ROCm · iotamudelta · Aug 6, 2018 · Aug 3, 2018 · Aug 3, 2018 · Aug 3, 2018
diff --git a/.gitignore b/.gitignore
@@ -141,10 +141,6 @@ docs/source/scripts/activation_images/
 # PyCharm files
 .idea
 
-# Visual Studio Code files
-.vscode
-.vs
-
 # OSX dir files
 .DS_Store
 
@@ -195,3 +191,11 @@ caffe2.egg-info
 
 # Atom/Watchman required file
 .watchmanconfig
+
+# BEGIN NOT-CLEAN-FILES (setup.py handles this marker. Do not change.)
+#
+# Below files are not deleted by "setup.py clean".
+
+# Visual Studio Code files
+.vscode
+.vs
diff --git a/README.md b/README.md
@@ -148,7 +148,9 @@ If you want to compile with CUDA support, install
 If you want to disable CUDA support, export environment variable `NO_CUDA=1`.
 Other potentially useful environment variables may be found in `setup.py`.
 
-If you want to build on Windows, Visual Studio 2017 and NVTX are also needed.
+If you want to build on Windows, Visual Studio 2017 14.11 toolset and NVTX are also needed.
+Especially, for CUDA 8 build on Windows, there will be an additional requirement for VS 2015 Update 3 and a patch for it.
+The details of the patch can be found out [here](https://support.microsoft.com/en-gb/help/4020481/fix-link-exe-crashes-with-a-fatal-lnk1000-error-when-you-use-wholearch).
 
 #### Install optional dependencies
 
@@ -196,9 +198,10 @@ On Windows
 set "VS150COMNTOOLS=C:\Program Files (x86)\Microsoft Visual Studio\2017\Enterprise\VC\Auxiliary\Build"
 set CMAKE_GENERATOR=Visual Studio 15 2017 Win64
 set DISTUTILS_USE_SDK=1
-REM The following line is needed for Python 2.7, but the support for it is very experimental.
+REM The following two lines are needed for Python 2.7, but the support for it is very experimental.
 set MSSdk=1
-REM As for CUDA 8, VS2015 Update 2 or up is required to build PyTorch. Use the following two lines.
+set FORCE_PY27_BUILD=1
+REM As for CUDA 8, VS2015 Update 3 is also required to build PyTorch. Use the following two lines.
 set "PREBUILD_COMMAND=%VS140COMNTOOLS%\..\..\VC\vcvarsall.bat"
 set PREBUILD_COMMAND_ARGS=x64
 

diff --git a/aten/doc/Type.h b/aten/doc/Type.h
@@ -5,14 +5,14 @@
 #include <functional>
 
 #include "ATen/ATenGeneral.h"
-#include "ATen/ArrayRef.h"
+#include "ATen/Allocator.h"
 #include "ATen/Generator.h"
-#include "ATen/Half.h"
-#include "ATen/SparseTensorRef.h"
-#include "ATen/ScalarType.h"
 #include "ATen/Scalar.h"
+#include "ATen/ScalarType.h"
+#include "ATen/SparseTensorRef.h"
 #include "ATen/Tensor.h"
-#include "ATen/Allocator.h"
+#include "ATen/core/ArrayRef.h"
+#include "ATen/core/Half.h"
 
 // To solve the conflict of s_addr in inaddr.h
 #ifdef _MSC_VER

diff --git a/aten/src/ATen/ATen.h b/aten/src/ATen/ATen.h
@@ -1,25 +1,25 @@
 #pragma once
 
 #include "ATen/ATenGeneral.h"
-#include "ATen/CPUGeneral.h"
 #include "ATen/Allocator.h"
-#include "ATen/Scalar.h"
-#include "ATen/Type.h"
-#include "ATen/Generator.h"
+#include "ATen/CPUGeneral.h"
+#include "ATen/CUDAGuard.h"
 #include "ATen/Context.h"
+#include "ATen/Device.h"
+#include "ATen/DeviceGuard.h"
+#include "ATen/DimVector.h"
+#include "ATen/Dispatch.h"
+#include "ATen/Formatting.h"
+#include "ATen/Functions.h"
+#include "ATen/Generator.h"
+#include "ATen/Layout.h"
+#include "ATen/OptionsGuard.h"
+#include "ATen/Scalar.h"
 #include "ATen/Storage.h"
 #include "ATen/Tensor.h"
-#include "ATen/Device.h"
 #include "ATen/TensorGeometry.h"
-#include "ATen/Functions.h"
-#include "ATen/Formatting.h"
-#include "ATen/TensorOperators.h"
 #include "ATen/TensorMethods.h"
-#include "ATen/Dispatch.h"
-#include "ATen/DimVector.h"
-#include "ATen/DeviceGuard.h"
+#include "ATen/TensorOperators.h"
 #include "ATen/TensorOptions.h"
-#include "ATen/Layout.h"
-#include "ATen/OptionsGuard.h"
-#include "ATen/CUDAGuard.h"
-#include "ATen/Error.h"
+#include "ATen/Type.h"
+#include "ATen/core/Error.h"
diff --git a/aten/src/ATen/ATenGeneral.cpp b/aten/src/ATen/ATenGeneral.cpp
@@ -0,0 +1 @@
+#include <ATen/ATenGeneral.h>
diff --git a/aten/src/ATen/AccumulateType.h b/aten/src/ATen/AccumulateType.h
@@ -1,6 +1,6 @@
 #pragma once
 #include "ATen/Config.h"
-#include "ATen/Half.h"
+#include "ATen/core/Half.h"
 
 // Defines the accumulation type for a scalar type.
 // Example:

diff --git a/aten/src/ATen/Allocator.h b/aten/src/ATen/Allocator.h
@@ -3,9 +3,9 @@
 #include <memory>
 #include <stddef.h>
 
-#include <ATen/Error.h>
-#include <ATen/Retainable.h>
 #include <ATen/Device.h>
+#include <ATen/Retainable.h>
+#include <ATen/core/Error.h>
 #include <ATen/core/UniqueVoidPtr.h>
 
 namespace at {

diff --git a/aten/src/ATen/ArrayRef.cpp b/aten/src/ATen/ArrayRef.cpp
diff --git a/aten/src/ATen/CPUFixedAllocator.h b/aten/src/ATen/CPUFixedAllocator.h
@@ -1,7 +1,7 @@
 #pragma once
 
+#include "ATen/core/Error.h"
 #include "TH/TH.h"
-#include "ATen/Error.h"
 
 // This file creates a fake allocator that just throws exceptions if
 // it is actually used.

diff --git a/aten/src/ATen/CheckGenerator.h b/aten/src/ATen/CheckGenerator.h
@@ -1,8 +1,8 @@
 #pragma once
 
-#include "ATen/Error.h"
 #include "ATen/Generator.h"
 #include "ATen/Utils.h"
+#include "ATen/core/Error.h"
 
 namespace at {
 

diff --git a/aten/src/ATen/Context.cpp b/aten/src/ATen/Context.cpp
@@ -37,6 +37,8 @@ Context::Context()
   Type::registerCPU(this);
 }
 
+// TODO: This could be bad juju if someone calls globalContext() in the
+// destructor of an object with static lifetime.
 Context & globalContext() {
   static Context globalContext_;
   return globalContext_;

diff --git a/aten/src/ATen/Context.h b/aten/src/ATen/Context.h
@@ -1,13 +1,13 @@
 #pragma once
 
-#include "ATen/ATenGeneral.h"
 #include <ATen/CPUGeneral.h>
+#include "ATen/ATenGeneral.h"
+#include "ATen/CUDAStream.h"
 #include "ATen/Generator.h"
 #include "ATen/Type.h"
 #include "ATen/Utils.h"
-#include "ATen/Error.h"
+#include "ATen/core/Error.h"
 #include "ATen/detail/CUDAHooksInterface.h"
-#include "ATen/CUDAStream.h"
 
 // This is temporary
 #include "ATen/core/ATenCoreTest.h"

diff --git a/aten/src/ATen/Device.cpp b/aten/src/ATen/Device.cpp
@@ -1,6 +1,6 @@
 #include <ATen/Device.h>
 
-#include <ATen/Error.h>
+#include <ATen/core/Error.h>
 
 #include <exception>
 #include <ostream>
@@ -77,20 +77,6 @@ Device::Device(const std::string& device_string) : Device(Type::CPU) {
 
 } // namespace at
 
-std::ostream& operator<<(std::ostream& stream, at::Device::Type type) {
-  switch (type) {
-    case at::Device::Type::CPU: {
-      stream << "cpu";
-      break;
-    }
-    case at::Device::Type::CUDA: {
-      stream << "cuda";
-      break;
-    }
-  }
-  return stream;
-}
-
 std::ostream& operator<<(std::ostream& stream, const at::Device& device) {
   stream << device.type();
   if (device.has_index()) {

diff --git a/aten/src/ATen/Device.h b/aten/src/ATen/Device.h
@@ -1,7 +1,8 @@
 #pragma once
 
-#include <ATen/Error.h>
 #include <ATen/ScalarType.h>
+#include <ATen/core/Error.h>
+#include <ATen/core/DeviceType.h>
 
 #include <cstddef>
 #include <iosfwd>
@@ -20,26 +21,25 @@ namespace at {
 /// represents a specific, concrete device,
 /// 2. When the device type is CPU, the device index must be zero.
 struct Device {
-  /// The possible values of the device *type*.
-  enum class Type { CPU, CUDA };
+  using Type = at::DeviceType;
 
-  /// Converts a `Backend` to a `Device::Type` if possible.
-  static Type backend_to_type(Backend backend) {
+  /// Converts a `Backend` to a `DeviceType` if possible.
+  static DeviceType backend_to_type(Backend backend) {
     switch (backend) {
       case kCPU:
       case kSparseCPU:
-        return Type::CPU;
+        return DeviceType::CPU;
       case kCUDA:
       case kSparseCUDA:
-        return Type::CUDA;
+        return DeviceType::CUDA;
       default:
         AT_ERROR(
             "Invalid backend ", toString(backend), " for Device construction");
     }
   }
 
-  /// Constructs a new `Device` from a `Type` and an optional device index.
-  /* implicit */ Device(Type type, int32_t index = -1)
+  /// Constructs a new `Device` from a `DeviceType` and an optional device index.
+  /* implicit */ Device(DeviceType type, int32_t index = -1)
       : type_(type), index_(index) {
     AT_CHECK(
         index == -1 || index >= 0,
@@ -59,7 +59,7 @@ struct Device {
   /* implicit */ Device(const std::string& device_string);
 
   /// Constructs a new `Device` from a `Backend` (which is converted to a
-  /// `Type`, if possible) and an optional device index.
+  /// `DeviceType`, if possible) and an optional device index.
   /* implicit */ Device(Backend backend, int32_t index = -1)
       : Device(backend_to_type(backend), index) {}
 
@@ -81,7 +81,7 @@ struct Device {
   }
 
   /// Returns the type of device this is.
-  Type type() const noexcept {
+  DeviceType type() const noexcept {
     return type_;
   }
 
@@ -97,21 +97,20 @@ struct Device {
 
   /// Return true if the device is of CUDA type.
   bool is_cuda() const noexcept {
-    return type_ == Type::CUDA;
+    return type_ == DeviceType::CUDA;
   }
 
   /// Return true if the device is of CPU type.
   bool is_cpu() const noexcept {
-    return type_ == Type::CPU;
+    return type_ == DeviceType::CPU;
   }
 
  private:
-  Type type_;
+  DeviceType type_;
   int32_t index_ = -1;
 };
 } // namespace at
 
-AT_API std::ostream& operator<<(std::ostream& stream, at::Device::Type type);
 AT_API std::ostream& operator<<(std::ostream& stream, const at::Device& device);
 
 namespace std {

diff --git a/aten/src/ATen/DeviceGuard.h b/aten/src/ATen/DeviceGuard.h
@@ -1,9 +1,9 @@
 #pragma once
 
 #include <ATen/Device.h>
-#include <ATen/Error.h>
 #include <ATen/ScalarType.h>
 #include <ATen/Tensor.h>
+#include <ATen/core/Error.h>
 #include <ATen/detail/CUDAHooksInterface.h>
 
 #include <cstddef>

diff --git a/aten/src/ATen/DimVector.h b/aten/src/ATen/DimVector.h
@@ -1,6 +1,6 @@
 #pragma once
 
-#include "SmallVector.h"
+#include <ATen/core/SmallVector.h>
 #include <stdint.h>
 
 namespace at {

diff --git a/aten/src/ATen/Dispatch.h b/aten/src/ATen/Dispatch.h
@@ -1,8 +1,8 @@
 #pragma once
 
-#include <ATen/Error.h>
-#include <ATen/Half.h>
 #include <ATen/Type.h>
+#include <ATen/core/Error.h>
+#include <ATen/core/Half.h>
 
 #define AT_PRIVATE_CASE_TYPE(enum_type, type, ...) \
   case enum_type: {                                \

diff --git a/aten/src/ATen/ExpandUtils.h b/aten/src/ATen/ExpandUtils.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include "ATen/Tensor.h"
-#include "ATen/Error.h"
+#include "ATen/core/Error.h"
 
 #include <functional>
 #include <sstream>

diff --git a/aten/src/ATen/Layout.h b/aten/src/ATen/Layout.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <ATen/ScalarType.h>
-#include <ATen/Error.h>
+#include <ATen/core/Error.h>
 
 #include <iostream>
 

diff --git a/aten/src/ATen/MatrixRef.h b/aten/src/ATen/MatrixRef.h
@@ -1,6 +1,6 @@
 #pragma once
-#include <ATen/ArrayRef.h>
 #include <ATen/Utils.h>
+#include <ATen/core/ArrayRef.h>
 
 #include <vector>
 

diff --git a/aten/src/ATen/OptionsGuard.cpp b/aten/src/ATen/OptionsGuard.cpp
@@ -1,5 +1,5 @@
 #include <ATen/OptionsGuard.h>
-#include <ATen/optional.h>
+#include <ATen/core/optional.h>
 
 namespace at {
 

diff --git a/aten/src/ATen/OptionsGuard.h b/aten/src/ATen/OptionsGuard.h
@@ -4,7 +4,7 @@
 #include <ATen/Layout.h>
 #include <ATen/ScalarType.h>
 #include <ATen/TensorOptions.h>
-#include <ATen/optional.h>
+#include <ATen/core/optional.h>
 
 namespace at {
 

diff --git a/aten/src/ATen/Registry.h b/aten/src/ATen/Registry.h
@@ -18,8 +18,8 @@
 #include <string>
 #include <vector>
 
-#include <ATen/Backtrace.h>
 #include <ATen/ATenGeneral.h>
+#include <ATen/core/Backtrace.h>
 
 namespace at {
 

diff --git a/aten/src/ATen/Scalar.h b/aten/src/ATen/Scalar.h
@@ -7,10 +7,9 @@
 #include <utility>
 
 #include "ATen/ATenGeneral.h"
-#include "ATen/Half.h"
 #include "ATen/ScalarType.h"
 #include "ATen/TensorBase.h"
-
+#include "ATen/core/Half.h"
 
 namespace at {
 

diff --git a/aten/src/ATen/ScalarType.h b/aten/src/ATen/ScalarType.h
@@ -1,8 +1,8 @@
 #pragma once
 
-#include "ATen/ArrayRef.h"
 #include "ATen/ATenGeneral.h"
-#include "ATen/Half.h"
+#include "ATen/core/ArrayRef.h"
+#include "ATen/core/Half.h"
 
 #include <cstdint>
 #include <iostream>