debug the device context

cmake/external/gflags.cmake

@@ -47,6 +47,10 @@ ExternalProject_Add(
                      -DCMAKE_POSITION_INDEPENDENT_CODE:BOOL=ON
                      -DCMAKE_BUILD_TYPE:STRING=${THIRD_PARTY_BUILD_TYPE}
 )
+
+ADD_LIBRARY(gflags STATIC IMPORTED GLOBAL)
+SET_PROPERTY(TARGET gflags PROPERTY IMPORTED_LOCATION ${GFLAGS_LIBRARIES})
+ADD_DEPENDENCIES(gflags extern_gflags)
 IF(WIN32)
   IF(NOT EXISTS "${GFLAGS_INSTALL_DIR}/lib/libgflags.lib")
     add_custom_command(TARGET extern_gflags POST_BUILD
@@ -54,9 +58,6 @@ IF(WIN32)
   )
   ENDIF()
 ENDIF(WIN32)
-ADD_LIBRARY(gflags STATIC IMPORTED GLOBAL)
-SET_PROPERTY(TARGET gflags PROPERTY IMPORTED_LOCATION ${GFLAGS_LIBRARIES})
-ADD_DEPENDENCIES(gflags extern_gflags)
 
 LIST(APPEND external_project_dependencies gflags)
 

paddle/fluid/inference/api/api_impl.cc

@@ -61,7 +61,7 @@ bool NativePaddlePredictor::Init(
     platform::EnableProfiler(tracking_device);
   }
 #endif
-
+  VLOG(3) << "before Place";
   if (config_.use_gpu) {
     place_ = paddle::platform::CUDAPlace(config_.device);
   } else {
@@ -73,10 +73,12 @@ bool NativePaddlePredictor::Init(
     sub_scope_ = &(parent_scope->NewScope());
     PADDLE_ENFORCE_NOT_NULL(sub_scope_, "create sub scope fail");
   } else {
+    VLOG(3) << "Before InitDevices";
     paddle::framework::InitDevices(false);
+    VLOG(3) << "after InitDevices";
     scope_.reset(new paddle::framework::Scope());
   }
-  VLOG(3) << "after scope"
+  VLOG(3) << "after scope";
   executor_.reset(new paddle::framework::Executor(place_));
   VLOG(3) << "executor";
   // Initialize the inference program

paddle/fluid/platform/init.cc

@@ -94,7 +94,9 @@ void InitDevices(bool init_p2p, const std::vector<int> devices) {
   int count = 0;
 #ifdef PADDLE_WITH_CUDA
   try {
+    VLOG(3) << "get cuda count";   
     count = platform::GetCUDADeviceCount();
+    VLOG(3) << "get cuda pass";
   } catch (const std::exception &exp) {
     LOG(WARNING) << "Compiled with WITH_GPU, but no GPU found in runtime.";
   }
@@ -107,11 +109,14 @@ void InitDevices(bool init_p2p, const std::vector<int> devices) {
     }
     places.emplace_back(platform::CUDAPlace(devices[i]));
   }
+  VLOG(3) << "before p2p";
   if (init_p2p) {
     InitP2P(devices);
   }
+  VLOG(3) << "p2p pass";
   places.emplace_back(platform::CPUPlace());
   platform::DeviceContextPool::Init(places);
+  VLOG(3) << "init pass";
 #ifndef PADDLE_WITH_MKLDNN
   platform::SetNumThreads(FLAGS_paddle_num_threads);
 #endif