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.

inner_session.cc 24 kB

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
4 years ago
4 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
5 years ago
4 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
4 years ago
5 years ago
4 years ago
5 years ago
4 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
5 years ago
5 years ago
4 years ago
5 years ago
4 years ago
4 years ago
4 years ago
5 years ago
5 years ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543
  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 "session/inner_session.h"
  17. #include <map>
  18. #include <memory>
  19. #include <vector>
  20. #include "analyzer/analyzer.h"
  21. #include "adx_datadump_server.h"
  22. #include "common/dump/dump_properties.h"
  23. #include "common/dump/dump_manager.h"
  24. #include "framework/common/util.h"
  25. #include "framework/common/debug/ge_log.h"
  26. #include "graph/ge_context.h"
  27. #include "graph/ge_global_options.h"
  28. #include "graph/ge_local_context.h"
  29. #include "graph/common/local_context.h"
  30. #include "graph/load/model_manager/model_manager.h"
  31. #include "graph/manager/graph_var_manager.h"
  32. #include "graph/manager/graph_mem_manager.h"
  33. #include "graph/utils/tensor_adapter.h"
  34. #include "runtime/mem.h"
  35. #include "ir_build/option_utils.h"
  36. namespace ge {
  37. namespace {
  38. const int32_t kDumpStatus = 0;
  39. Status CheckReuseMemoryOption(const std::map<string, string> &options) {
  40. auto iter = options.find(OPTION_EXEC_DISABLE_REUSED_MEMORY);
  41. if (iter != options.end()) {
  42. if (iter->second == "0") {
  43. GELOGD("%s=0, reuse memory is open", OPTION_EXEC_DISABLE_REUSED_MEMORY);
  44. } else if (iter->second == "1") {
  45. GELOGD("%s=1, reuse memory is close", OPTION_EXEC_DISABLE_REUSED_MEMORY);
  46. } else {
  47. GELOGE(PARAM_INVALID, "[CheckReuse][MemoryOption]option %s=%s is invalid",
  48. OPTION_EXEC_DISABLE_REUSED_MEMORY, iter->second.c_str());
  49. REPORT_INNER_ERROR("E19999", "CheckReuseMemoryOption failed because option %s=%s is invalid.",
  50. OPTION_EXEC_DISABLE_REUSED_MEMORY, iter->second.c_str());
  51. return FAILED;
  52. }
  53. }
  54. return SUCCESS;
  55. }
  56. }
  57. static std::mutex mutex_; // BuildGraph and RunGraph use
  58. bool InnerSession::is_dump_server_inited_ = false;
  59. InnerSession::InnerSession(uint64_t session_id, const std::map<string, string> &options)
  60. : init_flag_(false), session_id_(session_id), options_(options) {}
  61. Status InnerSession::Initialize() {
  62. if (init_flag_) {
  63. GELOGW("[InnerSession:%lu] session already initialize.", session_id_);
  64. return SUCCESS;
  65. }
  66. // If the global options and the session options are duplicated, the session options is preferred.
  67. auto all_options = options_;
  68. all_options.insert(GetMutableGlobalOptions().begin(), GetMutableGlobalOptions().end());
  69. Status ret = CheckReuseMemoryOption(all_options);
  70. if (ret != SUCCESS) {
  71. GELOGE(ret, "[CheckReuse][MemoryOption] failed, [InnerSession:%lu].", session_id_);
  72. REPORT_CALL_ERROR("E19999", "CheckReuseMemoryOption failed, InnerSession=%lu.", session_id_);
  73. return ret;
  74. }
  75. //Check option OP_PRECISION_MODE
  76. auto iter = all_options.find(ge::OP_PRECISION_MODE);
  77. if (iter != all_options.end() && !iter->second.empty() && !ge::CheckInputPathValid(iter->second)) {
  78. REPORT_INPUT_ERROR("E10001", std::vector<std::string>({"parameter", "value", "reason"}),
  79. std::vector<std::string>({ge::OP_PRECISION_MODE, iter->second,
  80. "path is not found"}));
  81. GELOGE(PARAM_INVALID, "[Check][OP_PRECISION_MODE] %s not found", iter->second.c_str());
  82. return FAILED;
  83. }
  84. // Check option modify_mixlist
  85. if (ge::CheckModifyMixlistParamValid(all_options) != ge::SUCCESS) {
  86. return FAILED;
  87. }
  88. UpdateThreadContext(std::map<std::string, std::string>{});
  89. // session device id set here
  90. std::string str_session_device_id;
  91. if (GetContext().GetOption("ge.session_device_id", str_session_device_id) == SUCCESS) {
  92. GELOGI("Option session device id has set, value is %s.", str_session_device_id.c_str());
  93. uint32_t session_device_id = 0;
  94. try {
  95. session_device_id = static_cast<uint32_t>(std::stoi(str_session_device_id.c_str()));
  96. // session device id has priority
  97. GetContext().SetCtxDeviceId(session_device_id);
  98. } catch (std::invalid_argument &) {
  99. GELOGW("session device id %s transform to int failed.", str_session_device_id.c_str());
  100. } catch (std::out_of_range &) {
  101. GELOGW("session device id %s transform to int failed.", str_session_device_id.c_str());
  102. }
  103. }
  104. GE_CHK_RT_RET(rtSetDevice(GetContext().DeviceId()));
  105. DumpProperties dump_properties;
  106. dump_properties.InitByOptions();
  107. GE_CHK_STATUS_RET(AddDumpProperties(dump_properties), "[Add][DumpProperties] failed.");
  108. ret = graph_manager_.Initialize(options_);
  109. if (ret != SUCCESS) {
  110. GELOGE(ret, "[Init][GraphManager] failed, InnerSession:%lu.", session_id_);
  111. REPORT_CALL_ERROR("E19999", "GraphManager initialize failed, InnerSession:%lu.", session_id_);
  112. GE_CHK_STATUS(RemoveDumpProperties(), "[Remove][DumpProperties] failed.");
  113. return ret;
  114. }
  115. ret = VarManager::Instance(session_id_)->SetMemoryMallocSize(all_options);
  116. if (ret != SUCCESS) {
  117. GELOGE(ret, "[Set][MemoryMallocSize] failed.");
  118. REPORT_CALL_ERROR("E19999", "VarManager SetMemoryMallocSize failed, InnerSession:%lu.", session_id_);
  119. (void)graph_manager_.Finalize();
  120. GE_CHK_STATUS(RemoveDumpProperties(), "[Remove][DumpProperties] failed.");
  121. GE_CHK_RT(rtDeviceReset(static_cast<int32_t>(GetContext().DeviceId())));
  122. return ret;
  123. }
  124. int32_t version = static_cast<int32_t>(SessionVersion::ClOUD_VERSION);
  125. const int DEFAULT_DEVICE_ID = 0;
  126. const int DEFAULT_JOB_ID = 0;
  127. ret = VarManager::Instance(session_id_)->Init(version, session_id_, DEFAULT_DEVICE_ID, DEFAULT_JOB_ID);
  128. if (ret != SUCCESS) {
  129. GELOGE(ret, "[Init][VarManager] failed.");
  130. REPORT_CALL_ERROR("E19999", "VarManager init failed, InnerSession:%lu.", session_id_);
  131. GE_CHK_STATUS(RemoveDumpProperties(), "[Remove][DumpProperties] failed.");
  132. }
  133. init_flag_ = true;
  134. return SUCCESS;
  135. }
  136. Status InnerSession::Finalize() {
  137. std::lock_guard<std::mutex> lock(resource_mutex_);
  138. if (!init_flag_) {
  139. GELOGW("[InnerSession:%lu] session does not initialize.", session_id_);
  140. return SUCCESS;
  141. }
  142. UpdateThreadContext(std::map<std::string, std::string>{});
  143. Status ret = graph_manager_.Finalize();
  144. if (ret != SUCCESS) {
  145. // Subsequent code execution is required, so no return is required
  146. GELOGE(ret, "[Finalize][GraphManager] failed, InnerSession:%lu.", session_id_);
  147. REPORT_CALL_ERROR("E19999", "GraphManager Finalize failed, InnerSession:%lu.", session_id_);
  148. }
  149. ModelManager::GetInstance()->DestroyAicpuSession(session_id_);
  150. init_flag_ = false;
  151. // release var memory
  152. GELOGI("VarManager free var memory.");
  153. (void)VarManager::Instance(session_id_)->FreeVarMemory();
  154. for (auto memory_type : MemManager::Instance().GetAllMemoryType()) {
  155. (void)MemManager::Instance().SessionScopeMemInstance(memory_type).Free(session_id_);
  156. }
  157. // release analyzer saved info(Session Level)
  158. Analyzer::GetInstance()->DestroySessionJsonObject(session_id_);
  159. GE_CHK_RT(rtDeviceReset(static_cast<int32_t>(GetContext().DeviceId())));
  160. GE_CHK_STATUS_RET(RemoveDumpProperties(), "[Remove][DumpProperties] failed.");
  161. return ret;
  162. }
  163. Status InnerSession::GetVariable(const std::string &name, Tensor &val) {
  164. UpdateThreadContext(std::map<std::string, std::string>{});
  165. return graph_manager_.GetVariable(name, val);
  166. }
  167. Status InnerSession::AddGraph(uint32_t graph_id, const Graph &graph) {
  168. std::map<std::string, std::string> options;
  169. return AddGraph(graph_id, graph, options);
  170. }
  171. Status InnerSession::AddGraph(uint32_t graph_id, const Graph &graph,
  172. const std::map<std::string, std::string> &options) {
  173. std::lock_guard<std::mutex> lock(resource_mutex_);
  174. if (!init_flag_) {
  175. GELOGE(GE_SESS_INIT_FAILED, "[Add][Graph] failed because GraphManager not init, InnerSession:%lu, graph_id:%u.",
  176. session_id_, graph_id);
  177. REPORT_INNER_ERROR("E19999", "AddGraph failed because GraphManager not init, InnerSession:%lu, graph_id:%u.",
  178. session_id_, graph_id);
  179. return GE_SESS_INIT_FAILED;
  180. }
  181. UpdateThreadContext(options);
  182. Status ret = graph_manager_.AddGraph(graph_id, graph, options, domi::GetContext());
  183. if (ret != SUCCESS) {
  184. GELOGE(ret, "[Add][Graph] failed, InnerSession:%lu graphid: %u.", session_id_, graph_id);
  185. REPORT_CALL_ERROR("E19999", "GraphManager AddGraph failed, InnerSession:%lu graphid: %u.", session_id_, graph_id);
  186. return ret;
  187. }
  188. GELOGI("[InnerSession:%lu] add graph success, graph_id=%u.", session_id_, graph_id);
  189. return SUCCESS;
  190. }
  191. Status InnerSession::AddGraphWithCopy(uint32_t graph_id, const Graph &graph,
  192. const std::map<std::string, std::string> &options) {
  193. std::lock_guard<std::mutex> lock(resource_mutex_);
  194. if (!init_flag_) {
  195. GELOGE(GE_SESS_INIT_FAILED, "[Add][Graph] failed because GraphManager not init, InnerSession:%lu, graph_id:%u.",
  196. session_id_, graph_id);
  197. REPORT_INNER_ERROR("E19999",
  198. "AddGraphWithCopy failed because GraphManager not init, InnerSession:%lu, graph_id:%u.",
  199. session_id_, graph_id);
  200. return GE_SESS_INIT_FAILED;
  201. }
  202. UpdateThreadContext(options);
  203. Status ret = graph_manager_.AddGraphWithCopy(graph_id, graph, options, domi::GetContext());
  204. if (ret != SUCCESS) {
  205. GELOGE(ret, "[Add][Graph] failed, InnerSession:%lu graphid: %u.", session_id_, graph_id);
  206. REPORT_CALL_ERROR("E19999",
  207. "GraphManager AddGraphWithCopy failed, InnerSession:%lu graphid: %u.", session_id_, graph_id);
  208. return ret;
  209. }
  210. GELOGI("[InnerSession:%lu] add graph success, graph_id=%u.", session_id_, graph_id);
  211. return SUCCESS;
  212. }
  213. Status InnerSession::RunGraph(uint32_t graph_id, const std::vector<Tensor> &inputs, std::vector<Tensor> &outputs) {
  214. GELOGI("[InnerSession:%lu] run graph on session, graph_id=%u.", session_id_, graph_id);
  215. if (mutex_.try_lock()) {
  216. std::lock_guard<std::mutex> lock(mutex_, std::adopt_lock);
  217. if (!init_flag_) {
  218. GELOGE(GE_SESS_INIT_FAILED, "[Run][Graph]failed because GraphManager not Init, InnerSession:%lu, graph_id:%u.",
  219. session_id_, graph_id);
  220. REPORT_INNER_ERROR("E19999", "RunGraph failed because GraphManager not Init, InnerSession:%lu, graph_id:%u.",
  221. session_id_, graph_id);
  222. return GE_SESS_INIT_FAILED;
  223. }
  224. UpdateThreadContext(graph_id);
  225. vector<GeTensor> geInputs;
  226. for (auto &item : inputs) {
  227. geInputs.push_back(TensorAdapter::AsGeTensor(item));
  228. }
  229. vector<GeTensor> geOutputs;
  230. Status ret = graph_manager_.RunGraph(graph_id, geInputs, geOutputs, session_id_);
  231. domi::GetContext().out_nodes_map.clear();
  232. domi::GetContext().user_out_nodes.clear();
  233. if (ret != SUCCESS) {
  234. GELOGE(ret, "[Run][Graph]failed, InnerSession:%lu graph_id=%u.", session_id_, graph_id);
  235. REPORT_CALL_ERROR("E19999",
  236. "GraphManager RunGraph failed, InnerSession:%lu graph_id=%u.", session_id_, graph_id);
  237. return ret;
  238. }
  239. outputs.clear();
  240. for (auto &item : geOutputs) {
  241. outputs.push_back(TensorAdapter::AsTensor(item));
  242. }
  243. GELOGI("[InnerSession:%lu] run graph success, graph_id=%u.", session_id_, graph_id);
  244. return SUCCESS;
  245. } else {
  246. GELOGE(GE_SESS_ALREADY_RUNNING, "[Run][Graph]failed, InnerSession:%lu, graph_id=%u.", session_id_, graph_id);
  247. REPORT_INNER_ERROR("E19999",
  248. "RunGraph failed because mutex try_lock false, InnerSession:%lu, graph_id=%u.",
  249. session_id_, graph_id);
  250. return GE_SESS_ALREADY_RUNNING;
  251. }
  252. }
  253. Status InnerSession::RunGraphWithStreamAsync(uint32_t graph_id, rtStream_t stream,
  254. const std::vector<Tensor> &inputs, std::vector<Tensor> &outputs) {
  255. GELOGI("Run graph with stream, session id = %lu, graph id = %u, stream = %p.",
  256. session_id_, graph_id, stream);
  257. if (mutex_.try_lock()) {
  258. std::lock_guard<std::mutex> lock(mutex_, std::adopt_lock);
  259. if (!init_flag_) {
  260. GELOGE(GE_SESS_INIT_FAILED, "[Run][GraphWithStream]failed because GraphManager not Init,"
  261. "session id = %lu, graph id = %u, stream = %p.", session_id_, graph_id, stream);
  262. REPORT_INNER_ERROR("E19999", "RunGraphWithStreamAsync failed because GraphManager not Init,"
  263. "session id = %lu, graph id = %u, stream = %p.", session_id_, graph_id, stream);
  264. return GE_SESS_INIT_FAILED;
  265. }
  266. UpdateThreadContext(graph_id);
  267. vector<GeTensor> ge_inputs;
  268. for (auto &item : inputs) {
  269. ge_inputs.emplace_back(TensorAdapter::AsGeTensor(item));
  270. }
  271. vector<GeTensor> ge_outputs;
  272. for (auto &item : outputs) {
  273. ge_outputs.emplace_back(TensorAdapter::AsGeTensor(item));
  274. }
  275. Status ret = graph_manager_.RunGraphWithStreamAsync(graph_id, stream, session_id_, ge_inputs, ge_outputs);
  276. domi::GetContext().out_nodes_map.clear();
  277. domi::GetContext().user_out_nodes.clear();
  278. if (ret != SUCCESS) {
  279. GELOGE(ret, "[Run][GraphWithStreamAsync]failed,"
  280. "session id = %lu, graph id = %u, stream = %p.", session_id_, graph_id, stream);
  281. REPORT_CALL_ERROR("E19999", "GraphManager RunGrapWithStreamhAsync failed,"
  282. "session id = %lu, graph id = %u, stream = %p.", session_id_, graph_id, stream);
  283. return ret;
  284. }
  285. GELOGI("Run graph with stream success, session id = %lu, graph id = %u, stream = %p.",
  286. session_id_, graph_id, stream);
  287. return SUCCESS;
  288. } else {
  289. GELOGE(GE_SESS_ALREADY_RUNNING, "[Run][GraphWithStreamAsync]failed because mutex try_lock false,"
  290. "session id = %lu, graph id = %u, stream = %p.", session_id_, graph_id, stream);
  291. REPORT_INNER_ERROR("E19999", "[Run][GraphWithStreamAsync]failed failed because mutex try_lock false,"
  292. "session id = %lu, graph id = %u, stream = %p.", session_id_, graph_id, stream);
  293. return GE_SESS_ALREADY_RUNNING;
  294. }
  295. }
  296. Status InnerSession::RemoveGraph(uint32_t graph_id) {
  297. std::lock_guard<std::mutex> lock(resource_mutex_);
  298. if (!init_flag_) {
  299. GELOGE(GE_SESS_INIT_FAILED,
  300. "[Remove][Graph] failed because GraphManager not init, InnerSession:%lu, graph_id=%u.",
  301. session_id_, graph_id);
  302. REPORT_INNER_ERROR("E19999",
  303. "RemoveGraph failed, because GraphManager not init, InnerSession:%lu, graph_id=%u.",
  304. session_id_, graph_id);
  305. return GE_SESS_INIT_FAILED;
  306. }
  307. UpdateThreadContext(graph_id);
  308. Status ret = graph_manager_.RemoveGraph(graph_id);
  309. if (ret != SUCCESS) {
  310. GELOGE(ret, "[Remove][Graph] failed, InnerSession:%lu, graph_id=%u.", session_id_, graph_id);
  311. REPORT_CALL_ERROR("E19999",
  312. "GraphManager RemoveGraph failed, InnerSession:%lu, graph_id=%u.", session_id_, graph_id);
  313. return ret;
  314. }
  315. GELOGI("[InnerSession:%lu] remove graph success, graph_id=%u.", session_id_, graph_id);
  316. return SUCCESS;
  317. }
  318. Status InnerSession::RegisterCallBackFunc(
  319. const std::string &key,
  320. const std::function<Status(uint32_t, const std::map<std::string, ge::Tensor> &)> &callback) {
  321. std::lock_guard<std::mutex> lock(resource_mutex_);
  322. if (!init_flag_) {
  323. GELOGE(GE_SESS_INIT_FAILED,
  324. "[Register][CallBackFunc] failed because GraphManager not initialize, InnerSession:%lu.", session_id_);
  325. REPORT_INNER_ERROR("E19999",
  326. "RegisterCallBackFunc failed because GraphManager not init, InnerSession:%lu.", session_id_);
  327. return GE_SESS_INIT_FAILED;
  328. }
  329. UpdateThreadContext(std::map<std::string, std::string>{});
  330. Status ret = graph_manager_.RegisterCallBackFunc(key, callback);
  331. if (ret != SUCCESS) {
  332. GELOGE(ret, "[Register][CallBackFunc] failed, InnerSession:%lu register %s.", session_id_, key.c_str());
  333. REPORT_CALL_ERROR("E19999",
  334. "GraphManager RegisterCallBackFunc failed, InnerSession:%lu register %s.",
  335. session_id_, key.c_str());
  336. return ret;
  337. }
  338. GELOGI("[InnerSession:%lu] register %s callback function success.", session_id_, key.c_str());
  339. return SUCCESS;
  340. }
  341. Status InnerSession::RegisterCallBackFunc(
  342. const std::string &key,
  343. const std::function<Status(uint32_t, const std::map<AscendString, ge::Tensor> &)> &callback) {
  344. std::lock_guard<std::mutex> lock(resource_mutex_);
  345. if (!init_flag_) {
  346. GELOGE(GE_SESS_INIT_FAILED,
  347. "[Register][CallBackFunc]failed because GraphManager not initialize, InnerSession:%lu.", session_id_);
  348. REPORT_INNER_ERROR("E19999",
  349. "RegisterCallBackFunc failed because GraphManager not initialize, InnerSession:%lu.",
  350. session_id_);
  351. return GE_SESS_INIT_FAILED;
  352. }
  353. UpdateThreadContext(std::map<std::string, std::string>{});
  354. Status ret = graph_manager_.RegisterCallBackFunc(key, callback);
  355. if (ret != SUCCESS) {
  356. GELOGE(ret, "[Register][CallBackFunc] failed, InnerSession:%lu register %s.", session_id_, key.c_str());
  357. REPORT_CALL_ERROR("E19999",
  358. "GraphManager RegisterCallBackFunc failed, InnerSession:%lu register %s.",
  359. session_id_, key.c_str());
  360. return ret;
  361. }
  362. GELOGI("[InnerSession:%lu] register %s callback function success.", session_id_, key.c_str());
  363. return SUCCESS;
  364. }
  365. Status InnerSession::BuildGraph(uint32_t graph_id, const std::vector<InputTensorInfo> &inputs) {
  366. UpdateThreadContext(graph_id);
  367. GELOGI("[InnerSession:%lu] build graph on session, graph_id=%u.", session_id_, graph_id);
  368. std::vector<ge::GeTensor> ge_inputs;
  369. for (auto const &input : inputs) {
  370. std::vector<int64_t> input_dims;
  371. std::transform(input.dims.begin(), input.dims.end(), std::back_inserter(input_dims),
  372. [](int64_t x) -> int64_t { return x; });
  373. GeShape input_shape(input_dims);
  374. GeTensorDesc input_tensor_desc;
  375. input_tensor_desc.SetShape(input_shape);
  376. input_tensor_desc.SetDataType(static_cast<ge::DataType>(input.data_type));
  377. ge_inputs.emplace_back(input_tensor_desc);
  378. }
  379. GeRootModelPtr ge_root_model = nullptr;
  380. Status ret = graph_manager_.BuildGraph(graph_id, ge_inputs, ge_root_model, session_id_, true);
  381. if (ret != SUCCESS) {
  382. GELOGE(ret, "[Build][Graph] failed, InnerSession:%lu graph_id=%u.", session_id_, graph_id);
  383. REPORT_CALL_ERROR("E19999",
  384. "GraphManager BuildGraph failed, InnerSession:%lu graph_id=%u.", session_id_, graph_id);
  385. return ret;
  386. }
  387. GELOGI("[InnerSession:%lu] build graph success, graph_id=%u.", session_id_, graph_id);
  388. return ret;
  389. }
  390. Status InnerSession::BuildGraph(uint32_t graph_id, const std::vector<ge::Tensor> &inputs) {
  391. UpdateThreadContext(graph_id);
  392. GELOGI("[InnerSession:%lu] build graph on session, graph_id=%u.", session_id_, graph_id);
  393. std::vector<ge::GeTensor> ge_inputs;
  394. for (const auto &input : inputs) {
  395. ge_inputs.emplace_back(TensorAdapter::AsGeTensor(input));
  396. }
  397. GeRootModelPtr ge_root_model = nullptr;
  398. Status ret = graph_manager_.BuildGraph(graph_id, ge_inputs, ge_root_model, session_id_, true);
  399. if (ret != SUCCESS) {
  400. GELOGE(ret, "[Build][Graph] failed, InnerSession:%lu graph_id=%u.", session_id_, graph_id);
  401. REPORT_CALL_ERROR("E19999",
  402. "GraphManager BuildGraph failed, InnerSession:%lu graph_id=%u.", session_id_, graph_id);
  403. return ret;
  404. }
  405. GELOGI("[InnerSession:%lu] build graph success, graph_id=%u.", session_id_, graph_id);
  406. return ret;
  407. }
  408. Status InnerSession::RunGraphAsync(uint32_t graph_id, const std::vector<ge::Tensor> &inputs,
  409. RunAsyncCallback callback) {
  410. UpdateThreadContext(graph_id);
  411. GELOGI("[InnerSession:%lu] run graph on session, graph_id=%u.", session_id_, graph_id);
  412. Status ret = graph_manager_.RunGraphAsync(graph_id, inputs, session_id_, callback);
  413. if (ret != SUCCESS) {
  414. GELOGE(ret, "[Run][GraphAsync]failed, InnerSession:%lu graph_id=%u.", session_id_, graph_id);
  415. REPORT_CALL_ERROR("E19999",
  416. "GraphManager RunGraphAsync failed, InnerSession:%lu graph_id=%u.", session_id_, graph_id);
  417. return ret;
  418. }
  419. GELOGI("[InnerSession:%lu] run graph success, graph_id=%u.", session_id_, graph_id);
  420. return ret;
  421. }
  422. const GraphManager &InnerSession::getGraphManagerObj() const { return graph_manager_; }
  423. void InnerSession::UpdateThreadContext(const std::map<std::string, std::string> &options) {
  424. GetThreadLocalContext().SetGlobalOption(GetMutableGlobalOptions());
  425. GetThreadLocalContext().SetSessionOption(options_);
  426. GetThreadLocalContext().SetGraphOption(options);
  427. GetContext().SetSessionId(session_id_);
  428. SetRtSocVersion();
  429. }
  430. void InnerSession::UpdateThreadContext(uint32_t graph_id) {
  431. auto options = graph_manager_.GetGraphOptions(graph_id);
  432. if (options == nullptr) {
  433. GELOGW("graph level options is null.");
  434. UpdateThreadContext(std::map<std::string, std::string>{});
  435. } else {
  436. UpdateThreadContext(*options);
  437. }
  438. }
  439. bool InnerSession::IsGraphNeedRebuild(uint32_t graph_id) {
  440. UpdateThreadContext(graph_id);
  441. return graph_manager_.IsGraphNeedRebuild(graph_id);
  442. }
  443. Status InnerSession::GetAllVariables(std::map<std::string, GeTensorDesc> &all_variables) {
  444. return VarManager::Instance(session_id_)->GetAllVariables(all_variables);
  445. }
  446. Status InnerSession::GenCheckPointGraph(const std::map<std::string, GeTensorDesc> &all_variables, Graph &graph) {
  447. return graph_manager_.GenCheckPointGraph(all_variables, graph);
  448. }
  449. Status InnerSession::SaveVariables(const Graph &graph, const std::vector<std::string> &var_names,
  450. const std::vector<Tensor> &outputs, std::vector<Tensor> &var_values) {
  451. return graph_manager_.SaveVariables(graph, var_names, outputs, var_values);
  452. }
  453. Status InnerSession::AddDumpProperties(const DumpProperties &dump_properties) {
  454. if (!is_dump_server_inited_) {
  455. if (dump_properties.IsDumpOpen() || dump_properties.IsOpDebugOpen()) {
  456. GE_IF_BOOL_EXEC(AdxDataDumpServerInit() != kDumpStatus,
  457. GELOGE(PARAM_INVALID, "[Init][AdxDataDumpServer] failed, session_id:%lu.", session_id_);
  458. return PARAM_INVALID)
  459. GELOGI("Init adx data dump server success");
  460. is_dump_server_inited_ = true;
  461. }
  462. }
  463. DumpManager::GetInstance().AddDumpProperties(session_id_, dump_properties);
  464. return SUCCESS;
  465. }
  466. Status InnerSession::RemoveDumpProperties() {
  467. DumpManager::GetInstance().RemoveDumpProperties(session_id_);
  468. if (is_dump_server_inited_ && DumpManager::GetInstance().GetDumpPropertiesMap().empty()) {
  469. GE_IF_BOOL_EXEC(AdxDataDumpServerUnInit() != kDumpStatus,
  470. GELOGE(PARAM_INVALID, "[UnInit][AdxDataDumpServer] failed, session_id:%lu.", session_id_);
  471. REPORT_INNER_ERROR("E19999", "RemoveDumpProperties failed because AdxDataDumpServerUnInit failed,"
  472. "session_id:%lu", session_id_);
  473. return PARAM_INVALID)
  474. GELOGI("UnInit adx data dump server success");
  475. is_dump_server_inited_ = false;
  476. }
  477. return SUCCESS;
  478. }
  479. void InnerSession::SetRtSocVersion() {
  480. const auto &global_options = GetMutableGlobalOptions();
  481. auto it = global_options.find(ge::SOC_VERSION);
  482. if (it != global_options.end()) {
  483. const char *soc_version = it->second.c_str();
  484. rtError_t rt_ret = rtSetSocVersion(soc_version);
  485. if (rt_ret != RT_ERROR_NONE) {
  486. GELOGW("Set soc version %s failed. ret:0x%X", soc_version, rt_ret);
  487. }
  488. GELOGI("Set soc version %s success.", soc_version);
  489. }
  490. }
  491. } // 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