!1548 optimize performance

From: @zhou_lili
Reviewed-by: @xchu42,@youui
Signed-off-by: @youui

ge/generator/ge_generator.cc

@@ -822,7 +822,7 @@ Status GeGenerator::BuildSingleOp(OpDescPtr &op_desc, const vector<GeTensor> &in
     auto node = comp_graph->FindNode(op_desc->GetName());
     Status ret = CheckEngineTypeSupport(node, engine_type);
     if (ret != SUCCESS) {
-      GELOGE(ret, "[Check][EngineType]value:%d for node:%s not support", engine_type, node->GetName().c_str());
+      GELOGE(ret, "[Check][EngineType]not support node:%s with engine of %d.", node->GetName().c_str(), engine_type);
       return ret;
     }
   }
@@ -848,6 +848,11 @@ Status GeGenerator::BuildSingleOp(OpDescPtr &op_desc, const vector<GeTensor> &in
 
   bool all_shape = false;
   (void)AttrUtils::GetBool(op_desc, kAicpuAllshape, all_shape);
+  GELOGD("Node: %s, all_shape is %d, compile_flag is %d.", op_desc->GetName().c_str(), all_shape, compile_flag);
+  (void)AttrUtils::SetInt(ge_model, ATTR_NAME_BUILD_MODE, fuzz_compile_flag);
+  if (all_shape) {
+    (void)AttrUtils::SetBool(ge_model, kAicpuAllshape, all_shape);
+  }
   if (all_shape && CheckNoAicore(root_graph)) {
     GELOGD("Get aicpu all_shape kernel!");
     vector<GeTensor> inputs_dynamic;
@@ -857,8 +862,6 @@ Status GeGenerator::BuildSingleOp(OpDescPtr &op_desc, const vector<GeTensor> &in
     GE_CHK_STATUS_RET_NOLOG(
       impl_->SaveParams(ge_model, op_desc_tmp->GetType(), op_attrs, inputs_dynamic, outputs_dynamic));
   } else if (fuzz_compile_flag) {
-    GELOGD("Get fuzz build result of %s.", op_desc->GetName().c_str());
-    (void)AttrUtils::SetInt(ge_model, ATTR_NAME_BUILD_MODE, fuzz_compile_flag);
     GeAttrValue::LIST_NAMED_ATTRS fuzz_build_attrs;
     if (GetFuzzBuildAttrs(op_desc, ge_root_model, fuzz_build_attrs) != SUCCESS) {
       GELOGE(FAILED, "[Get][FuzzRet]Failed to get fuzz build result of %s.", op_desc->GetName().c_str());

ge/hybrid/executor/worker/execution_engine.cc

@@ -364,20 +364,28 @@ Status ExecutionEngine::ExecuteAsync(NodeState &node_state,
                                      GraphExecutionContext &execution_context) {
   GELOGI("[%s] Node is ready for execution", task_context->GetNodeName());
   RECORD_EXECUTION_EVENT(&execution_context, task_context->GetNodeName(), "Start");
-  auto cb = std::shared_ptr<NodeDoneCallback>(new(std::nothrow) NodeDoneCallback(&execution_context, task_context));
-  GE_CHECK_NOTNULL(cb);
-  auto callback = [task_context, cb]() {
-    auto ret = cb->OnNodeDone();
-    if (ret != SUCCESS) {
-      task_context->OnError(ret);
-    }
-  };
-
+  std::function<void()> callback = nullptr;
+  GE_CHK_STATUS_RET_NOLOG(InitCallback(task_context, execution_context, callback));
   GE_CHK_STATUS_RET_NOLOG(DoExecuteAsync(node_state, *task_context, execution_context, callback));
   GE_CHK_STATUS_RET_NOLOG(PropagateOutputs(*node_state.GetNodeItem(), *task_context, execution_context));
   return SUCCESS;
 }
 
+Status ExecutionEngine::InitCallback(const std::shared_ptr<TaskContext> &task_context,
+                                     GraphExecutionContext &execution_context, std::function<void()> &callback) {
+  if (task_context->NeedCallback()) {
+    auto cb = std::shared_ptr<NodeDoneCallback>(new(std::nothrow) NodeDoneCallback(&execution_context, task_context));
+    GE_CHECK_NOTNULL(cb);
+    callback = [task_context, cb]() {
+      auto ret = cb->OnNodeDone();
+      if (ret != SUCCESS) {
+        task_context->OnError(ret);
+      }
+    };
+  }
+  return SUCCESS;
+}
+
 Status ExecutionEngine::DoExecuteAsync(NodeState &node_state,
                                        TaskContext &task_context,
                                        GraphExecutionContext &context,
@@ -385,7 +393,7 @@ Status ExecutionEngine::DoExecuteAsync(NodeState &node_state,
   const auto &task = node_state.GetKernelTask();
   if (task == nullptr) {
     GELOGE(INTERNAL_ERROR, "[Get][KernelTask] of [%s] is null.", node_state.GetName().c_str());
-    REPORT_INNER_ERROR("E19999", "GetKernelTask of %s is null.", node_state.GetName().c_str());
+    REPORT_INNER_ERROR("E19999", "GetKernelTask of %s failed.", node_state.GetName().c_str());
     return INTERNAL_ERROR;
   }
 

ge/hybrid/executor/worker/execution_engine.h

@@ -35,6 +35,8 @@ class ExecutionEngine {
                                TaskContext &task_context,
                                GraphExecutionContext &context,
                                const std::function<void()> &callback);
+  static Status InitCallback(const std::shared_ptr<TaskContext> &task_context,
+                             GraphExecutionContext &execution_context, std::function<void()> &callback);
 };
 }  // namespace hybrid
 }  // namespace ge

ge/hybrid/node_executor/task_context.cc

@@ -561,7 +561,8 @@ const DumpProperties &TaskContext::GetDumpProperties() const {
 }
 
 bool TaskContext::NeedCallback() {
-  return node_item_->has_observer || IsDumpEnabled() || execution_context_->profiling_level > 0;
+  return node_item_->has_observer || IsDumpEnabled() || execution_context_->profiling_level > 0 ||
+         !execution_context_->model->IsSingleOp();
 }
 
 Status TaskContext::Synchronize() {

tests/ut/ge/CMakeLists.txt

@@ -815,6 +815,7 @@ set(PROFILING_MNG_TEST_FILES
 set(HYBRID_TEST_FILES
     "hybrid/ge_hybrid_unittest.cc"
     "hybrid/known_node_executor_unittest.cc"
+	"hybrid/executor/worker/execution_engine_unittest.cc"
 )
 
 set(OTHERS_TEST_FILES

tests/ut/ge/hybrid/executor/worker/execution_engine_unittest.cc

@@ -0,0 +1,119 @@
+/**
+ * Copyright 2021 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * 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 <gtest/gtest.h>
+#include <gmock/gmock.h>
+#include <vector>
+#include "runtime/rt.h"
+
+#define protected public
+#define private public
+#include "hybrid/model/hybrid_model.h"
+#include "hybrid/node_executor/node_executor.h"
+#include "hybrid/executor/hybrid_execution_context.h"
+#include "hybrid/executor/hybrid_model_executor.h"
+#include "hybrid/executor/worker/execution_engine.h"
+#undef private
+#undef protected
+
+using namespace std;
+using namespace testing;
+using namespace ge;
+using namespace hybrid;
+
+
+class UtestExecutionEngine : public testing::Test {
+ protected:
+  void SetUp() {}
+
+  void TearDown() {
+  }
+};
+namespace {
+const int kIntBase = 10;
+}
+static ge::OpDescPtr CreateOpDesc(string name = "", string type = "") {
+  auto op_desc = std::make_shared<ge::OpDesc>(name, type);
+  op_desc->SetStreamId(0);
+  op_desc->SetId(0);
+  op_desc->SetWorkspace({});
+  op_desc->SetWorkspaceBytes({});
+  op_desc->SetInputOffset({});
+  op_desc->SetOutputOffset({});
+
+  ge::AttrUtils::SetStr(op_desc, ge::TVM_ATTR_NAME_MAGIC, "RT_DEV_BINARY_MAGIC_ELF_AIVEC");
+  bool support_dynamic = true;
+  ge::AttrUtils::GetBool(op_desc, "support_dynamicshape", support_dynamic);
+  return op_desc;
+}
+
+TEST_F(UtestExecutionEngine, ExecuteAsync_without_kernel_task) {
+  auto graph = make_shared<ComputeGraph>("graph");
+  OpDescPtr op_desc = CreateOpDesc("Add", "Add");
+  GeShape shape({2, 16});
+  GeTensorDesc tensor_desc(shape);
+  op_desc->AddInputDesc(tensor_desc);
+  op_desc->AddOutputDesc(tensor_desc);
+  auto node = graph->AddNode(op_desc);
+  std::unique_ptr<NodeItem> node_item;
+  NodeItem::Create(node, node_item);
+  ASSERT_TRUE(node_item != nullptr);
+  node_item->input_start = 0;
+  node_item->output_start = 0;
+
+  GraphExecutionContext execution_context;
+  execution_context.profiling_level = 1;
+  SubgraphContext subgraph_context(nullptr, &execution_context);
+
+  NodeState node_state(*node_item, &subgraph_context);
+  auto task_context = TaskContext::Create(&node_state, &execution_context, &subgraph_context);
+  auto shared_task_context = std::shared_ptr<TaskContext>(task_context.release());
+  node_state.SetTaskContext(shared_task_context);
+
+  ExecutionEngine execution_engine;
+  ASSERT_TRUE(node_state.GetTaskContext() != nullptr);
+  EXPECT_EQ(execution_engine.ExecuteAsync(node_state, node_state.GetTaskContext(), execution_context), INTERNAL_ERROR);
+}
+
+TEST_F(UtestExecutionEngine, ExecuteAsync_without_callback_and_kernel_task) {
+  auto graph = make_shared<ComputeGraph>("graph");
+  OpDescPtr op_desc = CreateOpDesc("Add", "Add");
+  GeShape shape({2, 16});
+  GeTensorDesc tensor_desc(shape);
+  op_desc->AddInputDesc(tensor_desc);
+  op_desc->AddOutputDesc(tensor_desc);
+  auto node = graph->AddNode(op_desc);
+  std::unique_ptr<NodeItem> node_item;
+  NodeItem::Create(node, node_item);
+  ASSERT_TRUE(node_item != nullptr);
+  node_item->input_start = 0;
+  node_item->output_start = 0;
+
+  GraphExecutionContext execution_context;
+  GeRootModelPtr ge_root_model = make_shared<GeRootModel>(graph);
+  HybridModel hybrid_model(ge_root_model);
+  execution_context.model = &hybrid_model;
+  SubgraphContext subgraph_context(nullptr, &execution_context);
+
+  NodeState node_state(*node_item, &subgraph_context);
+  auto task_context = TaskContext::Create(&node_state, &execution_context, &subgraph_context);
+  auto shared_task_context = std::shared_ptr<TaskContext>(task_context.release());
+  node_state.SetTaskContext(shared_task_context);
+
+  ExecutionEngine execution_engine;
+  ASSERT_TRUE(node_state.GetTaskContext() != nullptr);
+  EXPECT_EQ(execution_engine.ExecuteAsync(node_state, node_state.GetTaskContext(), execution_context), INTERNAL_ERROR);
+}