optimize updateoutput in gpu

mindspore/ccsrc/backend/session/ascend_session.cc

@@ -426,7 +426,12 @@ py::tuple AscendSession::RunOp(const OpRunInfo &op_run_info, const GraphInfo &gr
   if (op_run_info.value != nullptr) {
     std::vector<tensor::TensorPtr> pre_output_tensors;
     TensorValueToTensor(op_run_info.value, &pre_output_tensors);
-    std::copy(pre_output_tensors.begin(), pre_output_tensors.end(), std::back_inserter(outputs));
+    for (auto &pre_output : pre_output_tensors) {
+      tensor::TensorPtr tensor = std::make_shared<tensor::Tensor>(pre_output->data_type(), pre_output->shape());
+      tensor->set_device_address(pre_output->device_address());
+      tensor->set_dirty(false);
+      outputs.emplace_back(tensor);
+    }
   } else {
     UpdateOutputs(graph, &outputs, input_tensors);
   }

mindspore/ccsrc/backend/session/gpu_session.cc

@@ -300,7 +300,18 @@ py::tuple GPUSession::RunOp(const OpRunInfo &op_run_info, const GraphInfo &graph
   }
   // Fetch outputs
   VectorRef outputs;
-  UpdateOutputs(kernel_graph, &outputs, input_tensors);
+  if (op_run_info.value != nullptr) {
+    std::vector<tensor::TensorPtr> pre_output_tensors;
+    TensorValueToTensor(op_run_info.value, &pre_output_tensors);
+    for (auto &pre_output : pre_output_tensors) {
+      tensor::TensorPtr tensor = std::make_shared<tensor::Tensor>(pre_output->data_type(), pre_output->shape());
+      tensor->set_device_address(pre_output->device_address());
+      tensor->set_dirty(false);
+      outputs.emplace_back(tensor);
+    }
+  } else {
+    UpdateOutputs(kernel_graph, &outputs, input_tensors);
+  }
   // Trans output to tuple
   auto output_tensors = TransformBaseRefListToTuple(outputs);
   if (!utils::isa<PyObjectRef>(output_tensors) ||

mindspore/ccsrc/pipeline/pynative/pynative_execute.cc

@@ -565,9 +565,9 @@ py::object RunOpInMs(const OpExecInfoPtr &op_exec_info, PynativeStatusCode *stat
 
   if (session == nullptr) {
     session = session::SessionFactory::Get().Create(device_target);
+    MS_EXCEPTION_IF_NULL(session);
+    session->Init(ms_context->device_id());
   }
-  MS_EXCEPTION_IF_NULL(session);
-  session->Init(ms_context->device_id());
 
   std::vector<tensor::TensorPtr> input_tensors;
   std::vector<int> tensors_mask;