Pre Merge pull request !774 from 李磊/development

ge/graph/load/new_model_manager/data_dumper.cc

@@ -830,6 +830,13 @@ Status DataDumper::UnloadDumpInfo() {
   return SUCCESS;
 }
 
+void DataDumper::DumpShrink() {
+  compute_graph_.reset();
+  input_map_.clear();
+  ref_info_.clear();
+  op_list_.clear();
+}
+
 void DataDumper::PrintCheckLog(string &dump_list_key) {
   std::set<std::string> model_list = dump_properties_.GetAllDumpModel();
   if (model_list.empty()) {

ge/graph/load/new_model_manager/data_dumper.h

@@ -83,6 +83,8 @@ class DataDumper {
 
   Status UnloadDumpInfo();
 
+  void DumpShrink();
+
   void SetDumpProperties(const DumpProperties &dump_properties) { dump_properties_ = dump_properties; }
   const DumpProperties &GetDumpProperties() const { return dump_properties_; }
   bool GetOpDescInfo(uint32_t stream_id, uint32_t task_id, OpDescInfo &op_desc_info) const;
@@ -112,18 +114,18 @@ class DataDumper {
   struct InnerInputMapping;
 
   std::vector<OpDescInfo> op_desc_info_;
-  std::vector<InnerDumpInfo> op_list_;
+  std::vector<InnerDumpInfo> op_list_;  // release after DavinciModel::Init
   uint32_t end_graph_task_id_ = 0;
   uint32_t end_graph_stream_id_ = 0;
   bool is_end_graph_ = false;
-  std::multimap<std::string, InnerInputMapping> input_map_;
+  std::multimap<std::string, InnerInputMapping> input_map_;  // release after DavinciModel::Init
   bool load_flag_;
   uint32_t device_id_;
   uintptr_t global_step_;
   uintptr_t loop_per_iter_;
   uintptr_t loop_cond_;
-  ComputeGraphPtr compute_graph_;
-  std::map<OpDescPtr, void *> ref_info_;
+  ComputeGraphPtr compute_graph_;  // release after DavinciModel::Init
+  std::map<OpDescPtr, void *> ref_info_;  // release after DavinciModel::Init
   void *l1_fusion_addr_ = nullptr;
 
 

ge/graph/load/new_model_manager/davinci_model.cc

@@ -148,14 +148,7 @@ DavinciModel::~DavinciModel() {
       GELOGW("UnloadDumpInfo failed, ret: %u.", ret);
     }
 
-    for (const auto &op_and_addr : saved_task_addrs_) {
-      auto addr = op_and_addr.second;
-      if (addr != nullptr) {
-        GE_CHK_RT(rtFree(addr));
-      }
-      addr = nullptr;
-    }
-    saved_task_addrs_.clear();
+    ClearTaskAddrs();
 
     GE_CHK_STATUS(ModelRunStop());
 
@@ -220,6 +213,17 @@ DavinciModel::~DavinciModel() {
   }
 }
 
+void DavinciModel::ClearTaskAddrs() {
+  for (const auto &op_and_addr : saved_task_addrs_) {
+    auto addr = op_and_addr.second;
+    if (addr != nullptr) {
+      GE_CHK_RT(rtFree(addr));
+    }
+    addr = nullptr;
+  }
+  saved_task_addrs_.clear();
+}
+
 void DavinciModel::UnbindHcomStream() {
   if (!all_hccl_stream_list_.empty()) {
     for (size_t i = 0; i < all_hccl_stream_list_.size(); i++) {
@@ -261,7 +265,10 @@ Status DavinciModel::Assign(const GeModelPtr &ge_model) {
 /// @return: void
 ///
 void DavinciModel::Shrink() {
+  //DumperShrink();
   ge_model_.reset();  // delete object.
+  //op_list_.clear();
+  //ClearTaskAddrs();
 }
 
 Status DavinciModel::InitWeightMem(void *dev_ptr, void *weight_ptr, size_t weight_size) {
@@ -591,6 +598,7 @@ void DavinciModel::OpDebugUnRegister() {
 // initialize op sequence and call initialization function of each op respectively
 Status DavinciModel::Init(void *dev_ptr, size_t mem_size, void *weight_ptr, size_t weight_size) {
   // validating params
+
   GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(priority_ < 0 || priority_ > 7, return PARAM_INVALID,
                                  "Priority must between 0-7, now is %d", priority_);
   GE_CHK_BOOL_RET_STATUS(ge_model_ != nullptr, PARAM_INVALID, "GeModel is null.");
@@ -735,7 +743,6 @@ Status DavinciModel::ReportProfilingData() {
   }
   ProfilingManager::Instance().ReportProfilingData(model_id_, GetTaskDescInfo(), compute_graph_desc_info);
   GE_CHK_STATUS(SinkModelProfile(), "Sink model profiler failed.");
-  op_list_.clear();
 
   return SUCCESS;
 }
@@ -934,11 +941,15 @@ Status DavinciModel::InitInputOutputForDynamic(const ComputeGraphPtr &compute_gr
     }
     if (IsDataOp(op_desc->GetType())) {
       GELOGD("init data op %s", op_desc->GetName().c_str());
-      data_op_list_.push_back(op_desc);
+      auto data_op = AttrUtils::CopyOpDesc(op_desc);
+      GE_CHECK_NOTNULL(data_op);
+      data_op_list_.push_back(data_op);
     }
     if (op_desc->GetType() == NETOUTPUT) {
       GELOGD("init netouput op %s", op_desc->GetName().c_str());
-      output_op_list_.push_back(op_desc);
+      auto output_op = AttrUtils::CopyOpDesc(op_desc);
+      GE_CHECK_NOTNULL(output_op);
+      output_op_list_.push_back(output_op);
     }
   }
   return SUCCESS;
@@ -976,7 +987,9 @@ Status DavinciModel::InitDataOp(const NodePtr &node, uint32_t &data_op_index, ma
     return SUCCESS;
   }
 
-  data_op_list_.push_back(op_desc);
+  auto data_op = AttrUtils::CopyOpDesc(op_desc);
+  GE_CHECK_NOTNULL(data_op);
+  data_op_list_.push_back(data_op);
 
   // Make information for copy input data.
   const vector<int64_t> output_size_list = ModelUtils::GetOutputSize(op_desc);
@@ -1031,7 +1044,9 @@ void DavinciModel::AdjustDataOpList(const map<uint32_t, OpDescPtr> &data_by_inde
 
   data_op_list_.clear();
   for (auto &item : data_by_index) {
-    data_op_list_.emplace_back(item.second);
+    auto data_op = AttrUtils::CopyOpDesc(item.second);
+    GE_CHECK_NOTNULL_EXEC(data_op, return);
+    data_op_list_.emplace_back(data_op);
   }
 }
 
@@ -1063,7 +1078,9 @@ Status DavinciModel::InitNetOutput(const NodePtr &node) {
     return SUCCESS;
   }
 
-  output_op_list_.push_back(op_desc);
+  auto output_op = AttrUtils::CopyOpDesc(op_desc);
+  GE_CHECK_NOTNULL(output_op);
+  output_op_list_.push_back(output_op);
   // Make information for copy output data.
   const vector<int64_t> input_size_list = ModelUtils::GetInputSize(op_desc);
   const vector<void *> virtual_addr_list = ModelUtils::GetInputDataAddrs(runtime_param_, op_desc);

ge/graph/load/new_model_manager/davinci_model.h

@@ -458,6 +458,10 @@ class DavinciModel {
     data_dumper_.SaveDumpTask(task_id, stream_id, op_desc, args);
   }
 
+  void DumperShrink() {
+    data_dumper_.DumpShrink();
+  }
+
   void SetEndGraphId(uint32_t task_id, uint32_t stream_id);
   DavinciModel &operator=(const DavinciModel &model) = delete;
 
@@ -620,6 +624,8 @@ class DavinciModel {
 
   void ReleaseTask();
 
+  void ClearTaskAddrs();
+
   void UnbindTaskSinkStream();
 
   bool IsAicpuKernelConnectSpecifiedLayer();
@@ -842,12 +848,12 @@ class DavinciModel {
   string om_name_;
 
   uint32_t version_;
-  GeModelPtr ge_model_;
+  GeModelPtr ge_model_;  // release after DavinciModel::Init
 
   bool need_destroy_aicpu_kernel_{false};
   vector<std::string> out_node_name_;
 
-  map<uint32_t, OpDescPtr> op_list_;
+  map<uint32_t, OpDescPtr> op_list_;  // release after DavinciModel::Init
 
   // data op_desc
   vector<OpDescPtr> data_op_list_;
@@ -944,7 +950,7 @@ class DavinciModel {
   DataDumper data_dumper_;
   uint64_t iterator_count_;
   bool is_l1_fusion_enable_;
-  std::map<OpDescPtr, void *> saved_task_addrs_;
+  std::map<OpDescPtr, void *> saved_task_addrs_;  // release after DavinciModel::Init
   void *l1_fusion_addr_ = nullptr;
 
   bool known_node_ = false;

metadef

@@ -1 +1 @@
-Subproject commit c14d2be38171eed63416e71178774103faf1f5cd
+Subproject commit 03ad8a2899d7ac4f06e3ce5e21f533a48aa77895

tests/depends/error_manager/src/error_manager_stub.cc

@@ -58,7 +58,7 @@
   /// @param [in] value: vector parameter value
   ///
   void ErrorManager::ATCReportErrMessage(std::string error_code, const std::vector<std::string> &key,
-                                         const std::vector<std::string> &value) { 
+                                         const std::vector<std::string> &value) {
   }
 
   ///
@@ -66,13 +66,8 @@
   /// @param [in] msg: failed message map, key is error code, value is op_name
   /// @return int 0(success) -1(fail)
   ///
-  int ErrorManager::ReportMstuneCompileFailedMsg(const std::map<std::string, std::string> &msg) { return 0; }
-
-  ///
-  /// @brief save graph compile failed message from thread local map to global map
-  /// @param [in] graph_name: graph name
-  ///
-  void ErrorManager::SaveMstuneCompileFailedMsg(const std::string &graph_name) {}
+  int ErrorManager::ReportMstuneCompileFailedMsg(const std::string &root_graph_name,
+                                                 const std::map<std::string, std::string> &msg) { return 0; }
 
   ///
   /// @brief get graph compile failed message in mstune case