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.

set_input_output_offset_pass.h 1.5 kB

12345678910111213141516171819202122232425262728293031323334353637
  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_PASSES_SET_INPUT_OUTPUT_OFFSET_PASS_H_
  17. #define GE_GRAPH_PASSES_SET_INPUT_OUTPUT_OFFSET_PASS_H_
  18. #include "inc/graph_pass.h"
  19. namespace ge {
  20. class SetInputOutputOffsetPass : public GraphPass {
  21. public:
  22. Status Run(ComputeGraphPtr graph) override;
  23. private:
  24. Status SetInputOffset(const NodePtr &node, const vector<int> &connect_input);
  25. Status SetOutputOffset(const NodePtr &node, const vector<int> &connect_output);
  26. Status SetInputOffsetForFusion(const std::vector<int64_t> &memory_type, const ge::NodePtr &node);
  27. Status SetInputOffsetForHcom(const NodePtr &node, const vector<int> &connect_input);
  28. Status SetOutputOffsetForConcat(const NodePtr &node);
  29. Status SetOutputOffsetForHcom(const NodePtr &node, const vector<int> &connect_output);
  30. bool CheckBufferFusion(const NodePtr &node);
  31. };
  32. } // namespace ge
  33. #endif // GE_GRAPH_PASSES_SET_INPUT_OUTPUT_OFFSET_PASS_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