Upgrade CUTLASS 4.0 (#6336)

Co-authored-by: zhyncs <me@zhyncs.com>

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)

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 = {