update eigen version on Windows (#30573)

* update eigen version on Windows, test=develop

* add /bigobj for cl, test=develop

CMakeLists.txt

@@ -67,12 +67,17 @@ if(WIN32)
     set(CMAKE_SUPPRESS_REGENERATION ON)
     set(CMAKE_STATIC_LIBRARY_PREFIX lib)
 
+    set(CMAKE_C_FLAGS_DEBUG   "${CMAKE_C_FLAGS_DEBUG} /bigobj")
+    set(CMAKE_C_FLAGS_RELEASE  "${CMAKE_C_FLAGS_RELEASE} /bigobj")
+    set(CMAKE_CXX_FLAGS_DEBUG  "${CMAKE_CXX_FLAGS_DEBUG} /bigobj")
+    set(CMAKE_CXX_FLAGS_RELEASE   "${CMAKE_CXX_FLAGS_RELEASE} /bigobj")
+
     if (MSVC_STATIC_CRT)
         message(STATUS "Use static C runtime time, refer to https://docs.microsoft.com/en-us/cpp/c-runtime-library/crt-library-features?view=vs-2019")
-        set(CMAKE_C_FLAGS_DEBUG   "${CMAKE_C_FLAGS_DEBUG} /bigobj /MTd")
-        set(CMAKE_C_FLAGS_RELEASE  "${CMAKE_C_FLAGS_RELEASE} /bigobj /MT")
-        set(CMAKE_CXX_FLAGS_DEBUG  "${CMAKE_CXX_FLAGS_DEBUG} /bigobj /MTd")
-        set(CMAKE_CXX_FLAGS_RELEASE   "${CMAKE_CXX_FLAGS_RELEASE} /bigobj /MT")
+        set(CMAKE_C_FLAGS_DEBUG   "${CMAKE_C_FLAGS_DEBUG} /MTd")
+        set(CMAKE_C_FLAGS_RELEASE  "${CMAKE_C_FLAGS_RELEASE} /MT")
+        set(CMAKE_CXX_FLAGS_DEBUG  "${CMAKE_CXX_FLAGS_DEBUG} /MTd")
+        set(CMAKE_CXX_FLAGS_RELEASE   "${CMAKE_CXX_FLAGS_RELEASE} /MT")
         foreach(flag_var
             CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_DEBUG CMAKE_CXX_FLAGS_RELEASE
             CMAKE_CXX_FLAGS_MINSIZEREL CMAKE_CXX_FLAGS_RELWITHDEBINFO

cmake/external/eigen.cmake

@@ -20,21 +20,28 @@ set(EIGEN_SOURCE_DIR ${THIRD_PARTY_PATH}/eigen3/src/extern_eigen3)
 set(EIGEN_REPOSITORY https://gitlab.com/libeigen/eigen.git)
 set(EIGEN_TAG        4da2c6b1974827b1999bab652a3d4703e1992d26)
 
-# the recent version of eigen will cause compilation error on windows
-if(WIN32)
-    set(EIGEN_REPOSITORY ${GIT_URL}/eigenteam/eigen-git-mirror.git)
-    set(EIGEN_TAG        917060c364181f33a735dc023818d5a54f60e54c)
-endif()
-
 cache_third_party(extern_eigen3
     REPOSITORY    ${EIGEN_REPOSITORY}
     TAG           ${EIGEN_TAG}
     DIR           EIGEN_SOURCE_DIR)
 
 if(WIN32)
+    add_definitions(-DEIGEN_STRONG_INLINE=inline)
     file(TO_NATIVE_PATH ${PADDLE_SOURCE_DIR}/patches/eigen/Half.h native_src)
     file(TO_NATIVE_PATH ${EIGEN_SOURCE_DIR}/Eigen/src/Core/arch/CUDA/Half.h native_dst)
-    set(EIGEN_PATCH_COMMAND copy ${native_src} ${native_dst} /Y)
+    # For Windows
+    # which will cause a compilation error in Tensor:74:
+    # "can not open file 'unistd.h'"
+    # so use following patch to solve compilation error On Windows.
+    file(TO_NATIVE_PATH ${PADDLE_SOURCE_DIR}/patches/eigen/Tensor native_src2)
+    file(TO_NATIVE_PATH ${EIGEN_SOURCE_DIR}/unsupported/Eigen/CXX11/Tensor native_dst2)
+    # For VS2015
+    # which will cause a compilation error in TensorBlock.h:1028:
+    # "syntax error"
+    # so use following patch to solve compilation error On Windows.
+    file(TO_NATIVE_PATH ${PADDLE_SOURCE_DIR}/patches/eigen/TensorBlock.h native_src3)
+    file(TO_NATIVE_PATH ${EIGEN_SOURCE_DIR}/unsupported/Eigen/CXX11/src/Tensor/TensorBlock.h native_dst3)
+    set(EIGEN_PATCH_COMMAND copy ${native_src} ${native_dst} /Y && copy ${native_src2} ${native_dst2} /Y && copy ${native_src3} ${native_dst3} /Y)
 elseif(LINUX)
     # For gxx=4.8, __GXX_ABI_VERSION is less than 1004
     # which will cause a compilation error in Geometry_SSE.h:38:

paddle/fluid/platform/device_context.cc

@@ -278,26 +278,14 @@ class EigenCudaStreamDevice : public Eigen::StreamInterface {
 
   void* scratchpad() const override {
     if (scratch_ == NULL) {
-// windows use an old version of eigen that uses kCudaScratchSize,
-// once windows updates eigen to a recent version, the following code
-// can use kGpuScratchSize uniformly
-#ifdef _WIN32
-      scratch_ = allocate(Eigen::kCudaScratchSize + sizeof(unsigned int));
-#else
       scratch_ = allocate(Eigen::kGpuScratchSize + sizeof(unsigned int));
-#endif
     }
     return scratch_;
   }
 
   unsigned int* semaphore() const override {
     if (semaphore_ == NULL) {
-#ifdef _WIN32
-      char* scratch =
-          static_cast<char*>(scratchpad()) + Eigen::kCudaScratchSize;
-#else
       char* scratch = static_cast<char*>(scratchpad()) + Eigen::kGpuScratchSize;
-#endif
       semaphore_ = reinterpret_cast<unsigned int*>(scratch);
 #ifdef PADDLE_WITH_HIP
       PADDLE_ENFORCE_CUDA_SUCCESS(

patches/eigen/Tensor

+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2014 Benoit Steiner <benoit.steiner.goog@gmail.com>
+// Copyright (C) 2013 Christian Seiler <christian@iwakd.de>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+//#ifndef EIGEN_CXX11_TENSOR_MODULE
+//#define EIGEN_CXX11_TENSOR_MODULE
+
+#include "../../../Eigen/Core"
+
+#if EIGEN_HAS_CXX11
+
+#include "../SpecialFunctions"
+
+#include "../../../Eigen/src/Core/util/DisableStupidWarnings.h"
+#include "src/util/CXX11Meta.h"
+#include "src/util/MaxSizeVector.h"
+
+/** \defgroup CXX11_Tensor_Module Tensor Module
+  *
+  * This module provides a Tensor class for storing arbitrarily indexed
+  * objects.
+  *
+  * \code
+  * #include <Eigen/CXX11/Tensor>
+  * \endcode
+  *
+  * Much of the documentation can be found \ref eigen_tensors "here".
+  */
+
+#include <cmath>
+#include <cstddef>
+#include <cstring>
+#include <random>
+
+#ifdef _WIN32
+typedef __int16 int16_t;
+typedef unsigned __int16 uint16_t;
+typedef __int32 int32_t;
+typedef unsigned __int32 uint32_t;
+typedef __int64 int64_t;
+typedef unsigned __int64 uint64_t;
+#include <windows.h>
+#else
+#include <stdint.h>
+#include <unistd.h>
+#endif
+
+#ifdef _WIN32
+#include <windows.h>
+#elif defined(__APPLE__)
+#include <mach/mach_time.h>
+#else
+#include <time.h>
+#endif
+
+#if defined(EIGEN_USE_THREADS) || defined(EIGEN_USE_SYCL)
+#include "ThreadPool"
+#endif
+
+#ifdef EIGEN_USE_GPU
+  #include <iostream>
+  #if defined(EIGEN_USE_HIP)
+    #include <hip/hip_runtime.h>
+  #else
+    #include <cuda_runtime.h>
+  #endif
+  #include <atomic>
+#endif
+
+#include "src/Tensor/TensorMacros.h"
+#include "src/Tensor/TensorForwardDeclarations.h"
+#include "src/Tensor/TensorMeta.h"
+#include "src/Tensor/TensorFunctors.h"
+#include "src/Tensor/TensorCostModel.h"
+#include "src/Tensor/TensorDeviceDefault.h"
+#include "src/Tensor/TensorDeviceThreadPool.h"
+#include "src/Tensor/TensorDeviceGpu.h"
+#ifndef gpu_assert
+#define gpu_assert(x)
+#endif
+#include "src/Tensor/TensorDeviceSycl.h"
+#include "src/Tensor/TensorIndexList.h"
+#include "src/Tensor/TensorDimensionList.h"
+#include "src/Tensor/TensorDimensions.h"
+#include "src/Tensor/TensorInitializer.h"
+#include "src/Tensor/TensorTraits.h"
+#include "src/Tensor/TensorRandom.h"
+#include "src/Tensor/TensorUInt128.h"
+#include "src/Tensor/TensorIntDiv.h"
+#include "src/Tensor/TensorGlobalFunctions.h"
+
+#include "src/Tensor/TensorBase.h"
+#include "src/Tensor/TensorBlock.h"
+
+#include "src/Tensor/TensorEvaluator.h"
+#include "src/Tensor/TensorExpr.h"
+#include "src/Tensor/TensorReduction.h"
+#include "src/Tensor/TensorReductionGpu.h"
+#include "src/Tensor/TensorArgMax.h"
+#include "src/Tensor/TensorConcatenation.h"
+#include "src/Tensor/TensorContractionMapper.h"
+#include "src/Tensor/TensorContractionBlocking.h"
+#include "src/Tensor/TensorContraction.h"
+#include "src/Tensor/TensorContractionThreadPool.h"
+#include "src/Tensor/TensorContractionGpu.h"
+#include "src/Tensor/TensorConversion.h"
+#include "src/Tensor/TensorConvolution.h"
+#include "src/Tensor/TensorFFT.h"
+#include "src/Tensor/TensorPatch.h"
+#include "src/Tensor/TensorImagePatch.h"
+#include "src/Tensor/TensorVolumePatch.h"
+#include "src/Tensor/TensorBroadcasting.h"
+#include "src/Tensor/TensorChipping.h"
+#include "src/Tensor/TensorInflation.h"
+#include "src/Tensor/TensorLayoutSwap.h"
+#include "src/Tensor/TensorMorphing.h"
+#include "src/Tensor/TensorPadding.h"
+#include "src/Tensor/TensorReverse.h"
+#include "src/Tensor/TensorShuffling.h"
+#include "src/Tensor/TensorStriding.h"
+#include "src/Tensor/TensorCustomOp.h"
+#include "src/Tensor/TensorEvalTo.h"
+#include "src/Tensor/TensorForcedEval.h"
+#include "src/Tensor/TensorGenerator.h"
+#include "src/Tensor/TensorAssign.h"
+#include "src/Tensor/TensorScan.h"
+#include "src/Tensor/TensorTrace.h"
+
+#ifdef EIGEN_USE_SYCL
+#include "src/Tensor/TensorReductionSycl.h"
+#include "src/Tensor/TensorConvolutionSycl.h"
+#include "src/Tensor/TensorContractionSycl.h"
+#include "src/Tensor/TensorScanSycl.h"
+#endif
+
+#include "src/Tensor/TensorExecutor.h"
+#include "src/Tensor/TensorDevice.h"
+
+#include "src/Tensor/TensorStorage.h"
+#include "src/Tensor/Tensor.h"
+#include "src/Tensor/TensorFixedSize.h"
+#include "src/Tensor/TensorMap.h"
+#include "src/Tensor/TensorRef.h"
+
+#include "src/Tensor/TensorIO.h"
+
+#include "../../../Eigen/src/Core/util/ReenableStupidWarnings.h"
+
+#endif  // EIGEN_HAS_CXX11
+//#endif // EIGEN_CXX11_TENSOR_MODULE