[Intel GPU] oneDNN GPU GEMM support (pytorch#117202)

# Motivation

This PR is a part of RFC pytorch#114848, and it  is a successor PR of pytorch#116249 and pytorch#116019. This PR would depend on oneDNN compilation in pytorch#116249. Some runtime support is needed in pytorch#116019.

Aten operators like `addmm`, `baddmm` is defined in `Blas.cpp` in `aten/src/ATen/native/mkldnn/xpu/`.

Accompanied with these files provide core functionaliy, `BlasImpl.h`, `Utils.h` and other file provide basic utilities for them. For instance, `Utils.h` provide common memory descriptor query utils for `Matmul.h` and these utility function will also be used in other primitive, like `convolution`.  `BlasImpl.h` is a header file that provide helper for handling shape info processing in matmul related operators. It would not only help basic GEMM operator like `addmm, baddmm` but also help fusion operators used in `torch.compile` like `linear_pointwise` in pytorch#117824.

In next stage, we would continually complete the oneDNN support through enabling  `matmul fusion`  and `convolution` related code.

Co-authored-by: xiaolil1 <[email protected]>
Co-authored-by: lei,zhenyuan <[email protected]>
Pull Request resolved: pytorch#117202
Approved by: https://github.com/EikanWang, https://github.com/jgong5, https://github.com/malfet
ghstack dependencies: pytorch#117098, pytorch#117112

Author: ZhiweiYan-96

aten/CMakeLists.txt

@@ -18,6 +18,8 @@ cmake_policy(SET CMP0012 NEW)
 #############################################
 
 set(ATen_CPU_SRCS)
+set(ATen_XPU_SRCS)
+set(ATen_XPU_INCLUDE)
 set(ATen_CPU_TEST_SRCS)
 set(ATen_CPU_INCLUDE)
 set(ATen_THIRD_PARTY_INCLUDE)
@@ -39,6 +41,7 @@ set(ATen_XPU_INCLUDE)
 set(ATen_XPU_TEST_SRCS)
 set(ATen_VULKAN_TEST_SRCS)
 set(ATen_CPU_DEPENDENCY_LIBS)
+set(ATen_XPU_DEPENDENCY_LIBS)
 set(ATen_CUDA_DEPENDENCY_LIBS)
 set(ATen_HIP_DEPENDENCY_LIBS)
 set(ATen_PUBLIC_CUDA_DEPENDENCY_LIBS)
@@ -105,6 +108,8 @@ add_subdirectory(src/ATen)
 # Pass source, includes, and libs to parent
 set(ATen_CPU_SRCS ${ATen_CPU_SRCS} PARENT_SCOPE)
 set(ATen_CORE_SRCS ${ATen_CORE_SRCS} PARENT_SCOPE)
+set(ATen_XPU_SRCS ${ATen_XPU_SRCS} PARENT_SCOPE)
+set(ATen_XPU_INCLUDE ${ATen_XPU_INCLUDE} PARENT_SCOPE)
 set(ATen_CUDA_CU_SRCS ${ATen_CUDA_CU_SRCS} PARENT_SCOPE)
 set(ATen_CUDA_CPP_SRCS ${ATen_CUDA_CPP_SRCS} PARENT_SCOPE)
 set(ATen_CUDA_LINALG_SRCS ${ATen_CUDA_LINALG_SRCS} PARENT_SCOPE)
@@ -130,6 +135,7 @@ set(ATen_HIP_INCLUDE ${ATen_HIP_INCLUDE} PARENT_SCOPE)
 set(ATen_XPU_INCLUDE ${ATen_XPU_INCLUDE} PARENT_SCOPE)
 set(ATen_THIRD_PARTY_INCLUDE ${ATen_THIRD_PARTY_INCLUDE} PARENT_SCOPE)
 set(ATen_CPU_DEPENDENCY_LIBS ${ATen_CPU_DEPENDENCY_LIBS} PARENT_SCOPE)
+set(ATen_XPU_DEPENDENCY_LIBS ${ATen_XPU_DEPENDENCY_LIBS} PARENT_SCOPE)
 set(ATen_CUDA_DEPENDENCY_LIBS ${ATen_CUDA_DEPENDENCY_LIBS} PARENT_SCOPE)
 set(ATen_HIP_DEPENDENCY_LIBS ${ATen_HIP_DEPENDENCY_LIBS} PARENT_SCOPE)
 set(ATen_CORE_TEST_SRCS ${ATen_CORE_TEST_SRCS} PARENT_SCOPE)

aten/src/ATen/CMakeLists.txt

@@ -85,6 +85,8 @@ file(GLOB miopen_cpp "miopen/*.cpp")
 file(GLOB mkl_cpp "mkl/*.cpp")
 file(GLOB mkldnn_cpp "mkldnn/*.cpp")
 
+file(GLOB mkldnn_xpu_cpp "native/mkldnn/xpu/*.cpp" "native/mkldnn/xpu/detail/*.cpp")
+
 file(GLOB native_cpp "native/*.cpp")
 file(GLOB native_mkl_cpp "native/mkl/*.cpp")
 file(GLOB native_mkldnn_cpp "native/mkldnn/*.cpp")
@@ -238,6 +240,20 @@ else()
   set(all_cpu_cpp ${all_cpu_cpp} ${vulkan_cpp})
 endif()
 
+if(USE_XPU)
+  list(APPEND ATen_XPU_SRCS ${mkldnn_xpu_cpp})
+  list(APPEND ATen_XPU_DEPENDENCY_LIBS xpu_mkldnn)
+
+  list(APPEND ATen_XPU_DEPENDENCY_LIBS ${OCL_LIBRARY})
+  list(APPEND ATen_XPU_INCLUDE ${CMAKE_CURRENT_SOURCE_DIR}/native/mkldnn/xpu)
+  list(APPEND ATen_XPU_INCLUDE ${CMAKE_CURRENT_SOURCE_DIR}/native/mkldnn/xpu/detail)
+  list(APPEND ATen_XPU_INCLUDE ${PROJECT_SOURCE_DIR}/third_party/ideep/mkl-dnn/include)
+  list(APPEND ATen_XPU_INCLUDE ${XPU_MKLDNN_INCLUDE})
+
+  list(APPEND ATen_XPU_INCLUDE ${SYCL_INCLUDE_DIR})
+  list(APPEND ATen_XPU_DEPENDENCY_LIBS ${SYCL_LIBRARY})
+endif()
+
 # Metal
 if(USE_PYTORCH_METAL_EXPORT)
   # Add files needed from exporting metal models(optimized_for_mobile)
@@ -629,6 +645,7 @@ list(APPEND ATen_MOBILE_BENCHMARK_SRCS
 # Pass source, includes, and libs to parent
 set(ATen_CORE_SRCS ${ATen_CORE_SRCS} PARENT_SCOPE)
 set(ATen_CPU_SRCS ${ATen_CPU_SRCS} PARENT_SCOPE)
+set(ATen_XPU_SRCS ${ATen_XPU_SRCS} PARENT_SCOPE)
 set(ATen_CUDA_CU_SRCS ${ATen_CUDA_CU_SRCS} PARENT_SCOPE)
 set(ATen_CUDA_CPP_SRCS ${ATen_CUDA_CPP_SRCS} PARENT_SCOPE)
 set(ATen_CUDA_LINALG_SRCS ${ATen_CUDA_LINALG_SRCS} PARENT_SCOPE)
@@ -658,6 +675,7 @@ set(ATen_XPU_INCLUDE ${ATen_XPU_INCLUDE} PARENT_SCOPE)
 set(ATen_VULKAN_INCLUDE ${ATen_VULKAN_INCLUDE} PARENT_SCOPE)
 set(ATen_CPU_DEPENDENCY_LIBS ${ATen_CPU_DEPENDENCY_LIBS} PARENT_SCOPE)
 set(ATen_CUDA_DEPENDENCY_LIBS ${ATen_CUDA_DEPENDENCY_LIBS} PARENT_SCOPE)
+set(ATen_XPU_DEPENDENCY_LIBS ${ATen_XPU_DEPENDENCY_LIBS} PARENT_SCOPE)
 set(ATen_HIP_DEPENDENCY_LIBS ${ATen_HIP_DEPENDENCY_LIBS} PARENT_SCOPE)
 set(FLASH_ATTENTION_CUDA_SOURCES ${FLASH_ATTENTION_CUDA_SOURCES} PARENT_SCOPE)
 set(MEM_EFF_ATTENTION_CUDA_SOURCES ${MEM_EFF_ATTENTION_CUDA_SOURCES} PARENT_SCOPE)