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.

scope_pass_registry_impl.h 1.6 kB

12345678910111213141516171819202122232425262728293031323334353637383940
  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 REGISTER_SCOPE_SCOPE_REGISTRY_IMPL_H_
  17. #define REGISTER_SCOPE_SCOPE_REGISTRY_IMPL_H_
  18. #include "external/register/scope/scope_fusion_pass_register.h"
  19. #include <mutex>
  20. namespace ge {
  21. struct CreatePassFnPack;
  22. class ScopeFusionPassRegistry::ScopeFusionPassRegistryImpl {
  23. public:
  24. void RegisterScopeFusionPass(const std::string &pass_name, ScopeFusionPassRegistry::CreateFn create_fn,
  25. bool is_general);
  26. ScopeFusionPassRegistry::CreateFn GetCreateFn(const std::string &pass_name);
  27. std::unique_ptr<ScopeBasePass> CreateScopeFusionPass(const std::string &pass_name);
  28. std::vector<std::string> GetAllRegisteredPasses();
  29. bool SetPassEnableFlag(const std::string pass_name, const bool flag);
  30. private:
  31. std::mutex mu_;
  32. std::vector<std::string> pass_names_; // In the order of user registration
  33. std::map<std::string, CreatePassFnPack> create_fn_packs_;
  34. };
  35. } // namespace ge
  36. #endif // REGISTER_SCOPE_SCOPE_REGISTRY_IMPL_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