diff --git a/BUILD.bazel b/BUILD.bazel
index d8ed3cf73df662..47ba6b33028cce 100644
--- a/BUILD.bazel
+++ b/BUILD.bazel
@@ -524,7 +524,6 @@ cc_library(
     name = "aten_headers",
     hdrs = [
         "torch/csrc/Export.h",
-        "torch/csrc/WindowsTorchApiMacro.h",
         "torch/csrc/jit/frontend/function_schema_parser.h",
     ] + glob([
         "aten/src/**/*.h",
diff --git a/aten/src/ATen/SequenceNumber.h b/aten/src/ATen/SequenceNumber.h
index 44ae32a1e2d7db..dacfa551bda34d 100644
--- a/aten/src/ATen/SequenceNumber.h
+++ b/aten/src/ATen/SequenceNumber.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <cstdint>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <c10/macros/Export.h>
 
 namespace at {
 
diff --git a/aten/src/ATen/TracerMode.h b/aten/src/ATen/TracerMode.h
index 2b4527bd9cb995..154954ea4e34d2 100644
--- a/aten/src/ATen/TracerMode.h
+++ b/aten/src/ATen/TracerMode.h
@@ -2,7 +2,7 @@
 
 #include <c10/core/impl/LocalDispatchKeySet.h>
 #include <c10/macros/Macros.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <c10/macros/Export.h>
 
 // NOTE [Tracing Mode Switches]
 //
diff --git a/aten/src/ATen/TypeDefault.h b/aten/src/ATen/TypeDefault.h
index 7b5d77ba4d22ca..2297eaefe577ac 100644
--- a/aten/src/ATen/TypeDefault.h
+++ b/aten/src/ATen/TypeDefault.h
@@ -6,7 +6,7 @@
 #include <c10/core/MemoryFormat.h>
 #include <c10/util/ArrayRef.h>
 #include <c10/util/intrusive_ptr.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <c10/macros/Export.h>
 #include <ATen/Dimname.h>
 
 namespace c10 {
diff --git a/aten/src/ATen/core/Dict.h b/aten/src/ATen/core/Dict.h
index 1ad1f1bfda2c11..d5e127bf9edd6b 100644
--- a/aten/src/ATen/core/Dict.h
+++ b/aten/src/ATen/core/Dict.h
@@ -1,12 +1,12 @@
 #pragma once
 
 #include <c10/macros/Macros.h>
+#include <c10/macros/Export.h>
 #include <c10/util/TypeTraits.h>
 #include <c10/util/TypeList.h>
 #include <c10/util/intrusive_ptr.h>
 #include <c10/util/order_preserving_flat_hash_map.h>
 #include <c10/util/Optional.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
 #include <ATen/core/TensorBody.h>
 
 namespace c10 {
diff --git a/aten/src/ATen/core/List.h b/aten/src/ATen/core/List.h
index 6947d0185a185c..e0581daaa26db4 100644
--- a/aten/src/ATen/core/List.h
+++ b/aten/src/ATen/core/List.h
@@ -2,12 +2,12 @@
 
 #include <ATen/core/ivalue_to.h>
 #include <c10/macros/Macros.h>
+#include <c10/macros/Export.h>
 #include <c10/util/TypeTraits.h>
 #include <c10/util/TypeList.h>
 #include <c10/util/intrusive_ptr.h>
 #include <c10/util/ArrayRef.h>
 #include <c10/util/Optional.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
 #include <vector>
 
 namespace at {
diff --git a/aten/src/ATen/core/ivalue.h b/aten/src/ATen/core/ivalue.h
index 2e76ec8704b50d..428c2287e7a1a2 100644
--- a/aten/src/ATen/core/ivalue.h
+++ b/aten/src/ATen/core/ivalue.h
@@ -6,7 +6,7 @@
 #include <c10/util/C++17.h>
 #include <c10/util/MaybeOwned.h>
 #include <c10/util/intrusive_ptr.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <c10/macros/Export.h>
 #include <typeindex>
 
 namespace torch {
diff --git a/aten/src/ATen/templates/TensorBody.h b/aten/src/ATen/templates/TensorBody.h
index a8d1b7a5ea3ffd..f4a97e1e7b2859 100644
--- a/aten/src/ATen/templates/TensorBody.h
+++ b/aten/src/ATen/templates/TensorBody.h
@@ -19,12 +19,12 @@
 #include <c10/util/MaybeOwned.h>
 #include <c10/util/Optional.h>
 #include <c10/util/intrusive_ptr.h>
+#include <c10/macros/Export.h>
 #include <ATen/core/DeprecatedTypePropertiesRegistry.h>
 #include <ATen/core/DeprecatedTypeProperties.h>
 #include <ATen/core/NamedTensor.h>
 #include <ATen/core/QuantizerBase.h>
 #include <ATen/core/TensorBase.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
 
 namespace c10{
 template<class T> class List;
diff --git a/test/cpp/jit/torch_python_test.cpp b/test/cpp/jit/torch_python_test.cpp
index 76040da3416510..74b93436688c7d 100644
--- a/test/cpp/jit/torch_python_test.cpp
+++ b/test/cpp/jit/torch_python_test.cpp
@@ -1,6 +1,6 @@
 #include <ATen/core/ivalue.h>
 #include <c10/util/Exception.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/api/module.h>
 #include <torch/script.h>
 
diff --git a/tools/autograd/templates/VariableType.h b/tools/autograd/templates/VariableType.h
index 333e8a0d7ada5c..0a579cb6f45885 100644
--- a/tools/autograd/templates/VariableType.h
+++ b/tools/autograd/templates/VariableType.h
@@ -6,7 +6,7 @@
 
 #include <c10/util/intrusive_ptr.h>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/autograd/autograd_not_implemented_fallback.h>
 
 #include <cstdint> // for size_t
diff --git a/tools/build_variables.bzl b/tools/build_variables.bzl
index 3e1f24bc449aa9..aace0e14be1c51 100644
--- a/tools/build_variables.bzl
+++ b/tools/build_variables.bzl
@@ -72,7 +72,6 @@ def libtorch_generated_sources(gencode_pattern):
 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",
     "torch/csrc/jit/serialization/source_range_serialization.h",
diff --git a/torch/csrc/CudaIPCTypes.h b/torch/csrc/CudaIPCTypes.h
index 50398549e134b5..941716e6fbfa4d 100644
--- a/torch/csrc/CudaIPCTypes.h
+++ b/torch/csrc/CudaIPCTypes.h
@@ -1,6 +1,6 @@
 #pragma once
 #ifdef USE_CUDA
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <c10/core/Allocator.h>
 #include <c10/cuda/CUDACachingAllocator.h>
 #include <c10/cuda/CUDAException.h>
diff --git a/torch/csrc/Device.h b/torch/csrc/Device.h
index 32868120c06a1b..d85b92b1f2dc49 100644
--- a/torch/csrc/Device.h
+++ b/torch/csrc/Device.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/csrc/python_headers.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <ATen/Device.h>
 
diff --git a/torch/csrc/Dtype.h b/torch/csrc/Dtype.h
index 87b7d3993d54fd..7cecec68b1f0a5 100644
--- a/torch/csrc/Dtype.h
+++ b/torch/csrc/Dtype.h
@@ -2,7 +2,7 @@
 
 #include <ATen/ATen.h>
 #include <torch/csrc/python_headers.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 const int DTYPE_NAME_LEN = 64;
 
diff --git a/torch/csrc/WindowsTorchApiMacro.h b/torch/csrc/WindowsTorchApiMacro.h
deleted file mode 100644
index 8d829006f38e94..00000000000000
--- a/torch/csrc/WindowsTorchApiMacro.h
+++ /dev/null
@@ -1,2 +0,0 @@
-#pragma once
-#include <torch/csrc/Export.h>
diff --git a/torch/csrc/api/include/torch/cuda.h b/torch/csrc/api/include/torch/cuda.h
index a7e063b90af97c..3db1e128825817 100644
--- a/torch/csrc/api/include/torch/cuda.h
+++ b/torch/csrc/api/include/torch/cuda.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <cstdint>
 #include <cstddef>
diff --git a/torch/csrc/api/include/torch/data/datasets/mnist.h b/torch/csrc/api/include/torch/data/datasets/mnist.h
index 0b0c18af611f7c..3187eae9da176a 100644
--- a/torch/csrc/api/include/torch/data/datasets/mnist.h
+++ b/torch/csrc/api/include/torch/data/datasets/mnist.h
@@ -4,7 +4,7 @@
 #include <torch/data/example.h>
 #include <torch/types.h>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <cstddef>
 #include <string>
diff --git a/torch/csrc/api/include/torch/data/samplers/base.h b/torch/csrc/api/include/torch/data/samplers/base.h
index d0e00bb9e5076e..5469a8ea07a9ec 100644
--- a/torch/csrc/api/include/torch/data/samplers/base.h
+++ b/torch/csrc/api/include/torch/data/samplers/base.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/types.h>
 
 #include <cstddef>
diff --git a/torch/csrc/api/include/torch/data/samplers/custom_batch_request.h b/torch/csrc/api/include/torch/data/samplers/custom_batch_request.h
index a7c38977431bff..ec48c4450482ca 100644
--- a/torch/csrc/api/include/torch/data/samplers/custom_batch_request.h
+++ b/torch/csrc/api/include/torch/data/samplers/custom_batch_request.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <cstddef>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 namespace torch {
 namespace data {
diff --git a/torch/csrc/api/include/torch/data/samplers/distributed.h b/torch/csrc/api/include/torch/data/samplers/distributed.h
index d0f4bf25fb67cd..82eed0913100d1 100644
--- a/torch/csrc/api/include/torch/data/samplers/distributed.h
+++ b/torch/csrc/api/include/torch/data/samplers/distributed.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/data/samplers/base.h>
 
 #include <cstddef>
diff --git a/torch/csrc/api/include/torch/data/samplers/random.h b/torch/csrc/api/include/torch/data/samplers/random.h
index efdd992b84cd58..b5bc8ea0a11026 100644
--- a/torch/csrc/api/include/torch/data/samplers/random.h
+++ b/torch/csrc/api/include/torch/data/samplers/random.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/data/samplers/base.h>
 #include <torch/types.h>
 
diff --git a/torch/csrc/api/include/torch/data/samplers/sequential.h b/torch/csrc/api/include/torch/data/samplers/sequential.h
index 327c397d14159f..711d8421b23045 100644
--- a/torch/csrc/api/include/torch/data/samplers/sequential.h
+++ b/torch/csrc/api/include/torch/data/samplers/sequential.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/data/samplers/base.h>
 #include <torch/types.h>
 
diff --git a/torch/csrc/api/include/torch/data/samplers/stream.h b/torch/csrc/api/include/torch/data/samplers/stream.h
index fe6b206d3928fd..8d30c9e2e0d08f 100644
--- a/torch/csrc/api/include/torch/data/samplers/stream.h
+++ b/torch/csrc/api/include/torch/data/samplers/stream.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/data/samplers/base.h>
 #include <torch/data/samplers/custom_batch_request.h>
 #include <torch/types.h>
diff --git a/torch/csrc/api/include/torch/enum.h b/torch/csrc/api/include/torch/enum.h
index e19b0ebe34ccfc..529a48a6fe1199 100644
--- a/torch/csrc/api/include/torch/enum.h
+++ b/torch/csrc/api/include/torch/enum.h
@@ -5,7 +5,7 @@
 #include <ATen/core/Reduction.h>
 #include <c10/util/Exception.h>
 #include <c10/util/variant.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #define TORCH_ENUM_DECLARE(name) \
 namespace torch { \
diff --git a/torch/csrc/api/include/torch/jit.h b/torch/csrc/api/include/torch/jit.h
index 074ada4d69248c..f2364a14b1a75d 100644
--- a/torch/csrc/api/include/torch/jit.h
+++ b/torch/csrc/api/include/torch/jit.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/api/module.h>
 
 #include <string>
diff --git a/torch/csrc/api/include/torch/nn/init.h b/torch/csrc/api/include/torch/nn/init.h
index 1d6b5ec0acbba8..20a773d16a7675 100644
--- a/torch/csrc/api/include/torch/nn/init.h
+++ b/torch/csrc/api/include/torch/nn/init.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/enum.h>
 #include <torch/types.h>
 
diff --git a/torch/csrc/api/include/torch/nn/modules/activation.h b/torch/csrc/api/include/torch/nn/modules/activation.h
index 865914ec887b25..f36219e73c3d24 100644
--- a/torch/csrc/api/include/torch/nn/modules/activation.h
+++ b/torch/csrc/api/include/torch/nn/modules/activation.h
@@ -6,7 +6,7 @@
 #include <torch/nn/modules/common.h>
 #include <torch/nn/modules/linear.h>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 namespace torch {
 namespace nn {
diff --git a/torch/csrc/api/include/torch/nn/modules/container/functional.h b/torch/csrc/api/include/torch/nn/modules/container/functional.h
index eef942318b9b0f..024ef16d210d21 100644
--- a/torch/csrc/api/include/torch/nn/modules/container/functional.h
+++ b/torch/csrc/api/include/torch/nn/modules/container/functional.h
@@ -4,7 +4,7 @@
 #include <torch/nn/cloneable.h>
 #include <torch/nn/pimpl.h>
 #include <torch/types.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <functional>
 #include <utility>
diff --git a/torch/csrc/api/include/torch/nn/modules/conv.h b/torch/csrc/api/include/torch/nn/modules/conv.h
index f7c4cbef77875b..a644b52a9c7cc9 100644
--- a/torch/csrc/api/include/torch/nn/modules/conv.h
+++ b/torch/csrc/api/include/torch/nn/modules/conv.h
@@ -13,7 +13,7 @@
 #include <torch/nn/pimpl.h>
 #include <torch/types.h>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <cstddef>
 #include <vector>
diff --git a/torch/csrc/api/include/torch/nn/modules/distance.h b/torch/csrc/api/include/torch/nn/modules/distance.h
index bf0d2c39a41b36..053800d6ec7805 100644
--- a/torch/csrc/api/include/torch/nn/modules/distance.h
+++ b/torch/csrc/api/include/torch/nn/modules/distance.h
@@ -6,7 +6,7 @@
 #include <torch/nn/pimpl.h>
 #include <torch/types.h>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 namespace torch {
 namespace nn {
diff --git a/torch/csrc/api/include/torch/nn/modules/dropout.h b/torch/csrc/api/include/torch/nn/modules/dropout.h
index d10ed95f2e7f24..e6f2ef4ea61ad3 100644
--- a/torch/csrc/api/include/torch/nn/modules/dropout.h
+++ b/torch/csrc/api/include/torch/nn/modules/dropout.h
@@ -5,7 +5,7 @@
 #include <torch/nn/pimpl.h>
 #include <torch/types.h>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <cstddef>
 #include <vector>
diff --git a/torch/csrc/api/include/torch/nn/modules/loss.h b/torch/csrc/api/include/torch/nn/modules/loss.h
index 0d94697a5fa38a..9d6958b8ca3f41 100644
--- a/torch/csrc/api/include/torch/nn/modules/loss.h
+++ b/torch/csrc/api/include/torch/nn/modules/loss.h
@@ -7,7 +7,7 @@
 #include <torch/nn/pimpl.h>
 #include <torch/types.h>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <cstddef>
 #include <vector>
diff --git a/torch/csrc/api/include/torch/nn/modules/padding.h b/torch/csrc/api/include/torch/nn/modules/padding.h
index c873e2badfc634..5ad3087e4011c1 100644
--- a/torch/csrc/api/include/torch/nn/modules/padding.h
+++ b/torch/csrc/api/include/torch/nn/modules/padding.h
@@ -4,7 +4,7 @@
 #include <torch/nn/cloneable.h>
 #include <torch/nn/functional/padding.h>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 namespace torch {
 namespace nn {
diff --git a/torch/csrc/api/include/torch/nn/modules/pixelshuffle.h b/torch/csrc/api/include/torch/nn/modules/pixelshuffle.h
index 1bf27562143997..f24183ce364175 100644
--- a/torch/csrc/api/include/torch/nn/modules/pixelshuffle.h
+++ b/torch/csrc/api/include/torch/nn/modules/pixelshuffle.h
@@ -4,7 +4,7 @@
 #include <torch/nn/functional/pixelshuffle.h>
 #include <torch/nn/options/pixelshuffle.h>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 namespace torch {
 namespace nn {
diff --git a/torch/csrc/api/include/torch/nn/modules/pooling.h b/torch/csrc/api/include/torch/nn/modules/pooling.h
index 217110606d274a..bca96503b617e4 100644
--- a/torch/csrc/api/include/torch/nn/modules/pooling.h
+++ b/torch/csrc/api/include/torch/nn/modules/pooling.h
@@ -6,7 +6,7 @@
 #include <torch/nn/functional/pooling.h>
 #include <torch/nn/modules/common.h>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 namespace torch {
 namespace nn {
diff --git a/torch/csrc/api/include/torch/nn/modules/upsampling.h b/torch/csrc/api/include/torch/nn/modules/upsampling.h
index d859242bdab05e..4ac0b5177acaef 100644
--- a/torch/csrc/api/include/torch/nn/modules/upsampling.h
+++ b/torch/csrc/api/include/torch/nn/modules/upsampling.h
@@ -6,7 +6,7 @@
 #include <torch/nn/pimpl.h>
 #include <torch/types.h>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <cstddef>
 #include <ostream>
diff --git a/torch/csrc/api/include/torch/nn/options/activation.h b/torch/csrc/api/include/torch/nn/options/activation.h
index 4cd66ff443cde5..b06e501060e8c8 100644
--- a/torch/csrc/api/include/torch/nn/options/activation.h
+++ b/torch/csrc/api/include/torch/nn/options/activation.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/arg.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/types.h>
 
 namespace torch {
diff --git a/torch/csrc/api/include/torch/nn/options/adaptive.h b/torch/csrc/api/include/torch/nn/options/adaptive.h
index 7f73dbbd5a2583..3be918184af323 100644
--- a/torch/csrc/api/include/torch/nn/options/adaptive.h
+++ b/torch/csrc/api/include/torch/nn/options/adaptive.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/arg.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/types.h>
 
 namespace torch {
diff --git a/torch/csrc/api/include/torch/nn/options/batchnorm.h b/torch/csrc/api/include/torch/nn/options/batchnorm.h
index ea81afdff2f799..8f8412b14133f6 100644
--- a/torch/csrc/api/include/torch/nn/options/batchnorm.h
+++ b/torch/csrc/api/include/torch/nn/options/batchnorm.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/arg.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/types.h>
 
 namespace torch {
diff --git a/torch/csrc/api/include/torch/nn/options/conv.h b/torch/csrc/api/include/torch/nn/options/conv.h
index 86cdcb7ee71f67..adffd4e051019d 100644
--- a/torch/csrc/api/include/torch/nn/options/conv.h
+++ b/torch/csrc/api/include/torch/nn/options/conv.h
@@ -2,7 +2,7 @@
 
 #include <torch/arg.h>
 #include <torch/enum.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/expanding_array.h>
 #include <torch/types.h>
 
diff --git a/torch/csrc/api/include/torch/nn/options/distance.h b/torch/csrc/api/include/torch/nn/options/distance.h
index debb8c8c414801..302533d3cdf29a 100644
--- a/torch/csrc/api/include/torch/nn/options/distance.h
+++ b/torch/csrc/api/include/torch/nn/options/distance.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/arg.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/types.h>
 
 namespace torch {
diff --git a/torch/csrc/api/include/torch/nn/options/dropout.h b/torch/csrc/api/include/torch/nn/options/dropout.h
index 511690cd79907b..57f4bc9b5e73b8 100644
--- a/torch/csrc/api/include/torch/nn/options/dropout.h
+++ b/torch/csrc/api/include/torch/nn/options/dropout.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/arg.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/types.h>
 
 namespace torch {
diff --git a/torch/csrc/api/include/torch/nn/options/embedding.h b/torch/csrc/api/include/torch/nn/options/embedding.h
index 5f2e7ea9fb4c04..c304f04ead56ef 100644
--- a/torch/csrc/api/include/torch/nn/options/embedding.h
+++ b/torch/csrc/api/include/torch/nn/options/embedding.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/arg.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/types.h>
 #include <torch/enum.h>
 
diff --git a/torch/csrc/api/include/torch/nn/options/fold.h b/torch/csrc/api/include/torch/nn/options/fold.h
index ce99932e2df0ee..00edd84dfaebf8 100644
--- a/torch/csrc/api/include/torch/nn/options/fold.h
+++ b/torch/csrc/api/include/torch/nn/options/fold.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/arg.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/expanding_array.h>
 #include <torch/types.h>
 
diff --git a/torch/csrc/api/include/torch/nn/options/instancenorm.h b/torch/csrc/api/include/torch/nn/options/instancenorm.h
index 0ef2f7b8b1a512..ae9b3b6089a4da 100644
--- a/torch/csrc/api/include/torch/nn/options/instancenorm.h
+++ b/torch/csrc/api/include/torch/nn/options/instancenorm.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/arg.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/nn/options/batchnorm.h>
 #include <torch/types.h>
 
diff --git a/torch/csrc/api/include/torch/nn/options/linear.h b/torch/csrc/api/include/torch/nn/options/linear.h
index 24fdbca633217f..742a887e285b4c 100644
--- a/torch/csrc/api/include/torch/nn/options/linear.h
+++ b/torch/csrc/api/include/torch/nn/options/linear.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/arg.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/types.h>
 #include <c10/util/variant.h>
 
diff --git a/torch/csrc/api/include/torch/nn/options/loss.h b/torch/csrc/api/include/torch/nn/options/loss.h
index 1479de571d13eb..4961f4b8cb1242 100644
--- a/torch/csrc/api/include/torch/nn/options/loss.h
+++ b/torch/csrc/api/include/torch/nn/options/loss.h
@@ -2,7 +2,7 @@
 
 #include <torch/arg.h>
 #include <torch/enum.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/types.h>
 
 namespace torch {
diff --git a/torch/csrc/api/include/torch/nn/options/normalization.h b/torch/csrc/api/include/torch/nn/options/normalization.h
index c47f19129f70df..05ed1dcda954ab 100644
--- a/torch/csrc/api/include/torch/nn/options/normalization.h
+++ b/torch/csrc/api/include/torch/nn/options/normalization.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/arg.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/types.h>
 #include <vector>
 
diff --git a/torch/csrc/api/include/torch/nn/options/padding.h b/torch/csrc/api/include/torch/nn/options/padding.h
index 3265cd6438cefb..36c12cc16e3a90 100644
--- a/torch/csrc/api/include/torch/nn/options/padding.h
+++ b/torch/csrc/api/include/torch/nn/options/padding.h
@@ -3,7 +3,7 @@
 #include <c10/util/variant.h>
 #include <torch/arg.h>
 #include <torch/enum.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/expanding_array.h>
 #include <torch/types.h>
 
diff --git a/torch/csrc/api/include/torch/nn/options/pixelshuffle.h b/torch/csrc/api/include/torch/nn/options/pixelshuffle.h
index e28e0053e98b8c..89edb2a8bfa667 100644
--- a/torch/csrc/api/include/torch/nn/options/pixelshuffle.h
+++ b/torch/csrc/api/include/torch/nn/options/pixelshuffle.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/arg.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/types.h>
 
 namespace torch {
diff --git a/torch/csrc/api/include/torch/nn/options/pooling.h b/torch/csrc/api/include/torch/nn/options/pooling.h
index 3bceb46d1a3b68..38e20bdd02ce45 100644
--- a/torch/csrc/api/include/torch/nn/options/pooling.h
+++ b/torch/csrc/api/include/torch/nn/options/pooling.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/arg.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/expanding_array.h>
 #include <torch/types.h>
 
diff --git a/torch/csrc/api/include/torch/nn/options/rnn.h b/torch/csrc/api/include/torch/nn/options/rnn.h
index 09bbfe0fa2f4d4..eb1571ef8224b5 100644
--- a/torch/csrc/api/include/torch/nn/options/rnn.h
+++ b/torch/csrc/api/include/torch/nn/options/rnn.h
@@ -2,7 +2,7 @@
 
 #include <torch/arg.h>
 #include <torch/enum.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/types.h>
 
 namespace torch {
diff --git a/torch/csrc/api/include/torch/nn/options/transformer.h b/torch/csrc/api/include/torch/nn/options/transformer.h
index f5129734617ac2..bbdea49b3eaa35 100644
--- a/torch/csrc/api/include/torch/nn/options/transformer.h
+++ b/torch/csrc/api/include/torch/nn/options/transformer.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/arg.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/types.h>
 #include <torch/enum.h>
 
diff --git a/torch/csrc/api/include/torch/nn/options/transformercoder.h b/torch/csrc/api/include/torch/nn/options/transformercoder.h
index 6c5975ef5c12ac..d9e59b35eb1cb7 100644
--- a/torch/csrc/api/include/torch/nn/options/transformercoder.h
+++ b/torch/csrc/api/include/torch/nn/options/transformercoder.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/arg.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/types.h>
 #include <torch/enum.h>
 
diff --git a/torch/csrc/api/include/torch/nn/options/transformerlayer.h b/torch/csrc/api/include/torch/nn/options/transformerlayer.h
index 5f1930e4ed657b..aaad0040b3b6bd 100644
--- a/torch/csrc/api/include/torch/nn/options/transformerlayer.h
+++ b/torch/csrc/api/include/torch/nn/options/transformerlayer.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/arg.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/types.h>
 #include <torch/enum.h>
 
diff --git a/torch/csrc/api/include/torch/nn/options/upsampling.h b/torch/csrc/api/include/torch/nn/options/upsampling.h
index ce596635212f49..85ee9f47f1fa22 100644
--- a/torch/csrc/api/include/torch/nn/options/upsampling.h
+++ b/torch/csrc/api/include/torch/nn/options/upsampling.h
@@ -3,7 +3,7 @@
 #include <c10/util/variant.h>
 #include <torch/arg.h>
 #include <torch/enum.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/expanding_array.h>
 #include <torch/types.h>
 
diff --git a/torch/csrc/api/include/torch/nn/options/vision.h b/torch/csrc/api/include/torch/nn/options/vision.h
index b62698174afbff..1b06ab252b789a 100644
--- a/torch/csrc/api/include/torch/nn/options/vision.h
+++ b/torch/csrc/api/include/torch/nn/options/vision.h
@@ -2,7 +2,7 @@
 
 #include <torch/arg.h>
 #include <torch/enum.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/types.h>
 
 namespace torch {
diff --git a/torch/csrc/api/include/torch/nn/utils/clip_grad.h b/torch/csrc/api/include/torch/nn/utils/clip_grad.h
index a740aa9ad282b5..23b94f8466f254 100644
--- a/torch/csrc/api/include/torch/nn/utils/clip_grad.h
+++ b/torch/csrc/api/include/torch/nn/utils/clip_grad.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 namespace torch {
 namespace nn {
diff --git a/torch/csrc/api/include/torch/nn/utils/convert_parameters.h b/torch/csrc/api/include/torch/nn/utils/convert_parameters.h
index 212ec8bba2914d..8b730e0f0c00aa 100644
--- a/torch/csrc/api/include/torch/nn/utils/convert_parameters.h
+++ b/torch/csrc/api/include/torch/nn/utils/convert_parameters.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/types.h>
 
 namespace torch {
diff --git a/torch/csrc/api/include/torch/optim/optimizer.h b/torch/csrc/api/include/torch/optim/optimizer.h
index 71311011894027..010cc15e38a1c3 100644
--- a/torch/csrc/api/include/torch/optim/optimizer.h
+++ b/torch/csrc/api/include/torch/optim/optimizer.h
@@ -4,7 +4,7 @@
 #include <c10/util/flat_hash_map.h>
 #include <c10/util/Exception.h>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/arg.h>
 
 #include <algorithm>
diff --git a/torch/csrc/api/include/torch/optim/schedulers/lr_scheduler.h b/torch/csrc/api/include/torch/optim/schedulers/lr_scheduler.h
index 6ae21dca849bdb..21254ca49fee55 100644
--- a/torch/csrc/api/include/torch/optim/schedulers/lr_scheduler.h
+++ b/torch/csrc/api/include/torch/optim/schedulers/lr_scheduler.h
@@ -2,7 +2,7 @@
 
 #include <torch/optim/optimizer.h>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 namespace torch {
 namespace optim {
diff --git a/torch/csrc/api/include/torch/serialize.h b/torch/csrc/api/include/torch/serialize.h
index b7117feb9dc6fe..f92b6123d13966 100644
--- a/torch/csrc/api/include/torch/serialize.h
+++ b/torch/csrc/api/include/torch/serialize.h
@@ -3,7 +3,7 @@
 #include <c10/util/irange.h>
 #include <torch/serialize/archive.h>
 #include <torch/serialize/tensor.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <utility>
 
diff --git a/torch/csrc/api/include/torch/serialize/input-archive.h b/torch/csrc/api/include/torch/serialize/input-archive.h
index 5055b66bd8c1ec..9b2174c83df3ec 100644
--- a/torch/csrc/api/include/torch/serialize/input-archive.h
+++ b/torch/csrc/api/include/torch/serialize/input-archive.h
@@ -2,7 +2,7 @@
 
 #include <c10/util/Optional.h>
 #include <c10/core/Device.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/types.h>
 #include <torch/csrc/jit/api/module.h>
 
diff --git a/torch/csrc/api/include/torch/serialize/output-archive.h b/torch/csrc/api/include/torch/serialize/output-archive.h
index c125b27da0c86d..59335edf6f253a 100644
--- a/torch/csrc/api/include/torch/serialize/output-archive.h
+++ b/torch/csrc/api/include/torch/serialize/output-archive.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/api/module.h>
 
 #include <iosfwd>
diff --git a/torch/csrc/autograd/InferenceMode.h b/torch/csrc/autograd/InferenceMode.h
index 01727cea523468..4395082bb33173 100644
--- a/torch/csrc/autograd/InferenceMode.h
+++ b/torch/csrc/autograd/InferenceMode.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <c10/core/InferenceMode.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 namespace torch { namespace autograd {
 
diff --git a/torch/csrc/autograd/anomaly_mode.h b/torch/csrc/autograd/anomaly_mode.h
index a4e4210dabfe2f..95251dcbb8e4ac 100644
--- a/torch/csrc/autograd/anomaly_mode.h
+++ b/torch/csrc/autograd/anomaly_mode.h
@@ -2,7 +2,7 @@
 
 #include <string>
 #include <memory>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 namespace torch { namespace autograd {
 
diff --git a/torch/csrc/autograd/engine.h b/torch/csrc/autograd/engine.h
index f957ca14f31219..ae0b329321849a 100644
--- a/torch/csrc/autograd/engine.h
+++ b/torch/csrc/autograd/engine.h
@@ -6,7 +6,7 @@
 #include <ATen/Tensor.h>
 #include <ATen/core/ivalue.h>
 #include <ATen/ThreadLocalState.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/autograd/anomaly_mode.h>
 #include <torch/csrc/autograd/function.h>
 #include <torch/csrc/autograd/functions/basic_ops.h>
diff --git a/torch/csrc/autograd/function_hook.h b/torch/csrc/autograd/function_hook.h
index 61765fae53c19e..c09889af604840 100644
--- a/torch/csrc/autograd/function_hook.h
+++ b/torch/csrc/autograd/function_hook.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <vector>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <ATen/Tensor.h>
 
 // A hook that's called on gradients
diff --git a/torch/csrc/autograd/functions/accumulate_grad.h b/torch/csrc/autograd/functions/accumulate_grad.h
index 383e9aaf8086e3..4d094371cfae2d 100644
--- a/torch/csrc/autograd/functions/accumulate_grad.h
+++ b/torch/csrc/autograd/functions/accumulate_grad.h
@@ -3,7 +3,7 @@
 #include <torch/csrc/autograd/function.h>
 #include <torch/csrc/autograd/variable.h>
 #include <torch/csrc/autograd/utils/grad_layout_contract.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <ATen/BatchedTensorImpl.h>
 
 #include <mutex>
diff --git a/torch/csrc/autograd/functions/basic_ops.h b/torch/csrc/autograd/functions/basic_ops.h
index 8dfe0c3ab1fad0..7831ee71329593 100644
--- a/torch/csrc/autograd/functions/basic_ops.h
+++ b/torch/csrc/autograd/functions/basic_ops.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/autograd/function.h>
 #include <torch/csrc/autograd/variable.h>
 
diff --git a/torch/csrc/autograd/functions/comm.h b/torch/csrc/autograd/functions/comm.h
index 67878be6d2d4a3..1a89919d1d631b 100644
--- a/torch/csrc/autograd/functions/comm.h
+++ b/torch/csrc/autograd/functions/comm.h
@@ -2,7 +2,7 @@
 
 #include <torch/csrc/autograd/function.h>
 #include <torch/csrc/autograd/variable.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <ATen/ATen.h>
 #include <ATen/cuda/CUDAContext.h>
diff --git a/torch/csrc/autograd/functions/tensor.h b/torch/csrc/autograd/functions/tensor.h
index 08daf88c589b95..8fc7a8dc554f3b 100644
--- a/torch/csrc/autograd/functions/tensor.h
+++ b/torch/csrc/autograd/functions/tensor.h
@@ -2,7 +2,7 @@
 
 #include <torch/csrc/autograd/function.h>
 #include <torch/csrc/autograd/variable.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <ATen/TensorGeometry.h>
 #include <ATen/core/DeprecatedTypeProperties.h>
diff --git a/torch/csrc/autograd/functions/utils.h b/torch/csrc/autograd/functions/utils.h
index 331db5d32cb79e..163414bc1a4796 100644
--- a/torch/csrc/autograd/functions/utils.h
+++ b/torch/csrc/autograd/functions/utils.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/autograd/autograd.h>
 #include <torch/csrc/autograd/function.h>
 #include <torch/csrc/autograd/variable.h>
diff --git a/torch/csrc/autograd/grad_mode.h b/torch/csrc/autograd/grad_mode.h
index ee6e5f1beab43b..72b75524179117 100644
--- a/torch/csrc/autograd/grad_mode.h
+++ b/torch/csrc/autograd/grad_mode.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <ATen/core/grad_mode.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 namespace torch { namespace autograd {
 
diff --git a/torch/csrc/autograd/profiler_legacy.h b/torch/csrc/autograd/profiler_legacy.h
index c94366e260c23b..50073e6f9ac34d 100644
--- a/torch/csrc/autograd/profiler_legacy.h
+++ b/torch/csrc/autograd/profiler_legacy.h
@@ -10,7 +10,7 @@
 #include <forward_list>
 #include <tuple>
 #include <ATen/ATen.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/autograd/profiler_utils.h>
 #ifndef _WIN32
 #include <ctime>
diff --git a/torch/csrc/autograd/saved_variable.h b/torch/csrc/autograd/saved_variable.h
index d231f17722f5e6..363b6bd30bda90 100644
--- a/torch/csrc/autograd/saved_variable.h
+++ b/torch/csrc/autograd/saved_variable.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/autograd/forward_grad.h>
 #include <torch/csrc/autograd/saved_variable_hooks.h>
 
diff --git a/torch/csrc/autograd/variable.h b/torch/csrc/autograd/variable.h
index 3a81cfb3e96eab..babf0421373f4d 100644
--- a/torch/csrc/autograd/variable.h
+++ b/torch/csrc/autograd/variable.h
@@ -2,7 +2,7 @@
 
 #include <torch/csrc/utils/python_stub.h>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/autograd/edge.h>
 #include <torch/csrc/autograd/function_hook.h>
 #include <torch/csrc/autograd/cpp_hook.h>
diff --git a/torch/csrc/cuda/comm.h b/torch/csrc/cuda/comm.h
index b2ed512fd39287..7dd99e64141d82 100644
--- a/torch/csrc/cuda/comm.h
+++ b/torch/csrc/cuda/comm.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <ATen/ATen.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <ATen/cuda/ATenCUDAGeneral.h>
 #include <ATen/cuda/CUDAContext.h>
 #include <c10/util/Optional.h>
diff --git a/torch/csrc/distributed/autograd/rpc_messages/autograd_metadata.h b/torch/csrc/distributed/autograd/rpc_messages/autograd_metadata.h
index 41067cac9acc07..1d5aefbd2010a7 100644
--- a/torch/csrc/distributed/autograd/rpc_messages/autograd_metadata.h
+++ b/torch/csrc/distributed/autograd/rpc_messages/autograd_metadata.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <cstdint>
 
 namespace torch {
diff --git a/torch/csrc/distributed/rpc/profiler/remote_profiler_manager.h b/torch/csrc/distributed/rpc/profiler/remote_profiler_manager.h
index d2d68d646f9b61..d85ee5a3930780 100644
--- a/torch/csrc/distributed/rpc/profiler/remote_profiler_manager.h
+++ b/torch/csrc/distributed/rpc/profiler/remote_profiler_manager.h
@@ -1,6 +1,6 @@
 #pragma once
 #include <c10/util/Optional.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/distributed/rpc/types.h>
 #include <mutex>
 #include <unordered_map>
diff --git a/torch/csrc/jit/api/compilation_unit.h b/torch/csrc/jit/api/compilation_unit.h
index 0c07d2639b529e..00d193a9c7bc58 100644
--- a/torch/csrc/jit/api/compilation_unit.h
+++ b/torch/csrc/jit/api/compilation_unit.h
@@ -7,7 +7,7 @@
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/jit/runtime/graph_executor.h>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/utils/memory.h>
 
 #include <ATen/core/function_schema.h>
diff --git a/torch/csrc/jit/api/module.h b/torch/csrc/jit/api/module.h
index 8d37c11f88537f..75b041991c03eb 100644
--- a/torch/csrc/jit/api/module.h
+++ b/torch/csrc/jit/api/module.h
@@ -9,7 +9,7 @@
 #include <torch/csrc/jit/runtime/argument_spec.h>
 #include <torch/csrc/jit/runtime/graph_executor.h>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/api/include/torch/ordered_dict.h>
 #include <torch/csrc/jit/api/compilation_unit.h>
 #include <torch/csrc/utils/memory.h>
diff --git a/torch/csrc/jit/codegen/cuda/arith.h b/torch/csrc/jit/codegen/cuda/arith.h
index ab423619fd9b81..66b9b77e43468a 100644
--- a/torch/csrc/jit/codegen/cuda/arith.h
+++ b/torch/csrc/jit/codegen/cuda/arith.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/ir_interface_nodes.h>
 #include <torch/csrc/jit/codegen/cuda/type.h>
diff --git a/torch/csrc/jit/codegen/cuda/codegen.h b/torch/csrc/jit/codegen/cuda/codegen.h
index 5f9b4f269fbcac..2ffbb872155a79 100644
--- a/torch/csrc/jit/codegen/cuda/codegen.h
+++ b/torch/csrc/jit/codegen/cuda/codegen.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/codegen/cuda/kernel.h>
 
 #include <string>
diff --git a/torch/csrc/jit/codegen/cuda/compute_at.h b/torch/csrc/jit/codegen/cuda/compute_at.h
index 71e3950e083d8f..391225218db9a8 100644
--- a/torch/csrc/jit/codegen/cuda/compute_at.h
+++ b/torch/csrc/jit/codegen/cuda/compute_at.h
@@ -3,7 +3,7 @@
 #include <torch/csrc/jit/codegen/cuda/root_domain_map.h>
 
 #include <c10/util/Exception.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <deque>
 #include <unordered_map>
diff --git a/torch/csrc/jit/codegen/cuda/dispatch.h b/torch/csrc/jit/codegen/cuda/dispatch.h
index e83ac4e4a31b45..e6ceaf077b641d 100644
--- a/torch/csrc/jit/codegen/cuda/dispatch.h
+++ b/torch/csrc/jit/codegen/cuda/dispatch.h
@@ -3,7 +3,7 @@
 #include <torch/csrc/jit/codegen/cuda/utils.h>
 
 #include <c10/util/Exception.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <unordered_map>
 
diff --git a/torch/csrc/jit/codegen/cuda/expr_evaluator.h b/torch/csrc/jit/codegen/cuda/expr_evaluator.h
index 063737af793d45..ced4b59a78310a 100644
--- a/torch/csrc/jit/codegen/cuda/expr_evaluator.h
+++ b/torch/csrc/jit/codegen/cuda/expr_evaluator.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/codegen/cuda/ir_interface_nodes.h>
 #include <torch/csrc/jit/codegen/cuda/iter_visitor.h>
 
diff --git a/torch/csrc/jit/codegen/cuda/fusion.h b/torch/csrc/jit/codegen/cuda/fusion.h
index 1047ce6a916a50..2256b2278f575c 100644
--- a/torch/csrc/jit/codegen/cuda/fusion.h
+++ b/torch/csrc/jit/codegen/cuda/fusion.h
@@ -2,7 +2,7 @@
 
 #include <ATen/core/ivalue.h>
 #include <c10/util/Exception.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/ir_base_nodes.h>
 #include <torch/csrc/jit/codegen/cuda/iter_visitor.h>
diff --git a/torch/csrc/jit/codegen/cuda/index_reference_replay.h b/torch/csrc/jit/codegen/cuda/index_reference_replay.h
index 06d0c6eabb9b2c..c4626213e76b1a 100644
--- a/torch/csrc/jit/codegen/cuda/index_reference_replay.h
+++ b/torch/csrc/jit/codegen/cuda/index_reference_replay.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/compute_at_map.h>
 #include <torch/csrc/jit/codegen/cuda/index_compute.h>
diff --git a/torch/csrc/jit/codegen/cuda/instrumentation.cpp b/torch/csrc/jit/codegen/cuda/instrumentation.cpp
index b7d7beb05bb380..52e16b3a7afe9d 100644
--- a/torch/csrc/jit/codegen/cuda/instrumentation.cpp
+++ b/torch/csrc/jit/codegen/cuda/instrumentation.cpp
@@ -1,6 +1,6 @@
 #include <torch/csrc/jit/codegen/cuda/instrumentation.h>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #ifdef _WIN32
 #include <c10/util/win32-headers.h>
diff --git a/torch/csrc/jit/codegen/cuda/interface.h b/torch/csrc/jit/codegen/cuda/interface.h
index 2faf8cf0864c81..eebe6185661d13 100644
--- a/torch/csrc/jit/codegen/cuda/interface.h
+++ b/torch/csrc/jit/codegen/cuda/interface.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/jit/runtime/profiling_record.h>
 
diff --git a/torch/csrc/jit/codegen/cuda/ir_base_nodes.h b/torch/csrc/jit/codegen/cuda/ir_base_nodes.h
index ca100abca0c504..2e0fa0885bd6b5 100644
--- a/torch/csrc/jit/codegen/cuda/ir_base_nodes.h
+++ b/torch/csrc/jit/codegen/cuda/ir_base_nodes.h
@@ -3,7 +3,7 @@
 #include <c10/core/ScalarType.h>
 #include <c10/util/Exception.h>
 #include <c10/util/Optional.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/type.h>
 #include <torch/csrc/jit/codegen/cuda/utils.h>
diff --git a/torch/csrc/jit/codegen/cuda/ir_cloner.h b/torch/csrc/jit/codegen/cuda/ir_cloner.h
index b244231325204d..3d48280b93adbc 100644
--- a/torch/csrc/jit/codegen/cuda/ir_cloner.h
+++ b/torch/csrc/jit/codegen/cuda/ir_cloner.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/codegen/cuda/dispatch.h>
 
 #include <unordered_map>
diff --git a/torch/csrc/jit/codegen/cuda/ir_graphviz.h b/torch/csrc/jit/codegen/cuda/ir_graphviz.h
index 7bf74208a5b790..1144d95eb15215 100644
--- a/torch/csrc/jit/codegen/cuda/ir_graphviz.h
+++ b/torch/csrc/jit/codegen/cuda/ir_graphviz.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/codegen/cuda/dispatch.h>
 
 #include <sstream>
diff --git a/torch/csrc/jit/codegen/cuda/ir_interface_nodes.h b/torch/csrc/jit/codegen/cuda/ir_interface_nodes.h
index 0782a7e3888ab2..e03f7f28326445 100644
--- a/torch/csrc/jit/codegen/cuda/ir_interface_nodes.h
+++ b/torch/csrc/jit/codegen/cuda/ir_interface_nodes.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/fusion.h>
 #include <torch/csrc/jit/codegen/cuda/ir_base_nodes.h>
diff --git a/torch/csrc/jit/codegen/cuda/ir_internal_nodes.h b/torch/csrc/jit/codegen/cuda/ir_internal_nodes.h
index 7abb84ec43d2f4..313bda6ea36e04 100644
--- a/torch/csrc/jit/codegen/cuda/ir_internal_nodes.h
+++ b/torch/csrc/jit/codegen/cuda/ir_internal_nodes.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/fusion.h>
 #include <torch/csrc/jit/codegen/cuda/ir_base_nodes.h>
diff --git a/torch/csrc/jit/codegen/cuda/ir_iostream.h b/torch/csrc/jit/codegen/cuda/ir_iostream.h
index fde0fd2ef26938..5f0e4a40877eb1 100644
--- a/torch/csrc/jit/codegen/cuda/ir_iostream.h
+++ b/torch/csrc/jit/codegen/cuda/ir_iostream.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/dispatch.h>
 
diff --git a/torch/csrc/jit/codegen/cuda/ir_printer.h b/torch/csrc/jit/codegen/cuda/ir_printer.h
index 5c87cb192ae20e..a2c14386147e53 100644
--- a/torch/csrc/jit/codegen/cuda/ir_printer.h
+++ b/torch/csrc/jit/codegen/cuda/ir_printer.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/ir_iostream.h>
 #include <torch/csrc/jit/codegen/cuda/iter_visitor.h>
diff --git a/torch/csrc/jit/codegen/cuda/iter_visitor.h b/torch/csrc/jit/codegen/cuda/iter_visitor.h
index ef05d10f5e0f4c..d4aa56ea2fef30 100644
--- a/torch/csrc/jit/codegen/cuda/iter_visitor.h
+++ b/torch/csrc/jit/codegen/cuda/iter_visitor.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/dispatch.h>
 #include <torch/csrc/jit/codegen/cuda/type.h>
diff --git a/torch/csrc/jit/codegen/cuda/kernel.h b/torch/csrc/jit/codegen/cuda/kernel.h
index 14e8e699e06307..2dbccf753dfd7f 100644
--- a/torch/csrc/jit/codegen/cuda/kernel.h
+++ b/torch/csrc/jit/codegen/cuda/kernel.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/codegen/cuda/kernel_ir.h>
 #include <torch/csrc/jit/codegen/cuda/lower_thread_predicate.h>
 #include <torch/csrc/jit/codegen/cuda/lower_warp_reduce.h>
diff --git a/torch/csrc/jit/codegen/cuda/kernel_cache.h b/torch/csrc/jit/codegen/cuda/kernel_cache.h
index f0e454ba8e88d5..ae84c25e4f23ed 100644
--- a/torch/csrc/jit/codegen/cuda/kernel_cache.h
+++ b/torch/csrc/jit/codegen/cuda/kernel_cache.h
@@ -8,7 +8,7 @@
 #include <torch/csrc/jit/codegen/cuda/scheduler/registry.h>
 
 #include <c10/util/ArrayRef.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <mutex>
 #include <type_traits>
diff --git a/torch/csrc/jit/codegen/cuda/kernel_expr_evaluator.h b/torch/csrc/jit/codegen/cuda/kernel_expr_evaluator.h
index d8583c88968e52..647913875430cf 100644
--- a/torch/csrc/jit/codegen/cuda/kernel_expr_evaluator.h
+++ b/torch/csrc/jit/codegen/cuda/kernel_expr_evaluator.h
@@ -1,7 +1,7 @@
 
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/codegen/cuda/evaluator_common.h>
 #include <torch/csrc/jit/codegen/cuda/kernel_ir.h>
 
diff --git a/torch/csrc/jit/codegen/cuda/kernel_ir.h b/torch/csrc/jit/codegen/cuda/kernel_ir.h
index d9f517800e2236..4eec5c127e25cd 100644
--- a/torch/csrc/jit/codegen/cuda/kernel_ir.h
+++ b/torch/csrc/jit/codegen/cuda/kernel_ir.h
@@ -10,7 +10,7 @@
 #include <torch/csrc/jit/codegen/cuda/parallel_type_bitmap.h>
 
 #include <c10/util/Optional.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <cstdint>
 #include <string>
diff --git a/torch/csrc/jit/codegen/cuda/kernel_ir_builder.h b/torch/csrc/jit/codegen/cuda/kernel_ir_builder.h
index b0fb6d1d2565a5..17a095baf1202b 100644
--- a/torch/csrc/jit/codegen/cuda/kernel_ir_builder.h
+++ b/torch/csrc/jit/codegen/cuda/kernel_ir_builder.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/codegen/cuda/kernel.h>
 #include <torch/csrc/jit/codegen/cuda/kernel_ir.h>
 
diff --git a/torch/csrc/jit/codegen/cuda/kernel_ir_printer.h b/torch/csrc/jit/codegen/cuda/kernel_ir_printer.h
index c286a4b4184795..115901a031a911 100644
--- a/torch/csrc/jit/codegen/cuda/kernel_ir_printer.h
+++ b/torch/csrc/jit/codegen/cuda/kernel_ir_printer.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/kernel.h>
 #include <torch/csrc/jit/codegen/cuda/kernel_ir.h>
diff --git a/torch/csrc/jit/codegen/cuda/lower2device.h b/torch/csrc/jit/codegen/cuda/lower2device.h
index 9b36b6dd26fec7..79d05c0498c416 100644
--- a/torch/csrc/jit/codegen/cuda/lower2device.h
+++ b/torch/csrc/jit/codegen/cuda/lower2device.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/compute_at_map.h>
 #include <torch/csrc/jit/codegen/cuda/ir_all_nodes.h>
diff --git a/torch/csrc/jit/codegen/cuda/lower_alias_memory.h b/torch/csrc/jit/codegen/cuda/lower_alias_memory.h
index dfe75dbd221395..26b33b6d5dc713 100644
--- a/torch/csrc/jit/codegen/cuda/lower_alias_memory.h
+++ b/torch/csrc/jit/codegen/cuda/lower_alias_memory.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/dispatch.h>
 #include <torch/csrc/jit/codegen/cuda/ir_all_nodes.h>
diff --git a/torch/csrc/jit/codegen/cuda/lower_allocation.h b/torch/csrc/jit/codegen/cuda/lower_allocation.h
index e00c9ab83f2566..bc0344ca19f611 100644
--- a/torch/csrc/jit/codegen/cuda/lower_allocation.h
+++ b/torch/csrc/jit/codegen/cuda/lower_allocation.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/dispatch.h>
 #include <torch/csrc/jit/codegen/cuda/ir_all_nodes.h>
diff --git a/torch/csrc/jit/codegen/cuda/lower_index.h b/torch/csrc/jit/codegen/cuda/lower_index.h
index d6139e9691cab1..5eb27c78f283b5 100644
--- a/torch/csrc/jit/codegen/cuda/lower_index.h
+++ b/torch/csrc/jit/codegen/cuda/lower_index.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/instrumentation.h>
 #include <torch/csrc/jit/codegen/cuda/kernel_ir.h>
diff --git a/torch/csrc/jit/codegen/cuda/lower_insert_syncs.h b/torch/csrc/jit/codegen/cuda/lower_insert_syncs.h
index 7a9543417e4846..506183734484e4 100644
--- a/torch/csrc/jit/codegen/cuda/lower_insert_syncs.h
+++ b/torch/csrc/jit/codegen/cuda/lower_insert_syncs.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/ir_all_nodes.h>
 #include <torch/csrc/jit/codegen/cuda/kernel_ir.h>
diff --git a/torch/csrc/jit/codegen/cuda/lower_loops.h b/torch/csrc/jit/codegen/cuda/lower_loops.h
index 2c23fb91a7d441..ce3f49a01fa2fc 100644
--- a/torch/csrc/jit/codegen/cuda/lower_loops.h
+++ b/torch/csrc/jit/codegen/cuda/lower_loops.h
@@ -1,7 +1,7 @@
 
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/compute_at_map.h>
 #include <torch/csrc/jit/codegen/cuda/instrumentation.h>
diff --git a/torch/csrc/jit/codegen/cuda/lower_misaligned_vectorization.h b/torch/csrc/jit/codegen/cuda/lower_misaligned_vectorization.h
index db28adb9de3ba5..588d3787752b9d 100644
--- a/torch/csrc/jit/codegen/cuda/lower_misaligned_vectorization.h
+++ b/torch/csrc/jit/codegen/cuda/lower_misaligned_vectorization.h
@@ -1,5 +1,5 @@
 #pragma once
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/ir_all_nodes.h>
 
diff --git a/torch/csrc/jit/codegen/cuda/lower_predicate.h b/torch/csrc/jit/codegen/cuda/lower_predicate.h
index de70640f336e8a..393d0fa5c184e3 100644
--- a/torch/csrc/jit/codegen/cuda/lower_predicate.h
+++ b/torch/csrc/jit/codegen/cuda/lower_predicate.h
@@ -1,5 +1,5 @@
 #pragma once
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/ir_all_nodes.h>
 #include <torch/csrc/jit/codegen/cuda/kernel_ir.h>
diff --git a/torch/csrc/jit/codegen/cuda/lower_shift.h b/torch/csrc/jit/codegen/cuda/lower_shift.h
index ef568f217e77fb..378709ca443093 100644
--- a/torch/csrc/jit/codegen/cuda/lower_shift.h
+++ b/torch/csrc/jit/codegen/cuda/lower_shift.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/dispatch.h>
 #include <torch/csrc/jit/codegen/cuda/ir_all_nodes.h>
diff --git a/torch/csrc/jit/codegen/cuda/lower_thread_predicate.h b/torch/csrc/jit/codegen/cuda/lower_thread_predicate.h
index e68b6dde08c3c3..6c12263bb91f3f 100644
--- a/torch/csrc/jit/codegen/cuda/lower_thread_predicate.h
+++ b/torch/csrc/jit/codegen/cuda/lower_thread_predicate.h
@@ -1,7 +1,7 @@
 
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/ir_all_nodes.h>
 #include <torch/csrc/jit/codegen/cuda/lower_utils.h>
diff --git a/torch/csrc/jit/codegen/cuda/lower_trivial_reductions.h b/torch/csrc/jit/codegen/cuda/lower_trivial_reductions.h
index 3f5a94de9742c1..c16439ed4f03af 100644
--- a/torch/csrc/jit/codegen/cuda/lower_trivial_reductions.h
+++ b/torch/csrc/jit/codegen/cuda/lower_trivial_reductions.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/dispatch.h>
 #include <torch/csrc/jit/codegen/cuda/ir_all_nodes.h>
diff --git a/torch/csrc/jit/codegen/cuda/lower_unroll.h b/torch/csrc/jit/codegen/cuda/lower_unroll.h
index 31a46c09db4c81..bec4966dd94650 100644
--- a/torch/csrc/jit/codegen/cuda/lower_unroll.h
+++ b/torch/csrc/jit/codegen/cuda/lower_unroll.h
@@ -1,5 +1,5 @@
 #pragma once
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/kernel_ir.h>
 #include <torch/csrc/jit/codegen/cuda/lower_thread_predicate.h>
diff --git a/torch/csrc/jit/codegen/cuda/lower_utils.h b/torch/csrc/jit/codegen/cuda/lower_utils.h
index 238d0166851e3a..79c35c4aca942c 100644
--- a/torch/csrc/jit/codegen/cuda/lower_utils.h
+++ b/torch/csrc/jit/codegen/cuda/lower_utils.h
@@ -1,7 +1,7 @@
 
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/ir_all_nodes.h>
 #include <torch/csrc/jit/codegen/cuda/kernel_ir.h>
diff --git a/torch/csrc/jit/codegen/cuda/lower_validation.h b/torch/csrc/jit/codegen/cuda/lower_validation.h
index fac9642d418e9a..34fc468ed17468 100644
--- a/torch/csrc/jit/codegen/cuda/lower_validation.h
+++ b/torch/csrc/jit/codegen/cuda/lower_validation.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/ir_all_nodes.h>
 
diff --git a/torch/csrc/jit/codegen/cuda/manager.h b/torch/csrc/jit/codegen/cuda/manager.h
index 53eed90af8a832..39c97478effe19 100644
--- a/torch/csrc/jit/codegen/cuda/manager.h
+++ b/torch/csrc/jit/codegen/cuda/manager.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 
 /*
diff --git a/torch/csrc/jit/codegen/cuda/mutator.h b/torch/csrc/jit/codegen/cuda/mutator.h
index 9451ea6c47da72..2087a5b92fd37b 100644
--- a/torch/csrc/jit/codegen/cuda/mutator.h
+++ b/torch/csrc/jit/codegen/cuda/mutator.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/dispatch.h>
 #include <torch/csrc/jit/codegen/cuda/ir_base_nodes.h>
diff --git a/torch/csrc/jit/codegen/cuda/ops/composite.h b/torch/csrc/jit/codegen/cuda/ops/composite.h
index f130b274104ce4..1092f67900baee 100644
--- a/torch/csrc/jit/codegen/cuda/ops/composite.h
+++ b/torch/csrc/jit/codegen/cuda/ops/composite.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/ir_interface_nodes.h>
 #include <torch/csrc/jit/codegen/cuda/type.h>
diff --git a/torch/csrc/jit/codegen/cuda/ops/normalization.h b/torch/csrc/jit/codegen/cuda/ops/normalization.h
index 240d06637cd59d..dae58462b92925 100644
--- a/torch/csrc/jit/codegen/cuda/ops/normalization.h
+++ b/torch/csrc/jit/codegen/cuda/ops/normalization.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/ir_interface_nodes.h>
 #include <torch/csrc/jit/codegen/cuda/type.h>
diff --git a/torch/csrc/jit/codegen/cuda/parallel_type_bitmap.h b/torch/csrc/jit/codegen/cuda/parallel_type_bitmap.h
index a8ba625a214634..0bf8ae39277b39 100644
--- a/torch/csrc/jit/codegen/cuda/parallel_type_bitmap.h
+++ b/torch/csrc/jit/codegen/cuda/parallel_type_bitmap.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/codegen/cuda/type.h>
 
 #include <bitset>
diff --git a/torch/csrc/jit/codegen/cuda/parser.h b/torch/csrc/jit/codegen/cuda/parser.h
index 7fff8a3a95a7eb..4b2fcf50f9920a 100644
--- a/torch/csrc/jit/codegen/cuda/parser.h
+++ b/torch/csrc/jit/codegen/cuda/parser.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/jit/runtime/profiling_record.h>
 
diff --git a/torch/csrc/jit/codegen/cuda/partial_split_map.h b/torch/csrc/jit/codegen/cuda/partial_split_map.h
index 6548d0d374f1d8..be432bd5a161d4 100644
--- a/torch/csrc/jit/codegen/cuda/partial_split_map.h
+++ b/torch/csrc/jit/codegen/cuda/partial_split_map.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/dispatch.h>
 #include <torch/csrc/jit/codegen/cuda/ir_all_nodes.h>
diff --git a/torch/csrc/jit/codegen/cuda/partition.h b/torch/csrc/jit/codegen/cuda/partition.h
index 4ebac40a23baa6..0d8baca47007a5 100644
--- a/torch/csrc/jit/codegen/cuda/partition.h
+++ b/torch/csrc/jit/codegen/cuda/partition.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 
 /*
diff --git a/torch/csrc/jit/codegen/cuda/reference_tensor.h b/torch/csrc/jit/codegen/cuda/reference_tensor.h
index 883eda605bcf42..2220831dc09fbf 100644
--- a/torch/csrc/jit/codegen/cuda/reference_tensor.h
+++ b/torch/csrc/jit/codegen/cuda/reference_tensor.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/ir_interface_nodes.h>
 
diff --git a/torch/csrc/jit/codegen/cuda/root_domain_map.h b/torch/csrc/jit/codegen/cuda/root_domain_map.h
index 34e1f0b1936965..2bf20def43a3b8 100644
--- a/torch/csrc/jit/codegen/cuda/root_domain_map.h
+++ b/torch/csrc/jit/codegen/cuda/root_domain_map.h
@@ -5,7 +5,7 @@
 #include <torch/csrc/jit/codegen/cuda/iter_visitor.h>
 #include <torch/csrc/jit/codegen/cuda/utils.h>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 namespace torch {
 namespace jit {
diff --git a/torch/csrc/jit/codegen/cuda/transform_iter.h b/torch/csrc/jit/codegen/cuda/transform_iter.h
index 2fd9d862051d67..cde502d636e36e 100644
--- a/torch/csrc/jit/codegen/cuda/transform_iter.h
+++ b/torch/csrc/jit/codegen/cuda/transform_iter.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/ir_all_nodes.h>
 #include <torch/csrc/jit/codegen/cuda/ir_iostream.h>
diff --git a/torch/csrc/jit/codegen/cuda/transform_replay.h b/torch/csrc/jit/codegen/cuda/transform_replay.h
index 7264afa28bee04..92898b54ba7aa3 100644
--- a/torch/csrc/jit/codegen/cuda/transform_replay.h
+++ b/torch/csrc/jit/codegen/cuda/transform_replay.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <c10/util/Exception.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <algorithm>
 #include <unordered_map>
diff --git a/torch/csrc/jit/codegen/cuda/transform_rfactor.h b/torch/csrc/jit/codegen/cuda/transform_rfactor.h
index 781e4fcd183862..551f67905b0bb3 100644
--- a/torch/csrc/jit/codegen/cuda/transform_rfactor.h
+++ b/torch/csrc/jit/codegen/cuda/transform_rfactor.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/codegen/cuda/ir_all_nodes.h>
 #include <torch/csrc/jit/codegen/cuda/transform_iter.h>
diff --git a/torch/csrc/jit/codegen/cuda/type.h b/torch/csrc/jit/codegen/cuda/type.h
index 776350e2070119..63baeb32b3bda2 100644
--- a/torch/csrc/jit/codegen/cuda/type.h
+++ b/torch/csrc/jit/codegen/cuda/type.h
@@ -3,7 +3,7 @@
 #include <c10/core/ScalarType.h>
 #include <c10/util/Exception.h>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <array>
 #include <cstdint>
diff --git a/torch/csrc/jit/codegen/fuser/arg_spec.h b/torch/csrc/jit/codegen/fuser/arg_spec.h
index fe35f004b2c511..06a72c231b84a5 100644
--- a/torch/csrc/jit/codegen/fuser/arg_spec.h
+++ b/torch/csrc/jit/codegen/fuser/arg_spec.h
@@ -2,7 +2,7 @@
 #include <ATen/ATen.h>
 #include <ATen/core/functional.h> // fmap
 #include <c10/util/hash.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/codegen/fuser/tensor_desc.h>
 
 #include <cstdint>
diff --git a/torch/csrc/jit/codegen/fuser/codegen.h b/torch/csrc/jit/codegen/fuser/codegen.h
index 6abb6e995c95f3..2b96264549ff60 100644
--- a/torch/csrc/jit/codegen/fuser/codegen.h
+++ b/torch/csrc/jit/codegen/fuser/codegen.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/codegen/fuser/arg_spec.h>
 #include <torch/csrc/jit/codegen/fuser/partition_desc.h>
 #include <torch/csrc/jit/codegen/fuser/tensor_desc.h>
diff --git a/torch/csrc/jit/codegen/fuser/compiler.h b/torch/csrc/jit/codegen/fuser/compiler.h
index 4e1254647c3495..3d490f6940ab0d 100644
--- a/torch/csrc/jit/codegen/fuser/compiler.h
+++ b/torch/csrc/jit/codegen/fuser/compiler.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <ATen/core/stack.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/codegen/fuser/arg_spec.h>
 #include <torch/csrc/jit/codegen/fuser/fused_kernel.h>
 #include <torch/csrc/jit/codegen/fuser/interface.h>
diff --git a/torch/csrc/jit/codegen/fuser/cpu/fused_kernel.h b/torch/csrc/jit/codegen/fuser/cpu/fused_kernel.h
index 0949456b9d044a..ce5d6ee2c55461 100644
--- a/torch/csrc/jit/codegen/fuser/cpu/fused_kernel.h
+++ b/torch/csrc/jit/codegen/fuser/cpu/fused_kernel.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <ATen/ATen.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/codegen/fuser/fused_kernel.h>
 #include <torch/csrc/utils/disallow_copy.h>
 
diff --git a/torch/csrc/jit/codegen/fuser/cpu/temp_file.h b/torch/csrc/jit/codegen/fuser/cpu/temp_file.h
index f0dcc06a3e4913..080d76bde2225e 100644
--- a/torch/csrc/jit/codegen/fuser/cpu/temp_file.h
+++ b/torch/csrc/jit/codegen/fuser/cpu/temp_file.h
@@ -2,7 +2,7 @@
 
 #include <ATen/ATen.h>
 #include <c10/util/Exception.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/utils/disallow_copy.h>
 
 #ifdef _WIN32
diff --git a/torch/csrc/jit/codegen/fuser/cuda/fused_kernel.h b/torch/csrc/jit/codegen/fuser/cuda/fused_kernel.h
index 8cb9a4680ad306..4d39660532e358 100644
--- a/torch/csrc/jit/codegen/fuser/cuda/fused_kernel.h
+++ b/torch/csrc/jit/codegen/fuser/cuda/fused_kernel.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <ATen/ATen.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/codegen/fuser/fused_kernel.h>
 
 #include <cuda.h>
diff --git a/torch/csrc/jit/codegen/fuser/cuda/resource_strings.h b/torch/csrc/jit/codegen/fuser/cuda/resource_strings.h
index 8a89c6952b92ea..91ef0f0d7a34b9 100644
--- a/torch/csrc/jit/codegen/fuser/cuda/resource_strings.h
+++ b/torch/csrc/jit/codegen/fuser/cuda/resource_strings.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/frontend/code_template.h>
 
 namespace torch {
diff --git a/torch/csrc/jit/codegen/fuser/executor.h b/torch/csrc/jit/codegen/fuser/executor.h
index 407ab632e8c3e9..f454e3d6a1845d 100644
--- a/torch/csrc/jit/codegen/fuser/executor.h
+++ b/torch/csrc/jit/codegen/fuser/executor.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <ATen/core/stack.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/codegen/fuser/fused_kernel.h>
 #include <torch/csrc/jit/codegen/fuser/kernel_spec.h>
 
diff --git a/torch/csrc/jit/codegen/fuser/interface.h b/torch/csrc/jit/codegen/fuser/interface.h
index 4d6220dc9ed6f5..2c67064fe1359c 100644
--- a/torch/csrc/jit/codegen/fuser/interface.h
+++ b/torch/csrc/jit/codegen/fuser/interface.h
@@ -2,7 +2,7 @@
 
 #include <ATen/ATen.h>
 #include <ATen/core/stack.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 
 #include <cstdint>
diff --git a/torch/csrc/jit/codegen/fuser/kernel_cache.h b/torch/csrc/jit/codegen/fuser/kernel_cache.h
index e3a06eef7e4912..be5f9c355dad74 100644
--- a/torch/csrc/jit/codegen/fuser/kernel_cache.h
+++ b/torch/csrc/jit/codegen/fuser/kernel_cache.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <c10/util/Optional.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/codegen/fuser/kernel_spec.h>
 #include <torch/csrc/jit/ir/ir.h>
 
diff --git a/torch/csrc/jit/codegen/fuser/kernel_spec.h b/torch/csrc/jit/codegen/fuser/kernel_spec.h
index f54812c210ef32..57806ed4363111 100644
--- a/torch/csrc/jit/codegen/fuser/kernel_spec.h
+++ b/torch/csrc/jit/codegen/fuser/kernel_spec.h
@@ -3,7 +3,7 @@
 #include <ATen/ATen.h>
 #include <ATen/core/stack.h>
 #include <c10/util/Optional.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/codegen/fuser/arg_spec.h>
 #include <torch/csrc/jit/codegen/fuser/fused_kernel.h>
 #include <torch/csrc/jit/codegen/fuser/interface.h>
diff --git a/torch/csrc/jit/codegen/fuser/partition_desc.h b/torch/csrc/jit/codegen/fuser/partition_desc.h
index 59189471b266f2..87cf7986e5e91c 100644
--- a/torch/csrc/jit/codegen/fuser/partition_desc.h
+++ b/torch/csrc/jit/codegen/fuser/partition_desc.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <c10/util/Exception.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/codegen/fuser/tensor_desc.h>
 
 #include <cstdint>
diff --git a/torch/csrc/jit/codegen/fuser/tensor_desc.h b/torch/csrc/jit/codegen/fuser/tensor_desc.h
index cb48c384adc43f..992dd5f551cc50 100644
--- a/torch/csrc/jit/codegen/fuser/tensor_desc.h
+++ b/torch/csrc/jit/codegen/fuser/tensor_desc.h
@@ -4,7 +4,7 @@
 #include <ATen/core/jit_type.h>
 #include <c10/util/Exception.h>
 #include <c10/util/hash.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <algorithm>
 #include <iostream>
diff --git a/torch/csrc/jit/codegen/fuser/tensor_info.h b/torch/csrc/jit/codegen/fuser/tensor_info.h
index 3e30c46c4aa90d..9f3fcca1aca2e5 100644
--- a/torch/csrc/jit/codegen/fuser/tensor_info.h
+++ b/torch/csrc/jit/codegen/fuser/tensor_info.h
@@ -1,5 +1,5 @@
 #pragma once
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <cstdint>
 
diff --git a/torch/csrc/jit/frontend/builtin_functions.h b/torch/csrc/jit/frontend/builtin_functions.h
index 87831ab5566f60..2ff8d13758541e 100644
--- a/torch/csrc/jit/frontend/builtin_functions.h
+++ b/torch/csrc/jit/frontend/builtin_functions.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/api/module.h>
 
 namespace torch {
diff --git a/torch/csrc/jit/frontend/canonicalize_modified_loop.cpp b/torch/csrc/jit/frontend/canonicalize_modified_loop.cpp
index 5009482d69b02e..1c510c29acb736 100644
--- a/torch/csrc/jit/frontend/canonicalize_modified_loop.cpp
+++ b/torch/csrc/jit/frontend/canonicalize_modified_loop.cpp
@@ -2,7 +2,7 @@
 #include <memory>
 #include <string>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/frontend/canonicalize_modified_loop.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/jit/ir/ir_views.h>
diff --git a/torch/csrc/jit/frontend/canonicalize_modified_loop.h b/torch/csrc/jit/frontend/canonicalize_modified_loop.h
index 178206fb281663..f8a3a4ecc26901 100644
--- a/torch/csrc/jit/frontend/canonicalize_modified_loop.h
+++ b/torch/csrc/jit/frontend/canonicalize_modified_loop.h
@@ -1,7 +1,7 @@
 #pragma once
 #include <memory>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 namespace torch {
 namespace jit {
diff --git a/torch/csrc/jit/frontend/convert_to_ssa.h b/torch/csrc/jit/frontend/convert_to_ssa.h
index c9e44b64fd7ddd..787eae80578881 100644
--- a/torch/csrc/jit/frontend/convert_to_ssa.h
+++ b/torch/csrc/jit/frontend/convert_to_ssa.h
@@ -3,7 +3,7 @@
 #include <memory>
 #include <string>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 
 namespace torch {
diff --git a/torch/csrc/jit/frontend/edit_distance.h b/torch/csrc/jit/frontend/edit_distance.h
index 3d2cd89aa367c7..f0d999e83c1a2d 100644
--- a/torch/csrc/jit/frontend/edit_distance.h
+++ b/torch/csrc/jit/frontend/edit_distance.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <cstddef>
 
 namespace torch {
diff --git a/torch/csrc/jit/frontend/exit_transforms.h b/torch/csrc/jit/frontend/exit_transforms.h
index abffd65634ad41..84910c6bc1e4d6 100644
--- a/torch/csrc/jit/frontend/exit_transforms.h
+++ b/torch/csrc/jit/frontend/exit_transforms.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 
 namespace torch {
diff --git a/torch/csrc/jit/frontend/inline_loop_condition.cpp b/torch/csrc/jit/frontend/inline_loop_condition.cpp
index 4f471934261c94..0c5991fae3167c 100644
--- a/torch/csrc/jit/frontend/inline_loop_condition.cpp
+++ b/torch/csrc/jit/frontend/inline_loop_condition.cpp
@@ -2,7 +2,7 @@
 #include <memory>
 #include <string>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/frontend/inline_loop_condition.h>
 #include <torch/csrc/jit/ir/ir.h>
 
diff --git a/torch/csrc/jit/frontend/inline_loop_condition.h b/torch/csrc/jit/frontend/inline_loop_condition.h
index 29adac91434e06..c5efa0b40151ac 100644
--- a/torch/csrc/jit/frontend/inline_loop_condition.h
+++ b/torch/csrc/jit/frontend/inline_loop_condition.h
@@ -3,7 +3,7 @@
 #include <memory>
 #include <string>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 
 namespace torch {
diff --git a/torch/csrc/jit/frontend/lexer.h b/torch/csrc/jit/frontend/lexer.h
index 9a792435932fbf..dc5c615d4a2c18 100644
--- a/torch/csrc/jit/frontend/lexer.h
+++ b/torch/csrc/jit/frontend/lexer.h
@@ -2,7 +2,7 @@
 #include <ATen/core/Macros.h>
 #include <c10/util/C++17.h>
 #include <c10/util/Exception.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/frontend/parser_constants.h>
 #include <torch/csrc/jit/frontend/source_range.h>
 #include <torch/csrc/jit/frontend/strtod.h>
diff --git a/torch/csrc/jit/frontend/name_mangler.h b/torch/csrc/jit/frontend/name_mangler.h
index 22a85fbe39ed14..07a0352bce6910 100644
--- a/torch/csrc/jit/frontend/name_mangler.h
+++ b/torch/csrc/jit/frontend/name_mangler.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <ATen/core/qualified_name.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 namespace torch {
 namespace jit {
diff --git a/torch/csrc/jit/frontend/parser.h b/torch/csrc/jit/frontend/parser.h
index 7d1bfedf310945..8b2bb302a340bb 100644
--- a/torch/csrc/jit/frontend/parser.h
+++ b/torch/csrc/jit/frontend/parser.h
@@ -1,5 +1,5 @@
 #pragma once
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/frontend/tree.h>
 #include <torch/csrc/jit/frontend/tree_views.h>
 #include <memory>
diff --git a/torch/csrc/jit/frontend/schema_matching.h b/torch/csrc/jit/frontend/schema_matching.h
index fb6d1ab7f92e5e..52f9e6ef005048 100644
--- a/torch/csrc/jit/frontend/schema_matching.h
+++ b/torch/csrc/jit/frontend/schema_matching.h
@@ -1,5 +1,5 @@
 #pragma once
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/jit/ir/named_value.h>
 
diff --git a/torch/csrc/jit/frontend/script_type_parser.h b/torch/csrc/jit/frontend/script_type_parser.h
index 8f429e3e25447d..3a05af9c598abd 100644
--- a/torch/csrc/jit/frontend/script_type_parser.h
+++ b/torch/csrc/jit/frontend/script_type_parser.h
@@ -1,6 +1,6 @@
 #pragma once
 #include <ATen/core/jit_type.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/frontend/resolver.h>
 #include <torch/csrc/jit/frontend/tree_views.h>
 
diff --git a/torch/csrc/jit/frontend/tracer.h b/torch/csrc/jit/frontend/tracer.h
index e0c6967f48cd2e..eb49e1535b6621 100644
--- a/torch/csrc/jit/frontend/tracer.h
+++ b/torch/csrc/jit/frontend/tracer.h
@@ -4,7 +4,7 @@
 #include <ATen/core/jit_type.h>
 #include <ATen/core/stack.h>
 #include <c10/util/Exception.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/api/object.h>
 #include <torch/csrc/jit/frontend/source_range.h>
diff --git a/torch/csrc/jit/frontend/versioned_symbols.h b/torch/csrc/jit/frontend/versioned_symbols.h
index 314040af762f03..1c59708afd77d1 100644
--- a/torch/csrc/jit/frontend/versioned_symbols.h
+++ b/torch/csrc/jit/frontend/versioned_symbols.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/api/module.h>
 
 #include <cstdint>
diff --git a/torch/csrc/jit/ir/attributes.h b/torch/csrc/jit/ir/attributes.h
index 6c0938c94668d2..bfc9e50b5a498b 100644
--- a/torch/csrc/jit/ir/attributes.h
+++ b/torch/csrc/jit/ir/attributes.h
@@ -5,7 +5,7 @@
 
 #include <ATen/core/interned_strings.h>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 namespace c10 {
 struct Type;
diff --git a/torch/csrc/jit/ir/constants.h b/torch/csrc/jit/ir/constants.h
index cdef7ddd89d294..d9d11075dd2042 100644
--- a/torch/csrc/jit/ir/constants.h
+++ b/torch/csrc/jit/ir/constants.h
@@ -1,7 +1,7 @@
 #pragma once
 #include <ATen/core/ivalue.h>
 #include <ATen/core/jit_type.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/frontend/source_range.h>
 #include <torch/csrc/jit/ir/scope.h>
 
diff --git a/torch/csrc/jit/ir/ir.h b/torch/csrc/jit/ir/ir.h
index 07eadc7307f243..d19f872c2f6dfc 100644
--- a/torch/csrc/jit/ir/ir.h
+++ b/torch/csrc/jit/ir/ir.h
@@ -6,7 +6,7 @@
 #include <torch/csrc/jit/ir/scope.h>
 #include <torch/csrc/jit/runtime/operator.h>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/utils/disallow_copy.h>
 #include <torch/csrc/utils/python_stub.h>
 
diff --git a/torch/csrc/jit/ir/irparser.h b/torch/csrc/jit/ir/irparser.h
index 1566d0c64eba06..f49e8150521ab6 100644
--- a/torch/csrc/jit/ir/irparser.h
+++ b/torch/csrc/jit/ir/irparser.h
@@ -1,10 +1,10 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <string>
 #include <unordered_map>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 namespace torch {
 namespace jit {
diff --git a/torch/csrc/jit/ir/scope.h b/torch/csrc/jit/ir/scope.h
index 32c11042559918..b7385bb513cf9e 100644
--- a/torch/csrc/jit/ir/scope.h
+++ b/torch/csrc/jit/ir/scope.h
@@ -3,7 +3,7 @@
 #include <ATen/core/jit_type.h>
 #include <c10/util/Optional.h>
 #include <c10/util/intrusive_ptr.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/frontend/source_range.h>
 #include <unordered_map>
 
diff --git a/torch/csrc/jit/jit_log.h b/torch/csrc/jit/jit_log.h
index 8472df20e7f2b2..ba5c174ec2dac9 100644
--- a/torch/csrc/jit/jit_log.h
+++ b/torch/csrc/jit/jit_log.h
@@ -1,5 +1,5 @@
 #pragma once
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <memory>
 #include <ostream>
 #include <string>
diff --git a/torch/csrc/jit/jit_opt_limit.h b/torch/csrc/jit/jit_opt_limit.h
index a4f1b7c674c438..fe84f615bcc52d 100644
--- a/torch/csrc/jit/jit_opt_limit.h
+++ b/torch/csrc/jit/jit_opt_limit.h
@@ -1,5 +1,5 @@
 #pragma once
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <string>
 #include <unordered_map>
 
diff --git a/torch/csrc/jit/mobile/nnc/aot_compiler.h b/torch/csrc/jit/mobile/nnc/aot_compiler.h
index 4edac10542c726..aee92906fcc514 100644
--- a/torch/csrc/jit/mobile/nnc/aot_compiler.h
+++ b/torch/csrc/jit/mobile/nnc/aot_compiler.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/jit/mobile/nnc/context.h>
 
diff --git a/torch/csrc/jit/mobile/train/random.h b/torch/csrc/jit/mobile/train/random.h
index fd32a8a0b491b4..04a00859b382e1 100644
--- a/torch/csrc/jit/mobile/train/random.h
+++ b/torch/csrc/jit/mobile/train/random.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/data/samplers/base.h>
 #include <torch/types.h>
 
diff --git a/torch/csrc/jit/mobile/train/sequential.h b/torch/csrc/jit/mobile/train/sequential.h
index ca6749ae4e970f..d4960037d394dd 100644
--- a/torch/csrc/jit/mobile/train/sequential.h
+++ b/torch/csrc/jit/mobile/train/sequential.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/data/samplers/base.h>
 #include <torch/types.h>
 
diff --git a/torch/csrc/jit/passes/bailout_graph.h b/torch/csrc/jit/passes/bailout_graph.h
index 786b1afb0ddc78..266f54023e1a15 100644
--- a/torch/csrc/jit/passes/bailout_graph.h
+++ b/torch/csrc/jit/passes/bailout_graph.h
@@ -4,7 +4,7 @@
 #include <ATen/core/ivalue.h>
 #include <ATen/core/jit_type.h>
 #include <ATen/core/stack.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 
 #include <list>
diff --git a/torch/csrc/jit/passes/clear_profiling.h b/torch/csrc/jit/passes/clear_profiling.h
index b9ac975dee6967..7dee9bdb52ad6c 100644
--- a/torch/csrc/jit/passes/clear_profiling.h
+++ b/torch/csrc/jit/passes/clear_profiling.h
@@ -3,7 +3,7 @@
 #include <ATen/ATen.h>
 #include <ATen/core/ivalue.h>
 #include <ATen/core/jit_type.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 
 namespace torch {
diff --git a/torch/csrc/jit/passes/clear_undefinedness.h b/torch/csrc/jit/passes/clear_undefinedness.h
index 7e3c974b9c8b87..24add48764c58a 100644
--- a/torch/csrc/jit/passes/clear_undefinedness.h
+++ b/torch/csrc/jit/passes/clear_undefinedness.h
@@ -3,7 +3,7 @@
 #include <ATen/ATen.h>
 #include <ATen/core/ivalue.h>
 #include <ATen/core/jit_type.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 
 namespace torch {
diff --git a/torch/csrc/jit/passes/create_autodiff_subgraphs.h b/torch/csrc/jit/passes/create_autodiff_subgraphs.h
index 70870c72d80989..481b2aa352107b 100644
--- a/torch/csrc/jit/passes/create_autodiff_subgraphs.h
+++ b/torch/csrc/jit/passes/create_autodiff_subgraphs.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 
 #include <cstddef>
diff --git a/torch/csrc/jit/passes/create_functional_graphs.h b/torch/csrc/jit/passes/create_functional_graphs.h
index 6084e2fb2b9e5d..351816394d80c6 100644
--- a/torch/csrc/jit/passes/create_functional_graphs.h
+++ b/torch/csrc/jit/passes/create_functional_graphs.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 
 namespace torch {
diff --git a/torch/csrc/jit/passes/dtype_analysis.h b/torch/csrc/jit/passes/dtype_analysis.h
index cf2653a2b0eb49..b4bcdcdc7ae7ea 100644
--- a/torch/csrc/jit/passes/dtype_analysis.h
+++ b/torch/csrc/jit/passes/dtype_analysis.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <memory>
 
diff --git a/torch/csrc/jit/passes/guard_elimination.h b/torch/csrc/jit/passes/guard_elimination.h
index 0db3cb4eca0d32..03f13140e370df 100644
--- a/torch/csrc/jit/passes/guard_elimination.h
+++ b/torch/csrc/jit/passes/guard_elimination.h
@@ -4,7 +4,7 @@
 #include <ATen/core/ivalue.h>
 #include <ATen/core/jit_type.h>
 #include <ATen/core/stack.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 
 #include <list>
diff --git a/torch/csrc/jit/passes/inline_forked_closures.h b/torch/csrc/jit/passes/inline_forked_closures.h
index 0d3e58cdbbdf68..164c29f8b6557f 100644
--- a/torch/csrc/jit/passes/inline_forked_closures.h
+++ b/torch/csrc/jit/passes/inline_forked_closures.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 
 namespace torch {
diff --git a/torch/csrc/jit/passes/insert_guards.h b/torch/csrc/jit/passes/insert_guards.h
index a9a5035d05db91..28d9f168bf3d55 100644
--- a/torch/csrc/jit/passes/insert_guards.h
+++ b/torch/csrc/jit/passes/insert_guards.h
@@ -4,7 +4,7 @@
 #include <ATen/core/ivalue.h>
 #include <ATen/core/jit_type.h>
 #include <ATen/core/stack.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 
 #include <list>
diff --git a/torch/csrc/jit/passes/lift_closures.h b/torch/csrc/jit/passes/lift_closures.h
index 8549dc0e5f4f6e..c7cee8417fa457 100644
--- a/torch/csrc/jit/passes/lift_closures.h
+++ b/torch/csrc/jit/passes/lift_closures.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 
 namespace torch {
diff --git a/torch/csrc/jit/passes/liveness.h b/torch/csrc/jit/passes/liveness.h
index 970a5bf7c07a91..7b612dee962230 100644
--- a/torch/csrc/jit/passes/liveness.h
+++ b/torch/csrc/jit/passes/liveness.h
@@ -5,7 +5,7 @@
 #include <ATen/core/jit_type.h>
 #include <ATen/core/stack.h>
 #include <c10/util/sparse_bitset.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <list>
 #include <unordered_map>
diff --git a/torch/csrc/jit/passes/remove_mutation.h b/torch/csrc/jit/passes/remove_mutation.h
index 2c3974ee9f6a33..ebf56d386d8a41 100644
--- a/torch/csrc/jit/passes/remove_mutation.h
+++ b/torch/csrc/jit/passes/remove_mutation.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <c10/util/Exception.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/alias_analysis.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/utils/memory.h>
diff --git a/torch/csrc/jit/passes/requires_grad_analysis.h b/torch/csrc/jit/passes/requires_grad_analysis.h
index cbe09ff3f1520c..c7b80423dc5eaf 100644
--- a/torch/csrc/jit/passes/requires_grad_analysis.h
+++ b/torch/csrc/jit/passes/requires_grad_analysis.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <memory>
 
diff --git a/torch/csrc/jit/passes/restore_mutation.h b/torch/csrc/jit/passes/restore_mutation.h
index 6bc85f1ecc13ed..8c698be2bb8506 100644
--- a/torch/csrc/jit/passes/restore_mutation.h
+++ b/torch/csrc/jit/passes/restore_mutation.h
@@ -2,7 +2,7 @@
 
 #include <ATen/core/interned_strings.h>
 #include <c10/util/Exception.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/alias_analysis.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/utils/memory.h>
diff --git a/torch/csrc/jit/passes/shape_analysis.h b/torch/csrc/jit/passes/shape_analysis.h
index 9d65c664d310b5..1f0cb8c5b4507c 100644
--- a/torch/csrc/jit/passes/shape_analysis.h
+++ b/torch/csrc/jit/passes/shape_analysis.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <memory>
 
diff --git a/torch/csrc/jit/passes/symbolic_shape_analysis.h b/torch/csrc/jit/passes/symbolic_shape_analysis.h
index fdde10ab6f1712..c2dc78c7d7fb42 100644
--- a/torch/csrc/jit/passes/symbolic_shape_analysis.h
+++ b/torch/csrc/jit/passes/symbolic_shape_analysis.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <unordered_map>
 
diff --git a/torch/csrc/jit/passes/symbolic_shape_runtime_fusion.h b/torch/csrc/jit/passes/symbolic_shape_runtime_fusion.h
index 224b9cf41a3700..276834b92aa94f 100644
--- a/torch/csrc/jit/passes/symbolic_shape_runtime_fusion.h
+++ b/torch/csrc/jit/passes/symbolic_shape_runtime_fusion.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/jit/passes/symbolic_shape_analysis.h>
 #include <unordered_map>
diff --git a/torch/csrc/jit/passes/tensorexpr_fuser.h b/torch/csrc/jit/passes/tensorexpr_fuser.h
index 622b747dcce185..3b105bd129bf23 100644
--- a/torch/csrc/jit/passes/tensorexpr_fuser.h
+++ b/torch/csrc/jit/passes/tensorexpr_fuser.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/passes/pass_manager.h>
 #include <memory>
 
diff --git a/torch/csrc/jit/passes/utils/memory_dag.h b/torch/csrc/jit/passes/utils/memory_dag.h
index 47ac20f7288ab1..faf67dbfa20056 100644
--- a/torch/csrc/jit/passes/utils/memory_dag.h
+++ b/torch/csrc/jit/passes/utils/memory_dag.h
@@ -12,7 +12,7 @@
 #include <unordered_set>
 #include <vector>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 // Uses a compressed index representation for faster comparisons
 typedef c10::SparseBitVector<256> MemoryLocations;
diff --git a/torch/csrc/jit/passes/utils/op_registry.h b/torch/csrc/jit/passes/utils/op_registry.h
index 8b44578502c3fc..d68d1d6192d6c3 100644
--- a/torch/csrc/jit/passes/utils/op_registry.h
+++ b/torch/csrc/jit/passes/utils/op_registry.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <memory>
 
diff --git a/torch/csrc/jit/passes/utils/subgraph_utils.h b/torch/csrc/jit/passes/utils/subgraph_utils.h
index 784188616bc546..c073242e108d12 100644
--- a/torch/csrc/jit/passes/utils/subgraph_utils.h
+++ b/torch/csrc/jit/passes/utils/subgraph_utils.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/alias_analysis.h>
 #include <torch/csrc/jit/ir/ir.h>
 
diff --git a/torch/csrc/jit/python/pybind_utils.h b/torch/csrc/jit/python/pybind_utils.h
index a499ecff386d3f..1493501f4d0ece 100644
--- a/torch/csrc/jit/python/pybind_utils.h
+++ b/torch/csrc/jit/python/pybind_utils.h
@@ -9,10 +9,10 @@
 #include <pybind11/pytypes.h>
 #include <torch/csrc/Device.h>
 #include <torch/csrc/Dtype.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/Layout.h>
 #include <torch/csrc/QScheme.h>
 #include <torch/csrc/Stream.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
 #include <torch/csrc/jit/api/module.h>
 #include <torch/csrc/jit/frontend/schema_matching.h>
 #include <torch/csrc/jit/frontend/tracer.h>
diff --git a/torch/csrc/jit/python/update_graph_executor_opt.h b/torch/csrc/jit/python/update_graph_executor_opt.h
index f129f405c06259..9f0c4148e2b230 100644
--- a/torch/csrc/jit/python/update_graph_executor_opt.h
+++ b/torch/csrc/jit/python/update_graph_executor_opt.h
@@ -1,5 +1,5 @@
 #pragma once
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 namespace torch {
 namespace jit {
 TORCH_API void setGraphExecutorOptimize(bool o);
diff --git a/torch/csrc/jit/runtime/argument_spec.h b/torch/csrc/jit/runtime/argument_spec.h
index 560a0b92e1cfe6..95b0a58bade3b5 100644
--- a/torch/csrc/jit/runtime/argument_spec.h
+++ b/torch/csrc/jit/runtime/argument_spec.h
@@ -4,7 +4,7 @@
 #include <ATen/core/stack.h>
 #include <c10/util/hash.h>
 #include <c10/util/irange.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/autograd/variable.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <iostream>
diff --git a/torch/csrc/jit/runtime/autodiff.h b/torch/csrc/jit/runtime/autodiff.h
index 769b79da5ed001..8732ddd5c190c8 100644
--- a/torch/csrc/jit/runtime/autodiff.h
+++ b/torch/csrc/jit/runtime/autodiff.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 
 #include <ATen/ATen.h>
diff --git a/torch/csrc/jit/runtime/calculate_necessary_args.h b/torch/csrc/jit/runtime/calculate_necessary_args.h
index 07df670b010407..401353f74118ba 100644
--- a/torch/csrc/jit/runtime/calculate_necessary_args.h
+++ b/torch/csrc/jit/runtime/calculate_necessary_args.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/frontend/schema_matching.h>
 #include <cstddef>
 
diff --git a/torch/csrc/jit/runtime/interpreter.h b/torch/csrc/jit/runtime/interpreter.h
index 790c3282144bf2..31981ae3bb2cb8 100644
--- a/torch/csrc/jit/runtime/interpreter.h
+++ b/torch/csrc/jit/runtime/interpreter.h
@@ -6,7 +6,7 @@
 #include <ATen/ThreadLocalState.h>
 #include <ATen/core/ivalue.h>
 #include <ATen/core/jit_type.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/frontend/source_range.h>
 
 C10_DECLARE_bool(torch_jit_disable_warning_prints);
diff --git a/torch/csrc/jit/runtime/jit_exception.h b/torch/csrc/jit/runtime/jit_exception.h
index 529262a6aaa98f..974a18cc2b0605 100644
--- a/torch/csrc/jit/runtime/jit_exception.h
+++ b/torch/csrc/jit/runtime/jit_exception.h
@@ -2,7 +2,7 @@
 
 #include <stdexcept>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 namespace torch {
 namespace jit {
diff --git a/torch/csrc/jit/runtime/logging.h b/torch/csrc/jit/runtime/logging.h
index 1e31efee1c71dd..5499ecff5888a0 100644
--- a/torch/csrc/jit/runtime/logging.h
+++ b/torch/csrc/jit/runtime/logging.h
@@ -5,7 +5,7 @@
 #include <unordered_map>
 #include <vector>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 namespace torch {
 namespace jit {
diff --git a/torch/csrc/jit/runtime/print_handler.h b/torch/csrc/jit/runtime/print_handler.h
index 6e6b663002350b..d9ba851fa1abd0 100644
--- a/torch/csrc/jit/runtime/print_handler.h
+++ b/torch/csrc/jit/runtime/print_handler.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <atomic>
 #include <functional>
diff --git a/torch/csrc/jit/runtime/profiling_record.h b/torch/csrc/jit/runtime/profiling_record.h
index c8662f9eda425b..add579566fd830 100644
--- a/torch/csrc/jit/runtime/profiling_record.h
+++ b/torch/csrc/jit/runtime/profiling_record.h
@@ -4,7 +4,7 @@
 #include <ATen/core/ivalue.h>
 #include <ATen/core/jit_type.h>
 #include <ATen/core/stack.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 
 #include <list>
diff --git a/torch/csrc/jit/runtime/slice_indices_adjust.h b/torch/csrc/jit/runtime/slice_indices_adjust.h
index ea1e9511769db0..7a6b78d37e7d9a 100644
--- a/torch/csrc/jit/runtime/slice_indices_adjust.h
+++ b/torch/csrc/jit/runtime/slice_indices_adjust.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <cstddef>
 #include <cstdint>
 
diff --git a/torch/csrc/jit/runtime/symbolic_shape_registry.h b/torch/csrc/jit/runtime/symbolic_shape_registry.h
index 202c9c895cb564..4ff57141060dad 100644
--- a/torch/csrc/jit/runtime/symbolic_shape_registry.h
+++ b/torch/csrc/jit/runtime/symbolic_shape_registry.h
@@ -2,7 +2,7 @@
 // This file is temporary until native_functions.yaml and derivatives.yaml are
 // merged. Ideally this should all go into native_functions.yaml
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 
 namespace torch {
diff --git a/torch/csrc/jit/runtime/symbolic_shape_registry_util.h b/torch/csrc/jit/runtime/symbolic_shape_registry_util.h
index 9f9799e9aa21bf..b0ffbf0c71a1ae 100644
--- a/torch/csrc/jit/runtime/symbolic_shape_registry_util.h
+++ b/torch/csrc/jit/runtime/symbolic_shape_registry_util.h
@@ -2,7 +2,7 @@
 // This file is temporary until native_functions.yaml and derivatives.yaml are
 // merged. Ideally this should all go into native_functions.yaml
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 
 namespace torch {
diff --git a/torch/csrc/jit/serialization/pickle.cpp b/torch/csrc/jit/serialization/pickle.cpp
index c5e03aadc3fc3b..1a0270f288bef1 100644
--- a/torch/csrc/jit/serialization/pickle.cpp
+++ b/torch/csrc/jit/serialization/pickle.cpp
@@ -2,7 +2,7 @@
 
 #include <ATen/core/ivalue.h>
 #include <caffe2/serialize/inline_container.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/serialization/export.h>
 #include <torch/csrc/jit/serialization/import_read.h>
 
diff --git a/torch/csrc/jit/serialization/pickle.h b/torch/csrc/jit/serialization/pickle.h
index e4a1edd09d545e..d9d450ba4b135c 100644
--- a/torch/csrc/jit/serialization/pickle.h
+++ b/torch/csrc/jit/serialization/pickle.h
@@ -3,7 +3,7 @@
 #include <ATen/core/ivalue.h>
 #include <c10/util/ArrayRef.h>
 #include <caffe2/serialize/inline_container.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/serialization/pickler.h>
 #include <torch/csrc/jit/serialization/unpickler.h>
 
diff --git a/torch/csrc/jit/serialization/pickler.h b/torch/csrc/jit/serialization/pickler.h
index 3dc6bef9d9131d..86117c130aa5e1 100644
--- a/torch/csrc/jit/serialization/pickler.h
+++ b/torch/csrc/jit/serialization/pickler.h
@@ -8,7 +8,7 @@
 #include <ATen/core/ivalue.h>
 #include <ATen/core/jit_type.h>
 #include <c10/util/ArrayRef.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/utils/disallow_copy.h>
 
 namespace torch {
diff --git a/torch/csrc/jit/serialization/python_print.h b/torch/csrc/jit/serialization/python_print.h
index b0a05a37cce625..0e95d6de626af1 100644
--- a/torch/csrc/jit/serialization/python_print.h
+++ b/torch/csrc/jit/serialization/python_print.h
@@ -1,5 +1,5 @@
 #pragma once
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <iostream>
 #include <vector>
diff --git a/torch/csrc/jit/serialization/unpickler.h b/torch/csrc/jit/serialization/unpickler.h
index 586ff9cc4ae59b..76c6d820694d08 100644
--- a/torch/csrc/jit/serialization/unpickler.h
+++ b/torch/csrc/jit/serialization/unpickler.h
@@ -3,7 +3,7 @@
 #include <ATen/core/ivalue.h>
 #include <c10/util/ArrayRef.h>
 #include <caffe2/serialize/inline_container.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/serialization/pickler.h>
 
 namespace torch {
diff --git a/torch/csrc/jit/tensorexpr/bounds_inference.h b/torch/csrc/jit/tensorexpr/bounds_inference.h
index 8defed23aaa14c..fadf064bce4556 100644
--- a/torch/csrc/jit/tensorexpr/bounds_inference.h
+++ b/torch/csrc/jit/tensorexpr/bounds_inference.h
@@ -4,7 +4,7 @@
 #include <unordered_map>
 #include <vector>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/tensorexpr/mem_dependency_checker.h>
 
 namespace torch {
diff --git a/torch/csrc/jit/tensorexpr/exceptions.h b/torch/csrc/jit/tensorexpr/exceptions.h
index b6e97f7611583f..b5e656f4080c57 100644
--- a/torch/csrc/jit/tensorexpr/exceptions.h
+++ b/torch/csrc/jit/tensorexpr/exceptions.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/tensorexpr/fwd_decls.h>
 
 #include <sstream>
diff --git a/torch/csrc/jit/tensorexpr/external_functions.h b/torch/csrc/jit/tensorexpr/external_functions.h
index 995927d63fe40b..004578fb81a044 100644
--- a/torch/csrc/jit/tensorexpr/external_functions.h
+++ b/torch/csrc/jit/tensorexpr/external_functions.h
@@ -2,7 +2,7 @@
 
 #include <ATen/Functions.h>
 #include <c10/macros/Macros.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <cstdint>
 #include <vector>
 
diff --git a/torch/csrc/jit/tensorexpr/external_functions_registry.h b/torch/csrc/jit/tensorexpr/external_functions_registry.h
index a3b663b91721a5..935fdbf0412f2a 100644
--- a/torch/csrc/jit/tensorexpr/external_functions_registry.h
+++ b/torch/csrc/jit/tensorexpr/external_functions_registry.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <cstdint>
 #include <string>
 #include <unordered_map>
diff --git a/torch/csrc/jit/tensorexpr/ir_cloner.h b/torch/csrc/jit/tensorexpr/ir_cloner.h
index 5f516a02ffadb3..3cc2f4dcd6fe38 100644
--- a/torch/csrc/jit/tensorexpr/ir_cloner.h
+++ b/torch/csrc/jit/tensorexpr/ir_cloner.h
@@ -1,6 +1,6 @@
 #pragma once
 #include <c10/core/ScalarType.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <vector>
 
 #include <torch/csrc/jit/tensorexpr/ir_mutator.h>
diff --git a/torch/csrc/jit/tensorexpr/ir_mutator.h b/torch/csrc/jit/tensorexpr/ir_mutator.h
index 0a96876606dfb4..27f41185e75f7b 100644
--- a/torch/csrc/jit/tensorexpr/ir_mutator.h
+++ b/torch/csrc/jit/tensorexpr/ir_mutator.h
@@ -1,6 +1,6 @@
 #pragma once
 #include <c10/core/ScalarType.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/tensorexpr/fwd_decls.h>
 #include <vector>
 
diff --git a/torch/csrc/jit/tensorexpr/ir_visitor.h b/torch/csrc/jit/tensorexpr/ir_visitor.h
index e54786b2f90368..26101a2c8fde28 100644
--- a/torch/csrc/jit/tensorexpr/ir_visitor.h
+++ b/torch/csrc/jit/tensorexpr/ir_visitor.h
@@ -1,6 +1,6 @@
 #pragma once
 #include <c10/core/ScalarType.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/tensorexpr/fwd_decls.h>
 
 namespace torch {
diff --git a/torch/csrc/jit/tensorexpr/llvm_codegen.h b/torch/csrc/jit/tensorexpr/llvm_codegen.h
index b155ff8788445c..7ab506fa8fe1ec 100644
--- a/torch/csrc/jit/tensorexpr/llvm_codegen.h
+++ b/torch/csrc/jit/tensorexpr/llvm_codegen.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #ifdef TORCH_ENABLE_LLVM
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/tensorexpr/codegen.h>
 #include <torch/csrc/jit/tensorexpr/ir.h>
diff --git a/torch/csrc/jit/tensorexpr/llvm_jit.h b/torch/csrc/jit/tensorexpr/llvm_jit.h
index a837899cdce1d9..806c5d91767e6a 100644
--- a/torch/csrc/jit/tensorexpr/llvm_jit.h
+++ b/torch/csrc/jit/tensorexpr/llvm_jit.h
@@ -3,7 +3,7 @@
 #ifdef TORCH_ENABLE_LLVM
 #include <c10/util/Exception.h>
 #include <c10/util/Optional.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <llvm/ExecutionEngine/JITSymbol.h>
 #include <llvm/ExecutionEngine/Orc/Core.h>
diff --git a/torch/csrc/jit/tensorexpr/loopnest.h b/torch/csrc/jit/tensorexpr/loopnest.h
index 44885952e90d90..e7e29e699b183d 100644
--- a/torch/csrc/jit/tensorexpr/loopnest.h
+++ b/torch/csrc/jit/tensorexpr/loopnest.h
@@ -5,7 +5,7 @@
 #include <unordered_set>
 #include <vector>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/tensorexpr/fwd_decls.h>
 
 namespace torch {
diff --git a/torch/csrc/jit/tensorexpr/mem_dependency_checker.h b/torch/csrc/jit/tensorexpr/mem_dependency_checker.h
index 348edb55a5ff84..2f1032a1af654f 100644
--- a/torch/csrc/jit/tensorexpr/mem_dependency_checker.h
+++ b/torch/csrc/jit/tensorexpr/mem_dependency_checker.h
@@ -1,6 +1,6 @@
 #pragma once
 #include <c10/core/ScalarType.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <vector>
 
 #include <torch/csrc/jit/tensorexpr/bounds_overlap.h>
diff --git a/torch/csrc/jit/tensorexpr/registerizer.h b/torch/csrc/jit/tensorexpr/registerizer.h
index 75eface21016df..95ad24722c8023 100644
--- a/torch/csrc/jit/tensorexpr/registerizer.h
+++ b/torch/csrc/jit/tensorexpr/registerizer.h
@@ -1,7 +1,7 @@
 #pragma once
 #include <c10/core/ScalarType.h>
 #include <c10/util/irange.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/tensorexpr/hash_provider.h>
 #include <torch/csrc/jit/tensorexpr/ir_mutator.h>
diff --git a/torch/csrc/jit/tensorexpr/tensor.h b/torch/csrc/jit/tensorexpr/tensor.h
index bf9b7788c4415c..18244638be7875 100644
--- a/torch/csrc/jit/tensorexpr/tensor.h
+++ b/torch/csrc/jit/tensorexpr/tensor.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <functional>
 #include <vector>
 
diff --git a/torch/csrc/jit/tensorexpr/types.cpp b/torch/csrc/jit/tensorexpr/types.cpp
index 1e30803ae51b1f..d56ff125a845c5 100644
--- a/torch/csrc/jit/tensorexpr/types.cpp
+++ b/torch/csrc/jit/tensorexpr/types.cpp
@@ -1,6 +1,6 @@
 #include <torch/csrc/jit/tensorexpr/types.h>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/tensorexpr/exceptions.h>
 
 #include <c10/util/Logging.h>
diff --git a/torch/csrc/jit/tensorexpr/types.h b/torch/csrc/jit/tensorexpr/types.h
index 27be03f48ab023..ab77b1d8bdfa79 100644
--- a/torch/csrc/jit/tensorexpr/types.h
+++ b/torch/csrc/jit/tensorexpr/types.h
@@ -5,7 +5,7 @@
 
 #include <c10/core/ScalarType.h>
 #include <c10/util/Logging.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <torch/csrc/jit/tensorexpr/exceptions.h>
 
diff --git a/torch/csrc/jit/tensorexpr/unique_name_manager.h b/torch/csrc/jit/tensorexpr/unique_name_manager.h
index a52e75929f66d6..b6162e6168fd63 100644
--- a/torch/csrc/jit/tensorexpr/unique_name_manager.h
+++ b/torch/csrc/jit/tensorexpr/unique_name_manager.h
@@ -4,7 +4,7 @@
 #include <unordered_map>
 #include <unordered_set>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/tensorexpr/fwd_decls.h>
 
 namespace torch {
diff --git a/torch/csrc/jit/testing/file_check.cpp b/torch/csrc/jit/testing/file_check.cpp
index edbc1c4a8453f2..88ac4d61a66782 100644
--- a/torch/csrc/jit/testing/file_check.cpp
+++ b/torch/csrc/jit/testing/file_check.cpp
@@ -13,7 +13,7 @@
 #include <c10/util/Optional.h>
 #include <c10/util/StringUtil.h>
 #include <c10/util/irange.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/frontend/source_range.h>
 #include <torch/csrc/jit/ir/ir.h>
 #include <torch/csrc/jit/testing/file_check.h>
diff --git a/torch/csrc/jit/testing/file_check.h b/torch/csrc/jit/testing/file_check.h
index dbe2acdf06f160..313254ea42fda9 100644
--- a/torch/csrc/jit/testing/file_check.h
+++ b/torch/csrc/jit/testing/file_check.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <memory>
 #include <string>
 
diff --git a/torch/csrc/jit/testing/hooks_for_testing.h b/torch/csrc/jit/testing/hooks_for_testing.h
index e7d6e4595d29fc..108dea3f1f72d7 100644
--- a/torch/csrc/jit/testing/hooks_for_testing.h
+++ b/torch/csrc/jit/testing/hooks_for_testing.h
@@ -1,5 +1,5 @@
 #pragma once
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <torch/csrc/jit/api/compilation_unit.h>
 #include <functional>
 #include <memory>
diff --git a/torch/csrc/lazy/core/hash.h b/torch/csrc/lazy/core/hash.h
index e998764b7a3ec6..4299f324159538 100644
--- a/torch/csrc/lazy/core/hash.h
+++ b/torch/csrc/lazy/core/hash.h
@@ -12,7 +12,7 @@
 #include <ATen/Tensor.h>
 #include <c10/core/Scalar.h>
 #include <c10/util/int128.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 namespace torch {
 namespace lazy {
diff --git a/torch/csrc/utils.cpp b/torch/csrc/utils.cpp
index fc165008ea11d6..822497e4d6d778 100644
--- a/torch/csrc/utils.cpp
+++ b/torch/csrc/utils.cpp
@@ -6,7 +6,7 @@
 #include <torch/csrc/autograd/variable.h>
 #include <torch/csrc/DynamicTypes.h>
 
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 
 #include <algorithm>
 #include <cstdarg>
diff --git a/torch/csrc/utils/byte_order.h b/torch/csrc/utils/byte_order.h
index f4bef3804b8039..ab09a932c607ce 100644
--- a/torch/csrc/utils/byte_order.h
+++ b/torch/csrc/utils/byte_order.h
@@ -2,7 +2,7 @@
 
 #include <TH/THHalf.h>
 #include <c10/util/BFloat16.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <cstddef>
 #include <cstdint>
 
diff --git a/torch/csrc/utils/crash_handler.h b/torch/csrc/utils/crash_handler.h
index dc119451953726..93f03380945df9 100644
--- a/torch/csrc/utils/crash_handler.h
+++ b/torch/csrc/utils/crash_handler.h
@@ -1,5 +1,5 @@
 #pragma once
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <string>
 
 namespace torch {
diff --git a/torch/csrc/utils/tensor_flatten.h b/torch/csrc/utils/tensor_flatten.h
index 37ebb84b7f4ac9..cc39f4129bf7f5 100644
--- a/torch/csrc/utils/tensor_flatten.h
+++ b/torch/csrc/utils/tensor_flatten.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <ATen/core/functional.h>
-#include <torch/csrc/WindowsTorchApiMacro.h>
+#include <torch/csrc/Export.h>
 #include <ATen/ATen.h>
 #include <utility>
 #include <c10/core/TensorOptions.h>