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

name: TheRock CI Linux for rccl

on:
  workflow_call:
    inputs:
      amdgpu_families:
        type: string
      artifact_group:
        type: string
      event_name:
        type: string
        default: ''
      package_version:
        type: string
        default: ""
      cmake_options:
        type: string
      artifact_run_id:
        type: string
        default: ''
      test_scope:
        description: 'Test scope: smoke (22 curated tests) or all'
        type: string
        default: 'smoke'
      notify_email:
        type: string
        default: ''
      notify_webhook:
        type: string
        default: ''

permissions:
  contents: read

jobs:
  therock-build-linux:
    name: Build Linux Packages
    # Skipped when artifact_run_id is provided (reuse mode). Downstream jobs
    # use `!cancelled()` in their `if:` so they still run when this is skipped.
    if: ${{ inputs.artifact_run_id == '' }}
    runs-on: azure-linux-scale-rocm
    permissions:
      id-token: write
    container:
      image: ghcr.io/rocm/therock_build_manylinux_x86_64@sha256:48492540591673fdc8d51beb89bde41e7ba13cbb528f643c0a481ba42c4058f2 # 2026-04-16
    env:
      AMDGPU_FAMILIES: ${{ inputs.amdgpu_families }}
      TEATIME_FORCE_INTERACTIVE: 0
      CACHE_DIR: ${{ github.workspace }}/.container-cache
      # The ccache.conf will be written by setup_ccache.py before this gets used.
      CCACHE_CONFIGPATH: ${{ github.workspace }}/.ccache/ccache.conf
    steps:
      - name: "Checking out repository for rocm-systems"
        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0

      - name: Checkout TheRock repository
        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
        with:
          repository: "ROCm/TheRock"
          path: "TheRock"
          ref: release/therock-10.0

      - name: Install python deps
        run: |
          pip install -r TheRock/requirements.txt

      # safe.directory must be set before Runner Health Status
      - name: Adjust git config
        run: |
          git config --global --add safe.directory $PWD
          git config fetch.parallel 10

      - name: Setup ccache
        run: |
          ./TheRock/build_tools/setup_ccache.py \
            --config-preset "github-oss-dev" \
            --dir "$(dirname $CCACHE_CONFIGPATH)" \
            --local-path "$CACHE_DIR/ccache"

      - name: Runner health status
        run: |
          ./TheRock/build_tools/health_status.py

      - name: Fetch sources
        run: |
          ./TheRock/build_tools/fetch_sources.py --jobs 12 --no-include-rocm-systems

      - name: Patch rocm-systems
        run: |
          # Remove patches here if they cannot be applied cleanly, and they have not been deleted from TheRock repo
          # rm ./TheRock/patches/amd-mainline/rocm-systems/0003-Use-is_versioned-true-consistently-in-both-Comgr-Loa.patch

      - name: Configure Projects
        env:
          amdgpu_families: ${{ env.AMDGPU_FAMILIES }}
          package_version: ${{ inputs.package_version }}
          extra_cmake_options: "-DTHEROCK_ROCM_SYSTEMS_SOURCE_DIR=../ ${{ inputs.cmake_options }}"
          BUILD_DIR: build
        run: |
          python3 ./TheRock/build_tools/github_actions/build_configure.py

      - name: Build therock-dist
        run: cmake --build TheRock/build

      - name: Build therock-artifacts
        run: cmake --build TheRock/build --target therock-artifacts

      - name: Report
        #if: ${{ !cancelled() }}
        run: |
          echo "Full SDK du:"
          echo "------------"
          du -h -d 1 TheRock/build/dist/rocm
          echo "Artifacts:"
          echo "----------"
          du -h -d 1 TheRock/build/artifacts
          echo "CCache Stats:"
          echo "-------------"
          ccache -s -v
          tail -v -n +1 .ccache/compiler_check_cache/* > TheRock/build/logs/ccache_compiler_check_cache.log

      - name: Configure AWS Credentials for non-forked repos
        if: ${{ always() && !github.event.pull_request.head.repo.fork }}
        uses: aws-actions/configure-aws-credentials@7474bc4690e29a8392af63c5b98e7449536d5c3a # v4.3.1
        with:
          aws-region: us-east-2
          role-to-assume: arn:aws:iam::692859939525:role/therock-artifacts-external

      - name: Push artifacts
        run: |
          python3 TheRock/build_tools/artifact_manager.py push \
            --run-id ${{ github.run_id }} \
            --build-dir TheRock/build \
            --stage all

      - name: Post Build Upload
        if: always()
        run: |
          python3 TheRock/build_tools/github_actions/post_build_upload.py \
            --run-id ${{ github.run_id }} \
            --artifact-group ${{ env.AMDGPU_FAMILIES }} \
            --build-dir TheRock/build \
            --upload

  therock-test-linux-multi-node:
    name: "Test multi-node"
    # TODO: Re-enable gfx94X-dcgpu and gfx950-dcgpu multi-node tests once the clusters are stable again.
    if: false
    permissions:
      contents: read
      id-token: write
    needs: [therock-build-linux]
    uses: ./.github/workflows/therock-rccl-test-packages-multi-node.yml
    with:
      amdgpu_families: ${{ inputs.amdgpu_families }}
      artifact_group: ${{ inputs.artifact_group }}
      test_runs_on: ruby-linux-slurm-scale-runner
      artifact_run_id: ${{ inputs.artifact_run_id || github.run_id }}

  therock-test-linux-single-node:
    name: "Test single-node"
    # TODO(TheRock#3288): Re-enable gfx950-dcgpu runners once we get more capacity
    if: ${{ !cancelled() && inputs.amdgpu_families != 'gfx950-dcgpu' }}
    needs: [therock-build-linux]
    uses: ./.github/workflows/therock-rccl-test-packages-single-node.yml
    with:
      amdgpu_families: ${{ inputs.amdgpu_families }}
      artifact_group: ${{ inputs.artifact_group }}
      test_runs_on: linux-gfx942-8gpu-ossci-rocm
      artifact_run_id: ${{ inputs.artifact_run_id || github.run_id }}

  therock-test-pytorch-distributed:
    name: "Test PyTorch distributed (scheduled)"
    if: ${{ !cancelled() && inputs.amdgpu_families != 'gfx950-dcgpu' && (inputs.event_name == 'schedule' || inputs.event_name == 'workflow_dispatch') }}
    needs: [therock-build-linux]
    uses: ./.github/workflows/therock-rccl-test-pytorch-distributed.yml
    with:
      amdgpu_families: ${{ inputs.amdgpu_families }}
      artifact_group: ${{ inputs.artifact_group }}
      test_runs_on: linux-gfx942-8gpu-ossci-rocm
      artifact_run_id: ${{ inputs.artifact_run_id || github.run_id }}
      test_scope: ${{ inputs.test_scope }}
      notify_email: ${{ inputs.notify_email }}
      notify_webhook: ${{ inputs.notify_webhook }}

  therock-test-jax-collective:
    name: "Test JAX collective (scheduled)"
    if: ${{ !cancelled() && inputs.amdgpu_families != 'gfx950-dcgpu' && (inputs.event_name == 'schedule' || inputs.event_name == 'workflow_dispatch') }}
    needs: [therock-build-linux]
    uses: ./.github/workflows/therock-rccl-test-jax-collective.yml
    with:
      amdgpu_families: ${{ inputs.amdgpu_families }}
      artifact_group: ${{ inputs.artifact_group }}
      test_runs_on: linux-gfx942-8gpu-ossci-rocm
      artifact_run_id: ${{ inputs.artifact_run_id || github.run_id }}
      notify_email: ${{ inputs.notify_email }}
      notify_webhook: ${{ inputs.notify_webhook }}

  therock-test-rocprof:
    name: "Test rocprofv3 smoke"
    if: ${{ !cancelled() && inputs.amdgpu_families != 'gfx950-dcgpu' }}
    needs: [therock-build-linux]
    uses: ./.github/workflows/therock-rccl-test-rocprof.yml
    with:
      amdgpu_families: ${{ inputs.amdgpu_families }}
      artifact_group: ${{ inputs.artifact_group }}
      test_runs_on: linux-gfx942-8gpu-ossci-rocm
      artifact_run_id: ${{ inputs.artifact_run_id || github.run_id }}
      notify_email: ${{ inputs.notify_email }}
      notify_webhook: ${{ inputs.notify_webhook }}