diff --git a/.github/workflows/triton-benchmarks.yml b/.github/workflows/triton-benchmarks.yml index 9883b28437..2056d742b7 100644 --- a/.github/workflows/triton-benchmarks.yml +++ b/.github/workflows/triton-benchmarks.yml @@ -157,21 +157,8 @@ jobs: 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 \ - 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 + # FIXME: Remove after Grafana migration + 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}-dflt - name: Run Triton GEMM kernel benchmark - advanced path if: ${{ steps.install.outcome == 'success' && !cancelled() && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'gemm_benchmark.py_advanced') }} @@ -259,19 +246,8 @@ jobs: 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 + # FIXME: Remove after Grafana migration + 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}-dflt - 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') }} diff --git a/scripts/test-triton.sh b/scripts/test-triton.sh index 5516bbee05..86a06449e3 100755 --- a/scripts/test-triton.sh +++ b/scripts/test-triton.sh @@ -263,10 +263,7 @@ run_benchmark_gemm() { python setup.py install echo "Default path:" - TRITON_INTEL_ADVANCED_PATH=0 \ - IGC_VISAOptions=" -enableBCR -nolocalra" \ - IGC_DisableLoopUnroll=1 \ - python $TRITON_PROJ/benchmarks/triton_kernels_benchmark/gemm_benchmark.py + python $TRITON_PROJ/benchmarks/triton_kernels_benchmark/gemm_benchmark.py echo "Advanced path:" TRITON_INTEL_ADVANCED_PATH=1 \ @@ -283,10 +280,7 @@ run_benchmark_attention() { python setup.py install echo "Default path:" - TRITON_INTEL_ADVANCED_PATH=0 \ - TRITON_INTEL_ENABLE_ADDRESS_PAYLOAD_OPT=1 \ - IGC_VISAOptions=" -enableBCR" \ - python $TRITON_PROJ/benchmarks/triton_kernels_benchmark/flash_attention_fwd_benchmark.py + python $TRITON_PROJ/benchmarks/triton_kernels_benchmark/flash_attention_fwd_benchmark.py echo "Advanced path:" TRITON_INTEL_ADVANCED_PATH=1 \