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.

execution_engine.h 2.4 kB

5 years ago
5 years ago
5 years ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
  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_EXECUTOR_EXECUTION_ENGINE_H_
  17. #define GE_HYBRID_EXECUTOR_EXECUTOR_EXECUTION_ENGINE_H_
  18. #include "hybrid/executor/hybrid_execution_context.h"
  19. #include "hybrid/node_executor/task_context.h"
  20. #include "common/dump/dump_op.h"
  21. namespace ge {
  22. namespace hybrid {
  23. class NodeDoneCallback {
  24. public:
  25. NodeDoneCallback(GraphExecutionContext *graph_context, std::shared_ptr<TaskContext> task_context);
  26. ~NodeDoneCallback() = default;
  27. Status OnNodeDone();
  28. private:
  29. Status PrepareConstInputs(const NodeItem &node_item);
  30. Status DumpDynamicNode();
  31. Status ProfilingReport();
  32. Status SaveDumpOpInfo();
  33. Status GetTaskDescInfo(const NodePtr node, const HybridModel *model,
  34. std::vector<TaskDescInfo> &task_desc_info);
  35. GraphExecutionContext *graph_context_;
  36. std::shared_ptr<TaskContext> context_;
  37. DumpOp dump_op_;
  38. };
  39. class ExecutionEngine {
  40. public:
  41. static Status ExecuteAsync(NodeState &node_state,
  42. const std::shared_ptr<TaskContext> &task_context,
  43. GraphExecutionContext &execution_context,
  44. const std::function<void()> &callback);
  45. private:
  46. static Status ValidateInputTensors(const NodeState &node_state, const TaskContext &task_context);
  47. static Status PropagateOutputs(const NodeItem &node_item, TaskContext &task_context, GraphExecutionContext &context);
  48. static Status DoExecuteAsync(NodeState &node_state,
  49. TaskContext &task_context,
  50. GraphExecutionContext &context,
  51. const std::function<void()> &callback);
  52. };
  53. } // namespace hybrid
  54. } // namespace ge
  55. #endif // GE_HYBRID_EXECUTOR_EXECUTOR_EXECUTION_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