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.

taskdown_api.h 1.6 kB

3 years ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354
  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 TASKDOWN_API_H_
  17. #define TASKDOWN_API_H_
  18. #include <map>
  19. #include <vector>
  20. #include "cce/cce.h"
  21. #include "l2fusion_struct.hpp"
  22. #include "taskdown_common.hpp"
  23. namespace cce {
  24. #define CC_FUSION_OP_MAX 32
  25. typedef struct tagOpAddrsInfo {
  26. void *addrPos;
  27. uintptr_t addrData;
  28. } ccOpAddrsInfo;
  29. #ifdef __cplusplus
  30. extern "C" {
  31. #endif
  32. ccStatus_t ccUpdateKernelArgs(ccOpContext &opContext, uint64_t dataBaseAddr, uint64_t weightBaseAddr,
  33. uint64_t variableBaseAddr, void *argsAddr, uint64_t argsSize, void *l2ctrlAddr);
  34. #ifdef __cplusplus
  35. }
  36. #endif
  37. ccStatus_t ccGetKernelArgsAddrs(ccOpContext &opContext, void *argsAddr, uint64_t argsSize, void *l2ctrlAddr,
  38. std::vector<ccOpAddrsInfo> &opAddrsInfo);
  39. ccStatus_t ccSetKernelArgs(std::vector<ccOpAddrsInfo> &dateInfo);
  40. ccStatus_t ccGetKernelTypeByOpId(uint32_t opId, ccKernelType &kernelType);
  41. } // namespace cce
  42. #endif // TASKDOWN_API_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