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.

merge_pass.cc 8.7 kB

5 years ago
5 years ago
5 years ago
4 years ago
5 years ago
5 years ago
4 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
4 years ago
5 years ago
5 years ago
5 years ago
4 years ago
5 years ago
4 years ago
5 years ago
4 years ago
5 years ago
4 years ago
5 years ago
4 years ago
5 years ago
4 years ago
5 years ago
5 years ago
5 years ago
4 years ago
5 years ago
4 years ago
5 years ago
4 years ago
5 years ago
5 years ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224
  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 "graph/passes/merge_pass.h"
  17. #include <memory>
  18. #include <string>
  19. #include <vector>
  20. #include "framework/common/debug/ge_log.h"
  21. #include "common/ge/ge_util.h"
  22. #include "graph/common/omg_util.h"
  23. #include "graph/debug/ge_attr_define.h"
  24. #include "graph/utils/graph_utils.h"
  25. #include "graph/passes/pass_utils.h"
  26. namespace ge {
  27. const int kValueIndexOutputIndex = 1;
  28. const size_t kCaseNoInput = 0;
  29. const size_t kCaseOneInput = 1;
  30. Status MergePass::Run(NodePtr &node) {
  31. GELOGD("MergePass running");
  32. if (node == nullptr) {
  33. REPORT_INNER_ERROR("E19999", "Param node is nullptr, check invalid");
  34. GELOGE(PARAM_INVALID, "param [node] must not be null.");
  35. return PARAM_INVALID;
  36. }
  37. std::string op_type;
  38. GE_CHK_STATUS_RET(GetOriginalType(node, op_type), "get original type failed");
  39. if (op_type != MERGE) {
  40. return SUCCESS;
  41. }
  42. if (node->GetAllOutDataAnchors().empty()) {
  43. REPORT_INNER_ERROR("E19999", "Param node:%s(%s) all data anchor size is 0, check invalid",
  44. node->GetName().c_str(), node->GetType().c_str());
  45. GELOGE(PARAM_INVALID, "[%s] Merge node output anchor is empty", node->GetName().c_str());
  46. return PARAM_INVALID;
  47. }
  48. const auto &in_data_nodes = node->GetInDataNodes();
  49. switch (in_data_nodes.size()) {
  50. case kCaseNoInput: {
  51. /// Case A: input_count = 0, the output of merge node is inactive as well
  52. /// In which case the output branch can be removed
  53. /// until another merge node is met
  54. std::vector<NodePtr> del_nodes;
  55. std::vector<NodePtr> end_nodes;
  56. Status ret = PassUtils::RemoveBranch(node, del_nodes, end_nodes);
  57. for (auto &end_node : end_nodes) {
  58. AddRePassNode(end_node);
  59. }
  60. for (const auto &delete_node : del_nodes) {
  61. AddNodeDeleted(delete_node);
  62. }
  63. return ret;
  64. }
  65. case kCaseOneInput: { // Case B: input_count = 1, the merge node can be optimized out
  66. std::vector<int> merge_io_map = {PassUtils::GetUniqueInDataAnchorIndex(node), -1};
  67. if (merge_io_map[0] != -1 && IsNeedChangeIndexToConstant(node)) {
  68. int index = merge_io_map[0];
  69. if (ChangeIndexToConstant(node, index) != SUCCESS) {
  70. GELOGE(FAILED, "[%s] Change value index to be Constant failed.", node->GetName().c_str());
  71. return FAILED;
  72. }
  73. }
  74. auto in_node = in_data_nodes.at(0);
  75. if (IsMergeInputNeedOptimized(in_node)) {
  76. if (IsolateAndDeleteNode(in_node, {0}) != SUCCESS) {
  77. REPORT_CALL_ERROR("E19999", "Isolate and delete node:%s(%s) failed",
  78. in_node->GetName().c_str(), in_node->GetType().c_str());
  79. GELOGE(FAILED, "Isolate and delete node %s failed.", in_node->GetName().c_str());
  80. return FAILED;
  81. }
  82. }
  83. return IsolateAndDeleteNode(node, merge_io_map);
  84. }
  85. default: {
  86. // Case C: input_count > 1, the merge node can not be optimized
  87. return SUCCESS;
  88. }
  89. }
  90. }
  91. bool MergePass::IsNeedChangeIndexToConstant(NodePtr &node) const {
  92. /// value_index is the index 1 output of the Merge
  93. /// value_index link to other node, change it to be Constant
  94. GE_IF_BOOL_EXEC(node == nullptr, GELOGW("Node is nullptr"); return false);
  95. auto out_anchor = node->GetOutDataAnchor(kValueIndexOutputIndex);
  96. GE_IF_BOOL_EXEC(out_anchor == nullptr, GELOGW("Out_anchor is nullptr"); return false);
  97. for (const auto &peer_in_anchor : out_anchor->GetPeerInDataAnchors()) {
  98. if (peer_in_anchor != nullptr && peer_in_anchor->GetOwnerNode() != nullptr) {
  99. GELOGI(
  100. "[%s] MergePass, value_index link to other node, "
  101. "change it to be Constant.",
  102. node->GetName().c_str());
  103. return true;
  104. }
  105. }
  106. return false;
  107. }
  108. Status MergePass::ChangeIndexToConstant(NodePtr &node, int &value_index) {
  109. GE_CHECK_NOTNULL(node);
  110. ComputeGraphPtr graph = node->GetOwnerComputeGraph();
  111. if (graph == nullptr) {
  112. REPORT_INNER_ERROR("E19999", "Owner graph of node:%s(%s) is nullptr, check invalid",
  113. node->GetName().c_str(), node->GetType().c_str());
  114. GELOGE(FAILED, "[%s] The owner graph must not be null.", node->GetName().c_str());
  115. return FAILED;
  116. }
  117. OpDescPtr constant_op_desc = nullptr;
  118. if (CreateConstByValue(node, value_index, constant_op_desc) != SUCCESS) {
  119. return FAILED;
  120. }
  121. NodePtr const_node = graph->AddNode(constant_op_desc);
  122. if (const_node == nullptr) {
  123. REPORT_CALL_ERROR("E19999", "Add node:%s(%s) to graph:%s failed",
  124. constant_op_desc->GetName().c_str(), constant_op_desc->GetType().c_str(),
  125. graph->GetName().c_str());
  126. return FAILED;
  127. }
  128. // Change peer in anchors from value_index to new Constant node
  129. if (GraphUtils::ReplaceNodeAnchors(const_node, node, {}, {1}) != GRAPH_SUCCESS) {
  130. REPORT_CALL_ERROR("E19999", "Replace node:%s(%s) by node:%s(%s) failed",
  131. node->GetName().c_str(), node->GetType().c_str(),
  132. const_node->GetName().c_str(), const_node->GetType().c_str());
  133. GELOGE(FAILED, "[%s] ReplaceNodeAnchors failed.", node->GetName().c_str());
  134. return FAILED;
  135. }
  136. auto out_control_anchor = node->GetOutControlAnchor();
  137. GE_CHECK_NOTNULL(out_control_anchor);
  138. // Add control anchor between Merge and Constant
  139. if (out_control_anchor->LinkTo(const_node->GetInControlAnchor()) != GRAPH_SUCCESS) {
  140. REPORT_CALL_ERROR("E19999", "Op:%s(%s) link control to op:%s(%s) failed",
  141. node->GetName().c_str(), node->GetType().c_str(),
  142. const_node->GetName().c_str(), const_node->GetType().c_str());
  143. return FAILED;
  144. }
  145. return SUCCESS;
  146. }
  147. Status MergePass::CreateConstByValue(NodePtr &node, int value_index, OpDescPtr &op_desc) {
  148. std::string constant_name = node->GetName() + "_value_index";
  149. // 1. create Constant OpDesc
  150. op_desc = MakeShared<OpDesc>(constant_name, CONSTANT);
  151. if (op_desc == nullptr) {
  152. REPORT_CALL_ERROR("E19999", "New OpDesc failed");
  153. GELOGE(FAILED, "[%s] Make shared of Constant op desc failed.", constant_name.c_str());
  154. return FAILED;
  155. }
  156. // 2. get OpDesc of output number one of Merge(value_index)
  157. OpDescPtr original_op_desc = node->GetOpDesc();
  158. if (original_op_desc == nullptr) {
  159. REPORT_INNER_ERROR("E19999", "OpDesc in node is nullptr, check invalid");
  160. GELOGE(FAILED, "[%s] Op desc must not be null.", constant_name.c_str());
  161. return FAILED;
  162. }
  163. GeTensorDesc original_out_tensor_desc = original_op_desc->GetOutputDesc(1);
  164. original_out_tensor_desc.SetDataType(DT_INT32);
  165. // 3. create attr value of Constant, is a tensor
  166. GeTensorPtr const_tensor_ptr =
  167. MakeShared<GeTensor>(original_out_tensor_desc, reinterpret_cast<uint8_t *>(&value_index), sizeof(int));
  168. if (const_tensor_ptr == nullptr) {
  169. REPORT_CALL_ERROR("E19999", "New GeTensor failed");
  170. GELOGE(FAILED, "[%s] Make shared of Constant tensor failed.", constant_name.c_str());
  171. return FAILED;
  172. }
  173. GE_IF_BOOL_EXEC(!AttrUtils::SetTensor(op_desc, ATTR_NAME_WEIGHTS, const_tensor_ptr),
  174. REPORT_CALL_ERROR("E19999", "Set Attr:%s to op:%s(%s) failed",
  175. ATTR_NAME_WEIGHTS.c_str(),
  176. op_desc->GetName().c_str(), op_desc->GetType().c_str());
  177. GELOGE(FAILED, "get ATTR_NAME_WEIGHTS failed"); return FAILED);
  178. // 4. set Constant output desc
  179. GE_CHK_GRAPH_STATUS_RET(op_desc->AddOutputDesc(original_out_tensor_desc), "add out put desc failed");
  180. return SUCCESS;
  181. }
  182. bool MergePass::IsMergeInputNeedOptimized(NodePtr &node) const {
  183. if (node == nullptr) {
  184. return false;
  185. }
  186. // node is not inserted by MergeInputMemcpyPass
  187. if ((node->GetType() != MEMCPYASYNC) && (node->GetType() != MEMCPYADDRASYNC)) {
  188. return false;
  189. }
  190. if (node->GetInDataNodes().size() != 1) {
  191. return false;
  192. }
  193. auto in_node = node->GetInDataNodes().at(0);
  194. if (in_node == nullptr) {
  195. return false;
  196. }
  197. // in_node may be global_step var
  198. if ((in_node->GetType() == VARIABLE) || (in_node->GetType() == VARIABLEV2)) {
  199. return false;
  200. }
  201. return true;
  202. }
  203. } // 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