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.

ge_task_info.h 1.9 kB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576
  1. /**
  2. * Copyright 2020 Huawei Technologies Co., Ltd
  3. * Licensed under the Apache License, Version 2.0 (the "License");
  4. * you may not use this file except in compliance with the License.
  5. * You may obtain a copy of the License at
  6. * http://www.apache.org/licenses/LICENSE-2.0
  7. * Unless required by applicable law or agreed to in writing, software
  8. * distributed under the License is distributed on an "AS IS" BASIS,
  9. * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  10. * See the License for the specific language governing permissions and
  11. * limitations under the License.
  12. */
  13. #ifndef INC_COMMON_OPSKERNEL_GE_TASK_INFO_H_
  14. #define INC_COMMON_OPSKERNEL_GE_TASK_INFO_H_
  15. #include <runtime/rt.h>
  16. #include <stdint.h>
  17. #include <string>
  18. #include <vector>
  19. using std::string;
  20. namespace ge {
  21. // when need to eliminate GETaskKernelHcclInfo, so not need DAVINCI_TRAIN/DAVINCI_CLOUD
  22. struct GETaskKernelHcclInfo {
  23. string input_name;
  24. string hccl_type;
  25. void *inputDataAddr;
  26. void *outputDataAddr;
  27. void *workSpaceAddr;
  28. int32_t count;
  29. int32_t dataType;
  30. int32_t opType;
  31. int64_t rootId;
  32. uint64_t workSpaceMemSize;
  33. std::vector<int64_t> dims;
  34. std::vector<rtStream_t> hcclStreamList;
  35. };
  36. struct GETaskInfo {
  37. uint32_t id;
  38. uint16_t type;
  39. uint32_t streamID;
  40. void *stream; // rtKernelLaunch input argument
  41. void *event;
  42. void *privateDef;
  43. uint32_t privateDefLen;
  44. void *opsKernelStorePtr;
  45. std::vector<GETaskKernelHcclInfo> kernelHcclInfo;
  46. };
  47. struct HcomOpertion {
  48. std::string hcclType;
  49. void *inputPtr;
  50. void *outputPtr;
  51. uint64_t count;
  52. int32_t dataType;
  53. int32_t opType;
  54. int32_t root;
  55. };
  56. struct HcomRemoteAccessAddrInfo
  57. {
  58. uint32_t remotetRankID;
  59. uint64_t remoteAddr; // host embedding table address
  60. uint64_t localAddr; // device HBM address
  61. uint64_t length; // memory Length in Bytes
  62. };
  63. } // namespace ge
  64. #endif // INC_COMMON_OPSKERNEL_GE_TASK_INFO_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