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.

shape_inference_engine.h 1.9 kB

4 years ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051
  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_HYBRID_EXECUTOR_INFERSHAPE_SHAPE_INFERENCE_ENGINE_H_
  17. #define GE_HYBRID_EXECUTOR_INFERSHAPE_SHAPE_INFERENCE_ENGINE_H_
  18. #include "hybrid/executor/hybrid_execution_context.h"
  19. #include "hybrid/executor/subgraph_context.h"
  20. #include <mutex>
  21. namespace ge {
  22. namespace hybrid {
  23. class ShapeInferenceEngine {
  24. public:
  25. ShapeInferenceEngine(GraphExecutionContext *execution_context, SubgraphContext *subgraph_context);
  26. ~ShapeInferenceEngine() = default;
  27. Status InferShape(NodeState &node_state);
  28. Status InferShapeForSubgraph(const NodeItem &node_item, const FusedSubgraph &fused_subgraph);
  29. Status PropagateOutputShapes(NodeState &node_state);
  30. static Status CalcOutputTensorSizes(const NodeItem &node_item, bool fallback_with_range = false);
  31. private:
  32. static Status CanonicalizeShape(GeTensorDesc &tensor_desc, std::vector<int64_t> &shape, bool fallback_with_range);
  33. static Status CalcTensorSize(DataType data_type, const std::vector<int64_t> &shape, int64_t &tensor_size);
  34. static Status UpdatePeerNodeShape(const Node &node);
  35. Status AwaitDependentNodes(NodeState &node_state);
  36. GraphExecutionContext *execution_context_;
  37. SubgraphContext *subgraph_context_;
  38. std::mutex mu_;
  39. };
  40. } // namespace hybrid
  41. } // namespace ge
  42. #endif // GE_HYBRID_EXECUTOR_INFERSHAPE_SHAPE_INFERENCE_ENGINE_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