update code

Author: xuhancn

src/BuildOnWindows.cmake

@@ -26,22 +26,30 @@ endmacro()
 
 if(BUILD_SEPARATE_OPS)
   message("!!!!! BUILD_SEPARATE_OPS: ${BUILD_SEPARATE_OPS} - 1")
-  setup_common_libraries()
+  add_library(
+    torch_xpu_ops
+    STATIC
+    ${ATen_XPU_CPP_SRCS}
+    ${ATen_XPU_MKL_SRCS}
+    ${ATen_XPU_NATIVE_CPP_SRCS}
+    ${ATen_XPU_GEN_SRCS})
+  target_compile_definitions(torch_xpu_ops PRIVATE TORCH_XPU_BUILD_MAIN_LIB)
+
   foreach(sycl_src ${ATen_XPU_SYCL_SRCS})
     get_filename_component(name ${sycl_src} NAME_WLE REALPATH)
     set(sycl_lib torch-xpu-ops-sycl-${name})
     sycl_add_library(
       ${sycl_lib}
       SHARED
       SYCL_SOURCES ${sycl_src})
-    target_link_libraries(torch_xpu_ops_aten PUBLIC ${sycl_lib})
+    target_link_libraries(torch_xpu_ops PUBLIC ${sycl_lib})
     list(APPEND TORCH_XPU_OPS_LIBRARIES ${sycl_lib})
 
     # Decouple with PyTorch cmake definition.
     install(TARGETS ${sycl_lib} DESTINATION "${TORCH_INSTALL_LIB_DIR}")
   endforeach()
+  
   list(APPEND TORCH_XPU_OPS_LIBRARIES torch_xpu_ops)
-  list(APPEND TORCH_XPU_OPS_LIBRARIES torch_xpu_ops_aten)
 # Working with the compilers which don't support device code compression, we have to split kernels
 # into multiple libraries to meet the bin size limitation.
 elseif(BUILD_SPLIT_KERNEL_LIB OR __INTEL_LLVM_COMPILER LESS 20250004 OR ICX_DATE LESS 20241205)