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.

model_runner.h 2.3 kB

5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768
  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 INC_FRAMEWORK_GE_RUNTIME_MODEL_RUNNER_H_
  17. #define INC_FRAMEWORK_GE_RUNTIME_MODEL_RUNNER_H_
  18. #include <memory>
  19. #include <unordered_map>
  20. #include <vector>
  21. #include "common/ge_inner_error_codes.h"
  22. #include "common/ge_types.h"
  23. #include "ge_runtime/davinci_model.h"
  24. namespace ge {
  25. namespace model_runner {
  26. class RuntimeModel;
  27. using RuntimeInfo = std::tuple<uint32_t, uint32_t, void *>;
  28. class ModelRunner {
  29. public:
  30. static ModelRunner &Instance();
  31. bool LoadDavinciModel(uint32_t device_id, uint64_t session_id, uint32_t model_id,
  32. std::shared_ptr<DavinciModel> davinci_model, std::shared_ptr<ModelListener> listener);
  33. bool DistributeTask(uint32_t model_id);
  34. bool LoadModelComplete(uint32_t model_id);
  35. const std::vector<uint32_t> &GetTaskIdList(uint32_t model_id) const;
  36. const std::vector<uint32_t> &GetStreamIdList(uint32_t model_id) const;
  37. const std::map<std::string, std::shared_ptr<RuntimeInfo>> &GetRuntimeInfoMap(uint32_t model_id) const;
  38. void *GetModelHandle(uint32_t model_id) const;
  39. bool UnloadModel(uint32_t model_id);
  40. bool RunModel(uint32_t model_id, const InputData &input_data, OutputData *output_data);
  41. bool GetInputOutputDescInfo(uint32_t model_id, bool zero_copy, std::vector<InputOutputDescInfo> *input_desc,
  42. std::vector<InputOutputDescInfo> *output_desc, std::vector<uint32_t> *input_format,
  43. std::vector<uint32_t> *output_format);
  44. private:
  45. ModelRunner() = default;
  46. ~ModelRunner() = default;
  47. std::unordered_map<uint32_t, std::shared_ptr<RuntimeModel>> runtime_models_;
  48. };
  49. } // namespace model_runner
  50. } // namespace ge
  51. #endif // INC_FRAMEWORK_GE_RUNTIME_MODEL_RUNNER_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