From 74e5a98e2a8a4f5c71ca3ea9a1ffa8965bc863a0 Mon Sep 17 00:00:00 2001 From: zhyncs Date: Tue, 13 May 2025 15:05:40 -0700 Subject: [PATCH 1/2] chore: upgrade cutlass 4.0 --- sgl-kernel/CMakeLists.txt | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/sgl-kernel/CMakeLists.txt b/sgl-kernel/CMakeLists.txt index 300830f1c28a..5d52a13458f5 100755 --- a/sgl-kernel/CMakeLists.txt +++ b/sgl-kernel/CMakeLists.txt @@ -45,7 +45,7 @@ include(FetchContent) FetchContent_Declare( repo-cutlass GIT_REPOSITORY https://github.com/NVIDIA/cutlass - GIT_TAG ad7b2f5e84fcfa124cb02b91d5bd26d238c0459e + GIT_TAG f115c3f85467d5d9619119d1dbeb9c03c3d73864 GIT_SHALLOW OFF ) FetchContent_Populate(repo-cutlass) From 0353ec8065168a50cef312b0baadb753a6bec7f0 Mon Sep 17 00:00:00 2001 From: Elfie Guo Date: Thu, 15 May 2025 22:21:17 +0000 Subject: [PATCH 2/2] modify the template for upgrading to CUTLASS 4.0 --- sgl-kernel/csrc/gemm/fp8_blockwise_gemm_kernel.cu | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) mode change 100644 => 100755 sgl-kernel/csrc/gemm/fp8_blockwise_gemm_kernel.cu diff --git a/sgl-kernel/csrc/gemm/fp8_blockwise_gemm_kernel.cu b/sgl-kernel/csrc/gemm/fp8_blockwise_gemm_kernel.cu old mode 100644 new mode 100755 index 609134730f06..33e536f16f6a --- a/sgl-kernel/csrc/gemm/fp8_blockwise_gemm_kernel.cu +++ b/sgl-kernel/csrc/gemm/fp8_blockwise_gemm_kernel.cu @@ -140,7 +140,7 @@ void launch_sm90_fp8_blockwise_scaled_mm( LayoutSFB layout_sfb = ScaleConfig::tile_atom_to_shape_SFB(make_shape(m, n, k, 1)); typename GemmKernel::MainloopArguments mainloop_args{ - a_ptr, stride_a, b_ptr, stride_b, 4, a_s_ptr, layout_sfa, b_s_ptr, layout_sfb}; + a_ptr, stride_a, b_ptr, stride_b, a_s_ptr, layout_sfa, b_s_ptr, layout_sfb}; typename GemmKernel::EpilogueArguments epilogue_args{{}, nullptr, stride_d, o_ptr, stride_d}; typename Gemm::Arguments args = {