build changes to make cpu unified build working. (pytorch#10504)

Summary:
Properly annotated all apis for cpu front. Checked with cmake using

cmake -DUSE_ATEN=ON -DUSE_CUDA=OFF -DBUILD_ATEN=ON

and resulting libcaffe2.so has about 11k symbols.
Pull Request resolved: pytorch#10504

Reviewed By: ezyang

Differential Revision: D9316491

Pulled By: Yangqing

fbshipit-source-id: 215659abf350af7032e9a4b0f28a856babab2454

aten/src/ATen/ATenGeneral.h

@@ -1,11 +1,6 @@
 #pragma once
 
-#ifdef _WIN32
-# if defined(ATen_cpu_EXPORTS) || defined(caffe2_EXPORTS)
-#  define AT_API __declspec(dllexport)
-# else
-#  define AT_API __declspec(dllimport)
-# endif
-#else
-# define AT_API
-#endif
+#include "ATen/core/Macros.h"
+
+// TODO: Merge the *_API macros.
+#define AT_API AT_CORE_API

aten/src/ATen/core/Macros.h

@@ -9,17 +9,20 @@
 
 #ifdef _WIN32
 #if !defined(AT_CORE_STATIC_WINDOWS)
-#if defined(ATen_cpu_EXPORTS) || defined(caffe2_EXPORTS)
+// TODO: unfiy the controlling macros.
+#if defined(CAFFE2_BUILD_MAIN_LIBS) || defined(ATen_cpu_EXPORTS) || defined(caffe2_EXPORTS)
 #define AT_CORE_API __declspec(dllexport)
-#else
+#else // defined(CAFFE2_BUILD_MAIN_LIBS) || defined(ATen_cpu_EXPORTS) || defined(caffe2_EXPORTS)
 #define AT_CORE_API __declspec(dllimport)
-#endif
-#else
+#endif // defined(CAFFE2_BUILD_MAIN_LIBS) || defined(ATen_cpu_EXPORTS) || defined(caffe2_EXPORTS)
+#else // !defined(AT_CORE_STATIC_WINDOWS)
 #define AT_CORE_API
-#endif
-#else
-#define AT_CORE_API
-#endif
+#endif // !defined(AT_CORE_STATIC_WINDOWS)
+#else  // _WIN32
+#if defined(__GNUC__)
+#define AT_CORE_API __attribute__((__visibility__("default")))
+#endif // defined(__GNUC__)
+#endif  // _WIN32
 
 // Disable the copy and assignment operator for a class. Note that this will
 // disable the usage of the class in std containers.

aten/src/ATen/core/typeid.h

@@ -45,7 +45,7 @@ class TypeMeta;
  * use TypeIdentifier with custom types. This is for example used to store the
  * dtype of tensors.
  */
-class TypeIdentifier final : public at::IdWrapper<TypeIdentifier, uint16_t> {
+class AT_CORE_API TypeIdentifier final : public at::IdWrapper<TypeIdentifier, uint16_t> {
  public:
   static TypeIdentifier createTypeId();
 
@@ -82,14 +82,14 @@ inline std::ostream& operator<<(
 
 namespace caffe2 {
 
-std::unordered_map<TypeIdentifier, std::string>& gTypeNames();
-std::unordered_set<std::string>& gRegisteredTypeNames();
+AT_CORE_API std::unordered_map<TypeIdentifier, std::string>& gTypeNames();
+AT_CORE_API std::unordered_set<std::string>& gRegisteredTypeNames();
 
 // A utility function to return an exception std::string by prepending its
-// exception type before its what() content.
-std::string GetExceptionString(const std::exception& e);
+// exception type before its what() content
+AT_CORE_API std::string GetExceptionString(const std::exception& e);
 
-std::mutex& gTypeRegistrationMutex();
+AT_CORE_API std::mutex& gTypeRegistrationMutex();
 
 template <typename T>
 struct TypeNameRegisterer {
@@ -136,7 +136,7 @@ struct TypeNameRegisterer {
  * stores some additional data such as the item size and the name of the type
  * for run-time inspection.
  */
-class TypeMeta {
+class AT_CORE_API TypeMeta {
  public:
   using PlacementNew = void(void*, size_t);
   using TypedCopy = void(const void*, void*, size_t);
@@ -399,7 +399,10 @@ inline bool operator!=(const TypeMeta& lhs, const TypeMeta& rhs) noexcept {
 //   http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2017/p0537r0.html
 //   https://gcc.gnu.org/bugzilla/show_bug.cgi?id=51930
 // and as a result, we define these two macros slightly differently.
-
+// TODO(jiayq): AT_CORE_API below is not correct, because we may use the
+// definition in third party dependent libraries. The proper way is to use
+// CAFFE2_EXPORT (which explicitly requires dllexport). Marking this as a
+// todo item when the unified build is finished.
 #ifdef _MSC_VER
 #define CAFFE_KNOWN_TYPE(T)                                               \
   template <>                                                             \
@@ -425,10 +428,10 @@ inline bool operator!=(const TypeMeta& lhs, const TypeMeta& rhs) noexcept {
  * for your own types to allocate dynamic ids for them.
  */
 #ifdef _MSC_VER
-#define CAFFE_DECLARE_KNOWN_TYPE(PreallocatedId, T)     \
-  template <>                                           \
-  inline AT_CORE_API TypeIdentifier TypeMeta::Id<T>() { \
-    return TypeIdentifier(PreallocatedId);              \
+#define CAFFE_DECLARE_KNOWN_TYPE(PreallocatedId, T)       \
+  template <>                                             \
+  inline AT_CORE_API TypeIdentifier TypeMeta::Id<T>() {   \
+    return TypeIdentifier(PreallocatedId);                \
   }
 #else // _MSC_VER
 #define CAFFE_DECLARE_KNOWN_TYPE(PreallocatedId, T) \

aten/src/TH/THGeneral.h.in

@@ -32,18 +32,27 @@
 # define TH_EXTERNC extern
 #endif
 
+// Note(jiayq): copied from ATen/core/Macros.h. Because internal build of TH
+// and ATen are not unified yet, we need to duplicate code for now. Long term
+// we should merge macros.
 #ifdef _WIN32
-# if defined(ATen_cpu_EXPORTS) || defined(caffe2_EXPORTS)
-#  define TH_API TH_EXTERNC __declspec(dllexport)
-#  define TH_CPP_API __declspec(dllexport)
-# else
-#  define TH_API TH_EXTERNC __declspec(dllimport)
-#  define TH_CPP_API __declspec(dllimport)
-# endif
-#else
-# define TH_API TH_EXTERNC
-# define TH_CPP_API
-#endif
+#if !defined(AT_CORE_STATIC_WINDOWS)
+// TODO: unfiy the controlling macros.
+#if defined(CAFFE2_BUILD_MAIN_LIBS) || defined(ATen_cpu_EXPORTS) || defined(caffe2_EXPORTS)
+#define TH_CPP_API __declspec(dllexport)
+#else // defined(CAFFE2_BUILD_MAIN_LIBS) || defined(ATen_cpu_EXPORTS) || defined(caffe2_EXPORTS)
+#define TH_CPP_API __declspec(dllimport)
+#endif // defined(CAFFE2_BUILD_MAIN_LIBS) || defined(ATen_cpu_EXPORTS) || defined(caffe2_EXPORTS)
+#else // !defined(AT_CORE_STATIC_WINDOWS)
+#define TH_CPP_API
+#endif // !defined(AT_CORE_STATIC_WINDOWS)
+#else  // _WIN32
+#if defined(__GNUC__)
+#define TH_CPP_API __attribute__((__visibility__("default")))
+#endif // defined(__GNUC__)
+#endif  // _WIN32
+
+#define TH_API TH_EXTERNC TH_CPP_API
 
 #ifdef _WIN32
 # define TH_NO_RETURN __declspec(noreturn)

caffe2/CMakeLists.txt

@@ -229,6 +229,13 @@ else()
 target_compile_options(caffe2 INTERFACE "$<$<COMPILE_LANGUAGE:CXX>:-std=c++11>")
 endif()
 
+# Note(jiayq): This is not complete yet, but in the end we will need to deal with
+# explicit hidden visibility.
+# This line is here so that when testing build, we can enable it to properly test
+# annotation of public symbols. When finally doing proper build with all symbols
+# annotated, we will enable this line and have it wrapped with gcc/clang checks.
+# target_compile_options(caffe2 PRIVATE "-fvisibility=hidden")
+
 target_compile_options(caffe2 PRIVATE "-DCAFFE2_BUILD_MAIN_LIB")
 if (MSVC AND NOT BUILD_SHARED_LIBS)
   # Note [Supporting both static and dynamic libraries on Window]

caffe2/contrib/gloo/common.h

@@ -11,7 +11,7 @@
 namespace caffe2 {
 namespace gloo {
 
-void signalFailure(Blob* status_blob, std::exception& exception);
+CAFFE2_API void signalFailure(Blob* status_blob, std::exception& exception);
 
 struct createDeviceAttr {
     // "tcp" or "ibverbs"
@@ -22,7 +22,7 @@ struct createDeviceAttr {
     std::string interface;
 };
 
-std::shared_ptr<::gloo::transport::Device> createDevice(
+CAFFE2_API std::shared_ptr<::gloo::transport::Device> createDevice(
     const createDeviceAttr attr);
 
 // Captures the parameters passed to Gloo.

caffe2/contrib/gloo/store_handler.h

@@ -1,13 +1,14 @@
 #pragma once
 
+#include "caffe2/core/common.h"
 #include "caffe2/distributed/store_handler.h"
 
 #include <gloo/rendezvous/store.h>
 
 namespace caffe2 {
 namespace gloo {
 
-class StoreHandlerWrapper : public ::gloo::rendezvous::Store {
+class CAFFE2_API StoreHandlerWrapper : public ::gloo::rendezvous::Store {
  public:
   explicit StoreHandlerWrapper(StoreHandler& handler) : handler_(handler) {}
 

caffe2/core/allocator.h

@@ -17,10 +17,10 @@ constexpr size_t gCaffe2Alignment = 32;
 using MemoryDeleter = void (*)(void*);
 
 // A helper function that is basically doing nothing.
-void NoDelete(void*);
+CAFFE2_API void NoDelete(void*);
 
 // A virtual allocator class to do memory allocation and deallocation.
-struct CPUAllocator {
+struct CAFFE2_API CPUAllocator {
   CPUAllocator() {}
   virtual ~CPUAllocator() noexcept {}
   virtual std::pair<void*, MemoryDeleter> New(size_t nbytes) = 0;
@@ -29,7 +29,7 @@ struct CPUAllocator {
 
 // A virtual struct that is used to report Caffe2's memory allocation and
 // deallocation status
-class MemoryAllocationReporter {
+class CAFFE2_API MemoryAllocationReporter {
  public:
   MemoryAllocationReporter() : allocated_(0) {}
   void New(void* ptr, size_t nbytes);
@@ -41,7 +41,7 @@ class MemoryAllocationReporter {
   size_t allocated_;
 };
 
-struct DefaultCPUAllocator final : CPUAllocator {
+struct CAFFE2_API DefaultCPUAllocator final : CPUAllocator {
   DefaultCPUAllocator() {}
   ~DefaultCPUAllocator() override {}
   std::pair<void*, MemoryDeleter> New(size_t nbytes) override {
@@ -78,10 +78,10 @@ struct DefaultCPUAllocator final : CPUAllocator {
 };
 
 // Get the CPU Alloctor.
-CPUAllocator* GetCPUAllocator();
+CAFFE2_API CPUAllocator* GetCPUAllocator();
 // Sets the CPU allocator to the given allocator: the caller gives away the
 // ownership of the pointer.
-void SetCPUAllocator(CPUAllocator* alloc);
+CAFFE2_API void SetCPUAllocator(CPUAllocator* alloc);
 
 } // namespace caffe2
 

caffe2/core/blob.h

@@ -23,7 +23,7 @@ namespace caffe2 {
  * properly when the blob is deallocated or re-allocated with a new type. A blob
  * could contain anything, although the most common case is to contain a Tensor.
  */
-class Blob {
+class CAFFE2_API Blob {
  public:
   typedef void (*DestroyCall)(void*);
 

caffe2/core/blob_serialization.h

@@ -23,26 +23,13 @@ constexpr auto kTensorBlobType = "Tensor";
 // String used to separate chunk id from the blob name when storing in DB
 constexpr auto kChunkIdSeparator = "#%";
 
-// The Blob serialization registry and serializer creator functions.
-CAFFE_DECLARE_TYPED_REGISTRY(
-    BlobSerializerRegistry,
-    TypeIdentifier,
-    BlobSerializerBase,
-    std::unique_ptr);
-#define REGISTER_BLOB_SERIALIZER(id, ...) \
-  CAFFE_REGISTER_TYPED_CLASS(BlobSerializerRegistry, id, __VA_ARGS__)
-// Creates an operator with the given operator definition.
-inline unique_ptr<BlobSerializerBase> CreateSerializer(TypeIdentifier id) {
-  return BlobSerializerRegistry()->Create(id);
-}
-
 /**
  * @brief TensorSerializer is the serializer for Tensors.
  *
  * TensorSerializer takes in a blob that contains a Tensor, and serializes it
  * into a TensorProto protocol buffer.
  */
-class TensorSerializer : public BlobSerializerBase {
+class CAFFE2_API TensorSerializer : public BlobSerializerBase {
  public:
   TensorSerializer() {}
   ~TensorSerializer() override {}
@@ -73,25 +60,6 @@ class TensorSerializer : public BlobSerializerBase {
   unique_ptr<BaseContext> context_;
 };
 
-/**
- * @brief BlobDeserializerBase is an abstract class that deserializes a blob
- * from a BlobProto or a TensorProto.
- */
-class BlobDeserializerBase {
- public:
-  virtual ~BlobDeserializerBase() {}
-
-  // Deserializes from a BlobProto object.
-  virtual void Deserialize(const BlobProto& proto, Blob* blob) = 0;
-};
-
-CAFFE_DECLARE_REGISTRY(BlobDeserializerRegistry, BlobDeserializerBase);
-#define REGISTER_BLOB_DESERIALIZER(name, ...) \
-  CAFFE_REGISTER_CLASS(BlobDeserializerRegistry, name, __VA_ARGS__)
-// Creates an operator with the given operator definition.
-inline unique_ptr<BlobDeserializerBase> CreateDeserializer(const string& type) {
-  return BlobDeserializerRegistry()->Create(type);
-}
 
 /**
  * @brief TensorDeserializer is the deserializer for Tensors.
@@ -101,7 +69,7 @@ inline unique_ptr<BlobDeserializerBase> CreateDeserializer(const string& type) {
  * tensor, change the TensorProto's corresponding fields before calling
  * Deserialize.
  */
-class TensorDeserializer : public BlobDeserializerBase {
+class CAFFE2_API TensorDeserializer : public BlobDeserializerBase {
  public:
   void Deserialize(const BlobProto& proto, Blob* blob) override;
   void Deserialize(const TensorProto& proto, Tensor* tensor);

caffe2/core/blob_serializer_base.h

@@ -3,6 +3,10 @@
 #include <string>
 #include <functional>
 
+#include "caffe2/core/common.h"
+#include "caffe2/core/registry.h"
+#include "caffe2/proto/caffe2.pb.h"
+
 namespace caffe2 {
 
 class Blob;
@@ -52,4 +56,39 @@ class BlobSerializerBase {
   }
 };
 
+// The Blob serialization registry and serializer creator functions.
+CAFFE_DECLARE_TYPED_REGISTRY(
+    BlobSerializerRegistry,
+    TypeIdentifier,
+    BlobSerializerBase,
+    std::unique_ptr);
+#define REGISTER_BLOB_SERIALIZER(id, ...) \
+  CAFFE_REGISTER_TYPED_CLASS(BlobSerializerRegistry, id, __VA_ARGS__)
+// Creates an operator with the given operator definition.
+inline unique_ptr<BlobSerializerBase> CreateSerializer(TypeIdentifier id) {
+  return BlobSerializerRegistry()->Create(id);
+}
+
+
+/**
+ * @brief BlobDeserializerBase is an abstract class that deserializes a blob
+ * from a BlobProto or a TensorProto.
+ */
+class CAFFE2_API BlobDeserializerBase {
+ public:
+  virtual ~BlobDeserializerBase() {}
+
+  // Deserializes from a BlobProto object.
+  virtual void Deserialize(const BlobProto& proto, Blob* blob) = 0;
+};
+
+CAFFE_DECLARE_REGISTRY(BlobDeserializerRegistry, BlobDeserializerBase);
+#define REGISTER_BLOB_DESERIALIZER(name, ...) \
+  CAFFE_REGISTER_CLASS(BlobDeserializerRegistry, name, __VA_ARGS__)
+// Creates an operator with the given operator definition.
+inline unique_ptr<BlobDeserializerBase> CreateDeserializer(const string& type) {
+  return BlobDeserializerRegistry()->Create(type);
+}
+
+
 } // namespace caffe2