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.

hybrid_model_executor.h 2.1 kB

4 years ago
4 years ago
1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465
  1. /**
  2. * Copyright 2019-2020 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. #ifndef GE_HYBRID_EXECUTOR_HYBRID_MODEL_EXECUTOR_H_
  17. #define GE_HYBRID_EXECUTOR_HYBRID_MODEL_EXECUTOR_H_
  18. #include "common/thread_pool.h"
  19. #include "graph/load/model_manager/data_inputer.h"
  20. #include "hybrid/executor/hybrid_execution_context.h"
  21. #include "hybrid/executor/rt_callback_manager.h"
  22. #include "hybrid/executor/subgraph_executor.h"
  23. namespace ge {
  24. namespace hybrid {
  25. class HybridModelExecutor {
  26. public:
  27. struct ExecuteArgs {
  28. std::vector<TensorValue> inputs;
  29. std::vector<ConstGeTensorDescPtr> input_desc;
  30. std::vector<TensorValue> outputs;
  31. std::vector<ConstGeTensorDescPtr> output_desc;
  32. bool is_eos = false;
  33. int num_loops = 10;
  34. };
  35. HybridModelExecutor(HybridModel *model, uint32_t device_id, rtStream_t stream);
  36. ~HybridModelExecutor();
  37. Status Init();
  38. const GraphExecutionContext* GetContext() const {
  39. return &context_;
  40. }
  41. Status Execute(ExecuteArgs &args);
  42. private:
  43. Status ExecuteGraphInternal(ExecuteArgs &args);
  44. Status Cleanup();
  45. Status InitExecutionContext();
  46. static Status ResetExecutionContext(GraphExecutionContext &context);
  47. static Status CheckInputShapeByShapeRange(const GraphItem *graph_item, HybridModelExecutor::ExecuteArgs &args);
  48. HybridModel *model_;
  49. uint32_t device_id_;
  50. rtStream_t stream_;
  51. GraphExecutionContext context_;
  52. std::unique_ptr<SubgraphExecutor> root_graph_executor_;
  53. };
  54. } // namespace hybrid
  55. } // namespace ge
  56. #endif // GE_HYBRID_EXECUTOR_HYBRID_MODEL_EXECUTOR_H_

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