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.

spectral_ops.h 1.5 kB

4 years ago
12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152
  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 spectral_ops.h
  18. * \brief
  19. */
  20. #ifndef OPS_BUILT_IN_OP_PROTO_INC_SPECTRAL_OPS_H_
  21. #define OPS_BUILT_IN_OP_PROTO_INC_SPECTRAL_OPS_H_
  22. #include "graph/operator.h"
  23. #include "graph/operator_reg.h"
  24. namespace ge {
  25. /**
  26. *@brief Real-valued fast Fourier transform . \n
  27. *@par Inputs:
  28. *@li input: A float32 tensor.
  29. *@li fft_length: An int32 tensor of shape [1]. The FFT length . \n
  30. *@par Outputs:
  31. *@li y: A complex64 tensor of the same rank as `input`. The inner-most
  32. dimension of `input` is replaced with the `fft_length / 2 + 1` unique
  33. frequency components of its 1D Fourier transform . \n
  34. *@par Third-party framework compatibility
  35. * Compatible with TensorFlow RFFT operator.
  36. */
  37. REG_OP(RFFT)
  38. .INPUT(input, TensorType({DT_FLOAT}))
  39. .INPUT(fft_length, TensorType({DT_INT32}))
  40. .OUTPUT(y, TensorType({DT_COMPLEX64}))
  41. .OP_END_FACTORY_REG(RFFT)
  42. } // namespace ge
  43. #endif // OPS_BUILT_IN_OP_PROTO_INC_SPECTRAL_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