Browse Source

整改错误日志及上报

pull/1257/head
liudingyan 4 years ago
parent
commit
1fbc4e1fe1
7 changed files with 124 additions and 81 deletions
  1. +2
    -2
      ge/host_cpu_engine/engine/host_cpu_engine.cc
  2. +4
    -3
      ge/host_cpu_engine/ops_kernel_store/host_cpu_ops_kernel_builder.cc
  3. +31
    -17
      ge/init/gelib.cc
  4. +16
    -17
      ge/omm/csa_interact.cc
  5. +15
    -7
      ge/opskernel_manager/ops_kernel_builder_manager.cc
  6. +38
    -17
      ge/opskernel_manager/ops_kernel_manager.cc
  7. +18
    -18
      ge/plugin/engine/engine_manage.cc

+ 2
- 2
ge/host_cpu_engine/engine/host_cpu_engine.cc View File

@@ -34,8 +34,8 @@ Status HostCpuEngine::Initialize(const std::map<string, string> &options) {
if (ops_kernel_store_ == nullptr) {
ops_kernel_store_ = MakeShared<HostCpuOpsKernelInfoStore>();
if (ops_kernel_store_ == nullptr) {
GELOGE(FAILED, "Make HostCpuOpsKernelInfoStore failed.");
RRPORT_INNER_ERROR("E19999", "Initialize FAILED. Make HostCpuOpsKernelInfoStore failed.");
GELOGE(FAILED, "[init][HostCpuEngine] Initialize failed. as malloc shared_ptr failed.");
RRPORT_INNER_ERROR("E19999", "HostCpuEngine Initialize failed. as malloc shared_ptr failed.");
return FAILED;
}
}


+ 4
- 3
ge/host_cpu_engine/ops_kernel_store/host_cpu_ops_kernel_builder.cc View File

@@ -39,7 +39,8 @@ Status HostCpuOpsKernelBuilder::Initialize(const map<std::string, std::string> &
Status HostCpuOpsKernelBuilder::CalcOpRunningParam(Node &ge_node) {
OpDescPtr op_desc = ge_node.GetOpDesc();
if (op_desc == nullptr) {
GELOGE(FAILED, "CalcOpRunningParam failed, as op desc is null");
GELOGE(FAILED, "[get][opdes] OpDesc is nullptr, as input node's op desc is null.");
REPORT_INNER_ERROR("E19999", "CalcOpRunningParam failed, as input param node GetOpDesc return nullptr.");
return FAILED;
}

@@ -74,7 +75,7 @@ Status HostCpuOpsKernelBuilder::CalcOpRunningParam(Node &ge_node) {
if ((TensorUtils::CalcTensorMemSize(output_shape, format, data_type, output_mem_size) != GRAPH_SUCCESS) ||
(output_mem_size < 0)) {
GELOGE(FAILED, "CalcTensorMemSize failed.");
GELOGE(FAILED, "Calc op[%s:%s] out[%zu] mem size failed, mem_size=%ld, format=%s, data_type=%s.",
GELOGE(FAILED, "[Calc][TensorMemSize] op[%s:%s] out[%zu] mem size failed, mem_size=%ld, format=%s, data_type=%s.",
name.c_str(), type.c_str(), i, output_mem_size, TypeUtils::FormatToSerialString(format).c_str(),
TypeUtils::DataTypeToSerialString(data_type).c_str());
REPORT_CALL_ERROR("E19999", "CalcTensorMemSize failed. op[%s:%s] out[%zu] mem size failed, mem_size=%ld, format=%s, data_type=%s.",
@@ -89,7 +90,7 @@ Status HostCpuOpsKernelBuilder::CalcOpRunningParam(Node &ge_node) {
TensorUtils::SetSize(output_tensor, output_mem_size);
if (op_desc->UpdateOutputDesc(static_cast<uint32_t>(i), output_tensor) != GRAPH_SUCCESS) {
GELOGE(FAILED, "UpdateOutputDesc failed.");
GELOGE(FAILED, "Update op[%s:%s] out[%zu] desc failed, format=%s, data_type=%s.", name.c_str(), type.c_str(), i,
GELOGE(FAILED, "[Update][OutputDesc] op[%s:%s] out[%zu] desc failed, format=%s, data_type=%s.", name.c_str(), type.c_str(), i,
TypeUtils::FormatToSerialString(format).c_str(), TypeUtils::DataTypeToSerialString(data_type).c_str());
REPORT_CALL_ERROR("E19999", "UpdateOutputDesc failed. op[%s:%s] out[%zu] desc failed, format=%s, data_type=%s.", name.c_str(), type.c_str(), i,
TypeUtils::FormatToSerialString(format).c_str(), TypeUtils::DataTypeToSerialString(data_type).c_str());


+ 31
- 17
ge/init/gelib.cc View File

@@ -68,7 +68,8 @@ Status GELib::Initialize(const map<string, string> &options) {
// Multiple initializations are not allowed
instancePtr_ = MakeShared<GELib>();
if (instancePtr_ == nullptr) {
GELOGE(GE_CLI_INIT_FAILED, "GeLib initialize failed, malloc shared_ptr failed.");
GELOGE(GE_CLI_INIT_FAILED, "[init][GeLib] failed, as malloc shared_ptr failed.");
REPORT_INNER_ERROR("E19999", "GeLib initialize failed, as malloc shared_ptr failed.");
return GE_CLI_INIT_FAILED;
}

@@ -76,13 +77,15 @@ Status GELib::Initialize(const map<string, string> &options) {
map<string, string> new_options;
Status ret = instancePtr_->SetRTSocVersion(options, new_options);
if (ret != SUCCESS) {
GELOGE(ret, "GeLib initial failed.");
GELOGE(ret, "[set][RTSocVersion] failed, input options invalid.");
REPORT_INNER_ERROR("E19999", "GeLib initial failed, as SetRTSocVersion failed.");
return ret;
}

ret = instancePtr_->SetAiCoreNum(new_options);
if (ret != SUCCESS) {
GELOGE(ret, "GeLib initial: SetAiCoreNum failed.");
GELOGE(ret, "[Set][AiCoreNum] failed, input options invalid.");
REPORT_INNER_ERROR("E19999", "GeLib initial failed, as SetAiCoreNum failed.");
return ret;
}

@@ -97,7 +100,8 @@ Status GELib::Initialize(const map<string, string> &options) {
GE_TIMESTAMP_START(Init);
ret = instancePtr_->InnerInitialize(new_options);
if (ret != SUCCESS) {
GELOGE(ret, "GeLib initial failed.");
GELOGE(ret, "[init][GELib] failed, as options is invalid .");
REPORT_INNER_ERROR("E19999", "GeLib initial failed, as GELib::InnerInitialize failed.");
instancePtr_ = nullptr;
return ret;
}
@@ -118,7 +122,8 @@ Status GELib::InnerInitialize(const map<string, string> &options) {
Status initSystemStatus = SystemInitialize(options);
GE_TIMESTAMP_END(SystemInitialize, "InnerInitialize::SystemInitialize");
if (initSystemStatus != SUCCESS) {
GELOGE(initSystemStatus, "GE system initial failed.");
GELOGE(initSystemStatus, "[init][system]failed, as options invalid.");
REPORT_INNER_ERROR("E19999", "GE system initial failed, as SystemInitialize failed.");
RollbackInit();
return initSystemStatus;
}
@@ -129,7 +134,8 @@ Status GELib::InnerInitialize(const map<string, string> &options) {
Status initEmStatus = engineManager_.Initialize(options);
GE_TIMESTAMP_END(EngineInitialize, "InnerInitialize::EngineInitialize");
if (initEmStatus != SUCCESS) {
GELOGE(initEmStatus, "GE engine manager initial failed.");
GELOGE(initEmStatus, "[init][enginemanager] failed, options invalid. ");
REPORT_INNER_ERROR("E19999", "GE engine manager initial failed, as call DNNEngineManager::Initialize failed. ");
RollbackInit();
return initEmStatus;
}
@@ -140,7 +146,8 @@ Status GELib::InnerInitialize(const map<string, string> &options) {
Status initOpsStatus = opsManager_.Initialize(options);
GE_TIMESTAMP_END(OpsManagerInitialize, "InnerInitialize::OpsManagerInitialize");
if (initOpsStatus != SUCCESS) {
GELOGE(initOpsStatus, "GE ops manager initial failed.");
GELOGE(initOpsStatus, "[init][OpsKernelManager] failed, options invalid. ");
REPORT_INNER_ERROR("E19999", "GE ops manager initial failed. as call OpsKernelManager::Initialize failed. ");
RollbackInit();
return initOpsStatus;
}
@@ -151,7 +158,8 @@ Status GELib::InnerInitialize(const map<string, string> &options) {
Status initOpsBuilderStatus = OpsKernelBuilderManager::Instance().Initialize(options);
GE_TIMESTAMP_END(OpsKernelBuilderManagerInitialize, "InnerInitialize::OpsKernelBuilderManager");
if (initOpsBuilderStatus != SUCCESS) {
GELOGE(initOpsBuilderStatus, "GE ops builder manager initial failed.");
GELOGE(initOpsBuilderStatus, "[init][OpsKernelBuilderManager] failed, options invalid.");
REPORT_INNER_ERROR("E19999", "GE ops buildermanager initial failed, as call OpsKernelBuilderManager::Initialize failed.");
RollbackInit();
return initOpsBuilderStatus;
}
@@ -162,7 +170,8 @@ Status GELib::InnerInitialize(const map<string, string> &options) {
Status initSmStatus = sessionManager_.Initialize(options);
GE_TIMESTAMP_END(SessionManagerInitialize, "InnerInitialize::SessionManagerInitialize");
if (initSmStatus != SUCCESS) {
GELOGE(initSmStatus, "GE session manager initial failed.");
GELOGE(initSmStatus, "[init][SessionManager] failed, options invalid.");
REPORT_INNER_ERROR("E19999", "GE session manager initial failed. as SessionManager::Initialize failed.");
RollbackInit();
return initSmStatus;
}
@@ -205,7 +214,8 @@ Status GELib::SystemInitialize(const map<string, string> &options) {
auto model_manager = ModelManager::GetInstance();
GE_CHECK_NOTNULL(model_manager);
GE_IF_BOOL_EXEC(model_manager->EnableExceptionDump(options) != SUCCESS,
GELOGE(FAILED, "Enable exception dump failed");
REPORT_INNER_ERROR("E19999", "ModelManager EnableExceptionDump failed, options invalid.");
GELOGE(FAILED, "[check][ModelManager][EnableExceptionDump] failed, options invalid.");
return FAILED);
// 1.`is_train_mode_` means case: train
// 2.`(!is_train_mode_) && (options_.device_id != kDefaultDeviceIdForInfer)` means case: online infer
@@ -259,7 +269,9 @@ Status GELib::SetRTSocVersion(const map<string, string> &options, map<string, st
GELOGI("SOC_VERSION is not exist in options");
char version[kSocVersionLen] = {0};
rtError_t rt_ret = rtGetSocVersion(version, kSocVersionLen);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE, GELOGE(rt_ret, "rtGetSocVersion failed"); return FAILED;)
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE,
REPORT_CALL_ERROR("E19999","rtGetSocVersion failed.");
GELOGE(rt_ret, "[get][SocVersion] failed."); return FAILED;)
GELOGI("Succeeded in getting SOC_VERSION[%s] from runtime.", version);
new_options.insert(std::make_pair(ge::SOC_VERSION, version));
}
@@ -280,7 +292,8 @@ Status GELib::SetAiCoreNum(map<string, string> &options) {
options.emplace(std::make_pair(AICORE_NUM, std::to_string(aicore_num)));
return SUCCESS;
}
GELOGE(FAILED, "rtGetAiCoreCount failed.");
GELOGE(FAILED, "[get][AiCoreCount] failed.");
REPORT_CALL_ERROR("E19999", "rtGetAiCoreCount failed.");
return FAILED;
}

@@ -355,7 +368,7 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status GELib::InitSystemWithOpt
mem_type.push_back(RT_MEMORY_P2P_DDR);
Status initMmStatus = MemManager::Instance().Initialize(mem_type);
if (initMmStatus != SUCCESS) {
GELOGE(initMmStatus, "[Initialize] MemoryAllocatorManager initialize failed.");
GELOGE(initMmStatus, "[Init][MemManager] MemoryAllocatorManager initialize failed.");
return initMmStatus;
}

@@ -363,7 +376,7 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status GELib::InitSystemWithOpt
// Update CSA file
CsaInteract::GetInstance().Init(options.device_id, GetContext().TraceId());
Status ret = CsaInteract::GetInstance().WriteJobState(JOBSTATE_RUNNING, JOBSUBSTATE_ENV_INIT);
GE_LOGE_IF(ret != SUCCESS, "write job state failed, ret:%u", ret);
GE_LOGE_IF(ret != SUCCESS, "[write] [job state] failed, ret:%u", ret);

// set device id
GELOGI("set logical device id:%u", options.device_id);
@@ -394,7 +407,7 @@ Status GELib::SystemShutdownWithOptions(const Options &options) {

// Update CSA file
Status ret = CsaInteract::GetInstance().WriteJobState(JOBSTATE_SUCCEED);
GE_LOGE_IF(ret != SUCCESS, "write job state failed, ret:%u", ret);
GE_LOGE_IF(ret != SUCCESS, "[write] [job state] failed, ret:%u", ret);

is_system_inited = false;
is_shutdown = true;
@@ -410,7 +423,7 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status GELib::InitSystemWithout
mem_type.push_back(RT_MEMORY_P2P_DDR);
Status initMmStatus = MemManager::Instance().Initialize(mem_type);
if (initMmStatus != SUCCESS) {
GELOGE(initMmStatus, "[Initialize] MemoryAllocatorManager initialize failed.");
GELOGE(initMmStatus, "[Initialize][MemManager] MemoryAllocatorManager initialize failed.");
return initMmStatus;
}
GE_CHK_STATUS_RET(HostMemManager::Instance().Initialize());
@@ -506,7 +519,8 @@ Status GELib::Finalize() {
instancePtr_ = nullptr;
init_flag_ = false;
if (final_state != SUCCESS) {
GELOGE(FAILED, "finalization failed.");
GELOGE(FAILED, "[check][state]finalization failed.");
REPORT_INNER_ERROR("E19999", "GELib::Finalize finalization failed.");
return final_state;
}
GELOGI("finalization success.");


+ 16
- 17
ge/omm/csa_interact.cc View File

@@ -78,8 +78,8 @@ void CsaInteract::Init(int32_t dev_index, int64_t job_id) {
Status CsaInteract::WriteJobState(JobState job_state, JobSubState job_sub_state, uint32_t module_ret_errcode,
ErrorModule error_module) {
if (!is_init_) {
GELOGE(INTERNAL_ERROR, "CsaInteract has not init, can't WriteJobState");
REPORT_INNER_ERROR("E19999", "CsaInteract has not init, can't WriteJobState");
GELOGE(INTERNAL_ERROR, "[init][CsaInteract] obj has not init, can't WriteJobState");
REPORT_INNER_ERROR("E19999", "CsaInteracthas not init, can't WriteJobState");
return INTERNAL_ERROR;
}
if ((curr_state_ == JOBSTATE_FAILED) || (curr_state_ == JOBSTATE_KILLED)) {
@@ -108,9 +108,8 @@ Status CsaInteract::WriteJobState(JobState job_state, JobSubState job_sub_state,

content = content_json.dump();
} catch (const nlohmann::json::exception &e) {
GELOGE(INTERNAL_ERROR, "construct json object failed.");
GELOGE(INTERNAL_ERROR, "build jobstate content json string failed, exception:%s job_state:%u", e.what(), job_state);
REPORT_INNER_ERROR("E19999", "construct json object failed. exception:%s job_state:%u", e.what(), job_state);
GELOGE(INTERNAL_ERROR, "[create] [json object] exception:%s job_state:%u", e.what(), job_state);
REPORT_INNER_ERROR("E19999", "create json object failed. exception:%s job_state:%u", e.what(), job_state);
return INTERNAL_ERROR;
}

@@ -171,8 +170,8 @@ void CsaInteract::WriteInternalErrorCode() {
///
Status CsaInteract::WriteHcomDetection(const std::string &content) {
if (!is_init_) {
GELOGE(INTERNAL_ERROR, "CsaInteract has not init, can't WriteJobState");
REPORT_INNER_ERROR("E19999", "WriteHcomDetection failed. CsaInteract has not init, can't WriteJobState");
GELOGE(INTERNAL_ERROR, "[init][CsaInteract] obj has not init,can't WriteJobState");
REPORT_INNER_ERROR("E19999", "CsaInteracthas not init, can't WriteJobState");
return INTERNAL_ERROR;
}

@@ -197,15 +196,15 @@ Status CsaInteract::WriteFile(const std::string &file_name, const std::string &c
if (fd == EN_ERROR) {
if (MakePath(file_name) != SUCCESS) {
GELOGE(INTERNAL_ERROR, "MakePath failed.");
GELOGE(INTERNAL_ERROR, "csainteract create file path fail, errno is %d", errno);
REPORT_CALL_ERROR("E19999", "MakePath failed. create file path fail, errno is %d", errno);
GELOGE(INTERNAL_ERROR, "[create] [file path] errno is %d", errno);
REPORT_CALL_ERROR("E19999", "MakePath failed. errno is %d", errno);
return INTERNAL_ERROR;
}
fd = mmOpen2(file_name.c_str(), flags, M_IRUSR | M_IWUSR | M_UMASK_GRPREAD);
if (fd == EN_ERROR) {
GELOGE(INTERNAL_ERROR, "mmOpen2 failed.");
GELOGE(INTERNAL_ERROR, "open file fail, errno is %d", errno);
REPORT_CALL_ERROR("E19999", "mmOpen2 failed. open file fail, errno is %d", errno);
GELOGE(INTERNAL_ERROR, "[open] [file] errno is %d", errno);
REPORT_CALL_ERROR("E19999", "mmOpen2 failed. errno is %d", errno);
return INTERNAL_ERROR;
}
}
@@ -213,21 +212,21 @@ Status CsaInteract::WriteFile(const std::string &file_name, const std::string &c
mmSsize_t ret = mmWrite(fd, reinterpret_cast<void *>(const_cast<char *>(content.c_str())), content.length());
if (ret == EN_ERROR) {
GELOGE(INTERNAL_ERROR, "mmWrite failed.");
GELOGE(INTERNAL_ERROR, "write file fail, errno is %d", errno);
REPORT_CALL_ERROR("E19999", "mmWrite failed. write file fail, errno is %d", errno);
GELOGE(INTERNAL_ERROR, "[write] [file] errno is %d", errno);
REPORT_CALL_ERROR("E19999", "mmWrite failed. errno is %d", errno);
ret = mmClose(fd);
if (ret == EN_ERROR) {
GELOGE(INTERNAL_ERROR, "mmClose failed.");
GELOGE(INTERNAL_ERROR, "close file fail, error is %d", errno);
REPORT_CALL_ERROR("E19999", "mmClose failed. close file fail, error is %d", errno);
GELOGE(INTERNAL_ERROR, "[close] [file] error is %d", errno);
REPORT_CALL_ERROR("E19999", "mmClose failed. error is %d", errno);
}
return INTERNAL_ERROR;
}
ret = mmClose(fd);
if (ret == EN_ERROR) {
GELOGE(INTERNAL_ERROR, "mmClose failed.");
GELOGE(INTERNAL_ERROR, "close file fail, error is %d", errno);
REPORT_CALL_ERROR("E19999", "mmClose failed. close file fail, error is %d", errno);
GELOGE(INTERNAL_ERROR, "[close] [file] error is %d", errno);
REPORT_CALL_ERROR("E19999", "mmClose failed. error is %d", errno);
return INTERNAL_ERROR;
}



+ 15
- 7
ge/opskernel_manager/ops_kernel_builder_manager.cc View File

@@ -50,7 +50,8 @@ Status OpsKernelBuilderManager::Initialize(const map<std::string, std::string> &
GE_CHK_STATUS_RET_NOLOG(GetLibPaths(options, lib_paths));
plugin_manager_.reset(new (std::nothrow)PluginManager());
GE_CHECK_NOTNULL(plugin_manager_);
GE_CHK_STATUS_RET(plugin_manager_->LoadSo(lib_paths), "Failed to load libs");
GE_CHK_STATUS_RET(plugin_manager_->LoadSo(lib_paths),
"[load][libs]Failed, lib_paths=%s.", lib_paths.c_str());
}

auto &kernel_builders = OpsKernelBuilderRegistry::GetInstance().GetAll();
@@ -61,7 +62,7 @@ Status OpsKernelBuilderManager::Initialize(const map<std::string, std::string> &
GELOGI("Initialize ops kernel util for %s", kernel_lib_name.c_str());
GE_CHECK_NOTNULL(it.second);
GE_CHK_STATUS_RET(it.second->Initialize(options),
"Failed to invoke Initialize, kernel lib name = %s",
"[invoke][Initialize]failed, kernel lib name = %s",
kernel_lib_name.c_str());

ops_kernel_builders_.emplace(kernel_lib_name, it.second);
@@ -128,8 +129,11 @@ Status OpsKernelBuilderManager::CalcOpRunningParam(Node &node) const {
const std::string &lib_name = op_desc->GetOpKernelLibName();
auto it = ops_kernel_builders_.find(lib_name);
if (it == ops_kernel_builders_.end()) {
GELOGE(INTERNAL_ERROR,
"Failed to get OpKernelStore. libName = %s, node = %s",
GELOGE(INTERNAL_ERROR, "OpKernelLibName failed.");
GELOGE(INTERNAL_ERROR,"[Check][lib_name] libName = %s, node = %s not exist.",
lib_name.c_str(),
op_desc->GetName().c_str());
REPORT_INNER_ERROR("E19999", "CalcOpRunningParam failed, libName = %s, node = %s not exist.",
lib_name.c_str(),
op_desc->GetName().c_str());
return INTERNAL_ERROR;
@@ -137,7 +141,7 @@ Status OpsKernelBuilderManager::CalcOpRunningParam(Node &node) const {

GELOGD("To invoke CalcOpRunningParam, node = %s, lib name = %s", op_desc->GetName().c_str(), lib_name.c_str());
GE_CHK_STATUS_RET(it->second->CalcOpRunningParam(node),
"Failed to invoke CalcOpRunningParam, libName = %s, node = %s",
"[invoke] [CalcOpRunningParam]failed, libName = %s, node = %s",
lib_name.c_str(),
op_desc->GetName().c_str());
GELOGD("Done invoking CalcOpRunningParam successfully");
@@ -152,8 +156,12 @@ Status OpsKernelBuilderManager::GenerateTask(const Node &node,
const std::string &lib_name = op_desc->GetOpKernelLibName();
auto it = ops_kernel_builders_.find(lib_name);
if (it == ops_kernel_builders_.end()) {
GELOGE(INTERNAL_ERROR, "GenerateTask failed.");
GELOGE(INTERNAL_ERROR,
"Failed to get OpKernelStore. libName = %s, node = %s",
"[Check][lib_name] libName = %s, node = %s not exist.",
lib_name.c_str(),
op_desc->GetName().c_str());
REPORT_INNER_ERROR("E19999", "GenerateTask failed, libName = %s, node = %s not exist.",
lib_name.c_str(),
op_desc->GetName().c_str());
return INTERNAL_ERROR;
@@ -161,7 +169,7 @@ Status OpsKernelBuilderManager::GenerateTask(const Node &node,

GELOGD("To invoke GenerateTask, node = %s, lib name = %s", op_desc->GetName().c_str(), lib_name.c_str());
GE_CHK_STATUS_RET(it->second->GenerateTask(node, context, tasks),
"Failed to invoke GenerateTask, libName = %s, node = %s",
"[invoke][GenerateTask]failed, libName = %s, node = %s",
lib_name.c_str(),
op_desc->GetName().c_str());
GELOGD("Done invoking GenerateTask successfully");


+ 38
- 17
ge/opskernel_manager/ops_kernel_manager.cc View File

@@ -56,7 +56,8 @@ Status OpsKernelManager::Initialize(const map<string, string> &options_const) {
std::map<string, string> options(options_const);
Status ret = InitPluginOptions(options);
if (ret != SUCCESS) {
GELOGE(ret, "[OpsKernelManager] [Initialize] parse pluginFlag from ge options failed.");
GELOGE(ret, "[Init][PluginOptions] parse pluginFlag from ge options failed.");
REPORT_CALL_ERROR("E19999", "InitPluginOptions failed, options invalid.");
return ret;
}

@@ -85,7 +86,8 @@ Status OpsKernelManager::Initialize(const map<string, string> &options_const) {
initialize_ = options;
Status rst0 = plugin_manager_.InvokeAll<map<string, string> &, Status>(kInitialize, initialize_);
if (rst0 == FAILED) {
GELOGE(GE_OPS_GET_NO_VALID_SO, "There is invalid so about OpsKernelInfoStore.");
GELOGE(GE_OPS_GET_NO_VALID_SO, "[invoke][opskernelinfo]PluginManager InvokeAll failed.");
REPORT_INNER_ERROR("E19999", "PluginManager InvokeAll failed.");
return GE_OPS_GET_NO_VALID_SO;
}
Status rst1 =
@@ -114,13 +116,15 @@ Status OpsKernelManager::Initialize(const map<string, string> &options_const) {
}
ret = InitGraphOptimizerPriority();
if ((ret != SUCCESS)) {
GELOGE(ret, "Init graph optimizer priority failed.");
GELOGE(ret, "[Init][GraphOptimizerPriority] failed.");
REPORT_CALL_ERROR("E19999", "GraphOptimizerPriority failed.");
return ret;
}
init_flag_ = true;
return SUCCESS;
} else {
GELOGE(ret, "Failed to find any valid so file.");
GELOGE(ret, "[check][so file] not find any valid so file.");
REPORT_INNER_ERROR("E19999", "OpsKernelManager::Initialize failed, not find any valid so file.");
return ret;
}
}
@@ -175,20 +179,32 @@ Status OpsKernelManager::ParsePluginOptions(const map<string, string> &options,
} else if (flag == 1) {
enable_flag = true;
} else {
GELOGE(GE_GRAPH_OPTIONS_INVALID, "option_key:%s, its value %s is invalid, it must be 0 or 1.",
GELOGE(GE_GRAPH_OPTIONS_INVALID, "ParsePluginOptions failed.");
GELOGE(GE_GRAPH_OPTIONS_INVALID, "[check][data]option_key:%s, its value %s is invalid, it must be 0 or 1.",
plugin_name.c_str(), iter->second.c_str());
REPORT_INNER_ERROR("E19999", "ParsePluginOptions failed, option_key:%s, its value %s is invalid, it must be 0 or 1.",
plugin_name.c_str(), iter->second.c_str());
return GE_GRAPH_OPTIONS_INVALID;
}
} catch (std::invalid_argument &) {
GELOGE(GE_GRAPH_OPTIONS_INVALID, "option_key:ge.feFlag, its value %s is invalid_argument, it must be 0 or 1.",
GELOGE(GE_GRAPH_OPTIONS_INVALID, "ParsePluginOptions failed.");
GELOGE(GE_GRAPH_OPTIONS_INVALID, "[check][data]option_key:ge.feFlag, its value %s is invalid_argument, it must be 0 or 1.",
iter->second.c_str());
REPORT_INNER_ERROR("E19999", "ParsePluginOptions failed,its value %s is invalid_argument, it must be 0 or 1.",
iter->second.c_str());
return GE_GRAPH_OPTIONS_INVALID;
} catch (std::out_of_range &) {
GELOGE(GE_GRAPH_OPTIONS_INVALID, "option_key:ge.feFlag, its value %s is out of range, it must be 0 or 1.",
GELOGE(GE_GRAPH_OPTIONS_INVALID, "ParsePluginOptions failed.");
GELOGE(GE_GRAPH_OPTIONS_INVALID, "[check][data]option_key:ge.feFlag, its value %s is out of range, it must be 0 or 1.",
iter->second.c_str());
REPORT_INNER_ERROR("E19999", "ParsePluginOptions failed,its value %s is out of range, it must be 0 or 1.",
iter->second.c_str());
return GE_GRAPH_OPTIONS_INVALID;
} catch (...) {
GELOGE(GE_GRAPH_OPTIONS_INVALID, "option_key:%s, its value %s is invalid, it must be 0 or 1.",
GELOGE(GE_GRAPH_OPTIONS_INVALID, "ParsePluginOptions failed.");
GELOGE(GE_GRAPH_OPTIONS_INVALID, "[check][data]option_key:%s, its value %s is invalid, it must be 0 or 1.",
plugin_name.c_str(), iter->second.c_str());
REPORT_INNER_ERROR("E19999", "ParsePluginOptions failed,option_key:%s, its value %s is invalid, it must be 0 or 1.",
plugin_name.c_str(), iter->second.c_str());
return GE_GRAPH_OPTIONS_INVALID;
}
@@ -203,13 +219,13 @@ Status OpsKernelManager::ParsePluginOptions(const map<string, string> &options,
Status OpsKernelManager::CheckPluginPtr() const {
for (auto iter = ops_kernel_store_.begin(); iter != ops_kernel_store_.end(); ++iter) {
if (iter->second == nullptr) {
GELOGE(INTERNAL_ERROR, "CheckPluginPtr OpsKernelInfoStorePtr is null");
GELOGE(INTERNAL_ERROR, "[Check][PluginPtr] OpsKernelInfoStorePtr key=%s is null", iter->first.c_str());
return FAILED;
}
}
for (auto iter1 = graph_optimizers_.begin(); iter1 != graph_optimizers_.end(); ++iter1) {
if (iter1->second == nullptr) {
GELOGE(INTERNAL_ERROR, "CheckPluginPtr GraphOptimizerPtr is null");
GELOGE(INTERNAL_ERROR, "[Check][PluginPtr] GraphOptimizerPtr key=%s is null", iter1->first.c_str());
return FAILED;
}
}
@@ -222,7 +238,7 @@ Status OpsKernelManager::InitOpKernelInfoStores(const map<string, string> &optio
GELOGI("OpKernelInfoStore name: %s.", (it.first).c_str());
Status ret = it.second->Initialize(options);
if (ret != SUCCESS) {
GELOGE(GE_OPS_KERNEL_STORE_INIT_FAILED, "OpKernelInfoStore: %s initialize failed.", (it.first).c_str());
GELOGE(GE_OPS_KERNEL_STORE_INIT_FAILED, "[init][opKernelLib]OpKernelInfoStore: %s initialize failed.", (it.first).c_str());
return GE_OPS_KERNEL_STORE_INIT_FAILED;
}
}
@@ -247,7 +263,8 @@ void OpsKernelManager::InitOpsKernelInfo() {
}
std::shared_ptr<GELib> instance_ptr = ge::GELib::GetInstance();
if (instance_ptr == nullptr) {
GELOGE(GE_CLI_GE_NOT_INITIALIZED, "InitOpsKernelInfo failed.");
GELOGE(GE_CLI_GE_NOT_INITIALIZED, "[get][GELib]malloc instance_ptr failed.");
REPORT_INNER_ERROR("E19999", "GELib::GetInstance failed.");
return;
}
// sort opinfo of ops_kernel_info_
@@ -291,7 +308,8 @@ Status OpsKernelManager::InitGraphOptimzers(const map<string, string> &options)
GE_CHK_STATUS_RET(it.second->GetAttributes(attrs))
std::shared_ptr<GELib> instance_ptr = ge::GELib::GetInstance();
if (instance_ptr == nullptr) {
GELOGE(GE_CLI_GE_NOT_INITIALIZED, "InitGraphOptimzers failed.");
GELOGE(GE_CLI_GE_NOT_INITIALIZED, "[get][GELib]malloc instance_ptr failed.");
REPORT_INNER_ERROR("E19999", "GELib::GetInstance failed.");
return GE_CLI_GE_NOT_INITIALIZED;
}
if (!instance_ptr->DNNEngineManagerObj().IsEngineRegistered(attrs.engineName)) {
@@ -300,7 +318,8 @@ Status OpsKernelManager::InitGraphOptimzers(const map<string, string> &options)
}
Status ret = it.second->Initialize(options);
if (ret != SUCCESS) {
GELOGE(GE_OPS_GRAPH_OPTIMIZER_INIT_FAILED, "GraphOptimzer: %s initialize failed.", (it.first).c_str());
GELOGE(GE_OPS_GRAPH_OPTIMIZER_INIT_FAILED, "[init][GraphOptimzer]GraphOptimzer: %s initialize failed.", (it.first).c_str());
REPORT_INNER_ERROR("E19999", "InitGraphOptimzers failed. %s initialize failed.", (it.first).c_str());
return GE_OPS_GRAPH_OPTIMIZER_INIT_FAILED;
}
}
@@ -317,7 +336,8 @@ Status OpsKernelManager::Finalize() {
GELOGI("OpsKernelStore finalize, name: %s.", (iter->first).c_str());
Status status = iter->second->Finalize();
if (SUCCESS != status) {
GELOGE(status, "OpsKernelStore finalize failed, name: %s.", (iter->first).c_str());
GELOGE(status, "[check][status]OpsKernelStore finalize failed, name: %s.", (iter->first).c_str());
REPORT_INNER_ERROR("E19999", "OpsKernelStore finalize failed, name: %s.", (iter->first).c_str());
return status;
}
}
@@ -325,7 +345,8 @@ Status OpsKernelManager::Finalize() {
GELOGI("GraphOptimzers finalize, name: %s.", (iter->first).c_str());
Status status = iter->second->Finalize();
if (status != SUCCESS) {
GELOGE(status, "GraphOptimzers finalize failed, name: %s.", (iter->first).c_str());
GELOGE(status, "[check][status]GraphOptimzers finalize failed, name: %s.", (iter->first).c_str());
REPORT_INNER_ERROR("E19999", "GraphOptimzers finalize failed, name: %s.", (iter->first).c_str());
return status;
}
}
@@ -443,7 +464,7 @@ Status OpsKernelManager::FinalizeOpsKernel() {
GELOGI("ge invoke ops kernal finalize.");
Status ret = plugin_manager_.InvokeAll<Status>(kFinalize);
if (ret != SUCCESS) {
GELOGE(ret, "[Finalize] invoke Fe finalize failed.");
GELOGE(ret, "[Finalize][check][status] invoke Fe finalize failed.");
return ret;
}



+ 18
- 18
ge/plugin/engine/engine_manage.cc View File

@@ -29,8 +29,8 @@ std::unique_ptr<std::map<std::string, DNNEnginePtr>> EngineManager::engine_map_;

Status EngineManager::RegisterEngine(const std::string &engine_name, DNNEnginePtr engine_ptr) {
if (engine_ptr == nullptr) {
GELOGE(FAILED, "RegisterEngine failed. as input param engine_ptr is nullptr");
REPORT_INNER_ERROR("E19999", "RegisterEngine failed. as input param engine_ptr is nullptr");
GELOGE(FAILED, "[Register][Engine] failed, as input engine_ptr is nullptr");
REPORT_INNER_ERROR("E19999", "RegisterEngine failed, as input engine_ptr is nullptr");
return FAILED;
}

@@ -65,8 +65,8 @@ void RegisterAiCoreEngine() {
DNNEngineAttribute attr_aicore = {ai_core, mem_type_aicore, COST_0, DEVICE, FORMAT_RESERVED, FORMAT_RESERVED};
DNNEnginePtr aicore_engine_ptr = MakeShared<AICoreDNNEngine>(attr_aicore);
if (aicore_engine_ptr == nullptr) {
GELOGE(ge::FAILED, "RegisterAiCoreEngine failed. as make aiCoreEnginePtr failed");
REPORT_INNER_ERROR("E19999", "RegisterAiCoreEngine failed. as make aiCoreEnginePtr failed");
GELOGE(ge::FAILED, "[Register][AiCoreEngine] failed, as malloc shared_ptr failed.");
REPORT_INNER_ERROR("E19999", "RegisterAiCoreEngine failed, as malloc shared_ptr failed.");
return;
}
if (EngineManager::RegisterEngine(ai_core, aicore_engine_ptr) != SUCCESS) {
@@ -82,8 +82,8 @@ void RegisterVectorEngine() {
DEVICE, FORMAT_RESERVED, FORMAT_RESERVED};
DNNEnginePtr vectorcore_engine_ptr = MakeShared<VectorCoreDNNEngine>(attr_vector_core);
if (vectorcore_engine_ptr == nullptr) {
GELOGE(ge::FAILED, "RegisterVectorEngine failed. as make vectorCoreEnginePtr failed");
REPORT_INNER_ERROR("E19999", "RegisterVectorEngine failed. as make vectorCoreEnginePtr failed");
GELOGE(ge::FAILED, "[Register][VectorEngine] failed, as malloc shared_ptr failed.");
REPORT_INNER_ERROR("E19999", "RegisterVectorEngine failed, as malloc shared_ptr failed.");
return;
}
if (EngineManager::RegisterEngine(vector_core, vectorcore_engine_ptr) != SUCCESS) {
@@ -98,8 +98,8 @@ void RegisterAiCpuEngine() {
DNNEngineAttribute attr_aicpu = {vm_aicpu, mem_type_aicpu, COST_3, DEVICE, FORMAT_RESERVED, FORMAT_RESERVED};
DNNEnginePtr vm_engine_ptr = MakeShared<AICpuDNNEngine>(attr_aicpu);
if (vm_engine_ptr == nullptr) {
GELOGE(ge::FAILED, "RegisterAiCpuEngine failed. as make vm_engine_ptr failed");
REPORT_INNER_ERROR("E19999", "RegisterAiCpuEngine failed. as make vm_engine_ptr failed");
GELOGE(ge::FAILED, "[Register][AiCpuEngine] failed, as malloc shared_ptr failed.");
REPORT_INNER_ERROR("E19999", "RegisterAiCpuEngine failed, as malloc shared_ptr failed.");
return;
}
if (EngineManager::RegisterEngine(vm_aicpu, vm_engine_ptr) != SUCCESS) {
@@ -114,8 +114,8 @@ void RegisterAiCpuTFEngine() {
DNNEngineAttribute attr_aicpu_tf = {vm_aicpu_tf, mem_type_aicpu_tf, COST_2, DEVICE, FORMAT_RESERVED, FORMAT_RESERVED};
DNNEnginePtr vm_engine_ptr = MakeShared<AICpuTFDNNEngine>(attr_aicpu_tf);
if (vm_engine_ptr == nullptr) {
GELOGE(ge::FAILED, "RegisterAiCpuTFEngine failed. as make vm_engine_ptr failed");
REPORT_INNER_ERROR("E19999", "RegisterAiCpuTFEngine failed. as make vm_engine_ptr failed");
GELOGE(ge::FAILED, "[Register][AiCpuTFEngine] failed, as malloc shared_ptr failed.");
REPORT_INNER_ERROR("E19999", "RegisterAiCpuTFEngine failed, as malloc shared_ptr failed.");
return;
}
if (EngineManager::RegisterEngine(vm_aicpu_tf, vm_engine_ptr) != SUCCESS) {
@@ -131,8 +131,8 @@ void RegisterGeLocalEngine() {
DNNEngineAttribute attr_ge_local = {vm_ge_local, mem_type_ge_local, COST_9, DEVICE, FORMAT_RESERVED, FORMAT_RESERVED};
DNNEnginePtr ge_local_engine = MakeShared<GeLocalDNNEngine>(attr_ge_local);
if (ge_local_engine == nullptr) {
GELOGE(ge::FAILED, "RegisterGeLocalEngine failed. as make ge_local_engine failed");
REPORT_INNER_ERROR("E19999", "RegisterGeLocalEngine failed. as make ge_local_engine failed");
GELOGE(ge::FAILED, "[Register][GeLocalEngine] failed, as malloc shared_ptr failed.");
REPORT_INNER_ERROR("E19999", "RegisterGeLocalEngine failed, as malloc shared_ptr failed.");
return;
}
if (EngineManager::RegisterEngine(vm_ge_local, ge_local_engine) != SUCCESS) {
@@ -148,8 +148,8 @@ void RegisterHostCpuEngine() {
DNNEngineAttribute attr_host_cpu = {vm_host_cpu, mem_type_host_cpu, COST_10, HOST, FORMAT_RESERVED, FORMAT_RESERVED};
DNNEnginePtr host_cpu_engine = MakeShared<HostCpuDNNEngine>(attr_host_cpu);
if (host_cpu_engine == nullptr) {
GELOGE(ge::FAILED, "RegisterHostCpuEngine failed. as make host_cpu_engine failed");
REPORT_INNER_ERROR("E19999", "RegisterHostCpuEngine failed. as make host_cpu_engine failed");
GELOGE(ge::FAILED, "[Register][HostCpuEngine] failed, as malloc shared_ptr failed.");
REPORT_INNER_ERROR("E19999", "RegisterHostCpuEngine failed, as malloc shared_ptr failed.");
return;
}
if (EngineManager::RegisterEngine(vm_host_cpu, host_cpu_engine) != SUCCESS) {
@@ -164,8 +164,8 @@ void RegisterRtsEngine() {
DNNEngineAttribute attr_rts = {vm_rts, mem_type_rts, COST_1, DEVICE, FORMAT_RESERVED, FORMAT_RESERVED};
DNNEnginePtr rts_engine = MakeShared<RtsDNNEngine>(attr_rts);
if (rts_engine == nullptr) {
GELOGE(ge::FAILED, "RegisterRtsEngine failed. as make rts_engine failed");
REPORT_INNER_ERROR("E19999", "RegisterRtsEngine failed. as make rts_engine failed");
GELOGE(ge::FAILED, "[Register][RtsEngine] failed, as malloc shared_ptr failed.");
REPORT_INNER_ERROR("E19999", "RegisterRtsEngine failed, as malloc shared_ptr failed.");
return;
}
if (EngineManager::RegisterEngine(vm_rts, rts_engine) != SUCCESS) {
@@ -180,8 +180,8 @@ void RegisterHcclEngine() {
DNNEngineAttribute attr_hccl = {dnn_hccl, mem_type_hccl, COST_1, DEVICE, FORMAT_RESERVED, FORMAT_RESERVED};
DNNEnginePtr hccl_engine = MakeShared<HcclDNNEngine>(attr_hccl);
if (hccl_engine == nullptr) {
GELOGE(ge::FAILED, "RegisterHcclEngine failed. as make hccl_engine failed");
REPORT_INNER_ERROR("E19999", "RegisterHcclEngine failed. as make hccl_engine failed");
GELOGE(ge::FAILED, "[Register][HcclEngine] failed, as malloc shared_ptr failed.");
REPORT_INNER_ERROR("E19999", "RegisterHcclEngine failed, as malloc shared_ptr failed.");
return;
}
if (EngineManager::RegisterEngine(dnn_hccl, hccl_engine) != SUCCESS) {


Loading…
Cancel
Save