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.

range_vistor.h 1.5 kB

5 years ago
1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253
  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_RANGE_VISTOR_H_
  17. #define INC_GRAPH_RANGE_VISTOR_H_
  18. #include <vector>
  19. template <class E, class O>
  20. class RangeVistor {
  21. public:
  22. using Iterator = typename std::vector<E>::iterator;
  23. using ConstIterator = typename std::vector<E>::const_iterator;
  24. RangeVistor(O owner, const std::vector<E> &vs) : owner_(owner), elements_(vs) {}
  25. ~RangeVistor() {}
  26. Iterator begin() { return elements_.begin(); }
  27. Iterator end() { return elements_.end(); }
  28. ConstIterator begin() const { return elements_.begin(); }
  29. ConstIterator end() const { return elements_.end(); }
  30. std::size_t size() const { return elements_.size(); }
  31. bool empty() const { return elements_.empty(); }
  32. E &at(std::size_t index) { return elements_.at(index); }
  33. const E &at(std::size_t index) const { return elements_.at(index); }
  34. private:
  35. O owner_;
  36. std::vector<E> elements_;
  37. };
  38. #endif // INC_GRAPH_RANGE_VISTOR_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)