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.

kernel_fusion.cc 4.9 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125
  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. #include "kernel/kernel_fusion.h"
  17. #include <map>
  18. #include <string>
  19. #include <memory>
  20. #include <utility>
  21. #include "common/utils.h"
  22. #include "kernel/tbe/tbe_kernel_build.h"
  23. #include "kernel/tbe/tbe_kernel_parallel_build.h"
  24. #include "kernel/tbe/tbe_utils.h"
  25. #include "kernel/tbe/tbe_convert_utils.h"
  26. namespace mindspore {
  27. namespace kernel {
  28. using mindspore::kernel::tbe::TbeUtils;
  29. static bool GenPreBuildKernelJson(const std::vector<AnfNodePtr> &compute_nodes,
  30. std::vector<nlohmann::json> *prebuild_op_list) {
  31. MS_EXCEPTION_IF_NULL(prebuild_op_list);
  32. TbeKernelJsonCreator creator(PREBUILD);
  33. for (const auto &anf_node : compute_nodes) {
  34. nlohmann::json prebuild;
  35. if (!creator.GenTbeSingleKernelJson(anf_node, &prebuild)) {
  36. MS_LOG(ERROR) << "GenTbeSingleKernelJson failed";
  37. return false;
  38. }
  39. (*prebuild_op_list).push_back(prebuild);
  40. }
  41. return true;
  42. }
  43. std::map<int32_t, KernelModPtr> KernelFusion(const std::vector<FusionScopeInfo> &fusion_scopes) {
  44. MS_LOG(INFO) << "kernel fusion build start, scope size:" << fusion_scopes.size();
  45. std::map<int32_t, KernelModPtr> kernel_mod_ret;
  46. auto build_manger = std::make_shared<ParallelBuildManager>();
  47. MS_EXCEPTION_IF_NULL(build_manger);
  48. for (const auto &fusion_scope_iter : fusion_scopes) {
  49. auto scope_id = fusion_scope_iter.scope_id;
  50. nlohmann::json fusion_op;
  51. string fusion_kernel = "te_fusion";
  52. if (!TbeKernelBuild::GenFusionScopeJson(fusion_scope_iter.input_nodes, fusion_scope_iter.compute_nodes, &fusion_op,
  53. &fusion_kernel)) {
  54. continue;
  55. }
  56. // gen kernel_name & check cache
  57. std::string json_str = fusion_op.dump();
  58. size_t hash_id = std::hash<std::string>()(json_str);
  59. auto json_name = fusion_kernel.append("_").append(std::to_string(hash_id));
  60. fusion_op["fusion_op_name"] = json_name;
  61. // gen json for prebuild
  62. std::vector<nlohmann::json> prebuild_op_list;
  63. if (!GenPreBuildKernelJson(fusion_scope_iter.compute_nodes, &prebuild_op_list)) {
  64. continue;
  65. }
  66. // get io size
  67. std::vector<size_t> input_size_list;
  68. std::vector<size_t> output_size_list;
  69. if (!TbeKernelBuild::GetIOSize(fusion_op["op_list"], fusion_scope_iter.output_nodes, &input_size_list,
  70. &output_size_list)) {
  71. continue;
  72. }
  73. // search cache
  74. auto kernel_pack = TbeUtils::SearchCache(json_name, tbe::kProcessorAiCore);
  75. if (kernel_pack != nullptr) {
  76. MS_LOG(INFO) << "Use cached kernel, kernel json name: " << json_name;
  77. auto kernel_mod =
  78. build_manger->GenKernelMod(json_name, tbe::kProcessorAiCore, input_size_list, output_size_list, kernel_pack);
  79. if (kernel_mod != nullptr) {
  80. kernel_mod_ret[scope_id] = kernel_mod;
  81. continue;
  82. }
  83. }
  84. // fusion build
  85. nlohmann::json fusion_json;
  86. fusion_json["fusion_op"] = fusion_op;
  87. fusion_json["prebuild_ops"] = prebuild_op_list;
  88. auto task_id = build_manger->StartCompileOp(fusion_json);
  89. TbeUtils::SaveJsonInfo(json_name, fusion_json.dump());
  90. if (task_id < 0) {
  91. MS_EXCEPTION(ArgumentError) << "start compile failed.";
  92. }
  93. build_manger->SaveTaskInfo(task_id, nullptr, json_name, input_size_list, output_size_list, scope_id);
  94. }
  95. int build_failed_num = 0;
  96. while (!build_manger->IsAllTaskFinish()) {
  97. int task_id = -1;
  98. char *task_result = nullptr;
  99. char *pre_build_result = nullptr;
  100. auto ret = build_manger->WaitOne(&task_id, &task_result, &pre_build_result);
  101. if (!ret) {
  102. MS_EXCEPTION(ArgumentError) << "Build Failed. wait one ret:" << ret << ", task id:" << task_id;
  103. }
  104. if ((task_result != nullptr) && (strcmp(task_result, "Success") != 0)) {
  105. MS_LOG(INFO) << "Fusion warning: Fuison op build failed, err log: " << task_result
  106. << " change to single op build.";
  107. build_failed_num++;
  108. }
  109. auto kernel_mod_item = build_manger->TaskFinishProcess(task_id, false);
  110. if (kernel_mod_item.second != nullptr) {
  111. (void)kernel_mod_ret.emplace(kernel_mod_item);
  112. }
  113. }
  114. MS_LOG(INFO) << "Build Fusion Kernel Failed Num: " << build_failed_num;
  115. return kernel_mod_ret;
  116. }
  117. } // namespace kernel
  118. } // namespace mindspore