Triton benchmarks: Victor New Reduce Opt. #916
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| name: Triton benchmarks | |
| run-name: ${{ inputs.run_name }} | |
| on: | |
| workflow_dispatch: | |
| inputs: | |
| runner_label: | |
| description: Runner label, keep empty for default | |
| type: string | |
| default: "" | |
| tag: | |
| description: Tag for benchmark results | |
| type: string | |
| default: "test" | |
| benchmarking_method: | |
| description: The method used to obtain performance numbers | |
| type: choice | |
| options: | |
| - PYTORCH_LEGACY_PROFILER_USING_IPEX | |
| - ELAPSED_TIME | |
| - UPSTREAM_PYTORCH_PROFILER | |
| default: PYTORCH_LEGACY_PROFILER_USING_IPEX | |
| run_name: | |
| description: Run name | |
| type: string | |
| default: "Triton benchmarks" | |
| skip_benchmarks: | |
| description: JSON list of benchmarks to skip | |
| type: string | |
| default: "[]" | |
| schedule: | |
| - cron: "5 23 * * *" | |
| pull_request: | |
| branches: | |
| - main | |
| paths: | |
| - .github/workflows/triton-benchmarks.yml | |
| - benchmarks/** | |
| permissions: read-all | |
| env: | |
| PYTHON_VERSION: "3.10" | |
| BENCHMARKING_METHOD: ${{ inputs.benchmarking_method || 'PYTORCH_LEGACY_PROFILER_USING_IPEX' }} | |
| USE_IPEX: ${{ github.event_name != 'workflow_dispatch' && '1' || inputs.benchmarking_method == 'PYTORCH_LEGACY_PROFILER_USING_IPEX' && '1' || '0' }} | |
| TAG: ${{ inputs.tag || (github.event_name == 'pull_request' && 'pr') || (github.event_name == 'schedule' && 'ci') || 'test' }} | |
| jobs: | |
| build: | |
| name: Triton benchmarks | |
| runs-on: | |
| - ${{ inputs.runner_label || 'max1550' }} | |
| timeout-minutes: 720 | |
| defaults: | |
| run: | |
| shell: bash -noprofile --norc -eo pipefail -c "source /opt/intel/oneapi/setvars.sh > /dev/null; source {0}" | |
| steps: | |
| - name: Checkout repository | |
| uses: actions/checkout@v4 | |
| - name: Load pip cache | |
| id: pip-cache | |
| uses: ./.github/actions/load | |
| with: | |
| path: $HOME/.cache/pip | |
| # pip cache per commit id just to minimize network traffic | |
| key: pip-$PYTHON_VERSION-$GITHUB_SHA | |
| - name: Install Python | |
| uses: actions/setup-python@v5 | |
| with: | |
| python-version: ${{ env.PYTHON_VERSION }} | |
| - name: Install Python build dependencies | |
| run: | | |
| pip install wheel cmake | |
| - name: Setup PyTorch with IPEX | |
| if: ${{ env.USE_IPEX == '1' }} | |
| uses: ./.github/actions/setup-pytorch | |
| with: | |
| repository: Stonepia/pytorch | |
| - name: Setup PyTorch without IPEX | |
| if: ${{ env.USE_IPEX == '0' }} | |
| uses: ./.github/actions/setup-pytorch | |
| with: | |
| repository: pytorch/pytorch | |
| - name: Setup IPEX | |
| if: ${{ env.USE_IPEX == '1' }} | |
| uses: ./.github/actions/setup-ipex | |
| - name: Build Triton wheels | |
| uses: ./.github/actions/setup-triton | |
| with: | |
| command: DEBUG=1 python setup.py bdist_wheel | |
| - name: Install Triton | |
| run: | | |
| pip install python/dist/*.whl | |
| - name: Install benchmark dependencies | |
| run: | | |
| pip install matplotlib pandas tabulate | |
| - name: Create reports dir | |
| run: | | |
| mkdir reports | |
| echo "REPORTS=$PWD/reports" >> $GITHUB_ENV | |
| - name: Install benchmarks | |
| id: install | |
| run: | | |
| cd benchmarks | |
| python setup.py install | |
| - name: Run Triton Softmax kernel benchmark | |
| if: ${{ steps.install.outcome == 'success' && !cancelled() && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'fused_softmax.py') }} | |
| run: | | |
| cd benchmarks/triton_kernels_benchmark | |
| python fused_softmax.py --reports $REPORTS | |
| source ../../scripts/capture-hw-details.sh | |
| python ../../scripts/build_report.py $REPORTS/softmax-performance.csv $REPORTS/softmax-triton-report.csv --benchmark softmax --compiler triton --param_cols "N" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG | |
| python ../../scripts/build_report.py $REPORTS/softmax-performance.csv $REPORTS/softmax-xetla-report.csv --benchmark softmax --compiler xetla --param_cols "N" --tflops_col XeTLA-TFlops --hbm_col "XeTLA-GB/s" --tag $TAG | |
| - name: Run Triton GEMM kernel benchmark | |
| if: ${{ steps.install.outcome == 'success' && !cancelled() && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'gemm_benchmark.py') }} | |
| run: | | |
| cd benchmarks/triton_kernels_benchmark | |
| python gemm_benchmark.py --reports $REPORTS | |
| mv $REPORTS/matmul-performance.csv $REPORTS/matmul-performance-base.csv | |
| source ../../scripts/capture-hw-details.sh | |
| python ../../scripts/build_report.py $REPORTS/matmul-performance-base.csv $REPORTS/gemm-triton-report.csv --benchmark gemm --compiler triton --param_cols "B,M,K,N" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG | |
| python ../../scripts/build_report.py $REPORTS/matmul-performance-base.csv $REPORTS/gemm-xetla-report.csv --benchmark gemm --compiler xetla --param_cols "B,M,K,N" --tflops_col XeTLA-TFlops --hbm_col "XeTLA-GB/s" --tag $TAG | |
| - name: Run Triton GEMM kernel benchmark - default path | |
| if: ${{ steps.install.outcome == 'success' && !cancelled() && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'gemm_benchmark.py_default') }} | |
| run: | | |
| cd benchmarks/triton_kernels_benchmark | |
| # Default path: | |
| TRITON_INTEL_ADVANCED_PATH=0 \ | |
| TRITON_INTEL_ENABLE_ADDRESS_PAYLOAD_OPT=1 \ | |
| IGC_VISAOptions=" -enableBCR -nolocalra" \ | |
| IGC_DisableLoopUnroll=1 \ | |
| python gemm_benchmark.py --reports $REPORTS | |
| mv $REPORTS/matmul-performance.csv $REPORTS/matmul-performance-default-path.csv | |
| source ../../scripts/capture-hw-details.sh | |
| TAG="${TAG}-dflt" | |
| python ../../scripts/build_report.py $REPORTS/matmul-performance-default-path.csv $REPORTS/gemm-triton-default-report.csv --benchmark gemm --compiler triton --param_cols "B,M,K,N" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG | |
| - name: Run Triton GEMM kernel benchmark - advanced path | |
| if: ${{ steps.install.outcome == 'success' && !cancelled() && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'gemm_benchmark.py_advanced') }} | |
| run: | | |
| cd benchmarks/triton_kernels_benchmark | |
| # Advanced path: | |
| TRITON_INTEL_ADVANCED_PATH=1 \ | |
| TRITON_INTEL_ENABLE_ADDRESS_PAYLOAD_OPT=1 \ | |
| IGC_VISAOptions=" -enableBCR -nolocalra" \ | |
| IGC_DisableLoopUnroll=1 \ | |
| python gemm_benchmark.py --reports $REPORTS | |
| mv $REPORTS/matmul-performance.csv $REPORTS/matmul-performance-adv-path.csv | |
| source ../../scripts/capture-hw-details.sh | |
| TAG="${TAG}-adv" | |
| python ../../scripts/build_report.py $REPORTS/matmul-performance-adv-path.csv $REPORTS/gemm-triton-advanced-report.csv --benchmark gemm --compiler triton --param_cols "B,M,K,N" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG | |
| - name: Run Triton GEMM (A@B^t) kernel benchmark | |
| if: ${{ steps.install.outcome == 'success' && !cancelled() && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'gemm_benchmark.py_abt') }} | |
| run: | | |
| cd benchmarks/triton_kernels_benchmark | |
| TRANSPOSE_B=1 python gemm_benchmark.py --reports $REPORTS | |
| mv $REPORTS/matmul-performance.csv $REPORTS/matmul-performance-bt.csv | |
| source ../../scripts/capture-hw-details.sh | |
| python ../../scripts/build_report.py $REPORTS/matmul-performance-bt.csv $REPORTS/gemm-bt-triton-report.csv --benchmark gemm-bt --compiler triton --param_cols "B,M,K,N" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG | |
| python ../../scripts/build_report.py $REPORTS/matmul-performance-bt.csv $REPORTS/gemm-bt-onednn-report.csv --benchmark gemm-bt --compiler onednn --param_cols "B,M,K,N" --tflops_col onednn-TFlops --hbm_col "onednn-GB/s" --tag $TAG | |
| - name: Run Triton GEMM (A^t@B) kernel benchmark | |
| if: ${{ steps.install.outcome == 'success' && !cancelled() && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'gemm_benchmark.py_atb') }} | |
| run: | | |
| cd benchmarks/triton_kernels_benchmark | |
| TRANSPOSE_A=1 python gemm_benchmark.py --reports $REPORTS | |
| mv $REPORTS/matmul-performance.csv $REPORTS/matmul-performance-at.csv | |
| source ../../scripts/capture-hw-details.sh | |
| python ../../scripts/build_report.py $REPORTS/matmul-performance-at.csv $REPORTS/gemm-at-triton-report.csv --benchmark gemm-at --compiler triton --param_cols "B,M,K,N" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG | |
| python ../../scripts/build_report.py $REPORTS/matmul-performance-at.csv $REPORTS/gemm-at-onednn-report.csv --benchmark gemm-at --compiler onednn --param_cols "B,M,K,N" --tflops_col onednn-TFlops --hbm_col "onednn-GB/s" --tag $TAG | |
| - name: Run Triton GEMM (stream-k) kernel benchmark | |
| if: ${{ steps.install.outcome == 'success' && !cancelled() && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'gemm_streamk_benchmark.py') }} | |
| run: | | |
| cd benchmarks/triton_kernels_benchmark | |
| python gemm_streamk_benchmark.py --reports $REPORTS | |
| source ../../scripts/capture-hw-details.sh | |
| python ../../scripts/build_report.py $REPORTS/matmul-streamk-performance.csv $REPORTS/gemm-streamk-triton-report.csv --benchmark gemm-streamk --compiler triton --param_cols "M,K,N" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG | |
| - name: Run Triton GEMM (split-k) kernel benchmark | |
| if: ${{ steps.install.outcome == 'success' && !cancelled() && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'gemm_splitk_benchmark.py') }} | |
| run: | | |
| cd benchmarks/triton_kernels_benchmark | |
| python gemm_splitk_benchmark.py --reports $REPORTS | |
| source ../../scripts/capture-hw-details.sh | |
| python ../../scripts/build_report.py $REPORTS/matmul-splitk-performance.csv $REPORTS/gemm-splitk-triton-report.csv --benchmark gemm-splitk --compiler triton --param_cols "M,K,N" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG | |
| - name: Run Triton GEMM + PreOp (exp) kernel benchmark | |
| if: ${{ steps.install.outcome == 'success' && !cancelled() && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'gemm_preop_exp_benchmark.py') }} | |
| run: | | |
| cd benchmarks/triton_kernels_benchmark | |
| python gemm_preop_exp_benchmark.py --reports $REPORTS | |
| source ../../scripts/capture-hw-details.sh | |
| python ../../scripts/build_report.py $REPORTS/matmul-performance-preop-exp.csv $REPORTS/gemm-preop-exp-triton-report.csv --benchmark gemm-preop-exp --compiler triton --param_cols "B,M,K,N" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG | |
| - name: Run Triton GEMM + PostOp (Gelu) kernel benchmark | |
| if: ${{ steps.install.outcome == 'success' && !cancelled() && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'gemm_postop_gelu_benchmark.py') }} | |
| run: | | |
| cd benchmarks/triton_kernels_benchmark | |
| python gemm_postop_gelu_benchmark.py --reports $REPORTS | |
| source ../../scripts/capture-hw-details.sh | |
| python ../../scripts/build_report.py $REPORTS/matmul-performance-postop-gelu.csv $REPORTS/gemm-postop-gelu-triton-report.csv --benchmark gemm-postop-gelu --compiler triton --param_cols "B,M,K,N" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG | |
| - name: Run Triton GEMM + PostOp (add matrix) kernel benchmark | |
| if: ${{ steps.install.outcome == 'success' && !cancelled() && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'gemm_postop_addmatrix_benchmark.py') }} | |
| run: | | |
| cd benchmarks/triton_kernels_benchmark | |
| python gemm_postop_addmatrix_benchmark.py --reports $REPORTS | |
| source ../../scripts/capture-hw-details.sh | |
| python ../../scripts/build_report.py $REPORTS/matmul-performance-postop-addmatrix.csv $REPORTS/gemm-postop-addmatrix-triton-report.csv --benchmark gemm-postop-addmatrix --compiler triton --param_cols "B,M,K,N" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG | |
| - name: Run Triton FA kernel benchmark | |
| if: ${{ steps.install.outcome == 'success' && !cancelled() && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'flash_attention_fwd_benchmark.py') }} | |
| run: | | |
| cd benchmarks/triton_kernels_benchmark | |
| python flash_attention_fwd_benchmark.py --reports $REPORTS | |
| source ../../scripts/capture-hw-details.sh | |
| python ../../scripts/build_report.py $REPORTS/attn-performance.csv $REPORTS/attn-triton-report.csv --benchmark attn --compiler triton --param_cols "Z,H,N_CTX,D_HEAD,CAUSAL" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG | |
| python ../../scripts/build_report.py $REPORTS/attn-performance.csv $REPORTS/attn-xetla-report.csv --benchmark attn --compiler xetla --param_cols "Z,H,N_CTX,D_HEAD,CAUSAL" --tflops_col XeTLA-TFlops --hbm_col "XeTLA-GB/s" --tag $TAG | |
| - name: Run Triton FA kernel benchmark - default path | |
| if: ${{ steps.install.outcome == 'success' && !cancelled() && !contains(fromJson(inputs.skip_benchmark || '[]'), 'flash_attention_fwd_benchmark.py_default') }} | |
| run: | | |
| cd benchmarks/triton_kernels_benchmark | |
| TRITON_INTEL_ADVANCED_PATH=0 \ | |
| TRITON_INTEL_ENABLE_ADDRESS_PAYLOAD_OPT=1 \ | |
| IGC_VISAOptions=" -enableBCR" \ | |
| python flash_attention_fwd_benchmark.py --reports $REPORTS | |
| TAG="${TAG}-dflt" | |
| source ../../scripts/capture-hw-details.sh | |
| python ../../scripts/build_report.py $REPORTS/attn-performance.csv $REPORTS/attn-triton-default-report.csv --benchmark attn --compiler triton --param_cols "Z,H,N_CTX,D_HEAD,CAUSAL" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG | |
| - name: Run Triton FA kernel benchmark - advanced path | |
| if: ${{ steps.install.outcome == 'success' && !cancelled() && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'flash_attention_fwd_benchmark.py_advanced') }} | |
| run: | | |
| cd benchmarks/triton_kernels_benchmark | |
| TRITON_INTEL_ADVANCED_PATH=1 \ | |
| TRITON_INTEL_ENABLE_ADDRESS_PAYLOAD_OPT=1 \ | |
| IGC_VISAOptions=" -enableBCR" \ | |
| python flash_attention_fwd_benchmark.py --reports $REPORTS | |
| TAG="${TAG}-adv" | |
| source ../../scripts/capture-hw-details.sh | |
| python ../../scripts/build_report.py $REPORTS/attn-performance.csv $REPORTS/attn-triton-advanced-report.csv --benchmark attn --compiler triton --param_cols "Z,H,N_CTX,D_HEAD,CAUSAL" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG | |
| - name: Run Prefix Sums kernel benchmark | |
| if: ${{ steps.install.outcome == 'success' && !cancelled() && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'prefix_sums.py') }} | |
| run: | | |
| cd benchmarks/triton_kernels_benchmark | |
| python prefix_sums.py --reports $REPORTS | |
| source ../../scripts/capture-hw-details.sh | |
| python ../../scripts/build_report.py $REPORTS/prefix-sums.csv $REPORTS/prefix_sums-triton-report.csv --benchmark prefix_sums --compiler triton --param_cols "N" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG | |
| - name: Run micro benchmark | |
| if: ${{ steps.install.outcome == 'success' && !cancelled() && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'micro_benchmarks') }} | |
| run: | | |
| cd benchmarks/micro_benchmarks | |
| python run_benchmarks.py --reports $REPORTS | |
| - name: Save pip cache | |
| if: ${{ steps.pip-cache.outputs.status == 'miss' }} | |
| uses: ./.github/actions/save | |
| with: | |
| path: ${{ steps.pip-cache.outputs.path }} | |
| dest: ${{ steps.pip-cache.outputs.dest }} | |
| - name: Upload benchmark reports | |
| if: ${{ steps.install.outcome == 'success' && !cancelled() }} | |
| uses: actions/upload-artifact@v4 | |
| with: | |
| name: benchmark-reports | |
| path: reports |