job cleanup Docker containers on Linux
        if: ${{ runner.os == 'Linux' }}
        timeout-minutes: 5
        shell: bash
        run: |
          # Remove any stopped containers
          docker container prune -f || true
          # Remove dangling networks
          docker network prune -f || true

      - name: Run setup test environment workflow
        timeout-minutes: 15
        uses: './.github/actions/setup_test_environment'
        with:
          ARTIFACT_GROUP: ${{ inputs.artifact_group }}
          AMDGPU_TARGETS: ${{ inputs.amdgpu_targets }}
          ARTIFACT_RUN_ID: ${{ env.ARTIFACT_RUN_ID }}
          OUTPUT_ARTIFACTS_DIR: ${{ env.OUTPUT_ARTIFACTS_DIR }}
          VENV_DIR: ${{ env.VENV_DIR }}
          FETCH_ARTIFACT_ARGS: "--base-only"
          IS_PR_FROM_FORK: ${{ github.event.pull_request.head.repo.fork }}

      # The sanity checks run tools like 'offload-arch' which may search for
      # DLLs on multiple search paths (PATH, CWD, system32, etc.).
      # For typical "installs" of ROCm, the rocm/bin/ dir can be expected to be
      # added to PATH, so we do that here. If we don't do this, DLLs on test
      # runners in system32 may be picked up instead and the tests may not be
      # representative, see https://github.com/ROCm/TheRock/issues/2019 and
      # https://github.com/ROCm/TheRock/pull/3230#issuecomment-3844854922.
      - name: Set PATH and HIP_CLANG_PATH for windows
        if: ${{ runner.os == 'Windows' }}
        run: |
          echo "HIP_CLANG_PATH=${OUTPUT_ARTIFACTS_DIR}\lib\llvm\bin" >> $GITHUB_ENV
          echo "${OUTPUT_ARTIFACTS_DIR}\bin" >> $GITHUB_PATH

      - name: Driver / GPU sanity check
        timeout-minutes: 3
        run: |
          python ./build_tools/print_driver_gpu_info.py

      - name: Run ROCm Sanity Tests
        timeout-minutes: 5
        env:
          # Enable verbose logging, see
          # https://rocm.docs.amd.com/projects/HIP/en/latest/how-to/debugging.html
          AMD_LOG_LEVEL: 4
        run: |
          pytest tests/ --log-cli-level=info --timeout=300

      - name: Post-job cleanup processes on Windows
        if: ${{ always() && runner.os == 'Windows' }}
        timeout-minutes: 5
        shell: powershell
        run: . '${{ github.workspace }}\build_tools\github_actions\cleanup_processes.ps1'