From 5d6c3e7feb9bb61a9b046d61639dcd8bb944c45c Mon Sep 17 00:00:00 2001
From: "xinan.lin" <xinan.lin@intel.com>
Date: Mon, 20 May 2024 02:14:01 -0700
Subject: [PATCH] Expose TORCH_XPU_API to Pytorch

---
 src/aten/CMakeLists.txt  |  6 ++++++
 src/aten/EmptyTensor.cpp |  1 +
 src/aten/EmptyTensor.h   | 13 +++++++------
 3 files changed, 14 insertions(+), 6 deletions(-)

diff --git a/src/aten/CMakeLists.txt b/src/aten/CMakeLists.txt
index 63156cff5..9c2042944 100644
--- a/src/aten/CMakeLists.txt
+++ b/src/aten/CMakeLists.txt
@@ -1,6 +1,7 @@
 # ATen XPU sources
 
 file(GLOB xpu_cpp "*.cpp")
+file(GLOB xpu_h "*.h")
 file(GLOB xpu_sycl "sycl/*.cpp")
 
 list(APPEND ATen_XPU_CPP_SRCS ${xpu_cpp})
@@ -8,3 +9,8 @@ list(APPEND ATen_XPU_SYCL_SRCS ${xpu_sycl})
 
 set(ATen_XPU_CPP_SRCS ${ATen_XPU_CPP_SRCS} PARENT_SCOPE)
 set(ATen_XPU_SYCL_SRCS ${ATen_XPU_SYCL_SRCS} PARENT_SCOPE)
+
+
+foreach(HEADER  ${xpu_h})
+  install(FILES ${HEADER} DESTINATION "include/ATen/xpu")
+endforeach()
diff --git a/src/aten/EmptyTensor.cpp b/src/aten/EmptyTensor.cpp
index d990d3e9f..0528190f7 100644
--- a/src/aten/EmptyTensor.cpp
+++ b/src/aten/EmptyTensor.cpp
@@ -1,6 +1,7 @@
 #define TORCH_ASSERT_NO_OPERATORS
 #include <ATen/Context.h>
 #include <ATen/EmptyTensor.h>
+#include <ATen/xpu/EmptyTensor.h>
 #include <c10/core/DeviceGuard.h>
 #include <c10/xpu/XPUCachingAllocator.h>
 
diff --git a/src/aten/EmptyTensor.h b/src/aten/EmptyTensor.h
index 9f4343783..b268eacf3 100644
--- a/src/aten/EmptyTensor.h
+++ b/src/aten/EmptyTensor.h
@@ -4,13 +4,13 @@
 namespace at::detail {
 
 // XXX: add TORCH_XPU_API
-TensorBase empty_xpu(
+TORCH_XPU_API TensorBase empty_xpu(
     IntArrayRef size,
     ScalarType dtype,
     c10::optional<Device> device_opt,
     c10::optional<c10::MemoryFormat> memory_format_opt);
 
-TensorBase empty_xpu(
+TORCH_XPU_API TensorBase empty_xpu(
     IntArrayRef size,
     c10::optional<ScalarType> dtype_opt,
     c10::optional<Layout> layout_opt,
@@ -18,15 +18,16 @@ TensorBase empty_xpu(
     c10::optional<bool> pin_memory_opt,
     c10::optional<c10::MemoryFormat> memory_format_opt);
 
-TensorBase empty_xpu(IntArrayRef size, const TensorOptions& options);
+TORCH_XPU_API TensorBase
+empty_xpu(IntArrayRef size, const TensorOptions& options);
 
-TensorBase empty_strided_xpu(
+TORCH_XPU_API TensorBase empty_strided_xpu(
     IntArrayRef size,
     IntArrayRef stride,
     ScalarType dtype,
     c10::optional<Device> device_opt);
 
-TensorBase empty_strided_xpu(
+TORCH_XPU_API TensorBase empty_strided_xpu(
     IntArrayRef size,
     IntArrayRef stride,
     c10::optional<ScalarType> dtype_opt,
@@ -34,7 +35,7 @@ TensorBase empty_strided_xpu(
     c10::optional<Device> device_opt,
     c10::optional<bool> pin_memory_opt);
 
-TensorBase empty_strided_xpu(
+TORCH_XPU_API TensorBase empty_strided_xpu(
     IntArrayRef size,
     IntArrayRef stride,
     const TensorOptions& options);