kernel: add grouped gemm support for moe

src/kernels/CMakeLists.txt

@@ -1,16 +1,16 @@
 include(cc_library)
 
 cc_library(
-  NAME 
+  NAME
     kernels
-  HDRS 
+  HDRS
     reduce_kernel_utils.cuh
     activation_kernels.h
     layernorm_kernels.h
     pos_embedding_kernels.h
     kv_cache_kernels.h
     sampling/sampling_kernels.h
-  SRCS 
+  SRCS
     activation_kernels.cu
     layernorm_kernels.cu
     pos_embedding_kernels.cu
@@ -28,7 +28,7 @@ cc_library(
 
 add_subdirectory(attention)
 add_subdirectory(moe)
+add_subdirectory(gemm)
 add_subdirectory(quantization)
 add_subdirectory(playground)
-add_subdirectory(triton)
-
+# add_subdirectory(triton)

src/kernels/gemm/CMakeLists.txt

@@ -0,0 +1,24 @@
+include(cc_library)
+include(cc_test)
+
+cc_library(
+  NAME
+    gemm.kernels
+  HDRS
+    grouped_gemm_kernel_sm80.h
+  DEPS
+    cutlass
+)
+
+
+# cc_test(
+#   NAME
+#     gemm_kernel_test
+#   SRCS
+#     grouped_gemm_kernel_sm80_test.cu
+#   DEPS
+#     :gemm.kernels
+#     absl::random_random
+#     GTest::gtest_main
+#     torch
+# )

src/kernels/gemm/grouped_gemm_kernel_sm80.h

@@ -0,0 +1,3 @@
+#pragma once
+
+namespace llm {}  // namespace llm