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.

identity_kernel.cc 2.4 kB

4 years ago
1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465
  1. /**
  2. * Copyright 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. #include "identity_kernel.h"
  17. #include "inc/kernel_factory.h"
  18. #include "framework/common/types.h"
  19. namespace {
  20. constexpr uint32_t kInputDescIndex = 0;
  21. constexpr uint32_t kOutputDescIndex = 0;
  22. } // namespace
  23. namespace ge {
  24. Status IdentityKernel::Compute(const ge::OpDescPtr op_desc, const std::vector<ge::ConstGeTensorPtr> &input,
  25. std::vector<ge::GeTensorPtr> &v_output) {
  26. if (op_desc == nullptr) {
  27. GELOGE(PARAM_INVALID, "IdentityKernel op_desc is null.");
  28. return NOT_CHANGED;
  29. }
  30. if (input.empty()) {
  31. GELOGE(PARAM_INVALID, "Node [%s] inputs is empty.", op_desc->GetName().c_str());
  32. return NOT_CHANGED;
  33. }
  34. if (op_desc->GetOutputsSize() < 1) {
  35. GELOGE(PARAM_INVALID, "Node [%s] output is empty.", op_desc->GetName().c_str());
  36. return NOT_CHANGED;
  37. }
  38. GELOGD("IdentityKernel in: node[%s]", op_desc->GetName().c_str());
  39. auto out_tensor_desc = op_desc->GetOutputDesc(kOutputDescIndex);
  40. GeTensorPtr output_ptr = MakeShared<ge::GeTensor>(out_tensor_desc);
  41. if (output_ptr == nullptr) {
  42. GELOGE(OUT_OF_MEMORY, "Node [%s] make shared failed.", op_desc->GetName().c_str());
  43. return OUT_OF_MEMORY;
  44. }
  45. auto input_tensor_ptr = input.at(kInputDescIndex);
  46. if (input_tensor_ptr == nullptr) {
  47. GELOGE(PARAM_INVALID, "Node [%s] get input failed.", op_desc->GetName().c_str());
  48. return NOT_CHANGED;
  49. }
  50. if (output_ptr->SetData(input_tensor_ptr->GetData()) != GRAPH_SUCCESS) {
  51. GELOGW("Compute: SetData failed");
  52. return NOT_CHANGED;
  53. }
  54. v_output.emplace_back(output_ptr);
  55. GELOGD("IdentityKernel success: node[%s]", op_desc->GetName().c_str());
  56. return SUCCESS;
  57. }
  58. REGISTER_KERNEL(IDENTITY, IdentityKernel);
  59. REGISTER_KERNEL(PLACEHOLDERWITHDEFAULT, IdentityKernel);
  60. } // namespace ge

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