name: GPU Tests

on:
  push:
    branches: [ main ]
    paths:
      - 'crates/core/rtx-tensor/**'
      - 'crates/core/rtx-kernel/**'
      - 'crates/training/rtx-flash-attention/**'
      - 'crates/production/rtx-inference/**'
  pull_request:
    branches: [ main ]
    paths:
      - 'crates/core/rtx-tensor/**'
      - 'crates/core/rtx-kernel/**'
      - 'crates/training/rtx-flash-attention/**'
      - 'crates/production/rtx-inference/**'
  workflow_dispatch:

env:
  CARGO_TERM_COLOR: always
  RUST_BACKTRACE: 1

jobs:
  # Check if GPU runner is available
  check-gpu-availability:
    name: Check GPU Availability
    runs-on: ubuntu-latest
    outputs:
      has_gpu: ${{ steps.check.outputs.has_gpu }}
    steps:
    - name: Check for GPU runner
      id: check
      run: |
        # This would check if a self-hosted GPU runner is available
        echo "has_gpu=false" >> $GITEA_OUTPUT
        echo "GPU runner not configured yet - tests will be skipped"

  # CUDA tests (requires self-hosted runner with NVIDIA GPU)
  cuda-tests:
    name: CUDA Tests
    needs: check-gpu-availability
    if: needs.check-gpu-availability.outputs.has_gpu == 'true'
    runs-on: [self-hosted, gpu, cuda]
    strategy:
      fail-fast: false
      matrix:
        cuda_version: ['11.8', '12.1']
    steps:
    - uses: actions/checkout@v4

    - uses: dtolnay/rust-toolchain@stable

    - name: Check CUDA version
      run: nvidia-smi && nvcc --version

    - name: Run CUDA tests
      run: cargo test --features cuda
      env:
        CUDA_VERSION: ${{ matrix.cuda_version }}

  # Metal tests (requires macOS runner with Apple Silicon)
  metal-tests:
    name: Metal Tests
    needs: check-gpu-availability
    if: needs.check-gpu-availability.outputs.has_gpu == 'true'
    runs-on: [self-hosted, macos, metal]
    steps:
    - uses: actions/checkout@v4

    - uses: dtolnay/rust-toolchain@stable

    - name: Run Metal tests
      run: cargo test --features metal