TheRock/rocm-libraries/projects/hipthreads/.github/workflows/ci.yml

name: CI - Build and Test

on:
  # Run on push to main and develop branches
  push:
    branches: [ "main", "develop" ]
  
  # Run on pull requests to main and develop
  pull_request:
    branches: [ "main", "develop" ]
  
  # Run nightly at 5 AM UTC
  schedule:
    - cron: '0 5 * * *'
  
  # Allow manual triggering
  workflow_dispatch:

jobs:
  build-and-test:
    name: Build and Test
    if: github.repository == 'ArchSWTech/gpulib'
    runs-on: self-hosted
    env:
      HIP_VISIBLE_DEVICES: 0
    
    steps:
    - name: Checkout repository
      uses: actions/checkout@v4
      
    - name: Clean previous artifacts
      run: |
        echo "Cleaning up previous build artifacts..."
        rm -rf .venv libhipcxx build-debug
        # Clear pip cache to prevent accumulation
        rm -rf ~/.cache/pip
        echo "Cleanup complete"
      
    - name: Install system dependencies
      run: |
        # These deps are normally already present on the self-hosted runner.
        # Only install what's actually missing, and never let an unrelated
        # broken package sink the job: `apt-get install` implicitly runs
        # `dpkg --configure -a`, which retries any half-configured packages
        # (e.g. amdgpu-dkms failing to build against a freshly-pulled HWE
        # kernel) — a machine-level issue independent of hipThreads.
        missing=""
        for pkg in g++ cmake git python3-venv; do
          dpkg -s "$pkg" >/dev/null 2>&1 || missing="$missing $pkg"
        done
        if [ -n "$missing" ]; then
          sudo apt-get update
          sudo apt-get install -y $missing \
            || echo "::warning::apt-get reported errors (likely pre-existing broken packages on the runner); continuing"
        else
          echo "All system dependencies already installed; skipping apt-get."
        fi
        
    - name: Setup Python venv
      run: |
        python3 -m venv .venv
        source .venv/bin/activate
        pip install --upgrade pip
        
    - name: Install latest nightly ROCm
      run: |
        source .venv/bin/activate
        pip install \
          --index-url https://rocm.nightlies.amd.com/whl-multi-arch/ \
          "rocm[libraries,devel]"
        rocm-sdk init
        
    - name: Setup ROCm structure for hipcc
      run: |
        source .venv/bin/activate
        export ROCM_PATH=$VIRTUAL_ENV/lib/python3.12/site-packages/_rocm_sdk_devel
        
        # Create symlink to match standard ROCm structure
        cd $ROCM_PATH
        ln -s lib/llvm/amdgcn amdgcn
        
        # Verify symlink
        echo "Verifying symlink:"
        ls -la $ROCM_PATH/amdgcn
        
    - name: Verify ROCm Installation
      run: |
        source .venv/bin/activate
        echo "ROCm version:"
        hipcc --version
        echo ""
        echo "ROCm SDK info:"
        rocm-sdk version
        echo ""
        echo "Supported targets:"
        rocm-sdk targets
        echo ""
        echo "Checking for AMD GPUs:"
        rocminfo || echo "rocminfo failed"
        echo ""
        echo "Checking GPU devices:"
        ls -la /dev/kfd /dev/dri/render* || echo "GPU device nodes not found"
        echo ""
        echo "Checking GPU permissions:"
        groups
        
    - name: Install libhipcxx
      run: |
        source .venv/bin/activate
        export ROCM_PATH=$VIRTUAL_ENV/lib/python3.12/site-packages/_rocm_sdk_devel
        
        # Clone libhipcxx
        git clone https://github.com/ROCm/libhipcxx.git
        cd libhipcxx
        
        # Build and install to venv location
        mkdir build && cd build
        HIP_PLATFORM=amd cmake \
              -DCMAKE_HIP_COMPILER=$ROCM_PATH/lib/llvm/bin/clang++ \
              -DCMAKE_HIP_ARCHITECTURES=gfx1200 \
              -DCMAKE_INSTALL_PREFIX=$VIRTUAL_ENV \
              -DCMAKE_PREFIX_PATH="$VIRTUAL_ENV;$ROCM_PATH" \
              -DLIBCUDACXX_ENABLE_LIBCUDACXX_TESTS=OFF ..
        make -j$(nproc)
        make install
        
        # Verify installation
        echo "libhipcxx installed to $VIRTUAL_ENV"
        ls -la $VIRTUAL_ENV/include/hip/std/ || echo "Warning: libhipcxx headers not found"
        
    - name: Configure CMake (Debug with tests)
      run: |
        source .venv/bin/activate
        export ROCM_PATH=$VIRTUAL_ENV/lib/python3.12/site-packages/_rocm_sdk_devel
        mkdir -p build-debug
        cd build-debug
        HIP_PLATFORM=amd cmake .. \
          -DCMAKE_CXX_COMPILER=$VIRTUAL_ENV/bin/hipcc \
          -DCMAKE_BUILD_TYPE=Debug \
          -DDISABLE_WERROR=ON \
          -DCMAKE_CXX_FLAGS="-isystem $VIRTUAL_ENV/include" \
          -DCMAKE_HIP_FLAGS="-isystem $VIRTUAL_ENV/include" \
          -DCMAKE_PREFIX_PATH="$VIRTUAL_ENV;$ROCM_PATH"
        
    - name: Build and install library and tests
      run: |
        source .venv/bin/activate
        cd build-debug
        cmake --build . -j$(nproc)
        cmake --install . --prefix $VIRTUAL_ENV
        
    - name: Verify build outputs
      run: |
        echo "=== Library ==="
        ls -lh build-debug/lib/libhipthreads.a
        echo ""
        echo "=== Test Executables ==="
        ls -lh build-debug/bin/hip_thread_*_test
        
    - name: Run mutex test
      run: |
        source .venv/bin/activate
        export ROCM_PATH=$VIRTUAL_ENV/lib/python3.12/site-packages/_rocm_sdk_devel
        export LD_LIBRARY_PATH=$ROCM_PATH/lib:$LD_LIBRARY_PATH
        cd build-debug
        echo "Running hip_thread_mutex_test..."
        ./bin/hip_thread_mutex_test && echo "✓ hip_thread_mutex_test PASSED"
        
    - name: Run condvar test
      run: |
        source .venv/bin/activate
        export ROCM_PATH=$VIRTUAL_ENV/lib/python3.12/site-packages/_rocm_sdk_devel
        export LD_LIBRARY_PATH=$ROCM_PATH/lib:$LD_LIBRARY_PATH
        cd build-debug
        echo "Running hip_thread_condvar_test..."
        ./bin/hip_thread_condvar_test && echo "✓ hip_thread_condvar_test PASSED"
        
    - name: Run lit unit tests
      id: lit-tests
      continue-on-error: true
      run: |
        source .venv/bin/activate
        pip install lit
        export ROCM_PATH=$VIRTUAL_ENV/lib/python3.12/site-packages/_rocm_sdk_devel
        export LD_LIBRARY_PATH=$ROCM_PATH/lib:$LD_LIBRARY_PATH
        export HIPTHREADS_SOURCE_DIR=$GITHUB_WORKSPACE
        export HIPTHREADS_BUILD_DIR=$GITHUB_WORKSPACE/build-debug
        
        echo "Running lit unit tests..."
        cd $GITHUB_WORKSPACE
        lit -a -v --echo-all-commands -j 1 test/
        
    - name: "Benchmark: SAXPY (step3-simdize)"
      timeout-minutes: 4
      run: |
        source .venv/bin/activate
        export ROCM_PATH=$VIRTUAL_ENV/lib/python3.12/site-packages/_rocm_sdk_devel
        export LD_LIBRARY_PATH=$ROCM_PATH/lib:$LD_LIBRARY_PATH
        cd examples/saxpy/step3-simdize
        HIP_PLATFORM=amd cmake -B build \
          -DCMAKE_HIP_COMPILER=$ROCM_PATH/lib/llvm/bin/clang++ \
          -DCMAKE_HIP_ARCHITECTURES=gfx1201 \
          -DCMAKE_PREFIX_PATH="$VIRTUAL_ENV;$ROCM_PATH" \
          -DCMAKE_CXX_FLAGS="-isystem $VIRTUAL_ENV/include" \
          -DCMAKE_HIP_FLAGS="-isystem $VIRTUAL_ENV/include"
        cmake --build build -j$(nproc)
        ./build/bin/saxpy 2>&1 | tee /tmp/saxpy_output
        NS=$(grep -oP '= \K[0-9,]+(?=ns)' /tmp/saxpy_output | tr -d ',')
        THRESHOLD_NS=48742980  # 42385200ns (42.3852ms) * 1.15
        echo "SAXPY: ${NS}ns (threshold: ${THRESHOLD_NS}ns)"
        if [ "$NS" -gt "$THRESHOLD_NS" ]; then
          echo "::error::SAXPY perf regression: ${NS}ns exceeds 115% of 42385200ns baseline"
          exit 1
        fi
        
    - name: "Benchmark: IoWRT (step4-simdize)"
      timeout-minutes: 4
      run: |
        source .venv/bin/activate
        export ROCM_PATH=$VIRTUAL_ENV/lib/python3.12/site-packages/_rocm_sdk_devel
        export LD_LIBRARY_PATH=$ROCM_PATH/lib:$LD_LIBRARY_PATH
        cd examples/InOneWeekendRaytracer/step4-simdize
        HIP_PLATFORM=amd cmake -B build \
          -DCMAKE_HIP_COMPILER=$ROCM_PATH/lib/llvm/bin/clang++ \
          -DCMAKE_HIP_ARCHITECTURES=gfx1201 \
          -DCMAKE_PREFIX_PATH="$VIRTUAL_ENV;$ROCM_PATH" \
          -DCMAKE_CXX_FLAGS="-isystem $VIRTUAL_ENV/include" \
          -DCMAKE_HIP_FLAGS="-isystem $VIRTUAL_ENV/include"
        cmake --build build -j$(nproc)
        set +e
        TIMEFORMAT='%R'
        { time ./build/bin/inOneWeekend > image.ppm; } 2> /tmp/iowrt_time
        EXIT_CODE=$?
        set -e
        if [ $EXIT_CODE -ne 0 ]; then
          echo "::error::IoWRT binary crashed"
          exit 1
        fi
        ELAPSED=$(cat /tmp/iowrt_time)
        THRESHOLD=0.6522  # 0.5672s * 1.15
        echo "IoWRT: ${ELAPSED}s (threshold: ${THRESHOLD}s)"
        if [ "$(echo "$ELAPSED > $THRESHOLD" | bc)" -eq 1 ]; then
          echo "::error::IoWRT perf regression: ${ELAPSED}s exceeds 115% of 0.5672s baseline"
          exit 1
        fi
        
    - name: "Benchmark: SpMM (step3-hipthread-port)"
      timeout-minutes: 4
      run: |
        source .venv/bin/activate
        export ROCM_PATH=$VIRTUAL_ENV/lib/python3.12/site-packages/_rocm_sdk_devel
        export LD_LIBRARY_PATH=$ROCM_PATH/lib:$LD_LIBRARY_PATH
        cd examples/sparse-mat-mul/step3-hipthread-port
        HIP_PLATFORM=amd cmake -B build \
          -DCMAKE_HIP_COMPILER=$ROCM_PATH/lib/llvm/bin/clang++ \
          -DCMAKE_HIP_ARCHITECTURES=gfx1201 \
          -DCMAKE_PREFIX_PATH="$VIRTUAL_ENV;$ROCM_PATH" \
          -DCMAKE_CXX_FLAGS="-isystem $VIRTUAL_ENV/include" \
          -DCMAKE_HIP_FLAGS="-isystem $VIRTUAL_ENV/include"
        cmake --build build -j$(nproc)
        ./build/bin/spmm "/usr/local/share/hipthreads/pwtk.mtx" 2>&1 | tee /tmp/spmm_output
        NS=$(grep -oP '= \K[0-9,]+(?=ns)' /tmp/spmm_output | tr -d ',')
        THRESHOLD_NS=53384495000  # 46421300000ns (46.4213s) * 1.15
        echo "SpMM: ${NS}ns (threshold: ${THRESHOLD_NS}ns)"
        if [ "$NS" -gt "$THRESHOLD_NS" ]; then
          echo "::error::SpMM perf regression: ${NS}ns exceeds 115% of 46421300000ns baseline"
          exit 1
        fi
        
    - name: "Benchmark: llama3.c (step4-simdize)"
      timeout-minutes: 4
      run: |
        source .venv/bin/activate
        export ROCM_PATH=$VIRTUAL_ENV/lib/python3.12/site-packages/_rocm_sdk_devel
        export LD_LIBRARY_PATH=$ROCM_PATH/lib:$LD_LIBRARY_PATH
        cd examples/llama3.c/step4-simdize
        HIP_PLATFORM=amd cmake -B build \
          -DCMAKE_HIP_COMPILER=$ROCM_PATH/lib/llvm/bin/clang++ \
          -DCMAKE_HIP_ARCHITECTURES=gfx1201 \
          -DCMAKE_PREFIX_PATH="$VIRTUAL_ENV;$ROCM_PATH" \
          -DCMAKE_CXX_FLAGS="-isystem $VIRTUAL_ENV/include" \
          -DCMAKE_HIP_FLAGS="-isystem $VIRTUAL_ENV/include"
        cmake --build build -j$(nproc)
        ./build/bin/llama3 \
          /usr/local/share/hipthreads/llama3.2_3b_instruct.bin \
          -z /usr/local/share/hipthreads/tokenizer.bin \
          -s 42 -i "My car" -n 100 \
          2>&1 | tee /tmp/llama3_output
        TOKS=$(grep -oP 'achieved tok/s: \K[0-9.]+' /tmp/llama3_output)
        THRESHOLD=6.7323  # 7.9204 tok/s * 0.85
        echo "llama3.c: ${TOKS} tok/s (threshold: ${THRESHOLD} tok/s)"
        if [ "$(echo "$TOKS < $THRESHOLD" | bc)" -eq 1 ]; then
          echo "::error::llama3.c perf regression: ${TOKS} tok/s below 85% of 7.9204 tok/s baseline"
          exit 1
        fi
        
    - name: Fail if lit tests failed
      # always() ensures this check runs even if a prior step (e.g. a benchmark)
      # failed, so lit test failures are never silently hidden. Needed because
      # the lit-tests step uses continue-on-error: true.
      if: always() && steps.lit-tests.outcome == 'failure'
      run: |
        echo "::error::Lit unit tests failed (see 'Run lit unit tests' step for details)"
        exit 1
        
    - name: Upload build artifacts on failure
      if: failure()
      uses: actions/upload-artifact@v4
      with:
        name: build-logs-${{ github.sha }}
        path: |
          build-debug/CMakeFiles/*.log
          build-debug/**/*.ii
          build-debug/**/*.s
        retention-days: 7
        
    - name: Cleanup temporary files
      if: always()
      run: |
        echo "Cleaning up temporary files..."
        rm -rf libhipcxx
        rm -rf examples/*/step*/build examples/*/*/step*/build
        echo "Cleanup complete"
  
  notify:
    name: Send Email Notification
    runs-on: gpu-thread-code-scan
    needs: build-and-test
    # always() overrides the default behavior where a job with 'needs' is skipped
    # if the dependency failed. This ensures we send the email regardless of
    # whether build-and-test succeeded or failed (especially useful for failures).
    if: ${{ github.event_name == 'schedule' && always() }}
    
    steps:
    - name: Send success email
      if: needs.build-and-test.result == 'success'
      continue-on-error: true
      env:
        SMTP_SERVER: ${{ secrets.SMTP_SERVER_NAME }}
        MAIL_FROM: ${{ secrets.MAIL_USERNAME }}
        REPO: ${{ github.repository }}
        BRANCH: ${{ github.ref_name }}
        COMMIT: ${{ github.sha }}
        COMMIT_MSG: ${{ github.event.head_commit.message || github.event.pull_request.title || '' }}
        AUTHOR: ${{ github.actor }}
        RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
      run: |
        set -euo pipefail
        
        cat <<EOF | sed 's/$/\r/' > email.txt
        From: ${MAIL_FROM}
        To: asavic@amd.com,charboyd@amd.com,dmcintos@amd.com,kelvilui@amd.com,marsavic@amd.com
        Subject: =?UTF-8?B?4pyFIENJIEJ1aWxkIFN1Y2Nlc3NmdWw=?= - hipThreads [${BRANCH}]
        MIME-Version: 1.0
        Content-Type: text/plain; charset=UTF-8

        The CI workflow completed successfully.

        Repository: ${REPO}
        Branch: ${BRANCH}
        Commit: ${COMMIT}
        Commit Message: ${COMMIT_MSG}
        Author: ${AUTHOR}

        Workflow Run: ${RUN_URL}
        EOF
        
        cat email.txt
        
        curl --connect-timeout 10 --max-time 30 \
          --url "smtp://${SMTP_SERVER}:25" \
          --mail-from "${MAIL_FROM}" \
          --mail-rcpt "asavic@amd.com" \
          --mail-rcpt "charboyd@amd.com" \
          --mail-rcpt "dmcintos@amd.com" \
          --mail-rcpt "kelvilui@amd.com" \
          --mail-rcpt "marsavic@amd.com" \
          --upload-file email.txt
        
    - name: Send failure email
      if: needs.build-and-test.result == 'failure'
      continue-on-error: true
      env:
        SMTP_SERVER: ${{ secrets.SMTP_SERVER_NAME }}
        MAIL_FROM: ${{ secrets.MAIL_USERNAME }}
        REPO: ${{ github.repository }}
        BRANCH: ${{ github.ref_name }}
        COMMIT: ${{ github.sha }}
        COMMIT_MSG: ${{ github.event.head_commit.message || github.event.pull_request.title || '' }}
        AUTHOR: ${{ github.actor }}
        RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
      run: |
        set -euo pipefail
        
        cat <<EOF | sed 's/$/\r/' > email.txt
        From: ${MAIL_FROM}
        To: asavic@amd.com,charboyd@amd.com,dmcintos@amd.com,kelvilui@amd.com,marsavic@amd.com
        Subject: =?UTF-8?B?4p2MIENJIEJ1aWxkIEZhaWxlZA==?= - hipThreads [${BRANCH}]
        MIME-Version: 1.0
        Content-Type: text/plain; charset=UTF-8

        The CI workflow has failed.

        Repository: ${REPO}
        Branch: ${BRANCH}
        Commit: ${COMMIT}
        Commit Message: ${COMMIT_MSG}
        Author: ${AUTHOR}

        Workflow Run: ${RUN_URL}
        EOF
        
        cat email.txt
        
        curl --connect-timeout 10 --max-time 30 \
          --url "smtp://${SMTP_SERVER}:25" \
          --mail-from "${MAIL_FROM}" \
          --mail-rcpt "asavic@amd.com" \
          --mail-rcpt "charboyd@amd.com" \
          --mail-rcpt "dmcintos@amd.com" \
          --mail-rcpt "kelvilui@amd.com" \
          --mail-rcpt "marsavic@amd.com" \
          --upload-file email.txt