refactor: refine layers dir to support other cuda like devices. (#277)

Author: XuZhang99

xllm/core/framework/model/CMakeLists.txt

@@ -15,17 +15,12 @@ set(BASE_DEPS
 )
 
 if(USE_NPU)
-  # Modify dependencies for npu
-  list(APPEND BASE_DEPS torch_npu)
   list(APPEND BASE_DEPS :npu_layers)
   list(APPEND BASE_DEPS :platform_npu)
+else()
+  list(APPEND BASE_DEPS :common_layers)
 endif()
 
-if(USE_MLU)
-  # Modify dependencies for mlu
-  list(APPEND BASE_DEPS torch_mlu)
-  list(APPEND BASE_DEPS :mlu_layers)
-endif()
 
 # Define the library
 cc_library(

xllm/core/layers/CMakeLists.txt

@@ -79,7 +79,6 @@ cc_library(
 
 if(USE_NPU)
   add_subdirectory(npu)
-endif()
-if(USE_MLU)
-  add_subdirectory(mlu)
+else()
+  add_subdirectory(common)
 endif()

xllm/core/layers/mlu/CMakeLists.txt renamed to xllm/core/layers/common/CMakeLists.txt

@@ -4,7 +4,7 @@ include(cc_test)
 
 cc_library(
   NAME
-    mlu_layers 
+    common_layers 
   HDRS
     qwen3_attention.h
     attention.h
@@ -15,7 +15,7 @@ cc_library(
     qwen3_decoder_layer.h
     qwen3_moe_decoder_layer.h
     linear_impl.h
-    mlu_word_embedding_impl.h
+    word_embedding_impl.h
   SRCS
     qwen3_attention.cpp
     attention.cpp