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.

graph_loader.h 2.9 kB

5 years ago
5 years ago
5 years ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778
  1. /**
  2. * Copyright 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_GRAPH_LOAD_GRAPH_LOADER_H_
  17. #define GE_GRAPH_LOAD_GRAPH_LOADER_H_
  18. #include <memory>
  19. #include <string>
  20. #include <vector>
  21. #include "common/debug/log.h"
  22. #include "common/fmk_types.h"
  23. #include "common/ge_types.h"
  24. #include "graph/compute_graph.h"
  25. #include "graph/manager/graph_manager_utils.h"
  26. #include "graph/model.h"
  27. #include "runtime/mem.h"
  28. namespace ge {
  29. class GraphLoader {
  30. public:
  31. GraphLoader();
  32. virtual ~GraphLoader();
  33. GraphLoader(const GraphLoader &in) = delete;
  34. GraphLoader &operator=(const GraphLoader &in) = delete;
  35. static Status UnloadModel(uint32_t model_id);
  36. static Status GetMaxUsedMemory(uint32_t model_id, uint64_t &max_size);
  37. static Status LoadModel(const ModelData &model_data, const std::shared_ptr<ModelListener> &listener,
  38. uint32_t &model_id);
  39. static Status LoadModelFromFile(const std::string &path, const std::string &key_path, int32_t priority,
  40. const std::shared_ptr<ModelListener> &listener, uint32_t &model_id);
  41. static Status CommandHandle(const Command &command);
  42. static Status GetMemoryInfo(int64_t &free);
  43. static Status LoadDataFromFile(const std::string &path, const std::string &key_path, int32_t priority,
  44. ModelData &model_data);
  45. static Status LoadModelFromData(uint32_t &model_id, const ModelData &model_data, void *dev_ptr, size_t mem_size,
  46. void *weight_ptr, size_t weight_size);
  47. static Status LoadModelWithQ(uint32_t &model_id, const ModelData &model_data,
  48. const std::vector<uint32_t> &input_queue_ids,
  49. const std::vector<uint32_t> &output_queue_ids);
  50. static Status ExecuteModel(uint32_t model_id, rtStream_t stream, bool async_mode, const InputData &input_data,
  51. OutputData &output_data);
  52. static Status DestroyAicpuKernel(uint64_t session_id, uint32_t model_id);
  53. static Status DestroyAicpuSessionForInfer(uint32_t model_id);
  54. static Status LoadModelOnline(uint32_t &model_id, const std::shared_ptr<ge::GeRootModel> &ge_root_model,
  55. const std::shared_ptr<ModelListener> &listener);
  56. };
  57. } // namespace ge
  58. #endif // GE_GRAPH_LOAD_GRAPH_LOADER_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