support error_manager used in train

CMakeLists.txt

@@ -76,9 +76,7 @@ if (ENABLE_OPEN_SRC)
         find_module(runtime libruntime.so ${GE_LIB_PATH})
         find_module(runtime_compile libruntime_compile.so ${GE_LIB_PATH})
         find_module(resource libresource.so ${GE_LIB_PATH})
-        find_module(error_manager liberror_manager.so ${GE_LIB_PATH})
         find_module(ascend_hal_stub libascend_hal.so ${GE_LIB_PATH})
-        find_module(error_manager_static liberror_manager.a ${GE_LIB_PATH})
         find_module(msprofiler_fwk_ext libmsprofiler_fwk.a ${GE_LIB_PATH})
         #find_module(ascendcl_static libascendcl.a ${GE_LIB_PATH})
     elseif(ENABLE_GE_COV OR ENABLE_GE_UT)
@@ -86,11 +84,9 @@ if (ENABLE_OPEN_SRC)
     else()
         find_module(slog libalog.so ${ASCEND_ATC_DIR})
         find_module(static_mmpa libmmpa.a ${ASCEND_ATC_DIR})
-	find_module(error_manager liberror_manager.so ${ASCEND_ATC_DIR})
         if(PLATFORM STREQUAL "train")
             find_module(adump_server libadump_server.a ${ASCEND_RUNTIME_DIR})
             find_module(runtime libruntime.so ${ASCEND_RUNTIME_DIR})
-            find_module(error_manager liberror_manager.so ${ASCEND_RUNTIME_DIR})
             find_module(msprofiler_fwk_ext libmsprofiler_fwk.a ${ASCEND_RUNTIME_DIR})
             find_module(ascend_hal_stub libascend_hal.so ${ASCEND_DRIVER_DIR}/driver)
             if(PRODUCT STREQUAL "flr3")
@@ -100,8 +96,6 @@ if (ENABLE_OPEN_SRC)
             find_module(adump_server libadump_server.a ${ASCEND_ACL_DIR})
             find_module(runtime libruntime.so ${ASCEND_ACL_DIR})
 	    find_module(runtime_compile libruntime_compile.so ${ASCEND_ATC_DIR})
-            find_module(error_manager liberror_manager.so ${ASCEND_ATC_DIR})
-            find_module(error_manager_static liberror_manager.a ${ASCEND_ACL_DIR})
             find_module(msprofiler_ext libmsprofiler.a ${ASCEND_ACL_DIR})
             if(PRODUCT STREQUAL "flr3")
             elseif(PRODUCT STREQUAL "flr1")
@@ -114,11 +108,9 @@ if (ENABLE_OPEN_SRC)
         elseif(PLATFORM STREQUAL "all")
 	    find_module(adump_server libadump_server.a ${ASCEND_RUNTIME_DIR})
 	    find_module(runtime libruntime.so ${ASCEND_RUNTIME_DIR})
-	    find_module(error_manager liberror_manager.so ${ASCEND_RUNTIME_DIR})
             find_module(msprofiler_fwk_ext libmsprofiler_fwk.a ${ASCEND_RUNTIME_DIR})
 	    find_module(ascend_hal_stub libascend_hal.so ${ASCEND_DRIVER_DIR})
 	    find_module(runtime_compile libruntime_compile.so ${ASCEND_ATC_DIR})
-	    find_module(error_manager_static liberror_manager.a ${ASCEND_ACL_DIR})
 	    find_module(msprofiler_ext libmsprofiler.a ${ASCEND_ACL_DIR})
         else()
             message(STATUS "PLATFORM param is invalid, should be train or inference, you choose nothing!")
@@ -144,7 +136,6 @@ elseif (ENABLE_D OR ENABLE_ACL)
 
     # common libraries
     find_module(slog libalog.so ${ASCEND_MS_RUNTIME_PATH} ${ATLAS_MS_RUNTIME_PATH})
-    find_module(error_manager liberror_manager.so ${ASCEND_MS_RUNTIME_PATH} ${ATLAS_MS_RUNTIME_PATH})
     find_module(static_mmpa libmmpa.a ${ASCEND_MS_RUNTIME_PATH} ${ATLAS_MS_RUNTIME_PATH})
 
     if (ENABLE_D)
@@ -164,7 +155,6 @@ elseif(ENABLE_MS_TESTCASES)
 
     # common libraries
     find_module(slog libalog.so ${ASCEND_MS_RUNTIME_PATH} ${ATLAS_MS_RUNTIME_PATH})
-    find_module(error_manager liberror_manager.so ${ASCEND_MS_RUNTIME_PATH} ${ATLAS_MS_RUNTIME_PATH})
     find_module(static_mmpa libmmpa.a ${ASCEND_MS_RUNTIME_PATH} ${ATLAS_MS_RUNTIME_PATH})
 
     set(METADEF_DIR ${CMAKE_CURRENT_LIST_DIR}/metadef)

ge/client/ge_api.cc

@@ -32,6 +32,7 @@
 #include "graph/common/ge_call_wrapper.h"
 #include "register/op_registry.h"
 #include "common/ge/tbe_plugin_manager.h"
+#include "common/util/error_manager/error_manager.h"
 #include "toolchain/plog.h"
 
 using domi::OpRegistry;
@@ -79,6 +80,8 @@ Status CheckOptionsValid(const std::map<string, string> &options) {
 // Initialize GE, prepare for execution, call GELib::Initialize
 Status GEInitializeImpl(const std::map<string, string> &options) {
   GELOGT(TRACE_INIT, "GEInitialize start");
+
+  ErrorManager::GetInstance().GenWorkStreamIdDefault();
   // 0.check init status
   if (g_ge_initialized) {
     GELOGW("GEInitialize is called more than once");
@@ -157,6 +160,8 @@ Status GEInitialize(const std::map<AscendString, AscendString> &options) {
 // GE finalize, releasing all resources
 Status GEFinalize() {
   GELOGT(TRACE_INIT, "GEFinalize start");
+
+  ErrorManager::GetInstance().GenWorkStreamIdDefault();
   // check init status
   if (!g_ge_initialized) {
     GELOGW("GEFinalize is called before GEInitialize");
@@ -202,9 +207,19 @@ Status GEFinalize() {
   return ret;
 }
 
+std::string GEGetErrorMsg() {
+  return ErrorManager::GetInstance().GetErrorMessage();
+}
+
+std::string GEGetWarningMsg() {
+  return ErrorManager::GetInstance.GetWarningMessage();
+}
+
 // Initialize session，which calls innerSession
 Session::Session(const std::map<string, string> &options) {
   GELOGT(TRACE_INIT, "Session Constructor start");
+
+  ErrorManager::GetInstance().GenWorkStreamIdDefault();
   // check init status
   sessionId_ = 0;
   if (!g_ge_initialized) {
@@ -235,6 +250,8 @@ Session::Session(const std::map<string, string> &options) {
 
 Session::Session(const std::map<AscendString, AscendString> &options) {
   GELOGT(TRACE_INIT, "Session Constructor start");
+
+  ErrorManager::GetInstance().GenWorkStreamIdDefault();
   // check init status
   sessionId_ = 0;
   if (!g_ge_initialized) {
@@ -311,11 +328,13 @@ Session::~Session() {
 
 Status Session::AddGraph(uint32_t graph_id, const Graph &graph) {
   std::map<std::string, std::string> options;
+  ErrorManager::GetInstance().GenWorkStreamIdBySessionGraph(sessionId_, graph_id);
   return AddGraph(graph_id, graph, options);
 }
 
 Status Session::AddGraph(uint32_t graph_id, const Graph &graph, const std::map<std::string, std::string> &options) {
   GELOGT(TRACE_INIT, "Start to add graph in Session. graph_id: %u, session_id: %lu.", graph_id, sessionId_);
+  ErrorManager::GetInstance().GenWorkStreamIdBySessionGraph(sessionId_, graph_id);
   std::shared_ptr<GELib> instance_ptr = ge::GELib::GetInstance();
   if (instance_ptr == nullptr || !instance_ptr->InitFlag()) {
     GELOGE(GE_CLI_GE_NOT_INITIALIZED, "AddGraph failed in Session.");
@@ -334,6 +353,7 @@ Status Session::AddGraph(uint32_t graph_id, const Graph &graph, const std::map<s
 Status Session::AddGraph(uint32_t graph_id, const Graph &graph,
                          const std::map<AscendString, AscendString> &options) {
   GELOGT(TRACE_INIT, "Start to add graph in Session. graph_id: %u, session_id: %lu.", graph_id, sessionId_);
+  ErrorManager::GetInstance().GenWorkStreamIdBySessionGraph(sessionId_, graph_id);
   std::shared_ptr<GELib> instance_ptr = ge::GELib::GetInstance();
   if (instance_ptr == nullptr || !instance_ptr->InitFlag()) {
     GELOGE(GE_CLI_GE_NOT_INITIALIZED, "AddGraph failed in Session.");
@@ -360,6 +380,7 @@ Status Session::AddGraph(uint32_t graph_id, const Graph &graph,
 }
 
 Status Session::AddGraphWithCopy(uint32_t graph_id, const Graph &graph) {
+  ErrorManager::GetInstance().GenWorkStreamIdBySessionGraph(sessionId_, graph_id);
   std::map<AscendString, AscendString> options;
   return AddGraphWithCopy(graph_id, graph, options);
 }
@@ -367,6 +388,7 @@ Status Session::AddGraphWithCopy(uint32_t graph_id, const Graph &graph) {
 Status Session::AddGraphWithCopy(uint32_t graph_id, const Graph &graph,
                                  const std::map<AscendString, AscendString> &options) {
   GELOGT(TRACE_INIT, "Start to add graph in Session. graph_id: %u, session_id: %lu.", graph_id, sessionId_);
+  ErrorManager::GetInstance().GenWorkStreamIdBySessionGraph(sessionId_, graph_id);
   std::shared_ptr<GELib> instance_ptr = ge::GELib::GetInstance();
   if (instance_ptr == nullptr || !instance_ptr->InitFlag()) {
     GELOGE(GE_CLI_GE_NOT_INITIALIZED, "AddGraph failed in Session.");
@@ -389,6 +411,7 @@ Status Session::AddGraphWithCopy(uint32_t graph_id, const Graph &graph,
 Status Session::RemoveGraph(uint32_t graph_id) {
   GELOGT(TRACE_INIT, "Session RemoveGraph start");
 
+  ErrorManager::GetInstance().GenWorkStreamIdBySessionGraph(sessionId_, graph_id);
   // call RemoveGraph
   std::shared_ptr<GELib> instance_ptr = ge::GELib::GetInstance();
   if (!instance_ptr || !instance_ptr->InitFlag()) {
@@ -457,6 +480,7 @@ void PrintOutputResult(std::vector<Tensor> &outputs) {
 Status Session::RunGraph(uint32_t graph_id, const std::vector<Tensor> &inputs, std::vector<Tensor> &outputs) {
   GELOGT(TRACE_INIT, "Session RunGraph start");
 
+  ErrorManager::GetInstance().GenWorkStreamIdBySessionGraph(sessionId_, graph_id);
   std::vector<Tensor> graph_inputs = inputs;
   // call RunGraph
   std::shared_ptr<GELib> instance_ptr = ge::GELib::GetInstance();
@@ -483,10 +507,12 @@ Status Session::RunGraph(uint32_t graph_id, const std::vector<Tensor> &inputs, s
 }
 
 Status Session::RegisterCallBackFunc(const std::string &key, const pCallBackFunc &callback) {
+  ErrorManager::GetInstance().GenWorkStreamIdDefault();
   return ge::GELib::GetInstance()->SessionManagerObj().RegisterCallBackFunc(sessionId_, key, callback);
 }
 
 Status Session::RegisterCallBackFunc(const char *key, const session::pCallBackFunc &callback) {
+  ErrorManager::GetInstance().GenWorkStreamIdDefault();
   std::string str_key;
   if (key != nullptr) {
     str_key = key;
@@ -495,6 +521,7 @@ Status Session::RegisterCallBackFunc(const char *key, const session::pCallBackFu
 }
 
 Status Session::BuildGraph(uint32_t graph_id, const std::vector<InputTensorInfo> &inputs) {
+  ErrorManager::GetInstance().GenWorkStreamIdBySessionGraph(sessionId_, graph_id);
   std::shared_ptr<GELib> instance_ptr = ge::GELib::GetInstance();
   if (instance_ptr == nullptr || !instance_ptr->InitFlag()) {
     GELOGE(GE_CLI_GE_NOT_INITIALIZED, "SessionConstructor failed");
@@ -511,6 +538,7 @@ Status Session::BuildGraph(uint32_t graph_id, const std::vector<InputTensorInfo>
 
 Status Session::RunGraphAsync(uint32_t graph_id, const std::vector<InputTensorInfo> &inputs,
                               RunAsyncCallback callback) {
+  ErrorManager::GetInstance().GenWorkStreamIdBySessionGraph(sessionId_, graph_id);
   std::shared_ptr<GELib> instance_ptr = ge::GELib::GetInstance();
   if (instance_ptr == nullptr || !instance_ptr->InitFlag()) {
     GELOGE(GE_CLI_GE_NOT_INITIALIZED, "SessionConstructor failed");
@@ -529,6 +557,7 @@ Status Session::RunGraphAsync(uint32_t graph_id, const std::vector<InputTensorIn
 }
 
 Status Session::GetVariables(const std::vector<std::string> &var_names, std::vector<Tensor> &var_values) {
+  ErrorManager::GetInstance().GenWorkStreamIdDefault();
   auto instance_ptr = ge::GELib::GetInstance();
   if (instance_ptr == nullptr || !instance_ptr->InitFlag()) {
     GELOGE(GE_CLI_GE_NOT_INITIALIZED, "SessionConstructor failed");
@@ -544,6 +573,7 @@ Status Session::GetVariables(const std::vector<std::string> &var_names, std::vec
 }
 
 Status Session::GetVariables(const std::vector<AscendString> &var_names, std::vector<Tensor> &var_values) {
+  ErrorManager::GetInstance().GenWorkStreamIdDefault();
   auto instance_ptr = ge::GELib::GetInstance();
   if (instance_ptr == nullptr || !instance_ptr->InitFlag()) {
     GELOGE(GE_CLI_GE_NOT_INITIALIZED, "SessionConstructor failed");

ge/graph/load/model_manager/davinci_model.cc

@@ -2641,6 +2641,7 @@ void *DavinciModel::Run(DavinciModel *model) {
   bool seq_end_flag = false;
   uint32_t model_id = model->Id();
   uint32_t device_id = model->GetDeviceId();
+  GetContext().SetWorkStreamId(model->GetWorkStreamId());
 
   GELOGI("Model Run thread start, model_id:%u.", model_id);
   rtError_t rt_ret = rtSetDevice(static_cast<int32_t>(device_id));
@@ -2807,6 +2808,7 @@ Status DavinciModel::ModelRunStart() {
   int64_t maxDumpOpNum = std::strtol(opt.c_str(), nullptr, kDecimal);
   maxDumpOpNum_ = maxDumpOpNum;
 
+  work_stream_id_ = GetContext().WorkStreamId();
   CREATE_STD_THREAD(thread_id_, DavinciModel::Run, this);
   GELOGI("model tread create success, model id:%u.", model_id_);
   return SUCCESS;

ge/graph/load/model_manager/davinci_model.h

@@ -412,6 +412,8 @@ class DavinciModel {
   ///
   uint64_t GetSessionId() const { return session_id_; }
 
+  uint64_t GetWorkStreamId() const { return work_stream_id_; }
+
   ///
   /// @ingroup ge
   /// @brief SetDeviceId
@@ -960,6 +962,7 @@ class DavinciModel {
   vector<uintptr_t> output_mbuf_list_;  // output mbuf created by dequeue task.
 
   uint64_t session_id_;
+  uint64_t work_stream_id_;
 
   uint32_t device_id_;
 

ge/graph/manager/graph_manager.cc

@@ -541,7 +541,7 @@ Status GraphManager::OptimizeSubGraphWithMultiThreads(ComputeGraphPtr compute_gr
     }
     std::future<Status> f = executor.commit(GraphManager::ProcessSubGraphWithMultiThreads, this,
                                             compute_graph->GetGraphID(), subgraph,
-                                            compute_graph->GetName(), session_id,
+                                            compute_graph->GetName(), session_id, GetContext().WorkStreamId(),
                                             GetThreadLocalContext());
     if (!f.valid()) {
       GELOGE(FAILED, "Future is invalid");
@@ -557,7 +557,7 @@ Status GraphManager::OptimizeSubGraphWithMultiThreads(ComputeGraphPtr compute_gr
       }
       std::future<Status> f = executor.commit(GraphManager::ProcessSubGraphWithMultiThreads, this,
                                               compute_graph->GetGraphID(), subgraph,
-                                              compute_graph->GetName(), session_id,
+                                              compute_graph->GetName(), session_id, GetContext().WorkStreamId(),
                                               GetThreadLocalContext());
       if (!f.valid()) {
         GELOGE(FAILED, "Future is invalid");
@@ -2508,8 +2508,10 @@ Status GraphManager::ProcessSubGraphWithMultiThreads(GraphManager *graph_manager
                                                      const SubGraphInfoPtr &sub_graph_info_ptr,
                                                      const std::string &root_graph_name,
                                                      uint64_t session_id,
+                                                     uint64_t work_stream_id,
                                                      const GEThreadLocalContext &ge_context) {
   if (sub_graph_info_ptr != nullptr && graph_manager != nullptr) {
+    GetContext().SetWorkStreamId(work_stream_id);
     GetContext().SetSessionId(session_id);
     GetThreadLocalContext() = ge_context;
     graph_manager->UpdateLocalOmgContext(root_graph_id);
@@ -2643,6 +2645,7 @@ void GraphManager::PreRunThread(GraphManager *graph_manager) {
 
     GELOGI("A new loop start.");
 
+    GetContext().SetWorkStreamId(args.work_stream_id);
     GetContext().SetSessionId(args.session_id);
     GetThreadLocalContext() = args.context;
     graph_manager->UpdateLocalOmgContext(args.graph_id);
@@ -2724,8 +2727,8 @@ void GraphManager::PreRunThread(GraphManager *graph_manager) {
       ge_root_model = graph_node->GetGeRootModel();
     }
 
-    graph_manager->run_args_q_.Push(RunArgs( { graph_node, args.graph_id, args.session_id, args.input_tensor,
+    graph_manager->run_args_q_.Push(RunArgs( { graph_node, args.graph_id, args.session_id, args.work_stream_id,
-        ge_root_model, GetThreadLocalContext(), args.callback }));
+        args.input_tensor, ge_root_model, GetThreadLocalContext(), args.callback }));
     GELOGI("Loop end.");
   }
 }
@@ -2824,6 +2827,7 @@ void GraphManager::RunThread(GraphManager *graph_manager) {
 
     GELOGI("A new loop start.");
 
+    GetContext().SetWorkStreamId(args.work_stream_id);
     GetContext().SetSessionId(args.session_id);
     GetThreadLocalContext() = args.context;
     graph_manager->UpdateLocalOmgContext(args.graph_id);

ge/graph/manager/graph_manager.h

@@ -196,6 +196,7 @@ class GraphManager {
     GraphId graph_id;
     std::vector<ge::InputTensorInfo> input_tensor;
     uint64_t session_id;
+    uint64_t work_stream_id;
     GEThreadLocalContext context;
     RunAsyncCallback callback;
   };
@@ -204,6 +205,7 @@ class GraphManager {
     GraphNodePtr graph_node;
     GraphId graph_id;
     uint64_t session_id;
+    uint64_t work_stream_id;
     std::vector<ge::InputTensorInfo> input_tensor;
     GeRootModelPtr ge_root_model;
     GEThreadLocalContext context;
@@ -221,6 +223,7 @@ class GraphManager {
                                                 const SubGraphInfoPtr &sub_graph_info_ptr,
                                                 const std::string &root_graph_name,
                                                 uint64_t session_id,
+                                                uint64_t work_stream_id;
                                                 const GEThreadLocalContext &ge_context);
   Status ParseInputsDims(const std::vector<InputTensorInfo> &input_tensor);
   void ParseInputsDimsForData(const std::vector<InputTensorInfo> &input_tensor);

ge/offline/main.cc

@@ -1325,6 +1325,7 @@ int init(int argc, char* argv[]) {
     return ret;
   }
 
+  ErrorManager::GetInstance().GenWorkStreamIdDefault();
   return 0;
 }
 

inc/external/ge/ge_api.h

@@ -42,6 +42,10 @@ GE_FUNC_VISIBILITY Status GEInitialize(const std::map<AscendString, AscendString
 // Finalize GE, release all resources
 GE_FUNC_VISIBILITY Status GEFinalize();
 
+GE_FUNC_VISIBILITY std::string GEGetErrorMsg();
+
+GE_FUNC_VISIBILITY std::string GEGetWarningMsg();
+
 class GE_FUNC_VISIBILITY Session {
  public:
   ATTRIBUTED_DEPRECATED(Session(const std::map<AscendString, AscendString> &))

metadef

@@ -1 +1 @@
-Subproject commit b6de68fdf0f131fd5f8aa3a84245ad7779b348f5
+Subproject commit f982caa0981b1fdcc55a8ec27b4f4de9c58d33ba

parser

@@ -1 +1 @@
-Subproject commit 7a6311351f8294eb11033b10e9f7b2b993cc3c2a
+Subproject commit d2fc958450f7bd243eff8432aadeb9fa95fa2f61