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.

model_helper.cc 36 kB

5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
4 years ago
4 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 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
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876
  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 "framework/common/helper/model_helper.h"
  17. #include "common/model_parser/model_parser.h"
  18. #include "framework/omg/version.h"
  19. #include "graph/debug/ge_attr_define.h"
  20. #include "graph/utils/graph_utils.h"
  21. using std::string;
  22. using domi::ModelTaskDef;
  23. namespace {
  24. const int64_t kOriginalOmPartitionNum = 1;
  25. const uint32_t kStatiOmFileModelNum = 1;
  26. }
  27. namespace ge {
  28. FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY ModelHelper::~ModelHelper() { (void)ReleaseLocalModelData(); }
  29. Status ModelHelper::SaveModelPartition(std::shared_ptr<OmFileSaveHelper> &om_file_save_helper, ModelPartitionType type,
  30. const uint8_t *data, size_t size, size_t model_index) {
  31. if (size < 1 || size > UINT32_MAX) {
  32. GELOGE(PARAM_INVALID, "Add model partition failed, partition size %zu invalid", size);
  33. if (size > UINT32_MAX) {
  34. string item = "item";
  35. if (type == MODEL_DEF) {
  36. item = "model info";
  37. } else if (type == WEIGHTS_DATA) {
  38. item = "weight data";
  39. } else if (type == TASK_INFO) {
  40. item = "task info";
  41. } else if (type == TBE_KERNELS) {
  42. item = "tbe kernels";
  43. } else if (type == CUST_AICPU_KERNELS) {
  44. item = "aicpu kernels";
  45. }
  46. ErrorManager::GetInstance().ATCReportErrMessage("E19023", {"size", "item", "maxsize"},
  47. {std::to_string(size), item, std::to_string(UINT32_MAX)});
  48. }
  49. return PARAM_INVALID;
  50. }
  51. if (data == nullptr) {
  52. GELOGE(PARAM_INVALID, "Add model partition failed, data is null");
  53. return PARAM_INVALID;
  54. }
  55. ModelPartition partition_model;
  56. partition_model.data = const_cast<uint8_t *>(data);
  57. partition_model.size = static_cast<uint32_t>(size);
  58. partition_model.type = type;
  59. if (om_file_save_helper->AddPartition(partition_model, model_index) != SUCCESS) {
  60. GELOGE(PARAM_INVALID, "Add model partition failed, partition size %zu", size);
  61. return PARAM_INVALID;
  62. }
  63. return SUCCESS;
  64. }
  65. Status ModelHelper::SaveSizeToModelDef(const GeModelPtr &ge_model) {
  66. vector<int64_t> om_info;
  67. auto ge_model_weight = ge_model->GetWeight();
  68. GELOGD("SaveSizeToModelDef weight_data_size is %zu, %p", ge_model_weight.GetSize(), ge_model_weight.GetData());
  69. om_info.push_back(ge_model_weight.GetSize());
  70. TBEKernelStore tbe_kernel_store = ge_model->GetTBEKernelStore();
  71. GELOGD("SaveSizeToModelDef tbe_kernels_size is %zu", tbe_kernel_store.DataSize());
  72. om_info.push_back(tbe_kernel_store.DataSize());
  73. CustAICPUKernelStore cust_aicpu_kernel_store = ge_model->GetCustAICPUKernelStore();
  74. GELOGD("SaveSizeToModelDef cust aicpu kernels size is %zu", cust_aicpu_kernel_store.DataSize());
  75. om_info.push_back(cust_aicpu_kernel_store.DataSize());
  76. std::shared_ptr<ModelTaskDef> model_task_def = ge_model->GetModelTaskDefPtr();
  77. if (model_task_def == nullptr) {
  78. GELOGE(ACL_ERROR_GE_MEMORY_ALLOCATION, "Create model task def ptr failed");
  79. return ACL_ERROR_GE_MEMORY_ALLOCATION;
  80. }
  81. size_t partition_task_size = model_task_def->ByteSizeLong();
  82. GELOGD("SaveSizeToModelDef task_info_size is %zu", partition_task_size);
  83. om_info.push_back(partition_task_size);
  84. GE_CHK_BOOL_EXEC(ge::AttrUtils::SetListInt(*(ge_model.get()), "om_info_list", om_info),
  85. GELOGE(FAILED, "SetListInt of om_info_list failed.");
  86. return FAILED);
  87. return SUCCESS;
  88. }
  89. Status ModelHelper::SaveModelDef(std::shared_ptr<OmFileSaveHelper> &om_file_save_helper,
  90. const GeModelPtr &ge_model, ge::Buffer &model_buffer, size_t model_index) {
  91. ModelPtr model_tmp = ge::MakeShared<ge::Model>(ge_model->GetName(), ge_model->GetPlatformVersion());
  92. if (model_tmp == nullptr) {
  93. GELOGE(FAILED, "Create Model %s Ptr failed", ge_model->GetName().c_str());
  94. return FAILED;
  95. }
  96. model_tmp->SetGraph(ge_model->GetGraph());
  97. model_tmp->SetVersion(ge_model->GetVersion());
  98. model_tmp->SetAttr(ge_model->MutableAttrMap());
  99. Status ret = SaveSizeToModelDef(ge_model);
  100. if (ret != SUCCESS) {
  101. GELOGE(ret, "SaveSizeToModelDef failed");
  102. return ret;
  103. }
  104. (void)model_tmp->Save(model_buffer);
  105. GELOGD("MODEL_DEF size is %zu", model_buffer.GetSize());
  106. if (model_buffer.GetSize() > 0) {
  107. if (SaveModelPartition(om_file_save_helper, ModelPartitionType::MODEL_DEF, model_buffer.GetData(),
  108. model_buffer.GetSize(), model_index) != SUCCESS) {
  109. GELOGE(PARAM_INVALID, "Add model graph partition failed");
  110. return PARAM_INVALID;
  111. }
  112. }
  113. return SUCCESS;
  114. }
  115. Status ModelHelper::SaveModelWeights(std::shared_ptr<OmFileSaveHelper> &om_file_save_helper,
  116. const GeModelPtr &ge_model, size_t model_index) {
  117. auto ge_model_weight = ge_model->GetWeight();
  118. GELOGD("WEIGHTS_DATA size is %zu, %p", ge_model_weight.GetSize(), ge_model_weight.GetData());
  119. // weight is not necessary
  120. if (ge_model_weight.GetSize() > 0) {
  121. GE_CHK_STATUS_RET(SaveModelPartition(om_file_save_helper,
  122. ModelPartitionType::WEIGHTS_DATA,
  123. ge_model_weight.GetData(),
  124. ge_model_weight.GetSize(), model_index), "Add weight partition failed");
  125. }
  126. return SUCCESS;
  127. }
  128. Status ModelHelper::SaveModelTbeKernel(std::shared_ptr<OmFileSaveHelper> &om_file_save_helper,
  129. const GeModelPtr &ge_model, size_t model_index) {
  130. TBEKernelStore tbe_kernel_store = ge_model->GetTBEKernelStore();
  131. GELOGD("TBE_KERNELS size is %zu", tbe_kernel_store.DataSize());
  132. if (tbe_kernel_store.DataSize() > 0) {
  133. GE_CHK_STATUS_RET(
  134. SaveModelPartition(om_file_save_helper, ModelPartitionType::TBE_KERNELS,
  135. ge_model->GetTBEKernelStore().Data(), ge_model->GetTBEKernelStore().DataSize(),
  136. model_index), "Add tbe kernel partition failed");
  137. }
  138. // no need to check value, DATA->NetOutput
  139. (void)tbe_kernel_store.Load(tbe_kernel_store.Data(), tbe_kernel_store.DataSize());
  140. return SUCCESS;
  141. }
  142. Status ModelHelper::SaveModelCustAICPU(std::shared_ptr<OmFileSaveHelper> &om_file_save_helper,
  143. const GeModelPtr &ge_model, size_t model_index) {
  144. CustAICPUKernelStore cust_aicpu_kernel_store = ge_model->GetCustAICPUKernelStore();
  145. GELOGD("cust aicpu kernels size is %zu", cust_aicpu_kernel_store.DataSize());
  146. if (cust_aicpu_kernel_store.DataSize() > 0) {
  147. GE_CHK_STATUS_RET(SaveModelPartition(om_file_save_helper,
  148. ModelPartitionType::CUST_AICPU_KERNELS,
  149. ge_model->GetCustAICPUKernelStore().Data(),
  150. cust_aicpu_kernel_store.DataSize(), model_index),
  151. "Add cust aicpu kernel partition failed");
  152. }
  153. return SUCCESS;
  154. }
  155. Status ModelHelper::SaveModelTaskDef(std::shared_ptr<OmFileSaveHelper> &om_file_save_helper,
  156. const GeModelPtr &ge_model, ge::Buffer &task_buffer, size_t model_index) {
  157. std::shared_ptr<ModelTaskDef> model_task_def = ge_model->GetModelTaskDefPtr();
  158. if (model_task_def == nullptr) {
  159. GELOGE(ACL_ERROR_GE_MEMORY_ALLOCATION, "Create model task def ptr failed");
  160. return ACL_ERROR_GE_MEMORY_ALLOCATION;
  161. }
  162. size_t partition_task_size = model_task_def->ByteSizeLong();
  163. GE_IF_BOOL_EXEC(partition_task_size == 0 || partition_task_size > INT_MAX,
  164. GELOGE(FAILED, "Model_def's byte size (%zu) is invalid!", partition_task_size);
  165. return FAILED);
  166. task_buffer = ge::Buffer(partition_task_size);
  167. if (task_buffer.GetSize() == 0) {
  168. GELOGE(ACL_ERROR_GE_MEMORY_ALLOCATION, "Alloc model task def buffer failed");
  169. return ACL_ERROR_GE_MEMORY_ALLOCATION;
  170. }
  171. (void)model_task_def->SerializePartialToArray(task_buffer.GetData(), static_cast<int>(partition_task_size));
  172. GELOGD("TASK_INFO op_size:%d, stream_num:%u", model_task_def->op().size(), model_task_def->stream_num());
  173. GELOGD("TASK_INFO size is %zu", partition_task_size);
  174. if (SaveModelPartition(om_file_save_helper, ModelPartitionType::TASK_INFO, task_buffer.GetData(),
  175. partition_task_size, model_index) != SUCCESS) {
  176. GELOGE(PARAM_INVALID, "Add model task def partition failed");
  177. return PARAM_INVALID;
  178. }
  179. return SUCCESS;
  180. }
  181. Status ModelHelper::SaveModelHeader(std::shared_ptr<OmFileSaveHelper> &om_file_save_helper,
  182. const GeModelPtr &ge_model, size_t model_num) {
  183. // Save target/version to model_header
  184. ModelFileHeader &model_header = om_file_save_helper->GetModelFileHeader();
  185. model_header.platform_type = ge_model->GetPlatformType();
  186. model_header.om_ir_version = ge_model->GetVersion();
  187. model_header.model_num = model_num;
  188. std::string platform_version = ge_model->GetPlatformVersion();
  189. errno_t err;
  190. err = memcpy_s(model_header.platform_version, PLATFORM_VERSION_LEN, platform_version.c_str(),
  191. platform_version.size() + 1);
  192. if (err != EOK) {
  193. GELOGE(ACL_ERROR_GE_MEMORY_ALLOCATION,
  194. "ModelHelper SaveModel failed while allocating memory for platform_version.");
  195. return ACL_ERROR_GE_MEMORY_ALLOCATION;
  196. }
  197. string version = reinterpret_cast<char *>(model_header.platform_version);
  198. GELOGD("Platform version save: %s", version.c_str());
  199. size_t name_size = ge_model->GetName().size();
  200. name_size = name_size > (MODEL_NAME_LENGTH - 1) ? (MODEL_NAME_LENGTH - 1) : name_size;
  201. err = memcpy_s(model_header.name, MODEL_NAME_LENGTH, ge_model->GetName().c_str(), name_size);
  202. if (err != EOK) {
  203. GELOGE(ACL_ERROR_GE_MEMORY_ALLOCATION, "ModelHelper SaveModel failed while allocating memory for name");
  204. return ACL_ERROR_GE_MEMORY_ALLOCATION;
  205. }
  206. string model_name = reinterpret_cast<char *>(model_header.name);
  207. GELOGD("Model name save:%s", model_name.c_str());
  208. return SUCCESS;
  209. }
  210. Status ModelHelper::SaveAllModelPartiton(std::shared_ptr<OmFileSaveHelper>& om_file_save_helper,
  211. const GeModelPtr &ge_model, ge::Buffer &model_buffer,
  212. ge::Buffer &task_buffer, size_t model_index) {
  213. if (SaveModelDef(om_file_save_helper, ge_model, model_buffer, model_index) != SUCCESS) {
  214. GELOGE(FAILED, "save model def failed");
  215. return FAILED;
  216. }
  217. if (SaveModelWeights(om_file_save_helper, ge_model, model_index) != SUCCESS) {
  218. GELOGE(FAILED, "save model weights failed");
  219. return FAILED;
  220. }
  221. if (SaveModelTbeKernel(om_file_save_helper, ge_model, model_index) != SUCCESS) {
  222. GELOGE(FAILED, "save model tbe kernel failed");
  223. return FAILED;
  224. }
  225. if (SaveModelCustAICPU(om_file_save_helper, ge_model, model_index) != SUCCESS) {
  226. GELOGE(FAILED, "save model cust ai cpu failed");
  227. return FAILED;
  228. }
  229. if (SaveModelTaskDef(om_file_save_helper, ge_model, task_buffer, model_index) != SUCCESS) {
  230. GELOGE(FAILED, "save task def failed");
  231. return FAILED;
  232. }
  233. return SUCCESS;
  234. }
  235. FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelHelper::SaveToOmModel(const GeModelPtr &ge_model,
  236. const SaveParam &save_param,
  237. const std::string &output_file,
  238. ModelBufferData& model) {
  239. if (output_file.empty()) {
  240. GELOGE(FAILED, "GraphBuilder SaveModel received invalid file name prefix");
  241. return FAILED;
  242. }
  243. GE_IF_BOOL_EXEC(ge_model == nullptr, GELOGE(FAILED, "Ge_model is nullptr"); return FAILED);
  244. std::shared_ptr<OmFileSaveHelper> om_file_save_helper = ge::MakeShared<OmFileSaveHelper>();
  245. GE_CHECK_NOTNULL(om_file_save_helper);
  246. ge::Buffer model_buffer;
  247. ge::Buffer task_buffer;
  248. auto ret = SaveAllModelPartiton(om_file_save_helper, ge_model, model_buffer, task_buffer);
  249. if (ret != SUCCESS) {
  250. GELOGE(ret, "save all model partition failed");
  251. return ret;
  252. }
  253. ret = SaveModelHeader(om_file_save_helper, ge_model);
  254. if (ret != SUCCESS) {
  255. GELOGE(ret, "save model header failed");
  256. return ret;
  257. }
  258. ret = om_file_save_helper->SaveModel(save_param, output_file.c_str(), model, is_offline_);
  259. if (ret != SUCCESS) {
  260. GELOGE(FAILED, "OmFileSaveHelper SaveModel return fail.");
  261. return ret;
  262. }
  263. return SUCCESS;
  264. }
  265. FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelHelper::SaveToOmRootModel(
  266. const GeRootModelPtr &ge_root_model,
  267. const SaveParam &save_param,
  268. const std::string &output_file,
  269. ModelBufferData& model,
  270. bool is_unknown_shape) {
  271. GE_CHECK_NOTNULL(ge_root_model);
  272. GE_IF_BOOL_EXEC(ge_root_model == nullptr, GELOGE(FAILED, "Ge_root_model is nullptr"); return FAILED);
  273. auto &name_to_ge_model = ge_root_model->GetSubgraphInstanceNameToModel();
  274. GE_IF_BOOL_EXEC(name_to_ge_model.empty(), GELOGE(FAILED, "Ge_root_model has no sub model"); return FAILED);
  275. GE_IF_BOOL_EXEC(output_file.empty(),
  276. GELOGE(FAILED, "GraphBuilder SaveModel received invalid file name prefix");
  277. return FAILED);
  278. if (!is_unknown_shape) {
  279. auto &model_root = name_to_ge_model.begin()->second;
  280. return SaveToOmModel(model_root, save_param, output_file, model);
  281. }
  282. std::shared_ptr<OmFileSaveHelper> om_file_save_helper = ge::MakeShared<OmFileSaveHelper>();
  283. GE_CHECK_NOTNULL(om_file_save_helper);
  284. auto &first_ge_model = name_to_ge_model.at(ge_root_model->GetRootGraph()->GetName());
  285. // ge root model must be the first to be loaded
  286. vector<string> model_names{ge_root_model->GetRootGraph()->GetName()};
  287. for (auto &item : name_to_ge_model) {
  288. if (item.first != model_names.front()) {
  289. model_names.emplace_back(item.first);
  290. }
  291. }
  292. vector<ge::Buffer> model_buffers(model_names.size());
  293. vector<ge::Buffer> task_buffers(model_names.size());
  294. size_t cur_index = 0;
  295. if (model_names.size() > 1) {
  296. GELOGD("only save first model MODEL_DEF");
  297. if (SaveModelDef(om_file_save_helper, first_ge_model, model_buffers[cur_index], cur_index) != SUCCESS) {
  298. GELOGE(FAILED, "save model def failed");
  299. return FAILED;
  300. }
  301. ++cur_index;
  302. }
  303. for (; cur_index < model_names.size(); ++cur_index) {
  304. auto model_name = model_names[cur_index];
  305. GELOGD("cur model %s index is %zu", model_name.c_str(), cur_index);
  306. const GeModelPtr &ge_model = name_to_ge_model.at(model_name);
  307. auto ret = SaveAllModelPartiton(om_file_save_helper, ge_model, model_buffers[cur_index],
  308. task_buffers[cur_index], cur_index);
  309. if (ret != SUCCESS) {
  310. GELOGE(INTERNAL_ERROR, "Save model %s failed", model_name.c_str());
  311. return INTERNAL_ERROR;
  312. }
  313. }
  314. auto ret = SaveModelHeader(om_file_save_helper, first_ge_model, model_names.size());
  315. if (ret != SUCCESS) {
  316. GELOGE(INTERNAL_ERROR, "Save model %s header failed", first_ge_model->GetName().c_str());
  317. return INTERNAL_ERROR;
  318. }
  319. ret = om_file_save_helper->SaveRootModel(save_param, output_file.c_str(), model, is_offline_);
  320. if (ret != SUCCESS) {
  321. GELOGE(FAILED, "OmFileSaveHelper SaveModel return fail.");
  322. return FAILED;
  323. }
  324. return SUCCESS;
  325. }
  326. FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status
  327. ModelHelper::SaveOriginalGraphToOmModel(const ge::Graph &graph, const std::string &output_file) {
  328. if (output_file.empty()) {
  329. GELOGE(FAILED, "SaveModel received invalid file name prefix");
  330. return FAILED;
  331. }
  332. // Get computegraph from graph
  333. auto compute_graph = ge::GraphUtils::GetComputeGraph(graph);
  334. if (compute_graph == nullptr) {
  335. GELOGE(FAILED, "SaveModel fail for compute_graph null");
  336. return FAILED;
  337. }
  338. GE_DUMP(compute_graph, "OriginalGraph");
  339. // Model
  340. ModelPtr model_ptr = ge::MakeShared<ge::Model>();
  341. GE_CHECK_NOTNULL_EXEC(model_ptr, return MEMALLOC_FAILED);
  342. std::string original_model_name = compute_graph->GetName() + "_original";
  343. model_ptr->SetName(original_model_name);
  344. model_ptr->SetGraph(graph);
  345. model_ptr->SetVersion(static_cast<uint32_t>(OM_PROTO_VERSION));
  346. string framework_version;
  347. Status frame_rt = PlatformVersionManager::GetPlatformVersion(framework_version);
  348. if (frame_rt == SUCCESS) {
  349. uint32_t counter = 0;
  350. string model_framework_version = framework_version + "." + std::to_string(counter);
  351. model_ptr->SetPlatformVersion(model_framework_version);
  352. }
  353. // Model def
  354. ge::Buffer model_buffer;
  355. ge::graphStatus status = model_ptr->Save(model_buffer);
  356. if (status != ge::GRAPH_SUCCESS) {
  357. GELOGE(FAILED, "SaveModel fail for save buffer fail");
  358. return FAILED;
  359. }
  360. std::shared_ptr<OmFileSaveHelper> om_file_save_helper = ge::MakeShared<OmFileSaveHelper>();
  361. GE_CHECK_NOTNULL_EXEC(om_file_save_helper, return MEMALLOC_FAILED);
  362. ModelPartition partition_model;
  363. partition_model.data = model_buffer.GetData();
  364. partition_model.size = static_cast<uint32_t>(model_buffer.GetSize());
  365. partition_model.type = ModelPartitionType::MODEL_DEF;
  366. GELOGI("Original Model type[%u],size[%u]", partition_model.type, partition_model.size);
  367. if (partition_model.data != nullptr && partition_model.size > 0) {
  368. (void)om_file_save_helper->AddPartition(partition_model);
  369. // Condition of AddPartition is established, no need to check value
  370. }
  371. // Save target/version to model_header
  372. ModelFileHeader &model_header = om_file_save_helper->GetModelFileHeader();
  373. model_header.om_ir_version = model_ptr->GetVersion();
  374. model_header.headsize = MODEL_FILE_HEAD_LEN;
  375. std::string platform_version = model_ptr->GetPlatformVersion();
  376. errno_t err = memcpy_s(model_header.platform_version, PLATFORM_VERSION_LEN, platform_version.c_str(),
  377. platform_version.size() + 1);
  378. if (err != EOK) {
  379. GELOGE(FAILED, "ModelHelper SaveModel failed for platform_version");
  380. return FAILED;
  381. }
  382. size_t name_size = model_ptr->GetName().size();
  383. name_size = name_size > (MODEL_NAME_LENGTH - 1) ? (MODEL_NAME_LENGTH - 1) : name_size;
  384. err = memcpy_s(model_header.name, MODEL_NAME_LENGTH, model_ptr->GetName().c_str(), name_size);
  385. if (err != EOK) {
  386. GELOGE(FAILED, "ModelHelper SaveModel memory copy failed");
  387. return FAILED;
  388. }
  389. ModelBufferData model;
  390. Status ret = om_file_save_helper->SaveModelToFile(output_file.c_str(), model, is_offline_);
  391. return (ret == SUCCESS ? SUCCESS : FAILED);
  392. }
  393. FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelHelper::LoadModel(const ge::ModelData &model_data) {
  394. if (model_data.model_data == nullptr || model_data.model_len == 0) {
  395. GELOGE(ACL_ERROR_GE_EXEC_MODEL_DATA_SIZE_INVALID, "Model_data is nullptr, or model_data_size is 0");
  396. return ACL_ERROR_GE_EXEC_MODEL_DATA_SIZE_INVALID;
  397. }
  398. if (is_assign_model_) {
  399. GELOGE(ACL_ERROR_GE_EXEC_LOAD_MODEL_REPEATED, "Model helper has already loaded!");
  400. return ACL_ERROR_GE_EXEC_LOAD_MODEL_REPEATED;
  401. }
  402. if (ReleaseLocalModelData() != SUCCESS) {
  403. GELOGE(ACL_ERROR_GE_EXEC_RELEASE_MODEL_DATA, "ReleaseLocalModelData failed.");
  404. return ACL_ERROR_GE_EXEC_RELEASE_MODEL_DATA;
  405. }
  406. Status status = ModelParserBase::ParseModelContent(model_data, model_addr_tmp_, model_len_tmp_);
  407. if (status != SUCCESS) {
  408. GELOGE(ACL_ERROR_GE_PARAM_INVALID, "Parse model content failed!");
  409. return ACL_ERROR_GE_PARAM_INVALID;
  410. }
  411. file_header_ = reinterpret_cast<ModelFileHeader *>(model_data.model_data);
  412. OmFileLoadHelper om_load_helper;
  413. status = om_load_helper.Init(model_addr_tmp_, model_len_tmp_);
  414. if (status != SUCCESS) {
  415. GELOGE(status, "Om_load_helper init failed");
  416. model_addr_tmp_ = nullptr;
  417. return status;
  418. }
  419. auto partition_table = reinterpret_cast<ModelPartitionTable *>(model_addr_tmp_);
  420. if (partition_table->num == kOriginalOmPartitionNum) {
  421. model_addr_tmp_ = nullptr;
  422. GELOGE(ACL_ERROR_GE_PARAM_INVALID, "om model is error,please use executable om model");
  423. return ACL_ERROR_GE_PARAM_INVALID;
  424. }
  425. // Encrypt model need to del temp model/no encrypt model don't need to del model
  426. model_addr_tmp_ = nullptr;
  427. status = GenerateGeModel(om_load_helper);
  428. if (status != SUCCESS) {
  429. GELOGE(status, "GenerateGeModel failed");
  430. return status;
  431. }
  432. GELOGD("in ModelHelper::LoadModel, is_assign_model_ is setted to true!");
  433. is_assign_model_ = true;
  434. return SUCCESS;
  435. }
  436. FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelHelper::LoadRootModel(const ge::ModelData &model_data) {
  437. if (model_data.model_data == nullptr || model_data.model_len == 0) {
  438. GELOGE(ACL_ERROR_GE_EXEC_MODEL_DATA_SIZE_INVALID, "Model_data is nullptr, or model_data_size is 0");
  439. return ACL_ERROR_GE_EXEC_MODEL_DATA_SIZE_INVALID;
  440. }
  441. if (is_assign_model_) {
  442. GELOGE(ACL_ERROR_GE_EXEC_LOAD_MODEL_REPEATED, "Model helper has already loaded!");
  443. return ACL_ERROR_GE_EXEC_LOAD_MODEL_REPEATED;
  444. }
  445. if (ReleaseLocalModelData() != SUCCESS) {
  446. GELOGE(INTERNAL_ERROR, "ReleaseLocalModelData failed.");
  447. return INTERNAL_ERROR;
  448. }
  449. Status status = ModelParserBase::ParseModelContent(model_data, model_addr_tmp_, model_len_tmp_);
  450. if (status != SUCCESS) {
  451. GELOGE(ACL_ERROR_GE_PARAM_INVALID, "Parse model content failed!");
  452. return ACL_ERROR_GE_PARAM_INVALID;
  453. }
  454. file_header_ = reinterpret_cast<ModelFileHeader *>(model_data.model_data);
  455. //model verison 1.0 file header does not have model_num member
  456. is_unknown_shape_model_ = file_header_->version >= ge::MODEL_VERSION &&
  457. file_header_->model_num > kStatiOmFileModelNum;
  458. GELOGD("cur om model is ge root model or no %d, model version %u", is_unknown_shape_model_, file_header_->version);
  459. OmFileLoadHelper om_load_helper;
  460. if (is_unknown_shape_model_) {
  461. auto model_num = file_header_->model_num;
  462. status = om_load_helper.Init(model_addr_tmp_, model_len_tmp_, model_num);
  463. } else {
  464. status = om_load_helper.Init(model_addr_tmp_, model_len_tmp_);
  465. }
  466. if (status != SUCCESS) {
  467. GELOGE(status, "Om_load_helper init failed");
  468. model_addr_tmp_ = nullptr;
  469. return status;
  470. }
  471. // Encrypt model need to del temp model/no encrypt model don't need to del model
  472. model_addr_tmp_ = nullptr;
  473. status = GenerateGeRootModel(om_load_helper);
  474. if (status != SUCCESS) {
  475. GELOGE(status, "GenerateGeRootModel failed");
  476. return status;
  477. }
  478. GELOGD("in ModelHelper::LoadRootModel, is_assign_model_ is setted to true!");
  479. is_assign_model_ = true;
  480. return SUCCESS;
  481. }
  482. Status ModelHelper::GenerateGeModel(OmFileLoadHelper &om_load_helper) {
  483. model_ = ge::MakeShared<ge::GeModel>();
  484. GE_CHECK_NOTNULL(model_);
  485. Status ret = LoadModelData(om_load_helper);
  486. if (ret != SUCCESS) {
  487. return ACL_ERROR_GE_EXEC_LOAD_MODEL_PARTITION_FAILED;
  488. }
  489. ret = LoadWeights(om_load_helper);
  490. if (ret != SUCCESS) {
  491. return ACL_ERROR_GE_EXEC_LOAD_WEIGHT_PARTITION_FAILED;
  492. }
  493. ret = LoadTask(om_load_helper);
  494. if (ret != SUCCESS) {
  495. return ACL_ERROR_GE_EXEC_LOAD_TASK_PARTITION_FAILED;
  496. }
  497. ret = LoadTBEKernelStore(om_load_helper);
  498. if (ret != SUCCESS) {
  499. return ACL_ERROR_GE_EXEC_LOAD_KERNEL_PARTITION_FAILED;
  500. }
  501. ret = LoadCustAICPUKernelStore(om_load_helper);
  502. if (ret != SUCCESS) {
  503. return ACL_ERROR_GE_EXEC_LOAD_KERNEL_PARTITION_FAILED;
  504. }
  505. return SUCCESS;
  506. }
  507. Status ModelHelper::GenerateGeRootModel(OmFileLoadHelper &om_load_helper) {
  508. GELOGD("Begin to generate ge root model");
  509. root_model_ = ge::MakeShared<ge::GeRootModel>();
  510. GE_CHECK_NOTNULL(root_model_);
  511. if (!is_unknown_shape_model_) {
  512. if (GenerateGeModel(om_load_helper) != SUCCESS) {
  513. GELOGE(FAILED, "GenerateGeModel failed");
  514. return FAILED;
  515. }
  516. GE_CHECK_NOTNULL(model_);
  517. root_model_->SetRootGraph(GraphUtils::GetComputeGraph(model_->GetGraph()));
  518. return SUCCESS;
  519. }
  520. bool is_first_model = true;
  521. for (size_t mode_index = 0; mode_index < file_header_->model_num; ++mode_index) {
  522. GeModelPtr cur_model = ge::MakeShared<ge::GeModel>();
  523. Status ret = LoadModelData(om_load_helper, cur_model, mode_index);
  524. if (ret != SUCCESS) {
  525. return ACL_ERROR_GE_EXEC_LOAD_MODEL_PARTITION_FAILED;
  526. }
  527. if (is_first_model) {
  528. is_first_model = false;
  529. root_model_->SetRootGraph(GraphUtils::GetComputeGraph(cur_model->GetGraph()));
  530. root_model_->SetModelId(cur_model->GetModelId());
  531. model_ = cur_model;
  532. continue;
  533. }
  534. ret = LoadWeights(om_load_helper, cur_model, mode_index);
  535. if (ret != SUCCESS) {
  536. return ACL_ERROR_GE_EXEC_LOAD_WEIGHT_PARTITION_FAILED;
  537. }
  538. ret = LoadTBEKernelStore(om_load_helper, cur_model, mode_index);
  539. if (ret != SUCCESS) {
  540. return ACL_ERROR_GE_EXEC_LOAD_KERNEL_PARTITION_FAILED;
  541. }
  542. ret = LoadCustAICPUKernelStore(om_load_helper, cur_model, mode_index);
  543. if (ret != SUCCESS) {
  544. return ACL_ERROR_GE_EXEC_LOAD_KERNEL_PARTITION_FAILED;
  545. }
  546. ret = LoadTask(om_load_helper, cur_model, mode_index);
  547. if (ret != SUCCESS) {
  548. return ACL_ERROR_GE_EXEC_LOAD_TASK_PARTITION_FAILED;
  549. }
  550. root_model_->SetSubgraphInstanceNameToModel(cur_model->GetName(), cur_model);
  551. }
  552. return SUCCESS;
  553. }
  554. Status ModelHelper::LoadModelData(OmFileLoadHelper &om_load_helper) {
  555. ModelPartition partition_model_def;
  556. // no need to check value, DATA->NetOutput
  557. om_load_helper.GetModelPartition(ModelPartitionType::MODEL_DEF, partition_model_def);
  558. GELOGD("Model_def partition addr:%p,size:%u", partition_model_def.data, partition_model_def.size);
  559. ge::Model model;
  560. if (ge::Model::Load(partition_model_def.data, partition_model_def.size, model) != SUCCESS) {
  561. GELOGE(INTERNAL_ERROR, "Load model failed.");
  562. return INTERNAL_ERROR;
  563. }
  564. SetModelToGeModel(model_, model);
  565. return SUCCESS;
  566. }
  567. void ModelHelper::SetModelToGeModel(GeModelPtr &ge_model, Model &model) {
  568. ge_model->SetGraph(model.GetGraph());
  569. ge_model->SetName(model.GetName());
  570. ge_model->SetVersion(model.GetVersion());
  571. ge_model->SetPlatformVersion(model.GetPlatformVersion());
  572. ge_model->SetAttr(model.MutableAttrMap());
  573. }
  574. Status ModelHelper::LoadModelData(OmFileLoadHelper &om_load_helper, GeModelPtr &cur_model, size_t mode_index) {
  575. ModelPartition partition_model_def;
  576. // no need to check value, DATA->NetOutput
  577. om_load_helper.GetModelPartition(ModelPartitionType::MODEL_DEF, partition_model_def, mode_index);
  578. GELOGD("Model_def partition addr:%p,size:%u", partition_model_def.data, partition_model_def.size);
  579. ge::Model model;
  580. if (ge::Model::Load(partition_model_def.data, partition_model_def.size, model) != SUCCESS) {
  581. GELOGE(INTERNAL_ERROR, "Load model failed.");
  582. return INTERNAL_ERROR;
  583. }
  584. SetModelToGeModel(cur_model, model);
  585. return SUCCESS;
  586. }
  587. Status ModelHelper::LoadWeights(OmFileLoadHelper &om_load_helper) {
  588. ModelPartition partition;
  589. if (om_load_helper.GetModelPartition(ModelPartitionType::WEIGHTS_DATA, partition) != SUCCESS) {
  590. GELOGE(FAILED, "Get weight model partition failed.");
  591. return FAILED;
  592. }
  593. ge::Buffer weight = ge::Buffer::CopyFrom(partition.data, partition.size);
  594. model_->SetWeight(weight);
  595. GELOGD("GetWeight size:%u", partition.size);
  596. return SUCCESS;
  597. }
  598. Status ModelHelper::LoadWeights(OmFileLoadHelper &om_load_helper, GeModelPtr &cur_model, size_t mode_index) {
  599. ModelPartition partition;
  600. if (om_load_helper.GetModelPartition(ModelPartitionType::WEIGHTS_DATA, partition, mode_index) != SUCCESS) {
  601. GELOGE(FAILED, "Get weight model partition failed.");
  602. return FAILED;
  603. }
  604. ge::Buffer weight = ge::Buffer::CopyFrom(partition.data, partition.size);
  605. cur_model->SetWeight(weight);
  606. GELOGD("GetWeight size:%u", partition.size);
  607. return SUCCESS;
  608. }
  609. FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelHelper::LoadTask(OmFileLoadHelper &om_load_helper) {
  610. ModelPartition task_partition;
  611. if (om_load_helper.GetModelPartition(ModelPartitionType::TASK_INFO, task_partition) != SUCCESS) {
  612. GELOGE(FAILED, "Get task model partition failed.");
  613. return FAILED;
  614. }
  615. std::shared_ptr<ModelTaskDef> task = ge::MakeShared<ModelTaskDef>();
  616. GE_CHECK_NOTNULL(task);
  617. if (task_partition.size != 0) {
  618. if (!ReadProtoFromArray(task_partition.data, task_partition.size, task.get())) {
  619. GELOGE(INTERNAL_ERROR, "ReadProtoFromArray failed.");
  620. return INTERNAL_ERROR;
  621. }
  622. GELOGD("TASK_INFO op_size:%d, stream_num:%u", task->op().size(), task->stream_num());
  623. }
  624. model_->SetModelTaskDef(task);
  625. return SUCCESS;
  626. }
  627. FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelHelper::LoadTask(OmFileLoadHelper &om_load_helper,
  628. GeModelPtr &cur_model,
  629. size_t mode_index) {
  630. ModelPartition task_partition;
  631. if (om_load_helper.GetModelPartition(ModelPartitionType::TASK_INFO, task_partition, mode_index) != SUCCESS) {
  632. GELOGE(FAILED, "Get task model partition failed.");
  633. return FAILED;
  634. }
  635. std::shared_ptr<ModelTaskDef> task = ge::MakeShared<ModelTaskDef>();
  636. GE_CHECK_NOTNULL(task);
  637. if (task_partition.size != 0) {
  638. if (!ReadProtoFromArray(task_partition.data, task_partition.size, task.get())) {
  639. GELOGE(INTERNAL_ERROR, "ReadProtoFromArray failed.");
  640. return INTERNAL_ERROR;
  641. }
  642. GELOGD("TASK_INFO op_size:%d, stream_num:%u", task->op().size(), task->stream_num());
  643. }
  644. cur_model->SetModelTaskDef(task);
  645. return SUCCESS;
  646. }
  647. Status ModelHelper::LoadTBEKernelStore(OmFileLoadHelper &om_load_helper) {
  648. // Load tbe kernels
  649. ModelPartition partition_kernel_def;
  650. TBEKernelStore kernel_store;
  651. if (om_load_helper.GetModelPartition(ModelPartitionType::TBE_KERNELS, partition_kernel_def) == SUCCESS) {
  652. GELOGD("Kernels partition size:%u", partition_kernel_def.size);
  653. if (kernel_store.Load(partition_kernel_def.data, partition_kernel_def.size)) {
  654. GELOGD("Load tbe kernels success");
  655. } else {
  656. GELOGW("Load tbe kernels failed");
  657. }
  658. }
  659. model_->SetTBEKernelStore(kernel_store);
  660. return SUCCESS;
  661. }
  662. Status ModelHelper::LoadTBEKernelStore(OmFileLoadHelper &om_load_helper, GeModelPtr &cur_model, size_t mode_index) {
  663. // Load tbe kernels
  664. ModelPartition partition_kernel_def;
  665. TBEKernelStore kernel_store;
  666. if (om_load_helper.GetModelPartition(ModelPartitionType::TBE_KERNELS, partition_kernel_def, mode_index) ==
  667. SUCCESS) {
  668. GELOGD("Kernels partition size:%u", partition_kernel_def.size);
  669. if (kernel_store.Load(partition_kernel_def.data, partition_kernel_def.size)) {
  670. GELOGD("Load tbe kernels success");
  671. } else {
  672. GELOGW("Load tbe kernels failed");
  673. }
  674. }
  675. cur_model->SetTBEKernelStore(kernel_store);
  676. return SUCCESS;
  677. }
  678. Status ModelHelper::LoadCustAICPUKernelStore(OmFileLoadHelper &om_load_helper) {
  679. // Load cust aicpu kernels
  680. ModelPartition partition_kernel_def;
  681. CustAICPUKernelStore kernel_store;
  682. if (om_load_helper.GetModelPartition(ModelPartitionType::CUST_AICPU_KERNELS, partition_kernel_def) == SUCCESS) {
  683. GELOGD("Kernels partition size:%u", partition_kernel_def.size);
  684. if (kernel_store.Load(partition_kernel_def.data, partition_kernel_def.size)) {
  685. GELOGD("Load cust aicpu kernels success");
  686. } else {
  687. GELOGW("Load cust aicpu kernels failed");
  688. }
  689. }
  690. model_->SetCustAICPUKernelStore(kernel_store);
  691. return SUCCESS;
  692. }
  693. Status ModelHelper::LoadCustAICPUKernelStore(OmFileLoadHelper &om_load_helper,
  694. GeModelPtr &cur_model, size_t mode_index) {
  695. // Load cust aicpu kernels
  696. ModelPartition partition_kernel_def;
  697. CustAICPUKernelStore kernel_store;
  698. if (om_load_helper.GetModelPartition(ModelPartitionType::CUST_AICPU_KERNELS, partition_kernel_def, mode_index)
  699. == SUCCESS) {
  700. GELOGD("Kernels partition size:%u", partition_kernel_def.size);
  701. if (kernel_store.Load(partition_kernel_def.data, partition_kernel_def.size)) {
  702. GELOGD("Load cust aicpu kernels success");
  703. } else {
  704. GELOGW("Load cust aicpu kernels failed");
  705. }
  706. }
  707. cur_model->SetCustAICPUKernelStore(kernel_store);
  708. return SUCCESS;
  709. }
  710. FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY GeModelPtr ModelHelper::GetGeModel() {
  711. if (model_ != nullptr) {
  712. return model_;
  713. }
  714. GELOGD("Model has not been loaded!");
  715. std::shared_ptr<ge::GeModel> out_model = ge::MakeShared<ge::GeModel>();
  716. if (out_model == nullptr) {
  717. return nullptr;
  718. }
  719. return out_model;
  720. }
  721. FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY GeRootModelPtr ModelHelper::GetGeRootModel() {
  722. if (root_model_ != nullptr) {
  723. return root_model_;
  724. }
  725. GELOGD("Model has not been loaded!");
  726. std::shared_ptr<ge::GeRootModel> out_model = ge::MakeShared<ge::GeRootModel>();
  727. if (out_model == nullptr) {
  728. return nullptr;
  729. }
  730. return out_model;
  731. }
  732. Status ModelHelper::ReleaseLocalModelData() noexcept {
  733. Status result = SUCCESS;
  734. if (model_addr_tmp_ != nullptr) {
  735. errno_t ret = memset_s(static_cast<void *>(model_addr_tmp_), model_len_tmp_, 0, model_len_tmp_);
  736. if (ret != EOK) {
  737. GELOGE(FAILED, "Failed to memset memory, error-code %d", ret);
  738. result = FAILED;
  739. }
  740. delete[] model_addr_tmp_;
  741. model_addr_tmp_ = nullptr;
  742. model_len_tmp_ = 0;
  743. }
  744. return result;
  745. }
  746. FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelHelper::GetBaseNameFromFileName(
  747. const string &file_name, string &base_name) {
  748. GELOGD("Get base_name from file, file_name:%s", file_name.c_str());
  749. GE_CHK_BOOL_EXEC_WARN(!file_name.empty(), return FAILED, "File path may not valid, check params --output");
  750. size_t start_position = 0;
  751. // using output as base_name (ignore ".om")
  752. size_t filename_suffixes = 3;
  753. if (file_name.find_last_of('/') != string::npos) {
  754. start_position = file_name.find_last_of('/') + 1;
  755. }
  756. size_t end_position = file_name.length() - filename_suffixes;
  757. base_name = file_name.substr(start_position, end_position - start_position);
  758. GE_CHK_BOOL_EXEC_WARN(!base_name.empty(), return FAILED, "Get base_name failed, check params --output");
  759. return SUCCESS;
  760. }
  761. FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelHelper::GetModelNameFromMergedGraphName(
  762. const string &graph_name, string &model_name) {
  763. GELOGD("Get model_name from graph_name, graph_name:%s", graph_name.c_str());
  764. // this can only be used after merged graph(graph name will be append with "_x", x is index);
  765. GE_CHK_BOOL_EXEC_WARN(!graph_name.empty(), return FAILED, "File path may not valid, check params --output");
  766. size_t start_position = 0;
  767. size_t end_position = graph_name.length();
  768. // using graph as model_name (ignore "_x", x is the index of graph)
  769. if (graph_name.find_last_of('_') != string::npos) {
  770. end_position = graph_name.find_last_of('_');
  771. }
  772. model_name = graph_name.substr(start_position, end_position);
  773. GE_CHK_BOOL_EXEC_WARN(!model_name.empty(), return FAILED, "Get model_name failed, check params --output");
  774. return SUCCESS;
  775. }
  776. } // 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