Merge remote-tracking branch 'upstream/master' into ifu

Author: iotamudelta

CMakeLists.txt

@@ -404,24 +404,16 @@ else()
 endif()
 
 # ---[ Modules
-# TODO(orionr): Enable all of this for Windows DLL when we
-# can figure out how to get it to build
-if (NOT (MSVC AND BUILD_SHARED_LIBS))
 add_subdirectory(modules)
-endif()
 
 # ---[ Binaries
 # Binaries will be built after the Caffe2 main libraries and the modules
 # are built. For the binaries, they will be linked to the Caffe2 main
 # libraries, as well as all the modules that are built with Caffe2 (the ones
 # built in the previous Modules section above).
-# TODO(orionr): Enable all of this for Windows DLL when we
-# can figure out how to get it to build
-if (NOT (MSVC AND BUILD_SHARED_LIBS))
 if (BUILD_BINARY)
   add_subdirectory(binaries)
 endif()
-endif()
 
 include(cmake/Summary.cmake)
 caffe2_print_configuration_summary()

aten/src/ATen/cuda/ATenCUDAGeneral.h

@@ -1,7 +1,7 @@
 #pragma once
 
 #ifdef _WIN32
-# if defined(ATen_cuda_EXPORTS) || defined(caffe2_gpu_EXPORTS) || defined(CAFFE2_BUILD_MAIN_LIB)
+# if defined(ATen_cuda_EXPORTS) || defined(caffe2_gpu_EXPORTS) || defined(CAFFE2_CUDA_BUILD_MAIN_LIB)
 #  define AT_CUDA_API __declspec(dllexport)
 # else
 #  define AT_CUDA_API __declspec(dllimport)

aten/src/THC/THCGeneral.h.in

@@ -22,7 +22,7 @@
 #endif
 
 #ifdef _WIN32
-# if defined(ATen_cuda_EXPORTS) || defined(caffe2_gpu_EXPORTS) || defined(CAFFE2_BUILD_MAIN_LIB)
+# if defined(ATen_cuda_EXPORTS) || defined(caffe2_gpu_EXPORTS) || defined(CAFFE2_CUDA_BUILD_MAIN_LIB)
 #  define THC_API THC_EXTERNC __declspec(dllexport)
 #  define THC_CLASS __declspec(dllexport)
 # else

caffe2/CMakeLists.txt

@@ -65,11 +65,6 @@ endif()
 # ---[ Caffe2 build
 # Note: the folders that are being commented out have not been properly
 # addressed yet.
-# TODO(orionr): Enable all of this for Windows DLL when we
-# can figure out how to get it to build
-if (MSVC AND BUILD_SHARED_LIBS)
-add_subdirectory(utils)
-else()
 add_subdirectory(proto)
 add_subdirectory(contrib)
 add_subdirectory(core)
@@ -101,7 +96,6 @@ add_subdirectory(sgd)
 add_subdirectory(share)
 # add_subdirectory(test) # todo: use caffe2_gtest_main instead of gtest_main because we will need to call GlobalInit
 add_subdirectory(transforms)
-endif()
 
 # Advanced: if we have white list specified, we will do intersections for all
 # main lib srcs.
@@ -166,16 +160,8 @@ if (FALSE)
 endif()
 
 # ---[ List of libraries to link with
-# TODO(orionr): Enable all of this for Windows DLL when we
-# can figure out how to get it to build
-if (NOT (MSVC AND BUILD_SHARED_LIBS))
 add_library(caffe2_protos STATIC $<TARGET_OBJECTS:Caffe2_PROTO>)
 add_dependencies(caffe2_protos Caffe2_PROTO)
-else()
-# Do not include caffe2 or caffe protos, but rather have it only be
-# a library to attach local protobuf.
-add_library(caffe2_protos STATIC utils/dummy.cpp)
-endif()
 # If we are going to link protobuf locally inside caffe2 libraries, what we will do is
 # to create a helper static library that always contains libprotobuf source files, and
 # link the caffe2 related dependent libraries to it.
@@ -341,7 +327,7 @@ if(USE_CUDA)
   # NB: This must be target_compile_definitions, not target_compile_options,
   # as the latter is not respected by nvcc
   if (MSVC)
-      target_compile_definitions(caffe2_gpu PRIVATE "-DCAFFE2_BUILD_MAIN_LIB")
+	  target_compile_definitions(caffe2_gpu PRIVATE "-DCAFFE2_CUDA_BUILD_MAIN_LIB")
   endif()
 
   # Set standard properties on the target
@@ -401,9 +387,6 @@ if ($ENV{WERROR})
 endif()
 
 # ---[ Test binaries.
-# TODO(orionr): Enable all of this for Windows DLL when we
-# can figure out how to get it to build
-if (NOT (MSVC AND BUILD_SHARED_LIBS))
 if (BUILD_TEST)
   set(Caffe2_ALL_TEST_SRCS ${Caffe2_CPU_TEST_SRCS})
   if (USE_CUDA)
@@ -414,6 +397,8 @@ if (BUILD_TEST)
     get_filename_component(test_name ${test_src} NAME_WE)
     add_executable(${test_name} "${test_src}")
     target_link_libraries(${test_name} ${Caffe2_MAIN_LIBS} gtest_main)
+    target_include_directories(${test_name} PRIVATE $<INSTALL_INTERFACE:include>)
+    target_include_directories(${test_name} PRIVATE ${Caffe2_CPU_INCLUDE})
     if (${CMAKE_MAJOR_VERSION}.${CMAKE_MINOR_VERSION} GREATER 3.0)
       target_compile_features(${test_name} PRIVATE cxx_range_for)
     endif()
@@ -429,6 +414,8 @@ if (BUILD_TEST)
       get_filename_component(test_name ${test_src} NAME_WE)
       hip_add_executable(${test_name} "${test_src}")
       target_link_libraries(${test_name} ${Caffe2_MAIN_LIBS} gtest_main)
+      target_include_directories(${test_name} PRIVATE $<INSTALL_INTERFACE:include>)
+      target_include_directories(${test_name} PRIVATE ${Caffe2_CPU_INCLUDE})
       if (${CMAKE_MAJOR_VERSION}.${CMAKE_MINOR_VERSION} GREATER 3.0)
         target_compile_features(${test_name} PRIVATE cxx_range_for)
       endif()
@@ -439,13 +426,13 @@ if (BUILD_TEST)
     endforeach()
   endif()
 endif()
-endif()
 
 set(__aten_test_dir "test/aten")
 if (NOT USE_ROCM)
   foreach(test_src ${ATen_CPU_TEST_SRCS})
     get_filename_component(test_name ${test_src} NAME_WE)
     add_executable(${test_name} "${test_src}")
+    target_include_directories(${test_name} PRIVATE $<INSTALL_INTERFACE:include>)
     target_include_directories(${test_name} PRIVATE ${Caffe2_CPU_INCLUDE})
     target_include_directories(${test_name} SYSTEM PRIVATE ${Caffe2_DEPENDENCY_INCLUDE})
     target_link_libraries(${test_name} ${Caffe2_MAIN_LIBS})
@@ -457,6 +444,7 @@ if (NOT USE_ROCM)
     foreach(test_src ${ATen_CUDA_TEST_SRCS})
       get_filename_component(test_name ${test_src} NAME_WE)
       torch_cuda_based_add_executable(${test_name} "${test_src}")
+      target_include_directories(${test_name} PRIVATE $<INSTALL_INTERFACE:include>)
       target_include_directories(${test_name} PRIVATE ${Caffe2_CPU_INCLUDE})
       target_include_directories(${test_name} SYSTEM PRIVATE ${Caffe2_DEPENDENCY_INCLUDE})
       target_link_libraries(${test_name} ${Caffe2_MAIN_LIBS})
@@ -466,9 +454,6 @@ if (NOT USE_ROCM)
   endif()
 endif()
 
-# TODO(orionr): Enable all of this for Windows DLL when we
-# can figure out how to get it to build
-if (NOT (MSVC AND BUILD_SHARED_LIBS))
 if (BUILD_PYTHON)
   # Python site-packages
   # Get canonical directory for python site packages (relative to install
@@ -525,6 +510,9 @@ if (BUILD_PYTHON)
   if (APPLE)
     set_target_properties(caffe2_pybind11_state PROPERTIES LINK_FLAGS "-undefined dynamic_lookup")
   endif()
+  target_include_directories(caffe2_pybind11_state PRIVATE $<INSTALL_INTERFACE:include>)
+  target_include_directories(caffe2_pybind11_state PRIVATE ${Caffe2_CPU_INCLUDE})
+
   target_link_libraries(
       caffe2_pybind11_state caffe2_library)
   if (WIN32)
@@ -547,6 +535,8 @@ if (BUILD_PYTHON)
     if (APPLE)
       set_target_properties(caffe2_pybind11_state_gpu PROPERTIES LINK_FLAGS "-undefined dynamic_lookup")
     endif()
+    target_include_directories(caffe2_pybind11_state_gpu PRIVATE $<INSTALL_INTERFACE:include>)
+    target_include_directories(caffe2_pybind11_state_gpu PRIVATE ${Caffe2_CPU_INCLUDE})
     target_link_libraries(
         caffe2_pybind11_state_gpu caffe2_library caffe2_gpu_library)
     if (WIN32)
@@ -570,6 +560,8 @@ if (BUILD_PYTHON)
     if (APPLE)
       set_target_properties(caffe2_pybind11_state_hip PROPERTIES LINK_FLAGS "-undefined dynamic_lookup")
     endif()
+    target_include_directories(caffe2_pybind11_state_hip PRIVATE $<INSTALL_INTERFACE:include>)
+    target_include_directories(caffe2_pybind11_state_hip PRIVATE ${Caffe2_CPU_INCLUDE})
     target_link_libraries(
         caffe2_pybind11_state_hip caffe2_library caffe2_hip_library)
     if (WIN32)
@@ -657,7 +649,6 @@ if (BUILD_PYTHON)
   install(DIRECTORY ${CMAKE_BINARY_DIR}/caffe2 DESTINATION ${PYTHON_LIB_REL_PATH}
           FILES_MATCHING PATTERN "*.py")
 endif()
-endif()
 
 # Finally, set the Caffe2_MAIN_LIBS variable in the parent scope.
 set(Caffe2_MAIN_LIBS ${Caffe2_MAIN_LIBS} PARENT_SCOPE)

caffe2/core/common_gpu.h

@@ -25,6 +25,42 @@
 #include "caffe2/core/logging.h"
 #include "caffe2/core/common.h"
 
+// Defines CAFFE2_CUDA_EXPORT and CAFFE2_CUDA_IMPORT. On Windows, this corresponds to
+// different declarations (dllexport and dllimport). On Linux/Mac, it just
+// resolves to the same "default visibility" setting.
+#if defined(_MSC_VER)
+#if defined(CAFFE2_BUILD_SHARED_LIBS)
+#define CAFFE2_CUDA_EXPORT __declspec(dllexport)
+#define CAFFE2_CUDA_IMPORT __declspec(dllimport)
+#else
+#define CAFFE2_CUDA_EXPORT
+#define CAFFE2_CUDA_IMPORT
+#endif
+#else
+#if defined(__GNUC__)
+#define CAFFE2_CUDA_EXPORT __attribute__((__visibility__("default")))
+#else
+#define CAFFE2_CUDA_EXPORT
+#endif
+#define CAFFE2_CUDA_IMPORT CAFFE2_CUDA_EXPORT
+#endif
+
+// CAFFE2_CUDA_API is a macro that, depends on whether you are building the
+// main caffe2 library or not, resolves to either CAFFE2_CUDA_EXPORT or
+// CAFFE2_CUDA_IMPORT.
+//
+// This is used in e.g. Caffe2's protobuf files: when building the main library,
+// it is defined as CAFFE2_CUDA_EXPORT to fix a Windows global-variable-in-dll
+// issue, and for anyone dependent on Caffe2 it will be defined as
+// CAFFE2_CUDA_IMPORT.
+
+#ifdef CAFFE2_CUDA_BUILD_MAIN_LIB
+#define CAFFE2_CUDA_API CAFFE2_CUDA_EXPORT
+#else
+#define CAFFE2_CUDA_API CAFFE2_CUDA_IMPORT
+#endif
+
+
 // This is a macro defined for cuda fp16 support. In default, cuda fp16 is
 // supported by NVCC 7.5, but it is also included in the Tegra X1 platform with
 // a (custom?) NVCC 7.0. As a result, we would normally just check the cuda

caffe2/core/context_gpu.h

@@ -33,7 +33,7 @@ enum class CudaMemoryPoolType {
  *
  * The memory pool is set up during caffe2's global initialization time.
  */
-CAFFE2_API CudaMemoryPoolType GetCudaMemoryPoolType();
+CAFFE2_CUDA_API CudaMemoryPoolType GetCudaMemoryPoolType();
 
 /**
  * A struct to host thread-local cuda objects.
@@ -44,7 +44,7 @@ CAFFE2_API CudaMemoryPoolType GetCudaMemoryPoolType();
  * and deallocating these objects at the thread scope. This class is solely
  * used inside CUDAContext and should not be used externally.
  */
-class CAFFE2_API ThreadLocalCUDAObjects {
+class CAFFE2_CUDA_API ThreadLocalCUDAObjects {
   friend class CUDAContext;
 
  private:
@@ -135,9 +135,9 @@ class CAFFE2_API ThreadLocalCUDAObjects {
 #endif // CAFFE2_USE_CUDNN
 };
 
-CAFFE2_API BaseStaticContext* GetCUDAStaticContext();
+CAFFE2_CUDA_API BaseStaticContext* GetCUDAStaticContext();
 
-class CAFFE2_API CUDAContext final : public BaseContext {
+class CAFFE2_CUDA_API CUDAContext final : public BaseContext {
  public:
   // The default cuda context constructor.
   explicit CUDAContext(const int gpu_id = -1);
@@ -332,7 +332,7 @@ inline void CPUContext::CopyBytes<CPUContext, CUDAContext>(
  * GPU present during runtime, at global initialization time we will set
  * the CPU memory allocator to allocate pinned memory.
  */
-struct CAFFE2_API PinnedCPUAllocator final : CPUAllocator {
+struct CAFFE2_CUDA_API PinnedCPUAllocator final : CPUAllocator {
   PinnedCPUAllocator() {}
   ~PinnedCPUAllocator() override {}
   std::pair<void*, MemoryDeleter> New(size_t nbytes) override {
@@ -381,7 +381,7 @@ struct CAFFE2_API PinnedCPUAllocator final : CPUAllocator {
   DefaultCPUAllocator baseAllocator_;
 };
 
-class CAFFE2_API CUDAStaticContext final : public BaseStaticContext {
+class CAFFE2_CUDA_API CUDAStaticContext final : public BaseStaticContext {
  public:
   std::pair<void*, MemoryDeleter> New(size_t nbytes) const override;
 

caffe2/core/nomnigraph/CMakeLists.txt

@@ -3,29 +3,15 @@ file(GLOB_RECURSE NOMNI_SRCS *.cc)
 file(GLOB_RECURSE NOMNI_TEST_SRCS *Test.cc)
 exclude(NOMNI_SRCS "${NOMNI_SRCS}" "${NOMNI_TEST_SRCS}")
 
-# TODO(orionr): The nomnigraph source should likely just be included
-# in the Caffe2 source list, since this won't live separately
-add_library(nomnigraph STATIC "${NOMNI_SRCS}")
-target_compile_options(nomnigraph PRIVATE "-DCAFFE2_BUILD_MAIN_LIB")
-add_dependencies(nomnigraph Caffe2_PROTO)
+list(APPEND Caffe2_CPU_SRCS ${NOMNI_SRCS})
+list(APPEND Caffe2_CPU_INCLUDE ${CMAKE_CURRENT_SOURCE_DIR}/include)
+list(APPEND Caffe2_CPU_TEST_SRCS ${NOMNI_TEST_SRCS})
 
-target_include_directories(nomnigraph PUBLIC
-  $<BUILD_INTERFACE:${CMAKE_CURRENT_LIST_DIR}/include>
-  $<INSTALL_INTERFACE:include>)
-list(APPEND Caffe2_PUBLIC_DEPENDENCY_LIBS nomnigraph)
-set(Caffe2_PUBLIC_DEPENDENCY_LIBS ${Caffe2_PUBLIC_DEPENDENCY_LIBS} PARENT_SCOPE)
-
-install(TARGETS nomnigraph EXPORT Caffe2Targets DESTINATION lib)
-install(DIRECTORY ${CMAKE_CURRENT_LIST_DIR}/include
+install(DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}/include
         DESTINATION include
         FILES_MATCHING PATTERN "*.h")
 
-if (BUILD_TEST)
-  foreach(test_src ${NOMNI_TEST_SRCS})
-    get_filename_component(test_name ${test_src} NAME_WE)
-    add_executable(${test_name} "${test_src}")
-    target_link_libraries(${test_name} nomnigraph gtest_main)
-    add_test(NAME ${test_name} COMMAND $<TARGET_FILE:${test_name}>)
-    install(TARGETS ${test_name} DESTINATION test)
-  endforeach()
-endif()
+# ---[ Send the lists to the parent scope.
+set(Caffe2_CPU_SRCS ${Caffe2_CPU_SRCS} PARENT_SCOPE)
+set(Caffe2_CPU_INCLUDE ${Caffe2_CPU_INCLUDE} PARENT_SCOPE)
+set(Caffe2_CPU_TEST_SRCS ${Caffe2_CPU_TEST_SRCS} PARENT_SCOPE)