Fix the profiler's bug in multi-gpu mode

paddle/fluid/framework/executor.cc

@@ -25,7 +25,6 @@ limitations under the License. */
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/framework/reader.h"
 #include "paddle/fluid/platform/place.h"
-#include "paddle/fluid/platform/profiler.h"
 
 DECLARE_bool(benchmark);
 DEFINE_bool(check_nan_inf, false,
@@ -126,9 +125,6 @@ void Executor::Run(const ProgramDesc& pdesc, Scope* scope, int block_id,
   for (auto& op_desc : block.AllOps()) {
     auto op = paddle::framework::OpRegistry::CreateOp(*op_desc);
 
-    platform::DeviceContextPool& pool = platform::DeviceContextPool::Instance();
-    platform::RecordEvent record_event(op->Type(), pool.Get(place_));
-
     VLOG(3) << place_ << " " << op->DebugStringEx(local_scope);
     op->Run(*local_scope, place_);
 

paddle/fluid/framework/operator.cc

@@ -21,6 +21,7 @@ limitations under the License. */
 #include "paddle/fluid/framework/operator.h"
 #include "paddle/fluid/framework/shape_inference.h"
 #include "paddle/fluid/framework/var_type.h"
+#include "paddle/fluid/platform/profiler.h"
 
 DECLARE_bool(benchmark);
 
@@ -497,7 +498,8 @@ void OperatorWithKernel::RunImpl(const Scope& scope,
   this->InferShape(&infer_shape_ctx);
   platform::DeviceContextPool& pool = platform::DeviceContextPool::Instance();
   auto dev_ctx = pool.Get(place);
-
+  // profile
+  platform::RecordEvent record_event(Type(), dev_ctx);
   // check if op[type] has kernel registered.
   auto& all_op_kernels = AllOpKernels();
   auto kernels_iter = all_op_kernels.find(type_);