Merge branch 'develop' into disable_ut_test_parallel_executor_fetch_isolated_var

cmake/generic.cmake

@@ -446,6 +446,9 @@ function(nv_library TARGET_NAME)
         message(FATAL "Please specify source file or library in nv_library.")
       endif()
     endif(nv_library_SRCS)
+    if (WIN32)
+      set_target_properties(${TARGET_NAME} PROPERTIES VS_USER_PROPS ${WIN_PROPS})
+    endif(WIN32)
   endif()
 endfunction(nv_library)
 
@@ -461,6 +464,9 @@ function(nv_binary TARGET_NAME)
       add_dependencies(${TARGET_NAME} ${nv_binary_DEPS})
       common_link(${TARGET_NAME})
     endif()
+    if (WIN32)
+      set_target_properties(${TARGET_NAME} PROPERTIES VS_USER_PROPS ${WIN_PROPS})
+    endif(WIN32)
   endif()
 endfunction(nv_binary)
 
@@ -482,6 +488,9 @@ function(nv_test TARGET_NAME)
     set_property(TEST ${TARGET_NAME} PROPERTY ENVIRONMENT FLAGS_cpu_deterministic=true)
     set_property(TEST ${TARGET_NAME} PROPERTY ENVIRONMENT FLAGS_init_allocated_mem=true)
     set_property(TEST ${TARGET_NAME} PROPERTY ENVIRONMENT FLAGS_cudnn_deterministic=true)
+    if (WIN32)
+      set_target_properties(${TARGET_NAME} PROPERTIES VS_USER_PROPS ${WIN_PROPS})
+    endif(WIN32)
   endif()
 endfunction(nv_test)
 
@@ -712,6 +721,7 @@ function(proto_library TARGET_NAME)
   set(proto_hdrs)
   paddle_protobuf_generate_cpp(proto_srcs proto_hdrs ${proto_library_SRCS})
   cc_library(${TARGET_NAME} SRCS ${proto_srcs} DEPS ${proto_library_DEPS} protobuf)
+  add_dependencies(extern_xxhash ${TARGET_NAME})
 endfunction()
 
 function(py_proto_compile TARGET_NAME)

cmake/inference_lib.cmake

@@ -19,9 +19,8 @@ set(PADDLE_INSTALL_DIR "${CMAKE_BINARY_DIR}/paddle_install_dir" CACHE STRING
 set(PADDLE_INFERENCE_INSTALL_DIR "${CMAKE_BINARY_DIR}/paddle_inference_install_dir" CACHE STRING
   "A path setting paddle inference shared and static libraries")
   
-# TODO(zhaolong)
-# At present, the size of static lib in Windows exceeds the system limit,
-# so the generation of static lib is temporarily turned off.
+# At present, the size of static lib in Windows is very large,
+# so we need to crop the library size.
 if(WIN32)
     #todo: remove the option 
     option(WITH_STATIC_LIB "Compile demo with static/shared library, default use dynamic."   OFF)
@@ -196,7 +195,11 @@ set(PADDLE_INFERENCE_C_INSTALL_DIR "${CMAKE_BINARY_DIR}/paddle_inference_c_insta
 copy_part_of_thrid_party(inference_lib_dist ${PADDLE_INFERENCE_C_INSTALL_DIR})
 
 set(src_dir "${PADDLE_SOURCE_DIR}/paddle/fluid")
-set(paddle_fluid_c_lib ${PADDLE_BINARY_DIR}/paddle/fluid/inference/capi/libpaddle_fluid_c.*)
+if(WIN32)
+  set(paddle_fluid_c_lib ${PADDLE_BINARY_DIR}/paddle/fluid/inference/capi/${CMAKE_BUILD_TYPE}/paddle_fluid_c.*)
+else(WIN32)
+  set(paddle_fluid_c_lib ${PADDLE_BINARY_DIR}/paddle/fluid/inference/capi/libpaddle_fluid_c.*)
+endif(WIN32)
 
 copy(inference_lib_dist
       SRCS  ${src_dir}/inference/capi/paddle_c_api.h  ${paddle_fluid_c_lib}

cmake/init.cmake

@@ -26,4 +26,7 @@ if(WITH_GPU)
     set(CMAKE_CUDA_FLAGS_MINSIZEREL "-O1 -DNDEBUG")
 endif()
 
+if(WIN32)
+    set(WIN_PROPS ${CMAKE_SOURCE_DIR}/cmake/paddle_win.props)
+endif()
 

cmake/paddle_win.props

+<?xml version="1.0" encoding="utf-8"?>
+<Project xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
+    <ItemDefinitionGroup>
+        <CudaCompile>
+            <!-- Project schema: Host properties -->
+            <UseHostDefines>true</UseHostDefines>
+            <Emulation>false</Emulation>
+            <HostDebugInfo Condition="'$(Configuration)' == 'Debug'">true</HostDebugInfo>
+            <HostDebugInfo Condition="'$(Configuration)' != 'Debug'">false</HostDebugInfo>
+            <FastMath>false</FastMath>
+            <Optimization>InheritFromHost</Optimization>
+            <Runtime>InheritFromHost</Runtime>
+            <RuntimeChecks>InheritFromHost</RuntimeChecks>
+            <TypeInfo>InheritFromHost</TypeInfo>
+            <Warning>InheritFromHost</Warning>
+
+            <BaseCommandLineTemplate>-ccbin "%(VCBinDir)" -x cu [GenerateRelocatableDeviceCode] [Include] [RequiredIncludes] [InterleaveSourceInPTX] [GPUDebugInfo] [GenerateLineInfo] [Keep] [KeepDir] [MaxRegCount] [PtxAsOptionV] [TargetMachinePlatform] [NvccCompilation] [CudaRuntime] [AdditionalOptions]</BaseCommandLineTemplate>
+            <BuildCommandLineTemplate>--use-local-env</BuildCommandLineTemplate>
+            <BuildDynamicCommandLineTemplate>[CodeGeneration]</BuildDynamicCommandLineTemplate>
+            <CleanCommandLineTemplate>-clean</CleanCommandLineTemplate>
+            <!-- <HostCommandLineTemplate>-Xcompiler &quot;/EHsc [Warning] /nologo [Optimization] [ProgramDataBaseFileName] $(CudaForceSynchronousPdbWrites) /Zi [RuntimeChecks] [Runtime] [TypeInfo]&quot;</HostCommandLineTemplate> -->
+            <HostCommandLineTemplate>-Xcompiler &quot;/EHsc [Warning] /nologo [Optimization] [ProgramDataBaseFileName] $(CudaForceSynchronousPdbWrites) [RuntimeChecks] [Runtime] [TypeInfo]&quot;</HostCommandLineTemplate>
+
+            <DriverApiCommandLineTemplate>%(BaseCommandLineTemplate) [CompileOut] "%(FullPath)"</DriverApiCommandLineTemplate>
+            <RuntimeApiCommandLineTemplate>%(BaseCommandLineTemplate) [HostDebugInfo] [Emulation] [FastMath] [Defines] %(HostCommandLineTemplate) [CompileOut] "%(FullPath)"</RuntimeApiCommandLineTemplate>
+
+            <CommandLineTemplate>
+# (Approximate command-line.  Settings inherited from host are not visible below.)
+# (Please see the output window after a build for the full command-line)
+
+# Driver API (NVCC Compilation Type is .cubin, .gpu, or .ptx)
+set CUDAFE_FLAGS=--sdk_dir "$(WindowsSdkDir)"
+"$(CudaToolkitNvccPath)" %(BuildCommandLineTemplate) %(DriverApiCommandLineTemplate)
+
+# Runtime API (NVCC Compilation Type is hybrid object or .c file)
+set CUDAFE_FLAGS=--sdk_dir "$(WindowsSdkDir)"
+"$(CudaToolkitNvccPath)" %(BuildCommandLineTemplate) %(RuntimeApiCommandLineTemplate)
+            </CommandLineTemplate>
+            <ExecutionDescription>Compiling CUDA source file %(Identity)...</ExecutionDescription>
+            <ExclusionDescription>Skipping CUDA source file %(Identity) (excluded from build).</ExclusionDescription>
+
+            <!-- Miscellaneous -->
+            <PropsCacheOutputFile>%(Filename)%(Extension).cache</PropsCacheOutputFile>
+            <PropsCacheOutputPath>$(IntDir)%(PropsCacheOutputFile)</PropsCacheOutputPath>
+
+            <CudaCompileCoreProject>$(MSBuildProjectFullPath)</CudaCompileCoreProject>
+        </CudaCompile>
+
+        <CudaLink>
+            <PerformDeviceLink>true</PerformDeviceLink>
+            <LinkOut>$(IntDir)$(TargetName).device-link.obj</LinkOut>
+
+            <AdditionalLibraryDirectories></AdditionalLibraryDirectories>
+            <UseHostLibraryDirectories>true</UseHostLibraryDirectories>
+            <AdditionalDependencies></AdditionalDependencies>
+            <UseHostLibraryDependencies>true</UseHostLibraryDependencies>
+
+            <GPUDebugInfo>InheritFromProject</GPUDebugInfo>
+            <Optimization>InheritFromProject</Optimization>
+
+            <!-- Implicitly inherited from the project via @(CudaCompile) -->
+            <CodeGeneration></CodeGeneration>
+            <RuntimeChecks></RuntimeChecks>
+            <Runtime></Runtime>
+            <TargetMachinePlatform></TargetMachinePlatform>
+            <TypeInfo></TypeInfo>
+            <Warning></Warning>
+
+            <Inputs></Inputs>
+
+            <!-- <HostCommandLineTemplate>-Xcompiler &quot;/EHsc [Warning] /nologo [Optimization] /Zi [RuntimeChecks] [Runtime] [TypeInfo]&quot;</HostCommandLineTemplate> -->
+            <HostCommandLineTemplate>-Xcompiler &quot;/EHsc [Warning] /nologo [Optimization] [RuntimeChecks] [Runtime] [TypeInfo]&quot;</HostCommandLineTemplate>
+            <LinkCommandLineTemplate>"$(CudaToolkitNvccPath)" -dlink [LinkOut] %(HostCommandLineTemplate) [AdditionalLibraryDirectories] [AdditionalDependencies] [AdditionalOptions] [CodeGeneration] [GPUDebugInfo] [TargetMachinePlatform] [Inputs]</LinkCommandLineTemplate>
+            <CommandLineTemplate>
+# (Approximate command-line.  Settings inherited from host are not visible below.)
+# (Please see the output window after a build for the full command-line)
+
+%(LinkCommandLineTemplate)
+            </CommandLineTemplate>
+        </CudaLink>
+
+        <Link>
+            <AdditionalLibraryDirectories>%(AdditionalLibraryDirectories);$(CudaToolkitLibDir)</AdditionalLibraryDirectories>
+        </Link>
+
+        <ClCompile>
+            <AdditionalIncludeDirectories>%(AdditionalIncludeDirectories);$(CudaToolkitIncludeDir)</AdditionalIncludeDirectories>
+        </ClCompile>
+    </ItemDefinitionGroup>
+</Project>
+

paddle/fluid/framework/attribute.cc

@@ -14,8 +14,6 @@ limitations under the License. */
 
 #include "paddle/fluid/framework/attribute.h"
 
-#include <vector>
-
 namespace paddle {
 namespace framework {
 

paddle/fluid/framework/block_desc.h

@@ -30,6 +30,8 @@ namespace paddle {
 namespace framework {
 
 class ProgramDesc;
+class OpDesc;
+class VarDesc;
 
 // Each Protobuf Message, we provide a XXXBind class. In that class, we optimize
 // read/write speed. Only when we want the protobuf message, the local changes

paddle/fluid/framework/c/c_api.cc

@@ -12,17 +12,7 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include <string>
-#include <unordered_map>
-#include <unordered_set>
-#include <vector>
-
-#include "paddle/fluid/framework/block_desc.h"
 #include "paddle/fluid/framework/c/c_api.h"
-#include "paddle/fluid/framework/op_registry.h"
-#include "paddle/fluid/framework/program_desc.h"
-#include "paddle/fluid/framework/scope.h"
-#include "paddle/fluid/platform/init.h"
 
 extern "C" {
 

paddle/fluid/framework/c/c_api.h

@@ -24,6 +24,15 @@ limitations under the License. */
 #include "paddle/fluid/framework/op_info.h"
 #include "paddle/fluid/platform/device_context.h"
 
+namespace paddle {
+namespace framework {
+class OpInfoMap;
+}  // namespace framework
+namespace platform {
+class DeviceContextPool;
+}  // namespace platform
+}  // namespace paddle
+
 #ifdef __cplusplus
 extern "C" {
 #endif

paddle/fluid/framework/channel.h

@@ -277,7 +277,7 @@ class ChannelObject {
     size_t finished = 0;
     while (finished < n && WaitForWrite(lock)) {
       size_t m =
-          std::min(n - finished, capacity_ + reading_count_ - data_.size());
+          (std::min)(n - finished, capacity_ + reading_count_ - data_.size());
       for (size_t i = 0; i < m; i++) {
         data_.push_back(std::move(p[finished++]));
       }

paddle/fluid/framework/copy_same_tensor_test.cc

@@ -12,8 +12,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include <cstring>
 #include <random>
+
 #include "gflags/gflags.h"
 #include "gtest/gtest.h"
 #include "paddle/fluid/framework/tensor.h"

paddle/fluid/framework/data_device_transform.h

@@ -21,6 +21,8 @@ limitations under the License. */
 namespace paddle {
 namespace framework {
 
+class Tensor;
+
 void TransDataDevice(const Tensor& in, const platform::Place& dst_place,
                      Tensor* out);
 

paddle/fluid/framework/data_feed.h

@@ -41,6 +41,15 @@ limitations under the License. */
 #include "paddle/fluid/framework/variable.h"
 #include "paddle/fluid/string/string_helper.h"
 
+namespace paddle {
+namespace framework {
+class DataFeedDesc;
+class LoDTensor;
+class Scope;
+class Variable;
+}  // namespace framework
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 
@@ -418,6 +427,7 @@ class MultiSlotType {
 
   std::string DebugString() {
     std::stringstream ss;
+
     ss << "\ntype: " << type_ << "\n";
     ss << "offset: ";
     ss << "[";

paddle/fluid/framework/data_feed_factory.cc

@@ -17,10 +17,10 @@ limitations under the License. */
 #include <string>
 #include <unordered_map>
 
-#include "paddle/fluid/framework/data_feed.h"
-
 namespace paddle {
 namespace framework {
+class DataFeed;
+
 typedef std::shared_ptr<DataFeed> (*Createdata_feedFunction)();
 typedef std::unordered_map<std::string, Createdata_feedFunction> data_feedMap;
 data_feedMap g_data_feed_map;

paddle/fluid/framework/data_feed_factory.h

@@ -16,10 +16,13 @@ limitations under the License. */
 
 #include <memory>
 #include <string>
+
 #include "paddle/fluid/framework/data_feed.h"
 
 namespace paddle {
 namespace framework {
+class DataFeed;
+
 class DataFeedFactory {
  public:
   static std::string DataFeedTypeList();

paddle/fluid/framework/data_layout_transform.cc

@@ -13,8 +13,8 @@
 // limitations under the License.
 
 #include "paddle/fluid/framework/data_layout_transform.h"
+
 #include <string>
-#include <vector>
 
 #include "paddle/fluid/operators/math/math_function.h"
 #ifdef PADDLE_WITH_MKLDNN

paddle/fluid/framework/data_layout_transform.h

@@ -17,10 +17,18 @@
 #include <map>
 #include <unordered_map>
 #include <vector>
+
 #include "paddle/fluid/framework/op_kernel_type.h"
 #include "paddle/fluid/framework/tensor.h"
 #include "paddle/fluid/framework/variable.h"
 
+namespace paddle {
+namespace framework {
+class OpKernelType;
+class Tensor;
+}  // namespace framework
+}  // namespace paddle
+
 #ifdef PADDLE_WITH_MKLDNN
 #include "paddle/fluid/platform/mkldnn_helper.h"
 #endif

paddle/fluid/framework/data_layout_transform_test.cc

@@ -15,7 +15,6 @@
 #include "paddle/fluid/framework/data_layout_transform.h"
 
 #include "gtest/gtest.h"
-#include "paddle/fluid/platform/device_context.h"
 
 TEST(DataTransform, DataLayoutFunction) {
   auto place = paddle::platform::CPUPlace();

paddle/fluid/framework/data_transform.cc

@@ -18,8 +18,13 @@ limitations under the License. */
 #include "paddle/fluid/framework/data_layout_transform.h"
 #include "paddle/fluid/framework/data_type_transform.h"
 
+namespace paddle {
+namespace framework {
+class Variable;
+}  // namespace framework
+}  // namespace paddle
+
 #ifdef PADDLE_WITH_MKLDNN
-#include <algorithm>
 #include "paddle/fluid/platform/mkldnn_helper.h"
 #endif
 

paddle/fluid/framework/data_transform.h

@@ -30,6 +30,10 @@ limitations under the License. */
 namespace paddle {
 namespace framework {
 
+class OpKernelType;
+class Tensor;
+class Variable;
+
 void TransformData(const OpKernelType &expected_kernel_type,
                    const OpKernelType &kernel_type_for_var,
                    const Tensor &input_tensor, Tensor *out);

paddle/fluid/framework/data_type.cc

@@ -13,7 +13,6 @@
 // limitations under the License.
 
 #include "paddle/fluid/framework/data_type.h"
-#include <stdint.h>
 #include <string>
 #include <unordered_map>
 

paddle/fluid/framework/data_type.h

@@ -15,12 +15,19 @@ limitations under the License. */
 #pragma once
 #include <string>
 #include <typeindex>
-#include "paddle/fluid/framework/framework.pb.h"
-#include "paddle/fluid/platform/enforce.h"
 
+#include "paddle/fluid/framework/framework.pb.h"
 #include "paddle/fluid/platform/bfloat16.h"
+#include "paddle/fluid/platform/enforce.h"
 #include "paddle/fluid/platform/float16.h"
 
+namespace paddle {
+namespace platform {
+struct bfloat16;
+struct float16;
+}  // namespace platform
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 

paddle/fluid/framework/data_type_test.cc

@@ -14,9 +14,17 @@
 #include "paddle/fluid/framework/data_type.h"
 
 #include <string>
+
 #include "gtest/gtest.h"
 #include "paddle/fluid/framework/tensor.h"
 
+namespace paddle {
+namespace platform {
+struct bfloat16;
+struct float16;
+}  // namespace platform
+}  // namespace paddle
+
 TEST(DataType, float16) {
   using paddle::framework::Tensor;
   using paddle::platform::CPUPlace;

paddle/fluid/framework/data_type_transform.h

@@ -15,6 +15,7 @@ limitations under the License. */
 #pragma once
 
 #include <utility>
+
 #include "paddle/fluid/framework/op_kernel_type.h"
 #include "paddle/fluid/framework/tensor.h"
 #include "paddle/fluid/framework/variable.h"
@@ -23,6 +24,9 @@ limitations under the License. */
 namespace paddle {
 namespace framework {
 
+class OpKernelType;
+class Tensor;
+
 using KernelTypePair = std::pair<OpKernelType, OpKernelType>;
 
 void TransDataType(const OpKernelType& kernel_type_for_var,

paddle/fluid/framework/dataset_factory.cc

@@ -13,12 +13,9 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/framework/dataset_factory.h"
-#include <memory>
 #include <string>
 #include <unordered_map>
 
-#include "paddle/fluid/framework/data_set.h"
-
 namespace paddle {
 namespace framework {
 typedef std::unique_ptr<Dataset> (*CreateDatasetFunction)();

paddle/fluid/framework/dataset_factory.h

@@ -16,6 +16,7 @@ limitations under the License. */
 
 #include <memory>
 #include <string>
+
 #include "paddle/fluid/framework/data_set.h"
 
 namespace paddle {

paddle/fluid/framework/ddim.h

@@ -18,6 +18,7 @@ limitations under the License. */
 #include <stdexcept>
 #include <string>
 #include <vector>
+
 #include "paddle/fluid/framework/dim.h"
 
 namespace paddle {

paddle/fluid/framework/ddim_test.cc

@@ -12,7 +12,6 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 #include <sstream>
-#include <vector>
 
 #include "gtest/gtest.h"
 #include "paddle/fluid/framework/ddim.h"

paddle/fluid/framework/details/all_reduce_op_handle.h

@@ -20,6 +20,17 @@
 #include "paddle/fluid/framework/details/op_handle_base.h"
 #include "paddle/fluid/framework/lod_tensor.h"
 #include "paddle/fluid/framework/scope.h"
+
+namespace paddle {
+namespace framework {
+namespace ir {
+class Node;
+}  // namespace ir
+}  // namespace framework
+namespace platform {
+class NCCLCommunicator;
+}  // namespace platform
+}  // namespace paddle
 #if defined(PADDLE_WITH_NCCL)
 #include "paddle/fluid/framework/details/nccl_op_handle.h"
 #include "paddle/fluid/platform/nccl_helper.h"

paddle/fluid/framework/details/broadcast_op_handle.h

@@ -24,6 +24,20 @@
 #include "paddle/fluid/framework/selected_rows.h"
 #include "paddle/fluid/platform/device_context.h"
 
+namespace paddle {
+namespace framework {
+namespace details {
+struct VarHandle;
+}  // namespace details
+namespace ir {
+class Node;
+}  // namespace ir
+}  // namespace framework
+namespace platform {
+struct NCCLContextMap;
+}  // namespace platform
+}  // namespace paddle
+
 #if defined(PADDLE_WITH_NCCL)
 #include "paddle/fluid/platform/nccl_helper.h"
 #endif

paddle/fluid/framework/details/broadcast_op_handle_test.h

@@ -21,13 +21,15 @@
 
 #include "gtest/gtest.h"
 #include "paddle/fluid/framework/details/broadcast_op_handle.h"
-
 #include "paddle/fluid/platform/device_context.h"
 
 namespace paddle {
 namespace framework {
 namespace details {
 
+struct DummyVarHandle;
+struct VarHandle;
+
 namespace f = paddle::framework;
 namespace p = paddle::platform;
 

paddle/fluid/framework/details/build_strategy.h

@@ -27,6 +27,18 @@
 #include "paddle/fluid/platform/device_context.h"
 #include "paddle/fluid/platform/enforce.h"
 
+namespace paddle {
+namespace framework {
+namespace ir {
+class Graph;
+class PassBuilder;
+}  // namespace ir
+}  // namespace framework
+namespace platform {
+class NCCLCommunicator;
+}  // namespace platform
+}  // namespace paddle
+
 #if defined(PADDLE_WITH_NCCL)
 #include "paddle/fluid/platform/nccl_helper.h"
 #endif

paddle/fluid/framework/details/computation_op_handle.cc

@@ -19,6 +19,8 @@
 namespace paddle {
 namespace framework {
 namespace details {
+struct VarHandleBase;
+
 ComputationOpHandle::ComputationOpHandle(ir::Node *node, Scope *scope,
                                          platform::Place place,
                                          size_t scope_idx)

paddle/fluid/framework/details/computation_op_handle.h

@@ -24,9 +24,21 @@
 #include "paddle/fluid/framework/scope.h"
 #include "paddle/fluid/platform/device_context.h"
 
+namespace paddle {
+namespace framework {
+class OperatorBase;
+class Scope;
+namespace ir {
+class Node;
+}  // namespace ir
+}  // namespace framework
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 namespace details {
+struct VarHandleBase;
+
 class ComputationOpHandle : public OpHandleBase {
  public:
   ComputationOpHandle(ir::Node *node, Scope *scope, platform::Place place,

paddle/fluid/framework/details/eager_deletion_op_handle.h

@@ -19,12 +19,23 @@
 #include <string>
 #include <unordered_set>
 #include <vector>
+
 #include "paddle/fluid/framework/details/op_handle_base.h"
 #include "paddle/fluid/framework/ir/memory_optimize_pass/reference_count_pass_helper.h"
 
+namespace paddle {
+namespace platform {
+class CUDADeviceContext;
+}  // namespace platform
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 class Scope;
+class GarbageCollector;
+namespace ir {
+class Node;
+}  // namespace ir
 
 namespace ir {
 class MemOptVarInfo;

paddle/fluid/framework/details/exception_holder_test.cc

@@ -13,8 +13,7 @@
 // limitations under the License.
 
 #include "paddle/fluid/framework/details/exception_holder.h"
-#include <memory>
-#include <unordered_map>
+
 #include "gtest/gtest.h"
 #include "paddle/fluid/memory/allocation/allocator.h"
 

paddle/fluid/framework/details/fetch_async_op_handle.cc

@@ -15,9 +15,14 @@
 #include "paddle/fluid/framework/details/fetch_async_op_handle.h"
 #include <string>
 #include <utility>
-#include <vector>
 #include "paddle/fluid/platform/profiler.h"
 
+namespace paddle {
+namespace platform {
+class DeviceContext;
+}  // namespace platform
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 namespace details {

paddle/fluid/framework/details/fetch_async_op_handle.h

@@ -22,6 +22,18 @@
 #include "paddle/fluid/framework/scope.h"
 #include "paddle/fluid/platform/device_context.h"
 
+namespace paddle {
+namespace framework {
+class LoDTensor;
+namespace ir {
+class Node;
+}  // namespace ir
+}  // namespace framework
+namespace platform {
+class DeviceContext;
+}  // namespace platform
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 namespace details {

paddle/fluid/framework/details/fetch_barrier_op_handle.cc

@@ -19,6 +19,8 @@
 namespace paddle {
 namespace framework {
 namespace details {
+struct VarHandleBase;
+
 FetchBarrierOpHandle::FetchBarrierOpHandle(
     ir::Node *node, const std::vector<Scope *> &local_scopes,
     const std::vector<platform::Place> &places)

paddle/fluid/framework/details/fetch_barrier_op_handle.h

@@ -24,6 +24,15 @@
 #include "paddle/fluid/framework/scope.h"
 #include "paddle/fluid/platform/device_context.h"
 
+namespace paddle {
+namespace framework {
+class Scope;
+namespace ir {
+class Node;
+}  // namespace ir
+}  // namespace framework
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 namespace details {
@@ -32,6 +41,8 @@ namespace details {
 // all places if there are multiple places, must init with
 // multiple dev_ctxes_ !!!!
 
+struct VarHandleBase;
+
 struct FetchBarrierOpHandle : public OpHandleBase {
  public:
   FetchBarrierOpHandle(ir::Node *node, const std::vector<Scope *> &local_scopes,

paddle/fluid/framework/details/fetch_op_handle.h

@@ -22,6 +22,17 @@
 #include "paddle/fluid/framework/scope.h"
 #include "paddle/fluid/platform/device_context.h"
 
+namespace paddle {
+namespace framework {
+namespace ir {
+class Node;
+}  // namespace ir
+}  // namespace framework
+namespace platform {
+class DeviceContext;
+}  // namespace platform
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 namespace details {

paddle/fluid/framework/details/fused_all_reduce_op_handle.h

@@ -17,10 +17,22 @@
 #include <string>
 #include <utility>
 #include <vector>
+
 #include "paddle/fluid/framework/details/all_reduce_op_handle.h"
 #include "paddle/fluid/framework/details/op_handle_base.h"
 #include "paddle/fluid/framework/lod_tensor.h"
 #include "paddle/fluid/framework/scope.h"
+
+namespace paddle {
+namespace framework {
+namespace ir {
+class Node;
+}  // namespace ir
+}  // namespace framework
+namespace platform {
+class NCCLCommunicator;
+}  // namespace platform
+}  // namespace paddle
 #if defined(PADDLE_WITH_NCCL)
 #include "paddle/fluid/framework/details/nccl_op_handle.h"
 #include "paddle/fluid/platform/nccl_helper.h"

paddle/fluid/framework/details/fused_broadcast_op_handle.h

@@ -25,6 +25,17 @@
 #include "paddle/fluid/framework/selected_rows.h"
 #include "paddle/fluid/platform/device_context.h"
 
+namespace paddle {
+namespace framework {
+namespace ir {
+class Node;
+}  // namespace ir
+}  // namespace framework
+namespace platform {
+struct NCCLContextMap;
+}  // namespace platform
+}  // namespace paddle
+
 #if defined(PADDLE_WITH_NCCL)
 #include "paddle/fluid/platform/nccl_helper.h"
 #endif

paddle/fluid/framework/details/fused_broadcast_op_handle_test.cc

@@ -17,11 +17,20 @@
 #include <unordered_map>
 #include "gtest/gtest.h"
 #include "paddle/fluid/framework/details/broadcast_op_handle_test.h"
+#include "paddle/fluid/framework/details/op_handle_base.h"
+
+namespace paddle {
+namespace framework {
+class Scope;
+}  // namespace framework
+}  // namespace paddle
 
 namespace paddle {
 namespace framework {
 namespace details {
 
+struct VarHandle;
+
 struct TestFusedBroadcastOpHandle : TestBroadcastOpHandle {
   std::vector<std::string> out_varnames_;
   std::vector<std::unique_ptr<ir::Node>> nodes_;

paddle/fluid/framework/details/gather_op_handle.h

@@ -24,6 +24,14 @@
 #include "paddle/fluid/framework/selected_rows.h"
 #include "paddle/fluid/platform/device_context.h"
 
+namespace paddle {
+namespace framework {
+namespace ir {
+class Node;
+}  // namespace ir
+}  // namespace framework
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 namespace details {

paddle/fluid/framework/details/gather_op_handle_test.cc

@@ -17,11 +17,11 @@
 #include <unordered_map>
 #include "gtest/gtest.h"
 
-#include "paddle/fluid/platform/device_context.h"
-
 namespace paddle {
 namespace framework {
 namespace details {
+struct DummyVarHandle;
+
 namespace f = paddle::framework;
 namespace p = paddle::platform;
 

paddle/fluid/framework/details/multi_devices_helper.h

@@ -20,16 +20,21 @@
 #include <unordered_set>
 #include <utility>
 #include <vector>
+
 #include "paddle/fluid/framework/details/op_handle_base.h"
 #include "paddle/fluid/framework/details/var_handle.h"
-
+#include "paddle/fluid/framework/ir/graph.h"
+#include "paddle/fluid/framework/ir/pass.h"
 #include "paddle/fluid/framework/op_desc.h"
 #include "paddle/fluid/framework/op_proto_maker.h"
 #include "paddle/fluid/framework/program_desc.h"
 #include "paddle/fluid/platform/place.h"
 
-#include "paddle/fluid/framework/ir/graph.h"
-#include "paddle/fluid/framework/ir/pass.h"
+namespace paddle {
+namespace framework {
+class OpDesc;
+}  // namespace framework
+}  // namespace paddle
 
 namespace paddle {
 namespace framework {

paddle/fluid/framework/details/nan_inf_utils_detail.h

@@ -19,6 +19,12 @@
 #include "paddle/fluid/framework/scope.h"
 #include "paddle/fluid/platform/place.h"
 
+namespace paddle {
+namespace framework {
+class Tensor;
+}  // namespace framework
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 namespace details {

paddle/fluid/framework/details/op_handle_base.h

@@ -24,10 +24,22 @@
 #include "paddle/fluid/platform/device_context.h"
 #include "paddle/fluid/platform/macros.h"
 
+namespace paddle {
+namespace platform {
+class DeviceContext;
+}  // namespace platform
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 
 class Scope;
+namespace details {
+struct VarHandleBase;
+}  // namespace details
+namespace ir {
+class Node;
+}  // namespace ir
 
 namespace details {
 

paddle/fluid/framework/details/reduce_op_handle.h

@@ -24,6 +24,21 @@
 #include "paddle/fluid/framework/scope.h"
 #include "paddle/fluid/framework/selected_rows.h"
 #include "paddle/fluid/platform/device_context.h"
+
+namespace paddle {
+namespace framework {
+class SelectedRows;
+namespace details {
+struct VarHandle;
+}  // namespace details
+namespace ir {
+class Node;
+}  // namespace ir
+}  // namespace framework
+namespace platform {
+struct NCCLContextMap;
+}  // namespace platform
+}  // namespace paddle
 #if defined(PADDLE_WITH_NCCL)
 #include "paddle/fluid/platform/nccl_helper.h"
 #endif

paddle/fluid/framework/details/rpc_op_handle.h

@@ -24,6 +24,16 @@
 #include "paddle/fluid/framework/operator.h"
 #include "paddle/fluid/framework/scope.h"
 
+namespace paddle {
+namespace framework {
+class OpDesc;
+class Scope;
+namespace ir {
+class Node;
+}  // namespace ir
+}  // namespace framework
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 namespace details {

paddle/fluid/framework/details/scale_loss_grad_op_handle.cc

@@ -13,9 +13,17 @@
 // limitations under the License.
 
 #include "paddle/fluid/framework/details/scale_loss_grad_op_handle.h"
+
 #include <string>
+
 #include "paddle/fluid/platform/profiler.h"
 
+namespace paddle {
+namespace framework {
+class Tensor;
+}  // namespace framework
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 namespace details {

paddle/fluid/framework/details/scale_loss_grad_op_handle.h

@@ -21,6 +21,18 @@
 #include "paddle/fluid/framework/lod_tensor.h"
 #include "paddle/fluid/framework/scope.h"
 
+namespace paddle {
+namespace framework {
+class Scope;
+namespace ir {
+class Node;
+}  // namespace ir
+}  // namespace framework
+namespace platform {
+class DeviceContext;
+}  // namespace platform
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 namespace details {

paddle/fluid/framework/details/scope_buffered_monitor.h

@@ -17,7 +17,9 @@
 #include <unordered_set>
 #include <utility>
 #include <vector>
+
 #include "paddle/fluid/framework/scope.h"
+
 namespace paddle {
 namespace framework {
 namespace details {

paddle/fluid/framework/details/share_tensor_buffer_functor.cc

@@ -22,6 +22,17 @@
 #include "paddle/fluid/framework/selected_rows.h"
 #include "paddle/fluid/platform/enforce.h"
 
+namespace paddle {
+namespace framework {
+class Scope;
+class Tensor;
+class Variable;
+namespace ir {
+class MemOptVarInfo;
+}  // namespace ir
+}  // namespace framework
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 namespace details {

paddle/fluid/framework/details/share_tensor_buffer_functor.h

@@ -25,6 +25,15 @@
 #include "paddle/fluid/framework/scope.h"
 #include "paddle/fluid/framework/variable.h"
 
+namespace paddle {
+namespace framework {
+class Scope;
+namespace ir {
+class MemOptVarInfo;
+}  // namespace ir
+}  // namespace framework
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 namespace details {

paddle/fluid/framework/details/share_tensor_buffer_op_handle.cc

@@ -23,10 +23,20 @@
 #include "paddle/fluid/framework/selected_rows.h"
 #include "paddle/fluid/platform/enforce.h"
 
+namespace paddle {
+namespace framework {
+namespace ir {
+class MemOptVarInfo;
+}  // namespace ir
+}  // namespace framework
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 namespace details {
 
+class ComputationOpHandle;
+
 ComputationOpHandle *GetUniquePendingComputationOpHandle(
     ShareTensorBufferOpHandle *share_tensor_op) {
   ComputationOpHandle *result_op = nullptr;

paddle/fluid/framework/details/share_tensor_buffer_op_handle.h

@@ -22,10 +22,22 @@
 #include "paddle/fluid/framework/details/op_handle_base.h"
 #include "paddle/fluid/framework/details/share_tensor_buffer_functor.h"
 
+namespace paddle {
+namespace framework {
+class Scope;
+namespace ir {
+class MemOptVarInfo;
+class Node;
+}  // namespace ir
+}  // namespace framework
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 namespace details {
 
+class ComputationOpHandle;
+
 class ShareTensorBufferOpHandle : public OpHandleBase {
  public:
   ShareTensorBufferOpHandle(

paddle/fluid/framework/details/sparse_all_reduce_op_handle.h

@@ -23,6 +23,17 @@
 #include "paddle/fluid/framework/scope.h"
 #include "paddle/fluid/platform/nccl_helper.h"
 
+namespace paddle {
+namespace framework {
+namespace ir {
+class Node;
+}  // namespace ir
+}  // namespace framework
+namespace platform {
+class NCCLCommunicator;
+}  // namespace platform
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 namespace details {

paddle/fluid/framework/details/var_handle.h

@@ -24,6 +24,14 @@
 #include "paddle/fluid/platform/macros.h"
 #include "paddle/fluid/platform/place.h"
 
+namespace paddle {
+namespace framework {
+namespace ir {
+class Node;
+}  // namespace ir
+}  // namespace framework
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 namespace details {

paddle/fluid/framework/details/variable_visitor.cc

@@ -15,6 +15,14 @@
 #include "paddle/fluid/framework/details/variable_visitor.h"
 
 #include "paddle/fluid/framework/selected_rows.h"
+
+namespace paddle {
+namespace framework {
+class LoDTensor;
+class Variable;
+}  // namespace framework
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 namespace details {

paddle/fluid/framework/details/variable_visitor.h

@@ -17,6 +17,13 @@
 #include "paddle/fluid/framework/lod_tensor.h"
 #include "paddle/fluid/framework/variable.h"
 
+namespace paddle {
+namespace framework {
+class Tensor;
+class Variable;
+}  // namespace framework
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 namespace details {

paddle/fluid/framework/device_worker.cc

@@ -13,11 +13,13 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/framework/device_worker.h"
-#include "xxhash.h"  // NOLINT
 
 namespace paddle {
 namespace framework {
 
+class LoDTensor;
+class Scope;
+
 void DeviceWorker::SetRootScope(Scope* root_scope) { root_scope_ = root_scope; }
 
 void DeviceWorker::SetDataFeed(DataFeed* data_feed) {

paddle/fluid/framework/device_worker.h

@@ -39,6 +39,18 @@ limitations under the License. */
 #include "paddle/fluid/platform/port.h"
 #include "paddle/fluid/platform/timer.h"
 
+namespace paddle {
+namespace framework {
+class LoDTensor;
+class ProgramDesc;
+class Scope;
+class Tensor;
+}  // namespace framework
+namespace platform {
+class DeviceContext;
+}  // namespace platform
+}  // namespace paddle
+
 #if defined(PADDLE_WITH_NCCL)
 #include "paddle/fluid/platform/nccl_helper.h"
 #endif

paddle/fluid/framework/device_worker_factory.cc

@@ -20,6 +20,8 @@ limitations under the License. */
 namespace paddle {
 namespace framework {
 
+class DeviceWorker;
+
 typedef std::shared_ptr<DeviceWorker> (*Createdevice_workerFunction)();
 typedef std::unordered_map<std::string, Createdevice_workerFunction>
     device_workerMap;

paddle/fluid/framework/device_worker_factory.h

@@ -16,11 +16,14 @@ limitations under the License. */
 
 #include <memory>
 #include <string>
+
 #include "paddle/fluid/framework/device_worker.h"
 
 namespace paddle {
 namespace framework {
 
+class DeviceWorker;
+
 class DeviceWorkerFactory {
  public:
   static std::string DeviceWorkerTypeList();

paddle/fluid/framework/device_worker_test.cc

@@ -13,9 +13,10 @@
 // limitations under the License.
 
 #include "paddle/fluid/framework/device_worker.h"
+
 #include <gtest/gtest.h>
+
 #include "paddle/fluid/framework/lod_tensor.h"
-#include "paddle/fluid/framework/trainer.h"
 
 namespace paddle {
 namespace framework {

paddle/fluid/framework/dlpack_tensor.cc

@@ -11,10 +11,17 @@
 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 // See the License for the specific language governing permissions and
 // limitations under the License.
+#include "paddle/fluid/framework/dlpack_tensor.h"
 #include <unordered_map>
-
 #include "paddle/fluid/framework/data_type.h"
-#include "paddle/fluid/framework/dlpack_tensor.h"
+
+namespace paddle {
+namespace platform {
+struct bfloat16;
+struct float16;
+}  // namespace platform
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 

paddle/fluid/framework/dlpack_tensor.h

@@ -15,11 +15,14 @@
 #pragma once
 
 #include <dlpack/dlpack.h>
+
 #include "paddle/fluid/framework/tensor.h"
 
 namespace paddle {
 namespace framework {
 
+class Tensor;
+
 class DLPackTensor {
  public:
   using LaneType = decltype(::DLTensor::dtype.lanes);  // uint16_t

paddle/fluid/framework/dlpack_tensor_test.cc

@@ -17,6 +17,12 @@
 #include <gtest/gtest.h>
 #include <vector>
 
+namespace paddle {
+namespace platform {
+struct float16;
+}  // namespace platform
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 

paddle/fluid/framework/downpour_worker.cc

@@ -13,10 +13,14 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/framework/device_worker.h"
-#include "paddle/fluid/framework/device_worker_factory.h"
-#include "paddle/fluid/framework/fleet/fleet_wrapper.h"
 #include "paddle/fluid/platform/cpu_helper.h"
-#include "paddle/fluid/string/string_helper.h"
+
+namespace paddle {
+namespace framework {
+class LoDTensor;
+class Variable;
+}  // namespace framework
+}  // namespace paddle
 
 #if defined _WIN32 || defined __APPLE__
 #else

paddle/fluid/framework/downpour_worker_opt.cc

@@ -12,18 +12,16 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include <set>
-#include <unordered_map>
-#include "paddle/fluid/framework/data_type.h"
 #include "paddle/fluid/framework/device_worker.h"
-#include "paddle/fluid/framework/device_worker_factory.h"
-#include "paddle/fluid/framework/fleet/fleet_wrapper.h"
 #include "paddle/fluid/platform/cpu_helper.h"
-#include "paddle/fluid/platform/lodtensor_printer.h"
 
 namespace paddle {
 namespace framework {
 
+class OpDesc;
+class OperatorBase;
+class ProgramDesc;
+
 bool HasDependentOutput(const OpDesc& op_desc,
                         const std::unordered_set<std::string>& dependent_vars) {
   for (auto& var : op_desc.Outputs()) {

paddle/fluid/framework/eigen.h

@@ -14,6 +14,8 @@ limitations under the License. */
 
 #pragma once
 
+#include <stdint.h>
+
 #include "paddle/fluid/framework/tensor.h"
 #include "unsupported/Eigen/CXX11/Tensor"
 

paddle/fluid/framework/executor.h

@@ -19,6 +19,7 @@ limitations under the License. */
 #include <string>
 #include <unordered_map>
 #include <vector>
+
 #include "paddle/fluid/framework/data_set.h"
 #include "paddle/fluid/framework/executor_gc_helper.h"
 #include "paddle/fluid/framework/garbage_collector.h"
@@ -32,6 +33,11 @@ limitations under the License. */
 namespace paddle {
 namespace framework {
 
+class Dataset;
+class ProgramDesc;
+class Scope;
+class TrainerBase;
+
 struct ExecutorPrepareContext {
   ExecutorPrepareContext(const framework::ProgramDesc& prog, size_t block_id);
 

paddle/fluid/framework/executor_gc_helper.cc

@@ -13,16 +13,19 @@
 // limitations under the License.
 
 #include "paddle/fluid/framework/executor_gc_helper.h"
+
 #include <deque>
 #include <string>
-#include <unordered_map>
 #include <unordered_set>
 #include <utility>
-#include <vector>
+
 #include "glog/logging.h"
-#include "paddle/fluid/framework/lod_tensor.h"
-#include "paddle/fluid/framework/lod_tensor_array.h"
-#include "paddle/fluid/framework/selected_rows.h"
+#include "paddle/fluid/framework/block_desc.h"
+#include "paddle/fluid/framework/framework.pb.h"
+#include "paddle/fluid/framework/no_need_buffer_vars_inference.h"
+#include "paddle/fluid/framework/op_info.h"
+#include "paddle/fluid/framework/operator.h"
+#include "paddle/fluid/framework/var_desc.h"
 #include "paddle/fluid/platform/enforce.h"
 
 namespace paddle {

paddle/fluid/framework/executor_gc_helper.h

@@ -18,6 +18,7 @@
 #include <string>
 #include <unordered_map>
 #include <vector>
+
 #include "paddle/fluid/framework/garbage_collector.h"
 #include "paddle/fluid/framework/operator.h"
 #include "paddle/fluid/framework/scope.h"
@@ -26,6 +27,10 @@ namespace paddle {
 namespace framework {
 
 // Result map: op -> variable names that can be deleted after op runs
+class GarbageCollector;
+class OperatorBase;
+class Scope;
+
 std::unordered_map<const OperatorBase *, std::vector<std::string>>
 GetUnusedVars(const BlockDesc &block,
               const std::vector<std::unique_ptr<OperatorBase>> &ops,

paddle/fluid/framework/feed_fetch_method.cc

@@ -13,16 +13,17 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/framework/feed_fetch_method.h"
+
 #include <string>
-#include <vector>
+
 #include "glog/logging.h"
-#include "paddle/fluid/framework/var_type.h"
-#include "paddle/fluid/framework/variable.h"
-#include "paddle/fluid/platform/place.h"
 
 namespace paddle {
 namespace framework {
 
+class LoDTensor;
+class Variable;
+
 void SetFeedVariable(Scope* scope, const LoDTensor& input,
                      const std::string& var_name, size_t index) {
   // If var_name Variable is not found in GlobalScope, a new variable will

paddle/fluid/framework/feed_fetch_method.h

@@ -15,12 +15,16 @@ limitations under the License. */
 #pragma once
 
 #include <string>
+
 #include "paddle/fluid/framework/feed_fetch_type.h"
 #include "paddle/fluid/framework/scope.h"
 
 namespace paddle {
 namespace framework {
 
+class LoDTensor;
+class Scope;
+
 void SetFeedVariable(Scope* scope, const LoDTensor& input,
                      const std::string& var_name, size_t index);
 

paddle/fluid/framework/fleet/fleet_wrapper.cc

@@ -29,12 +29,6 @@ limitations under the License. */
 #include "paddle/fluid/framework/fleet/fleet_wrapper.h"
 #include <algorithm>
 #include <utility>
-#include "paddle/fluid/framework/channel.h"
-#include "paddle/fluid/framework/data_feed.h"
-#include "paddle/fluid/framework/io/fs.h"
-#include "paddle/fluid/framework/op_registry.h"
-#include "paddle/fluid/framework/scope.h"
-#include "paddle/fluid/platform/timer.h"
 
 namespace paddle {
 namespace framework {

paddle/fluid/framework/fleet/fleet_wrapper.h

@@ -35,6 +35,12 @@ limitations under the License. */
 #include "paddle/fluid/framework/variable_helper.h"
 #include "paddle/fluid/platform/macros.h"  // for DISABLE_COPY_AND_ASSIGN
 
+namespace paddle {
+namespace framework {
+class Scope;
+}  // namespace framework
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 

paddle/fluid/framework/fleet/gloo_wrapper.cc

@@ -10,10 +10,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/framework/fleet/gloo_wrapper.h"
-#include <thread>  // NOLINT
-#include <vector>
 #include "paddle/fluid/framework/io/fs.h"
-#include "paddle/fluid/platform/errors.h"
 #include "paddle/fluid/string/string_helper.h"
 
 namespace gloo {

paddle/fluid/framework/fleet/heter_wrapper.cc

@@ -27,15 +27,6 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/framework/fleet/heter_wrapper.h"
-#include <algorithm>
-#include <utility>
-#include "paddle/fluid/framework/channel.h"
-#include "paddle/fluid/framework/data_feed.h"
-#include "paddle/fluid/framework/device_worker.h"
-#include "paddle/fluid/framework/io/fs.h"
-#include "paddle/fluid/framework/op_registry.h"
-#include "paddle/fluid/framework/scope.h"
-#include "paddle/fluid/platform/timer.h"
 #ifdef PADDLE_WITH_PSLIB
 
 namespace paddle {
@@ -122,14 +113,15 @@ void HeterWrapper::SerializeToReq(const std::string& varname, Scope* scope,
   if (platform::is_cpu_place(tensor->place())) {
     memcpy(data_ptr, tensor->data<void>(),
            tensor->numel() * SizeOfType(tensor->type()));
-  }
 #ifdef PADDLE_WITH_CUDA
-  else {
+  } else {
     memory::Copy(platform::CPUPlace(), data_ptr,
                  BOOST_GET_CONST(platform::CUDAPlace, tensor->place()),
                  tensor->data<void>(),
                  tensor->numel() * SizeOfType(tensor->type()), nullptr);
   }
+#else
+  }
 #endif
 }
 
@@ -239,7 +231,7 @@ void HeterWrapper::CallRemoteXpu(std::shared_ptr<HeterTask> task,
   request.set_cur_batch(task->cur_batch_);
 
   OnHeterRpcDone* done = new OnHeterRpcDone([this, task, worker](void* done) {
-    auto* closure = (OnHeterRpcDone*)done;
+    auto* closure = reinterpret_cast<OnHeterRpcDone*>(done);
     if (closure->cntl.Failed()) {
       VLOG(0) << "call xpu fail: " << closure->cntl.ErrorText();
     } else {

paddle/fluid/framework/fleet/nccl_wrapper.cc

@@ -13,9 +13,6 @@
 // limitations under the License.
 
 #include "paddle/fluid/framework/fleet/nccl_wrapper.h"
-#include <utility>
-#include "paddle/fluid/framework/data_feed.h"
-#include "paddle/fluid/framework/scope.h"
 
 namespace paddle {
 namespace framework {

paddle/fluid/framework/fleet/nccl_wrapper.h

@@ -21,6 +21,7 @@ limitations under the License. */
 #include <random>
 #include <string>
 #include <vector>
+
 #include "paddle/fluid/framework/program_desc.h"
 #include "paddle/fluid/framework/scope.h"
 #include "paddle/fluid/framework/variable_helper.h"
@@ -29,6 +30,12 @@ limitations under the License. */
 #endif
 #include "paddle/fluid/platform/macros.h"  // for DISABLE_COPY_AND_ASSIGN
 
+namespace paddle {
+namespace framework {
+class Scope;
+}  // namespace framework
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 

paddle/fluid/framework/garbage_collector.h

@@ -19,9 +19,16 @@
 #include <memory>
 #include <mutex>  // NOLINT
 #include <utility>
+
 #include "gflags/gflags.h"
 #include "paddle/fluid/platform/device_context.h"
 
+namespace paddle {
+namespace platform {
+class DeviceContext;
+}  // namespace platform
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 

paddle/fluid/framework/generator.h

@@ -16,7 +16,6 @@ limitations under the License. */
 
 #include <glog/logging.h>
 #include <stdint.h>
-
 #include <atomic>
 #include <deque>
 #include <iostream>  // temp for debug

paddle/fluid/framework/hetercpu_worker.cc

@@ -12,13 +12,6 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include "paddle/fluid/framework/device_worker.h"
-#include "paddle/fluid/framework/device_worker_factory.h"
-#include "paddle/fluid/framework/fleet/fleet_wrapper.h"
-#include "paddle/fluid/framework/fleet/heter_wrapper.h"
-#include "paddle/fluid/platform/cpu_helper.h"
-#include "paddle/fluid/string/string_helper.h"
-
 #ifdef PADDLE_WITH_PSLIB
 
 #if defined _WIN32 || defined __APPLE__

paddle/fluid/framework/heterxpu_trainer.cc

@@ -12,16 +12,6 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include <cstdlib>
-#include <ctime>
-#include <string>
-#include <vector>
-#include "io/fs.h"
-#include "paddle/fluid/framework/data_feed_factory.h"
-#include "paddle/fluid/framework/data_set.h"
-#include "paddle/fluid/framework/device_worker_factory.h"
-#include "paddle/fluid/framework/fleet/fleet_wrapper.h"
-#include "paddle/fluid/framework/trainer.h"
 #if (defined PADDLE_WITH_CUDA) && (defined PADDLE_WITH_PSLIB)
 #include "paddle/fluid/platform/cuda_device_guard.h"
 
@@ -334,7 +324,7 @@ int HeterXpuTrainer::RunTask(const HeterRequest* request,
   std::shared_ptr<HeterServiceContext> context = object_pool_.Get();
 
   if (!context->scope_) {
-    int num = rand() % places_.size();
+    int num = rand_r() % places_.size();
     context->place_num_ = num;
     auto place = places_[num];
     context->scope_ = &(place_scopes_[num]->NewScope());

paddle/fluid/framework/inlined_vector.h

@@ -15,6 +15,7 @@
 #pragma once
 #include <cstdint>
 #include <vector>
+
 #include "paddle/fluid/platform/enforce.h"
 
 namespace paddle {

paddle/fluid/framework/inlined_vector_test.cc

@@ -13,10 +13,10 @@
 // limitations under the License.
 
 #include "paddle/fluid/framework/inlined_vector.h"
+
 #include <cstdlib>
 #include <ctime>
-#include <iostream>
-#include <vector>
+
 #include "gtest/gtest.h"
 
 namespace paddle {

paddle/fluid/framework/io/crypto/cipher_utils.cc

@@ -15,8 +15,6 @@
 #include "paddle/fluid/framework/io/crypto/cipher_utils.h"
 
 #include <cryptopp/osrng.h>
-
-#include <fstream>
 #include <sstream>
 
 #include "paddle/fluid/platform/enforce.h"

paddle/fluid/framework/io/crypto/cipher_utils.h

@@ -17,6 +17,7 @@
 #include <sstream>
 #include <string>
 #include <unordered_map>
+
 namespace paddle {
 namespace framework {
 

paddle/fluid/framework/io/fs.h

@@ -18,6 +18,7 @@
 #include <memory>
 #include <string>
 #include <vector>
+
 #include "glog/logging.h"
 #include "paddle/fluid/framework/io/shell.h"
 #include "paddle/fluid/string/string_helper.h"

paddle/fluid/framework/io/shell.h

@@ -32,6 +32,7 @@
 #include <string>
 #include <utility>
 #include <vector>
+
 #include "paddle/fluid/platform/port.h"
 #include "paddle/fluid/string/string_helper.h"
 

paddle/fluid/framework/ir/attention_lstm_fuse_pass.h

@@ -20,6 +20,8 @@ namespace paddle {
 namespace framework {
 namespace ir {
 
+class Graph;
+
 class AttentionLSTMFusePass : public FusePassBase {
  protected:
   void ApplyImpl(ir::Graph* graph) const override;

paddle/fluid/framework/ir/conv_affine_channel_fuse_pass.cc

@@ -13,19 +13,28 @@
 // limitations under the License.
 
 #include "paddle/fluid/framework/ir/conv_affine_channel_fuse_pass.h"
+
 #include <cmath>
-#include <functional>
-#include <string>
 #include <vector>
+
 #include "paddle/fluid/framework/lod_tensor.h"
 #include "paddle/fluid/framework/op_version_registry.h"
 #include "paddle/fluid/operators/math/cpu_vec.h"
 #include "paddle/fluid/platform/enforce.h"
 
+namespace paddle {
+namespace framework {
+class LoDTensor;
+class Scope;
+}  // namespace framework
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 namespace ir {
 
+class Node;
+
 #define GET_CONV_BN_NODES(pattern_name)                                    \
   /* OPERATORS */                                                          \
   GET_IR_NODE_FROM_SUBGRAPH(conv, conv, pattern_name);                     \

paddle/fluid/framework/ir/conv_affine_channel_fuse_pass.h

@@ -15,6 +15,7 @@
 #pragma once
 
 #include <string>
+
 #include "paddle/fluid/framework/ir/fuse_pass_base.h"
 #include "paddle/fluid/framework/ir/graph.h"
 #include "paddle/fluid/framework/ir/graph_pattern_detector.h"
@@ -26,6 +27,8 @@ namespace ir {
 /*
  * Fuse the Conv and ConvAffineChannel.
  */
+class Graph;
+
 class ConvAffineChannelFusePass : public FusePassBase {
  public:
   virtual ~ConvAffineChannelFusePass() {}

paddle/fluid/framework/ir/conv_bn_fuse_pass.cc

@@ -13,15 +13,22 @@
 // limitations under the License.
 
 #include "paddle/fluid/framework/ir/conv_bn_fuse_pass.h"
-#include <algorithm>
-#include <functional>
+
 #include <string>
 #include <vector>
+
 #include "paddle/fluid/framework/lod_tensor.h"
 #include "paddle/fluid/framework/op_version_registry.h"
 #include "paddle/fluid/operators/math/cpu_vec.h"
 #include "paddle/fluid/platform/enforce.h"
 
+namespace paddle {
+namespace framework {
+class LoDTensor;
+class Scope;
+}  // namespace framework
+}  // namespace paddle
+
 namespace paddle {
 namespace framework {
 namespace ir {

paddle/fluid/framework/ir/conv_bn_fuse_pass.h

@@ -15,6 +15,7 @@
 #pragma once
 
 #include <string>
+
 #include "paddle/fluid/framework/ir/fuse_pass_base.h"
 #include "paddle/fluid/framework/ir/graph.h"
 #include "paddle/fluid/framework/ir/graph_pattern_detector.h"
@@ -26,6 +27,8 @@ namespace ir {
 /*
  * Fuse the Conv and BatchNorm to a ConvBNMKLDNNOp.
  */
+class Graph;
+
 class ConvBNFusePass : public FusePassBase {
  public:
   virtual ~ConvBNFusePass() {}

paddle/fluid/framework/ir/conv_elementwise_add2_act_fuse_pass.h

@@ -20,6 +20,8 @@ namespace paddle {
 namespace framework {
 namespace ir {
 
+class Graph;
+
 class ConvElementwiseAdd2ActFusePass : public FusePassBase {
  public:
   virtual ~ConvElementwiseAdd2ActFusePass() {}

paddle/fluid/framework/ir/conv_elementwise_add_act_fuse_pass.cc

@@ -14,6 +14,7 @@
 
 #include "paddle/fluid/framework/ir/conv_elementwise_add_act_fuse_pass.h"
 #include <string>
+
 #include "paddle/fluid/framework/ir/graph_viz_pass.h"
 #include "paddle/fluid/framework/op_version_registry.h"