Add cutlass submodule for sgl-kernel (#2676)

.gitmodules

@@ -0,0 +1,3 @@
+[submodule "sgl-kernel/3rdparty/cutlass"]
+	path = sgl-kernel/3rdparty/cutlass
+	url = https://github.com/NVIDIA/cutlass.git

sgl-kernel/CMakeLists.txt

@@ -8,6 +8,8 @@ set(CMAKE_CXX_STANDARD_REQUIRED ON)
 set(CMAKE_CUDA_STANDARD 17)
 set(CMAKE_CUDA_STANDARD_REQUIRED ON)
 
+set(CUTLASS_DIR "3rdparty/cutlass")
+
 # Set CUDA architectures
 set(CMAKE_CUDA_ARCHITECTURES "75;80;86;89;90")
 message(STATUS "Building for CUDA architectures: ${CMAKE_CUDA_ARCHITECTURES}")
@@ -38,6 +40,8 @@ target_include_directories(_kernels
         ${CMAKE_CURRENT_SOURCE_DIR}/src/sgl-kernel/csrc
         ${CUDA_INCLUDE_DIRS}
         ${TORCH_INCLUDE_DIRS}
+        ${CUTLASS_DIR}/include
+        ${CUTLASS_DIR}/tools/util/include
 )
 
 target_link_libraries(_kernels

sgl-kernel/setup.py

@@ -58,6 +58,11 @@ def update_wheel_platform_tag():
     old_wheel.rename(new_wheel)
 
 
+cutlass = root / "3rdparty" / "cutlass"
+include_dirs = [
+    cutlass.resolve() / "include",
+    cutlass.resolve() / "tools" / "util" / "include",
+]
 nvcc_flags = [
     "-O3",
     "-Xcompiler",
@@ -82,6 +87,7 @@ ext_modules = [
             "src/sgl-kernel/csrc/moe_align_kernel.cu",
             "src/sgl-kernel/csrc/sgl_kernel_ops.cu",
         ],
+        include_dirs=include_dirs,
         extra_compile_args={
             "nvcc": nvcc_flags,
             "cxx": cxx_flags,