gpu-test.yml

name: GPU Kernel Tests
on:
  push:
    branches: [main, master]
  pull_request:

jobs:
  metal-test:
    runs-on: macos-latest
    steps:
      - uses: actions/checkout@v4
        with:
          submodules: recursive
      - name: Compile Metal shaders
        run: |
          cd kernels
          PASS=0; FAIL=0
          for f in *.metal; do
            if xcrun -sdk macosx metal -std=metal3.0 -Werror -c "$f" -o /dev/null 2>/dev/null; then
              PASS=$((PASS+1))
            else
              echo "FAIL: $f"
              FAIL=$((FAIL+1))
            fi
          done
          echo "$PASS Metal shaders pass, $FAIL fail"
          [ $FAIL -eq 0 ] || exit 1

  wgsl-validate:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: dtolnay/rust-toolchain@stable
      - name: Install naga
        run: cargo install naga-cli
      - name: Validate WGSL shaders
        run: |
          cd kernels
          PASS=0; FAIL=0
          for f in *.wgsl; do
            if naga "$f" 2>/dev/null; then
              PASS=$((PASS+1))
            else
              echo "FAIL: $f"
              FAIL=$((FAIL+1))
            fi
          done
          echo "$PASS WGSL shaders pass, $FAIL fail"
          [ $FAIL -eq 0 ] || exit 1

  cuda-compile:
    runs-on: ubuntu-latest
    container: nvidia/cuda:12.4.0-devel-ubuntu22.04
    steps:
      - uses: actions/checkout@v4
      - name: Compile CUDA kernels
        run: |
          cd kernels
          PASS=0; FAIL=0
          for f in *.cu; do
            if nvcc -std=c++17 --compile "$f" -o /dev/null 2>&1; then
              PASS=$((PASS+1))
            else
              echo "FAIL: $f"
              FAIL=$((FAIL+1))
            fi
          done
          echo "$PASS CUDA kernels compile, $FAIL fail"
          [ $FAIL -eq 0 ] || exit 1

  cuda-benchmark:
    runs-on: [self-hosted, gpu, cuda]
    if: github.event_name == 'push' && github.ref == 'refs/heads/main'
    needs: cuda-compile
    steps:
      - uses: actions/checkout@v4
        with:
          submodules: recursive
      - name: Build GPU library
        run: |
          mkdir build && cd build
          cmake .. -DCMAKE_BUILD_TYPE=Release -DLUX_GPU_CUDA_BACKEND=ON
          cmake --build . --parallel
      - name: Run CUDA benchmarks
        run: |
          cd build
          ./test/test_backends 2>&1
          echo "=== Benchmark complete ==="

  cuda-cpu-mode:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Compile CUDA kernels as C++ (CPU mode)
        run: |
          g++ -std=c++17 -O2 -I kernels -o test_cuda_cpu test/test_cuda_cpu.cpp
      - name: Run CUDA CPU-mode tests
        run: ./test_cuda_cpu