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.

internal_ops.h 2.4 kB

4 years ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384
  1. /**
  2. * Copyright 2019 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. /*!
  17. * \file internal_ops.h
  18. * \brief
  19. */
  20. #ifndef OPS_BUILT_IN_OP_PROTO_INC_INTERNAL_OPS_H_
  21. #define OPS_BUILT_IN_OP_PROTO_INC_INTERNAL_OPS_H_
  22. #include "graph/operator_reg.h"
  23. #include "graph/operator.h"
  24. namespace ge {
  25. /**
  26. *@brief aicpu assit help op for auxiliary matrix generation. \n
  27. *@par Inputs:
  28. *The input is dynamic for attribute func_name \n
  29. *@par Attributes:
  30. *@li func_name:An required param, for example "topkv2". \n
  31. *@par Outputs:
  32. *The output is dynamic for attribute func_name.
  33. */
  34. REG_OP(AssistHelp)
  35. .DYNAMIC_INPUT(x, TensorType({ DT_FLOAT, DT_FLOAT16, DT_INT8, DT_INT16, DT_UINT16,
  36. DT_UINT8, DT_INT32, DT_INT64, DT_UINT32, DT_UINT64, DT_BOOL, DT_DOUBLE }))
  37. .DYNAMIC_OUTPUT(y, TensorType({ DT_FLOAT, DT_FLOAT16, DT_INT8, DT_INT16, DT_UINT16,
  38. DT_UINT8, DT_INT32, DT_INT64, DT_UINT32, DT_UINT64, DT_BOOL, DT_DOUBLE}))
  39. . REQUIRED_ATTR (func_name, String)
  40. . OP_END_FACTORY_REG(AssistHelp)
  41. /**
  42. *@brief aicpu cache help for lhisi cache flush. \n
  43. *@par Inputs:
  44. *The input is dynamic for attribute func_name \n
  45. *@par Outputs:
  46. *The output is dynamic for attribute func_name.
  47. */
  48. REG_OP(CacheUpdate)
  49. .INPUT(x, TensorType::BasicType())
  50. .OUTPUT(x, TensorType::BasicType())
  51. .OP_END_FACTORY_REG(CacheUpdate)
  52. /**
  53. *@brief transfer data from L1 buffer to DDR or DDR to L1. \n
  54. *@par Inputs:
  55. *The input is dynamic for attribute func_name \n
  56. *@par Outputs:
  57. *The output is dynamic for attribute func_name.
  58. *@par Restrictions:
  59. *Warning: THIS FUNCTION IS EXPERIMENTAL. Please do not use.
  60. */
  61. REG_OP(InternalDataMove)
  62. .INPUT(x, TensorType::ALL())
  63. .OUTPUT(y, TensorType::ALL())
  64. .REQUIRED_ATTR(src_buf, String)
  65. .REQUIRED_ATTR(dst_buf, String)
  66. .OP_END_FACTORY_REG(InternalDataMove)
  67. } // namespace ge
  68. #endif // OPS_BUILT_IN_OP_PROTO_INC_INTERNAL_OPS_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