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.

runtime_model.h 3.4 kB

5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485
  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_GE_RUNTIME_RUNTIME_MODEL_H_
  17. #define GE_GE_RUNTIME_RUNTIME_MODEL_H_
  18. #include <map>
  19. #include <memory>
  20. #include <string>
  21. #include <vector>
  22. #include "ge_runtime/davinci_model.h"
  23. #include "common/ge_types.h"
  24. #include "runtime/base.h"
  25. #include "runtime/rt_model.h"
  26. namespace ge {
  27. namespace model_runner {
  28. class Task;
  29. class RuntimeModel {
  30. public:
  31. RuntimeModel() = default;
  32. ~RuntimeModel();
  33. bool Load(uint32_t device_id, uint64_t session_id, std::shared_ptr<DavinciModel> &davinci_model);
  34. const std::vector<uint32_t> &GetTaskIdList() const;
  35. bool Run();
  36. bool CopyInputData(const InputData &input_data);
  37. bool GetInputOutputDescInfo(bool zero_copy, std::vector<InputOutputDescInfo> *input_desc,
  38. std::vector<InputOutputDescInfo> *output_desc, std::vector<uint32_t> *input_format,
  39. std::vector<uint32_t> *output_format);
  40. private:
  41. bool InitResource(std::shared_ptr<DavinciModel> &davinci_model);
  42. void GenerateTask(uint32_t device_id, uint64_t session_id, std::shared_ptr<DavinciModel> &davinci_model);
  43. bool LoadTask();
  44. bool InitStream(std::shared_ptr<DavinciModel> &davinci_model);
  45. bool InitEvent(uint32_t event_num);
  46. bool InitLabel(uint32_t batch_num);
  47. bool InitDataInfo(std::shared_ptr<DavinciModel> &davinci_model);
  48. bool InitOutputInfo(std::shared_ptr<DavinciModel> &davinci_model);
  49. bool InitConstantInfo(std::shared_ptr<DavinciModel> &davinci_model);
  50. void RtModelUnbindStream() noexcept;
  51. void RtStreamDestory() noexcept;
  52. void RtModelDestory() noexcept;
  53. void RtLabelDestory() noexcept;
  54. void RtEventDestory() noexcept;
  55. bool CopyInputDataToModel(const std::vector<DataBuffer> &data, const std::shared_ptr<OpInfo> &data_info);
  56. bool CopyHostData(const std::vector<DataBuffer> &data, const std::shared_ptr<OpInfo> &data_info) const;
  57. bool CopyTransData(const std::vector<DataBuffer> &data, const std::shared_ptr<OpInfo> &data_info);
  58. bool GetInputDescInfo(std::vector<InputOutputDescInfo> *input_desc, std::vector<uint32_t> *formats);
  59. bool GetOutputDescInfo(std::vector<InputOutputDescInfo> *output_desc, std::vector<uint32_t> *formats);
  60. void CreateOutput(uint32_t index, const OpInfo &op_info, InputOutputDescInfo *output, uint32_t *format);
  61. rtModel_t rt_model_handle_{};
  62. rtStream_t rt_model_stream_{};
  63. std::vector<rtStream_t> stream_list_{};
  64. std::vector<rtLabel_t> label_list_{};
  65. std::vector<rtEvent_t> event_list_{};
  66. std::vector<std::shared_ptr<Task>> task_list_{};
  67. std::vector<std::shared_ptr<OpInfo>> data_info_list_{};
  68. std::vector<std::shared_ptr<OpInfo>> output_info_list_{};
  69. std::vector<std::shared_ptr<OpInfo>> constant_info_list_{};
  70. std::vector<uint32_t> task_id_list_{};
  71. };
  72. } // namespace model_runner
  73. } // namespace ge
  74. #endif // GE_GE_RUNTIME_RUNTIME_MODEL_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.