Recover the profiler

cmake/cuda.cmake

@@ -199,10 +199,13 @@ elseif(CMAKE_BUILD_TYPE  STREQUAL "MinSizeRel")
     list(APPEND CUDA_NVCC_FLAGS  ${CMAKE_CXX_FLAGS_RELEASE})
 endif()
 else(NOT WIN32)
+list(APPEND CUDA_NVCC_FLAGS  "--compiler-options;/bigobj")
 if(CMAKE_BUILD_TYPE  STREQUAL "Debug")
-    list(APPEND CUDA_NVCC_FLAGS  "-g -G --compiler-options;/bigobj")
+  list(APPEND CUDA_NVCC_FLAGS  "-g -G")
+  # match the cl's _ITERATOR_DEBUG_LEVEL
+  list(APPEND CUDA_NVCC_FLAGS  "-D_DEBUG")
 elseif(CMAKE_BUILD_TYPE STREQUAL "Release")
-  list(APPEND CUDA_NVCC_FLAGS "-O3 -DNDEBUG --compiler-options;/bigobj")
+  list(APPEND CUDA_NVCC_FLAGS "-O3 -DNDEBUG")
 else()
   message(FATAL "Windows only support Release or Debug build now. Please set visual studio build type to Release/Debug, x64 build.")
 endif()

paddle/fluid/framework/lod_tensor.cc

@@ -26,10 +26,8 @@ limitations under the License. */
 #include "paddle/fluid/memory/memcpy.h"
 #include "paddle/fluid/memory/memory.h"
 
-#if !defined(_WIN32)
 #include "paddle/fluid/recordio/scanner.h"
 #include "paddle/fluid/recordio/writer.h"
-#endif  // _WIN32
 
 namespace paddle {
 namespace framework {
@@ -305,7 +303,6 @@ void DeserializeFromStream(std::istream &is, LoDTensor *tensor,
   TensorFromStream(is, static_cast<Tensor *>(tensor), dev_ctx);
 }
 
-#if !defined(_WIN32)
 void WriteToRecordIO(recordio::Writer *writer,
                      const std::vector<LoDTensor> &tensor,
                      const platform::DeviceContext &dev_ctx) {
@@ -335,19 +332,7 @@ bool ReadFromRecordIO(recordio::Scanner *scanner,
 
   return true;
 }
-#else
+
-class Writer {};
-class Scanner {};
-void WriteToRecordIO(recordio::Writer *writer,
-                     const std::vector<LoDTensor> &tensor,
-                     const platform::DeviceContext &dev_ctx) {}
-bool ReadFromRecordIO(recordio::Scanner *scanner,
-                      const platform::DeviceContext &dev_ctx,
-                      std::vector<LoDTensor> *result_ptr) {
-  PADDLE_ENFORCE("windows didn't supported recordio!.");
-  return true;
-}
-#endif  // _WIN32
 std::vector<LoDTensor> LoDTensor::SplitLoDTensor(
     const std::vector<platform::Place> places) const {
   check_memory_size();

paddle/fluid/framework/lod_tensor_test.cc

@@ -274,7 +274,6 @@ TEST(LoD, ConvertToOffsetBasedLoD) {
   EXPECT_EQ(offset_lod, expected);
 }
 
-#if !defined(_WIN32)
 template <typename T>
 static void TestRecordIO() {
   LoDTensor tensor;
@@ -321,7 +320,6 @@ TEST(LoDTensor, RecordIO) {
   TestRecordIO<float>();
   TestRecordIO<double>();
 }
-#endif  // !defined(_WIN32)
 
 }  // namespace framework
 }  // namespace paddle

paddle/fluid/framework/operator.cc

@@ -153,17 +153,14 @@ void OperatorBase::Run(const Scope& scope, const platform::Place& place) {
 #endif
   }
 
-// The profile has a process-wide mutex, results in serious performance issue
+  // The profile has a process-wide mutex, results in serious performance issue
-// in concurrency scenerio. Here use an `if` to fix this issue.
+  // in concurrency scenerio. Here use an `if` to fix this issue.
-// Please not remove the `if`, ask @Superjomn if there are any concern.
+  // Please not remove the `if`, ask @Superjomn if there are any concern.
-#ifndef _WIN32
   if (platform::IsProfileEnabled()) {
     platform::DeviceContextPool& pool = platform::DeviceContextPool::Instance();
     platform::RecordEvent record_event(Type(), pool.Get(place));
     RunImpl(scope, place);
-  } else
+  } else {
-#endif
-  {
     RunImpl(scope, place);
   }
   VLOG(30) << place << " " << DebugStringEx(&scope);

paddle/fluid/inference/api/analysis_predictor.cc

@@ -56,7 +56,6 @@ bool AnalysisPredictor::Init(
     const std::shared_ptr<framework::Scope> &parent_scope,
     const std::shared_ptr<framework::ProgramDesc> &program) {
   VLOG(30) << "Predictor::init()";
-#if !defined(_WIN32)
   if (FLAGS_profile) {
     LOG(WARNING) << "Profiler is actived, might affect the performance";
     LOG(INFO) << "You can turn off by set gflags '-profile false'";
@@ -64,7 +63,6 @@ bool AnalysisPredictor::Init(
                                            : platform::ProfilerState::kCPU;
     platform::EnableProfiler(tracking_device);
   }
-#endif
 
   // no matter with or without MKLDNN
   paddle::platform::SetNumThreads(FLAGS_paddle_num_threads);
@@ -501,12 +499,10 @@ bool AnalysisPredictor::LoadParameters() {
 }
 
 AnalysisPredictor::~AnalysisPredictor() {
-#if !defined(_WIN32)
   if (FLAGS_profile) {
     platform::DisableProfiler(platform::EventSortingKey::kTotal,
                               "./profile.log");
   }
-#endif
   if (sub_scope_) {
     scope_->DeleteScope(sub_scope_);
   }

paddle/fluid/inference/api/api_impl.cc

@@ -64,7 +64,6 @@ void NativePaddlePredictor::PrepareFeedFetch() {
 bool NativePaddlePredictor::Init(
     std::shared_ptr<framework::Scope> parent_scope) {
   VLOG(3) << "Predictor::init()";
-#if !defined(_WIN32)
   if (FLAGS_profile) {
     LOG(WARNING) << "Profiler is actived, might affect the performance";
     LOG(INFO) << "You can turn off by set gflags '-profile false'";
@@ -73,7 +72,6 @@ bool NativePaddlePredictor::Init(
                                            : platform::ProfilerState::kCPU;
     platform::EnableProfiler(tracking_device);
   }
-#endif
 
   // no matter with or without MKLDNN
   paddle::platform::SetNumThreads(FLAGS_paddle_num_threads);
@@ -121,12 +119,10 @@ bool NativePaddlePredictor::Init(
 }
 
 NativePaddlePredictor::~NativePaddlePredictor() {
-#if !defined(_WIN32)
   if (FLAGS_profile) {
     platform::DisableProfiler(platform::EventSortingKey::kTotal,
                               "./profile.log");
   }
-#endif
   if (sub_scope_) {
     scope_->DeleteScope(sub_scope_);
   }

paddle/fluid/inference/tests/api/tester_helper.h

@@ -177,11 +177,9 @@ void TestOneThreadPrediction(
     warmup_timer.tic();
     predictor->Run(inputs[0], outputs, batch_size);
     PrintTime(batch_size, 1, 1, 0, warmup_timer.toc(), 1);
-#if !defined(_WIN32)
     if (FLAGS_profile) {
       paddle::platform::ResetProfiler();
     }
-#endif
   }
 
   LOG(INFO) << "Run " << num_times << " times...";
@@ -230,11 +228,9 @@ void TestMultiThreadPrediction(
         warmup_timer.tic();
         predictor->Run(inputs[0], outputs, batch_size);
         PrintTime(batch_size, 1, num_threads, tid, warmup_timer.toc(), 1);
-#if !defined(_WIN32)
         if (FLAGS_profile) {
           paddle::platform::ResetProfiler();
         }
-#endif
       }
 
       LOG(INFO) << "Thread " << tid << " run " << num_times << " times...";