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.

global_step_insert_pass.cc 4.0 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899
  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. #include "graph/passes/global_step_insert_pass.h"
  17. #include <memory>
  18. #include <string>
  19. #include <vector>
  20. #include "framework/common/debug/ge_log.h"
  21. #include "framework/common/util.h"
  22. #include "graph/debug/ge_attr_define.h"
  23. #include "common/ge/ge_util.h"
  24. #include "graph/manager/graph_var_manager.h"
  25. #include "graph/passes/pass_utils.h"
  26. namespace ge {
  27. NodePtr GlobalStepInsertPass::InsertOp(ComputeGraphPtr &compute_graph, const string &node_type, const string &node_name,
  28. const std::vector<GeTensorDesc> &input_list,
  29. const std::vector<GeTensorDesc> &output_list) {
  30. OpDescPtr op_desc = MakeShared<OpDesc>(node_name, node_type);
  31. GE_IF_BOOL_EXEC(op_desc == nullptr, GELOGE(FAILED, "Make OpDesc failed"); return nullptr);
  32. for (auto &input_desc : input_list) {
  33. graphStatus graph_status = op_desc->AddInputDesc(input_desc);
  34. if (graph_status != GRAPH_SUCCESS) {
  35. GELOGE(FAILED, "Add node:%s intput desc failed, error=%u.", node_name.c_str(), graph_status);
  36. return nullptr;
  37. }
  38. }
  39. for (auto &output_desc : output_list) {
  40. graphStatus graph_status = op_desc->AddOutputDesc(output_desc);
  41. if (graph_status != GRAPH_SUCCESS) {
  42. GELOGE(FAILED, "Add node:%s output desc failed, error=%u.", node_name.c_str(), graph_status);
  43. return nullptr;
  44. }
  45. }
  46. GE_IF_BOOL_EXEC(compute_graph == nullptr, GELOGE(FAILED, "compute_graph is nullptr"); return nullptr);
  47. NodePtr node = compute_graph->AddNode(op_desc);
  48. GE_IF_BOOL_EXEC(node == nullptr,
  49. GELOGE(FAILED, "add node failed, name:%s, type:%s.", node_name.c_str(), node_type.c_str());
  50. return nullptr);
  51. GELOGI("Insert op success, name:%s, type:%s.", node_name.c_str(), node_type.c_str());
  52. return node;
  53. }
  54. Status GlobalStepInsertPass::Run(ComputeGraphPtr compute_graph) {
  55. NodePtr output_node = compute_graph->FindFirstNodeMatchType(NETOUTPUT);
  56. if (output_node == nullptr) {
  57. GELOGD("Node type %s can't be found in graph %u", NETOUTPUT, compute_graph->GetGraphID());
  58. return SUCCESS;
  59. }
  60. if (compute_graph->GetParentGraph() != nullptr) {
  61. GELOGD("Subgraph %s no need global step variable.", compute_graph->GetName().c_str());
  62. return SUCCESS;
  63. }
  64. NodePtr exist_node = compute_graph->FindNode(NODE_NAME_GLOBAL_STEP);
  65. if (exist_node != nullptr) {
  66. GELOGD("Node %s already exist, no need add.", NODE_NAME_GLOBAL_STEP.c_str());
  67. return SUCCESS;
  68. }
  69. // set global step tensor desc
  70. GeTensorDesc tensor_desc(GeShape({1}), FORMAT_ND, DT_UINT64);
  71. std::vector<GeTensorDesc> input_desc_list = {};
  72. std::vector<GeTensorDesc> output_desc_list = {tensor_desc};
  73. NodePtr global_step = InsertOp(compute_graph, VARIABLE, NODE_NAME_GLOBAL_STEP, input_desc_list, output_desc_list);
  74. if (global_step == nullptr) {
  75. GELOGE(FAILED, "Add global_step node failed, global_step is null.");
  76. return FAILED;
  77. }
  78. // add ctrl edges
  79. graphStatus add_ret = GraphUtils::AddEdge(global_step->GetOutControlAnchor(), output_node->GetInControlAnchor());
  80. if (add_ret != GRAPH_SUCCESS) {
  81. GELOGE(FAILED, "Add global_step to netoutput edge failed, add_ret=%u.", add_ret);
  82. return FAILED;
  83. }
  84. GELOGD("Add global_step to netoutput edge in graph %u success", compute_graph->GetGraphID());
  85. return SUCCESS;
  86. }
  87. } // 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