|
| 1 | +name: Iris Performance Regression Test |
| 2 | + |
| 3 | +on: |
| 4 | + push: |
| 5 | + branches: [ main ] |
| 6 | + pull_request: |
| 7 | + branches: [ main ] |
| 8 | + workflow_dispatch: |
| 9 | + |
| 10 | +concurrency: |
| 11 | + group: ${{ github.workflow }}-${{ github.head_ref || github.ref }} |
| 12 | + cancel-in-progress: ${{ github.ref != 'refs/heads/main' }} |
| 13 | + |
| 14 | +jobs: |
| 15 | + build-apptainer-image: |
| 16 | + runs-on: [self-hosted, mi3008x] |
| 17 | + timeout-minutes: 20 |
| 18 | + |
| 19 | + steps: |
| 20 | + - name: Checkout repository |
| 21 | + uses: actions/checkout@v4 |
| 22 | + |
| 23 | + - name: Setup Apptainer |
| 24 | + run: | |
| 25 | + apt-get update && apt-get install -y software-properties-common |
| 26 | + add-apt-repository -y ppa:apptainer/ppa |
| 27 | + apt-get update && apt-get install -y apptainer |
| 28 | +
|
| 29 | + - name: Build Iris Apptainer container |
| 30 | + run: | |
| 31 | + # Create persistent Apptainer directory |
| 32 | + mkdir -p ~/apptainer |
| 33 | +
|
| 34 | + # Build Apptainer image from definition file (only if it doesn't exist) |
| 35 | + if [ ! -f ~/apptainer/iris-dev.sif ]; then |
| 36 | + echo "Building new Apptainer image..." |
| 37 | + apptainer build ~/apptainer/iris-dev.sif apptainer/iris.def |
| 38 | + else |
| 39 | + echo "Using existing Apptainer image" |
| 40 | + fi |
| 41 | +
|
| 42 | + performance-test: |
| 43 | + name: ${{ matrix.example_name }} |
| 44 | + needs: build-apptainer-image |
| 45 | + runs-on: [self-hosted, mi3008x] |
| 46 | + timeout-minutes: 30 |
| 47 | + strategy: |
| 48 | + fail-fast: false |
| 49 | + matrix: |
| 50 | + # Performance baselines measured on AMD Instinct MI325X (8 GPUs) |
| 51 | + include: |
| 52 | + - example_name: "GEMM All-Scatter WG Specialization" |
| 53 | + example_path: "10_gemm_all_scatter_wg_specialization" |
| 54 | + tflops_threshold: 1600 # Actual: ~2182 TFLOPs |
| 55 | + benchmark_args: "-m 16384 -n 16384 -k 16384 --BLK_M 128 --BLK_N 128 --BLK_K 64 --gsize_m 6 --gemm_sms 256" |
| 56 | + |
| 57 | + - example_name: "GEMM All-Scatter" |
| 58 | + example_path: "07_gemm_all_scatter" |
| 59 | + tflops_threshold: 1000 # Actual: ~1407 TFLOPs |
| 60 | + benchmark_args: "-m 16384 -n 16384 -k 16384 --BLK_M 256 --BLK_N 64 --BLK_K 64 --gsize_m 6 --gemm_sms 256" |
| 61 | + |
| 62 | + - example_name: "GEMM All-Scatter Producer-Consumer" |
| 63 | + example_path: "11_gemm_all_scatter_producer_consumer" |
| 64 | + tflops_threshold: 1600 # Actual: ~2190 TFLOPs |
| 65 | + benchmark_args: "-m 16384 -n 16384 -k 16384 --BLK_M 128 --BLK_N 128 --BLK_K 64 --gsize_m 6 --gemm_sms 256 --comm_sms 48" |
| 66 | + |
| 67 | + - example_name: "GEMM All-Scatter Bulk Synchronous" |
| 68 | + example_path: "12_gemm_all_scatter_bulk_synchronous" |
| 69 | + tflops_threshold: 900 # Actual: ~1262 TFLOPs |
| 70 | + benchmark_args: "-m 16384 -n 16384 -k 16384 --BLK_M 128 --BLK_N 128 --BLK_K 64 --gsize_m 6 --gemm_sms 256" |
| 71 | + |
| 72 | + steps: |
| 73 | + - name: Checkout repository |
| 74 | + uses: actions/checkout@v4 |
| 75 | + |
| 76 | + - name: Run ${{ matrix.example_name }} Benchmark (8 ranks) |
| 77 | + run: | |
| 78 | + bash .github/scripts/run_perf_benchmark.sh \ |
| 79 | + "${{ matrix.example_path }}" \ |
| 80 | + "${{ matrix.tflops_threshold }}" \ |
| 81 | + ${{ matrix.benchmark_args }} |
| 82 | +
|
0 commit comments