diff --git a/BUILD.bazel b/BUILD.bazel
index 32608b42b96..703abac16c7 100644
--- a/BUILD.bazel
+++ b/BUILD.bazel
@@ -542,6 +542,7 @@ header_template_rule(
 cc_library(
     name = "aten_headers",
     hdrs = [
+        "torch/csrc/Export.h",
         "torch/csrc/WindowsTorchApiMacro.h",
         "torch/csrc/jit/frontend/function_schema_parser.h",
     ] + glob([
diff --git a/tools/autograd/templates/Functions.h b/tools/autograd/templates/Functions.h
index 0540bb65b33..641488909e3 100644
--- a/tools/autograd/templates/Functions.h
+++ b/tools/autograd/templates/Functions.h
@@ -6,11 +6,10 @@
 #include <ATen/core/functional.h>
 #include <ATen/TensorGeometry.h>
 
-#include "torch/csrc/THP_export.h"
 #include "torch/csrc/autograd/function.h"
 #include "torch/csrc/autograd/variable.h"
 #include "torch/csrc/autograd/saved_variable.h"
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 namespace torch { namespace autograd { namespace generated {
 
diff --git a/tools/build_variables.bzl b/tools/build_variables.bzl
index a8add3034cd..6239c6e63dc 100644
--- a/tools/build_variables.bzl
+++ b/tools/build_variables.bzl
@@ -71,6 +71,7 @@ def libtorch_generated_sources(gencode_pattern):
 # copied from https://github.com/pytorch/pytorch/blob/f99a693cd9ff7a9b5fdc71357dac66b8192786d3/aten/src/ATen/core/CMakeLists.txt
 jit_core_headers = [
     "torch/csrc/utils/memory.h",
+    "torch/csrc/Export.h",
     "torch/csrc/WindowsTorchApiMacro.h",
     "torch/csrc/jit/frontend/source_range.h",
     "torch/csrc/jit/serialization/callstack_debug_info_serialization.h",
diff --git a/torch/csrc/Exceptions.h b/torch/csrc/Exceptions.h
index 36472a7eafe..f6cf5f341b0 100644
--- a/torch/csrc/Exceptions.h
+++ b/torch/csrc/Exceptions.h
@@ -9,10 +9,9 @@
 
 #include <c10/util/Exception.h>
 #include <pybind11/pybind11.h>
-#include <torch/csrc/THP_export.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/utils/auto_gil.h>
 #include <torch/csrc/jit/runtime/jit_exception.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
 #include <c10/util/StringUtil.h>
 #include <ATen/detail/FunctionTraits.h>
 
@@ -259,9 +258,9 @@ bool THPException_init(PyObject *module);
 
 namespace torch {
 
-THP_CLASS std::string processErrorMsg(std::string str);
+TORCH_PYTHON_API std::string processErrorMsg(std::string str);
 
-THP_API bool get_cpp_stacktraces_enabled();
+TORCH_PYTHON_API bool get_cpp_stacktraces_enabled();
 
 // Abstract base class for exceptions which translate to specific Python types
 struct PyTorchError : public std::exception {
diff --git a/torch/csrc/Export.h b/torch/csrc/Export.h
new file mode 100644
index 00000000000..4bcd910f127
--- /dev/null
+++ b/torch/csrc/Export.h
@@ -0,0 +1,9 @@
+#pragma once
+
+#include <c10/macros/Export.h>
+
+#ifdef THP_BUILD_MAIN_LIB
+#define TORCH_PYTHON_API C10_EXPORT
+#else
+#define TORCH_PYTHON_API C10_IMPORT
+#endif
diff --git a/torch/csrc/Generator.h b/torch/csrc/Generator.h
index bab668751ee..d2d03e585dc 100644
--- a/torch/csrc/Generator.h
+++ b/torch/csrc/Generator.h
@@ -1,9 +1,9 @@
 #pragma once
 
+#include <torch/csrc/Export.h>
 #include <torch/csrc/python_headers.h>
 #include <ATen/ATen.h>
 
-#include <torch/csrc/THP_export.h>
 
 // NOLINTNEXTLINE(cppcoreguidelines-pro-type-member-init)
 struct THPGenerator {
@@ -14,16 +14,16 @@ struct THPGenerator {
 // Creates a new Python object wrapping the default at::Generator. The reference is
 // borrowed. The caller should ensure that the at::Generator object lifetime
 // last at least as long as the Python wrapper.
-THP_API PyObject * THPGenerator_initDefaultGenerator(at::Generator cdata);
+TORCH_PYTHON_API PyObject * THPGenerator_initDefaultGenerator(at::Generator cdata);
 
 #define THPGenerator_Check(obj) \
   PyObject_IsInstance(obj, THPGeneratorClass)
 
-THP_API PyObject *THPGeneratorClass;
+TORCH_PYTHON_API extern PyObject *THPGeneratorClass;
 
 bool THPGenerator_init(PyObject *module);
 
-THP_API PyObject * THPGenerator_Wrap(at::Generator gen);
+TORCH_PYTHON_API PyObject * THPGenerator_Wrap(at::Generator gen);
 
 // Creates a new Python object for a Generator. The Generator must not already
 // have a PyObject* associated with it.
diff --git a/torch/csrc/THP.h b/torch/csrc/THP.h
index 9c0dd406ee6..146e6c532a7 100644
--- a/torch/csrc/THP.h
+++ b/torch/csrc/THP.h
@@ -5,7 +5,7 @@
 #include <TH/TH.h>
 #include <TH/THTensor.hpp>
 
-#include <torch/csrc/THP_export.h>
+#include <torch/csrc/Export.h>
 
 // Back-compatibility macros, Thanks to http://cx-oracle.sourceforge.net/
 // define PyInt_* macros for Python 3.x.  NB: We must include Python.h first,
diff --git a/torch/csrc/THP_export.h b/torch/csrc/THP_export.h
deleted file mode 100644
index 991d50d9292..00000000000
--- a/torch/csrc/THP_export.h
+++ /dev/null
@@ -1,17 +0,0 @@
-#ifndef THP_EXPORT_H
-#define THP_EXPORT_H
-
-#ifdef _WIN32
-# ifdef THP_BUILD_MAIN_LIB
-#  define THP_API extern __declspec(dllexport)
-#  define THP_CLASS __declspec(dllexport)
-# else
-#  define THP_API extern __declspec(dllimport)
-#  define THP_CLASS __declspec(dllimport)
-# endif
-#else
-# define THP_API extern
-# define THP_CLASS
-#endif
-
-#endif
diff --git a/torch/csrc/WindowsTorchApiMacro.h b/torch/csrc/WindowsTorchApiMacro.h
index 44ae3b3b818..8d829006f38 100644
--- a/torch/csrc/WindowsTorchApiMacro.h
+++ b/torch/csrc/WindowsTorchApiMacro.h
@@ -1,9 +1,2 @@
 #pragma once
-
-#include <c10/macros/Export.h>
-
-#ifdef _WIN32
-#define TORCH_PYTHON_API
-#else
-#define TORCH_PYTHON_API TORCH_API
-#endif
+#include <torch/csrc/Export.h>
diff --git a/torch/csrc/autograd/python_saved_variable_hooks.h b/torch/csrc/autograd/python_saved_variable_hooks.h
index 145301f6f7e..f8c215f555e 100644
--- a/torch/csrc/autograd/python_saved_variable_hooks.h
+++ b/torch/csrc/autograd/python_saved_variable_hooks.h
@@ -4,7 +4,7 @@
 #include <torch/csrc/autograd/python_variable.h>
 #include <torch/csrc/autograd/saved_variable_hooks.h>
 #include <torch/csrc/python_headers.h>
-#include <torch/csrc/THP_export.h>
+#include <torch/csrc/Export.h>
 #include <ATen/ATen.h>
 
 namespace py = pybind11;
diff --git a/torch/csrc/autograd/python_variable.h b/torch/csrc/autograd/python_variable.h
index 30bfb61bcbf..c3e2c1d0efa 100644
--- a/torch/csrc/autograd/python_variable.h
+++ b/torch/csrc/autograd/python_variable.h
@@ -5,7 +5,7 @@
 #include <ATen/ATen.h>
 
 #include <torch/csrc/autograd/variable.h>
-#include <torch/csrc/THP_export.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/Exceptions.h>
 
 // Python object that backs torch.autograd.Variable
@@ -21,11 +21,11 @@ struct THPVariable {
 
 TORCH_API void registerPythonTensorClass(const std::string& device, PyObject* python_tensor_class);
 
-THP_API PyObject *THPVariableClass;
-THP_API PyObject *ParameterClass;
+TORCH_PYTHON_API extern PyObject *THPVariableClass;
+TORCH_PYTHON_API extern PyObject *ParameterClass;
 
 bool THPVariable_initModule(PyObject *module);
-THP_API PyObject * THPVariable_Wrap(at::TensorBase var);
+TORCH_PYTHON_API PyObject * THPVariable_Wrap(at::TensorBase var);
 
 static inline bool THPVariable_CheckTypeExact(PyTypeObject* tp) {
   // Check that a python object is a `Tensor`, but not a `Tensor` subclass.
@@ -61,4 +61,4 @@ inline const at::Tensor& THPVariable_Unpack(PyObject* obj) {
   return THPVariable_Unpack(reinterpret_cast<THPVariable*>(obj));
 }
 
-THP_API c10::impl::PyInterpreter* getPyInterpreter();
+TORCH_PYTHON_API c10::impl::PyInterpreter* getPyInterpreter();
diff --git a/torch/csrc/autograd/python_variable_indexing.cpp b/torch/csrc/autograd/python_variable_indexing.cpp
index 19e74030fff..6393b82bd26 100644
--- a/torch/csrc/autograd/python_variable_indexing.cpp
+++ b/torch/csrc/autograd/python_variable_indexing.cpp
@@ -2,7 +2,7 @@
 
 #include <torch/csrc/DynamicTypes.h>
 #include <torch/csrc/Exceptions.h>
-#include <torch/csrc/THP_export.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/autograd/function.h>
 #include <torch/csrc/autograd/python_variable.h>
 #include <torch/csrc/autograd/utils/wrap_outputs.h>
diff --git a/torch/csrc/distributed/c10d/comm.hpp b/torch/csrc/distributed/c10d/comm.hpp
index 4690c355ce7..43ffdb07d78 100644
--- a/torch/csrc/distributed/c10d/comm.hpp
+++ b/torch/csrc/distributed/c10d/comm.hpp
@@ -3,6 +3,7 @@
 #include <ATen/ATen.h>
 #include <ATen/core/ivalue.h>
 #include <c10d/ProcessGroup.hpp>
+#include <torch/csrc/Export.h>
 
 namespace c10d {
 
@@ -85,7 +86,7 @@ class TORCH_API GradBucket {
 // Requires implementing 1) `runHook` method that communicates gradients
 // asynchronously, and 2) `parseHookResult` method that converts the hook
 // result into a tensor.
-class TORCH_PYTHON_API CommHookInterface {
+class TORCH_API CommHookInterface {
  public:
   virtual ~CommHookInterface() = default;
 
@@ -121,9 +122,8 @@ inline at::Tensor parseCppCommHookResult(
 
 // This CppCommHook interface only requires implementing runHook method that
 // potentially uses a state.
-// Still need TORCH_PYTHON_API instead of TORCH_API to support Windows platform.
 template <typename T>
-class TORCH_PYTHON_API CppCommHookInterface : public CommHookInterface {
+class CppCommHookInterface : public CommHookInterface {
  public:
   explicit CppCommHookInterface(T& state) : state_(state) {}
 
diff --git a/torch/csrc/generic/Storage.h b/torch/csrc/generic/Storage.h
index 572f84af0de..a2063593207 100644
--- a/torch/csrc/generic/Storage.h
+++ b/torch/csrc/generic/Storage.h
@@ -4,7 +4,7 @@
 
 #include <torch/csrc/StorageDefs.h>
 
-THP_API PyObject * THPStorage_(New)(c10::intrusive_ptr<c10::StorageImpl> ptr);
+TORCH_PYTHON_API PyObject * THPStorage_(New)(c10::intrusive_ptr<c10::StorageImpl> ptr);
 extern PyObject *THPStorageClass;
 
 #include <torch/csrc/Types.h>
diff --git a/torch/csrc/utils.h b/torch/csrc/utils.h
index cf9717c815b..fb0beb74101 100644
--- a/torch/csrc/utils.h
+++ b/torch/csrc/utils.h
@@ -126,8 +126,8 @@
 #define THPUtils_assert(cond, ...) THPUtils_assertRet(nullptr, cond, __VA_ARGS__)
 #define THPUtils_assertRet(value, cond, ...)                                   \
 if (THP_EXPECT(!(cond), 0)) { THPUtils_setError(__VA_ARGS__); return value; }
-THP_API void THPUtils_setError(const char *format, ...);
-THP_API void THPUtils_invalidArguments(
+TORCH_PYTHON_API void THPUtils_setError(const char *format, ...);
+TORCH_PYTHON_API void THPUtils_invalidArguments(
         PyObject *given_args, PyObject *given_kwargs,
         const char *function_name, size_t num_options, ...);