You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.

known_node_executor_unittest.cc 3.4 kB

4 years ago
4 years ago
12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091
  1. /**
  2. * Copyright 2019-2021 Huawei Technologies Co., Ltd
  3. *
  4. * Licensed under the Apache License, Version 2.0 (the "License");
  5. * you may not use this file except in compliance with the License.
  6. * You may obtain a copy of the License at
  7. *
  8. * http://www.apache.org/licenses/LICENSE-2.0
  9. *
  10. * Unless required by applicable law or agreed to in writing, software
  11. * distributed under the License is distributed on an "AS IS" BASIS,
  12. * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13. * See the License for the specific language governing permissions and
  14. * limitations under the License.
  15. */
  16. #include <gtest/gtest.h>
  17. #include <gmock/gmock.h>
  18. #include <vector>
  19. #include <memory>
  20. #define protected public
  21. #define private public
  22. #include "hybrid/node_executor/compiledsubgraph/known_node_executor.h"
  23. #include "common/dump/dump_manager.h"
  24. #undef private
  25. #undef protected
  26. #include "graph/manager/graph_mem_allocator.h"
  27. #include "../graph/passes/graph_builder_utils.h"
  28. using namespace std;
  29. using namespace testing;
  30. using namespace ge;
  31. using namespace hybrid;
  32. class UnknownNodeExecutorTest : public testing::Test {
  33. protected:
  34. void SetUp() {}
  35. void TearDown() {}
  36. };
  37. namespace {
  38. class KnownNodeTaskMock : public KnownNodeTask {
  39. public:
  40. KnownNodeTaskMock(std::shared_ptr<DavinciModel> davinci_model): KnownNodeTask(davinci_model) {};
  41. ~KnownNodeTaskMock() override = default;
  42. MOCK_METHOD2(DoInitDavinciModel, Status(void *, size_t));
  43. };
  44. }
  45. TEST_F(UnknownNodeExecutorTest, test_init_davinci_model) {
  46. auto davinci_model = std::make_shared<DavinciModel>(0, nullptr);
  47. davinci_model->SetDeviceId(0);
  48. davinci_model->SetKnownNode(true);
  49. auto ge_model = make_shared<GeModel>();
  50. AttrUtils::SetInt(ge_model, ATTR_MODEL_VAR_SIZE, 0);
  51. AttrUtils::SetInt(ge_model, ATTR_MODEL_MEMORY_SIZE, 1024);
  52. davinci_model->Assign(ge_model);
  53. HybridModel model(nullptr);
  54. KnownNodeTaskMock mock(davinci_model);
  55. DumpProperties dump_properties;
  56. dump_properties.enable_dump_ = "1";
  57. DumpManager::GetInstance().AddDumpProperties(model.GetSessionId(), dump_properties);
  58. EXPECT_CALL(mock, DoInitDavinciModel).WillRepeatedly(::testing::Return(SUCCESS));
  59. ASSERT_EQ(mock.InitDavinciModel(model, model.GetModelWeight("subgraph")), SUCCESS);
  60. int32_t buffer[8];
  61. model.weight_buffer_map_.emplace("subgraph", TensorBuffer::Create(buffer, sizeof(buffer)));
  62. ASSERT_EQ(mock.InitDavinciModel(model, model.GetModelWeight("subgraph")), SUCCESS);
  63. }
  64. TEST_F(UnknownNodeExecutorTest, TestParseAttrForAllocatingOutputs) {
  65. ut::GraphBuilder builder("test-graph");
  66. auto data_node = builder.AddNode("Data0", DATA, 1, 1);
  67. auto netoutput_node = builder.AddNode("NodeOutput", NETOUTPUT, 2, 2);
  68. builder.AddDataEdge(data_node, 0, netoutput_node, 0);
  69. auto const_node = builder.AddNode("Const0", CONSTANT, 0, 1);
  70. builder.AddDataEdge(const_node, 0, netoutput_node, 1);
  71. auto graph = builder.GetGraph();
  72. ut::GraphBuilder builder2("root-graph");
  73. auto partitioned_call = builder2.AddNode("Node0", PARTITIONEDCALL, 1, 2);
  74. NodeItem node_item(partitioned_call);
  75. ASSERT_EQ(KnownNodeExecutor::ParseAttrForAllocatingOutputs(node_item, *graph), SUCCESS);
  76. ASSERT_EQ(node_item.ref_outputs.size(), 1);
  77. ASSERT_EQ(node_item.ref_outputs[1], const_node);
  78. ASSERT_EQ(node_item.reuse_inputs.size(), 1);
  79. ASSERT_EQ(node_item.reuse_inputs[0], 0);
  80. }

图引擎模块(GE)是MindSpore的一个子模块,其代码由C++实现,位于前端模块ME和底层硬件之间,起到承接作用。图引擎模块以ME下发的图作为输入,然后进行一系列的深度图优化操作,最后输出一张可以在底层硬件上高效运行的图。GE针对昇腾AI处理器的硬件结构特点,做了特定的优化工作,以此来充分发挥出昇腾AI处理器的强大算力。在进行模型训练/推理时,GE会被自动调用而用户并不感知。GE主要由GE API和GE Core两部分组成,详细的架构图如下所示