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.

op_kernel_bin.h 1.5 kB

5 years ago
1234567891011121314151617181920212223242526272829303132333435363738394041424344454647
  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_GRAPH_OP_KERNEL_BIN_H_
  17. #define INC_GRAPH_OP_KERNEL_BIN_H_
  18. #include <memory>
  19. #include <string>
  20. #include <utility>
  21. #include <vector>
  22. namespace ge {
  23. class OpKernelBin {
  24. public:
  25. OpKernelBin(std::string name, std::vector<char> &&data) : name_(std::move(name)), data_(std::move(data)) {}
  26. ~OpKernelBin() = default;
  27. const std::string &GetName() const { return name_; }
  28. const uint8_t *GetBinData() const { return (const uint8_t *)data_.data(); }
  29. size_t GetBinDataSize() const { return data_.size(); }
  30. OpKernelBin(const OpKernelBin &) = delete;
  31. const OpKernelBin &operator=(const OpKernelBin &) = delete;
  32. private:
  33. std::string name_;
  34. std::vector<char> data_;
  35. };
  36. using OpKernelBinPtr = std::shared_ptr<OpKernelBin>;
  37. const char *const OP_EXTATTR_NAME_TBE_KERNEL = "tbeKernel";
  38. } // namespace ge
  39. #endif // INC_GRAPH_OP_KERNEL_BIN_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

Contributors (1)