!2130 upgrad Ascend software package 15 Jul 22

Merge pull request !2130 from yanghaoran/master

CMakeLists.txt

@@ -125,7 +125,6 @@ else ()
                 message(STATUS "PLATFORM param is invalid, should be train or inference, you choose nothing!")
             endif()
         endif()
-
         set(METADEF_DIR ${CMAKE_CURRENT_LIST_DIR}/metadef)
         set(PARSER_DIR ${CMAKE_CURRENT_LIST_DIR}/parser)
         set(GE_DEPEND_DIR ${CMAKE_CURRENT_LIST_DIR}/..)
@@ -158,6 +157,7 @@ else ()
     elseif(ENABLE_MS_TESTCASES)
         include(cmake/external_libs/protobuf_static.cmake)
         include(cmake/external_libs/protoc.cmake)
+        include(cmake/external_libs/json.cmake)
         include(cmake/external_libs/securec.cmake)
         include(cmake/FindModule.cmake)
         include(cmake/intf_pub_linux.cmake)
@@ -175,5 +175,4 @@ else ()
     endif()
 
     add_subdirectory(ge)
-
 endif ()

cmake/external_libs/json.cmake

@@ -9,10 +9,6 @@ if (GE_PB_PKG)
     set(REQ_URL "${GE_PB_PKG}/libs/ge_nlohmann_json/include.zip")
     set(MD5 "0dc903888211db3a0f170304cd9f3a89")
     set(JSON_INCLUDE_DIR ${JSON_SRC_DIR})
-#elseif (ENABLE_GITEE)
-#    set(REQ_URL "https://gitee.com/mirrors/JSON-for-Modern-CPP/repository/archive/v3.6.1.zip")
-#    set(MD5 "5bda78ce308e6cfcf614dcf1d5ff27a7")
-#set(JSON_INCLUDE_DIR "${JSON_SRC_DIR}/include")
 else()
     set(REQ_URL "https://github.com/nlohmann/json/releases/download/v3.6.1/include.zip")
     set(MD5 "0dc903888211db3a0f170304cd9f3a89")

ge/common/model/ge_model.cc

@@ -34,7 +34,6 @@ void GeModel::Init() {
 }
 
 GeModel::GeModel() {
-  attrs_.InitDefault();
   Init();
 }
 
@@ -78,12 +77,12 @@ void GeModel::SetPlatformVersion(const std::string &platform_version) { this->pl
 
 void GeModel::SetPlatformType(uint8_t platform_type) { this->platform_type_ = platform_type; }
 
-void GeModel::SetAttr(const ProtoAttrMapHelper &attrs) { attrs_ = attrs; }
+void GeModel::SetAttr(const ProtoAttrMap &attrs) { attrs_ = attrs; }
 
-ProtoAttrMapHelper GeModel::MutableAttrMap() { return attrs_; }
+ProtoAttrMap &GeModel::MutableAttrMap() { return attrs_; }
 
-ConstProtoAttrMapHelper GeModel::GetAttrMap() const {
-  return ConstProtoAttrMapHelper(attrs_.GetProtoOwner(), attrs_.GetProtoMsg());
+ConstProtoAttrMap &GeModel::GetAttrMap() const {
+  return attrs_;
 }
 
 Status GeModel::GetSessionId(uint32_t model_id, uint64_t &session_id) const {

ge/common/model/ge_model.h

@@ -17,24 +17,26 @@
 #ifndef GE_MODEL_GE_MODEL_H_
 #define GE_MODEL_GE_MODEL_H_
 
-#include <securec.h>
 #include <map>
 #include <memory>
 #include <string>
+
+#include "securec.h"
+#include "runtime/rt.h"
 #include "common/tbe_kernel_store.h"
 #include "common/cust_aicpu_kernel_store.h"
 #include "framework/common/debug/log.h"
 #include "framework/common/fmk_error_codes.h"
+#include "framework/common/ge_types.h"
 #include "graph/buffer.h"
 #include "external/graph/graph.h"
 #include "proto/task.pb.h"
 
 namespace ge {
-const uint32_t INVALID_MODEL_ID = 0xFFFFFFFFUL;
 class GeModel : public AttrHolder {
  public:
   GeModel();
-  ~GeModel() = default;
+  ~GeModel() override = default;
   GeModel(const GeModel &other) = delete;
   GeModel &operator=(const GeModel &other) = delete;
 
@@ -55,34 +57,34 @@ class GeModel : public AttrHolder {
   void SetCustAICPUKernelStore(const CustAICPUKernelStore &cust_aicpu_kernal_store);
   void SetWeight(const Buffer &weights_buffer);
 
+  bool LoadTBEKernelStore(const uint8_t *const data, const size_t len);
+  bool loadAICPUKernelStore(const uint8_t *const data, const size_t len);
+
   void SetName(const std::string &name);
-  void SetVersion(uint32_t version);
+  void SetVersion(const uint32_t version);
   void SetPlatformVersion(const std::string &platform_version);
-  void SetPlatformType(uint8_t platform_type);
+  void SetPlatformType(const uint8_t platform_type);
 
-  void SetAttr(const ProtoAttrMapHelper &attrs);
+  void SetAttr(const ProtoAttrMap &attrs);
 
-  ProtoAttrMapHelper MutableAttrMap() override;
+  ProtoAttrMap &MutableAttrMap() override;
 
   using AttrHolder::SetAttr;
   using AttrHolder::GetAllAttrs;
   using AttrHolder::GetAllAttrNames;
 
-  void SetModelId(uint32_t model_id) { model_id_ = model_id; }
+  void SetModelId(const uint32_t model_id) { model_id_ = model_id; }
   uint32_t GetModelId() const { return model_id_; }
 
-  Status GetSessionId(uint32_t model_id, uint64_t &session_id) const;
-  void InsertSessionMap(uint32_t model_id, uint64_t session_id) {
-    model_id_to_session_id_map_.insert({model_id, session_id});
-  }
+  Status GetSessionId(const uint32_t model_id, uint64_t &session_id) const;
 
  protected:
-  ConstProtoAttrMapHelper GetAttrMap() const override;
+  ConstProtoAttrMap &GetAttrMap() const override;
 
  private:
   void Init();
 
-  ProtoAttrMapHelper attrs_;  /*lint !e148*/
+  ProtoAttrMap attrs_;  /*lint !e148*/
 
   Graph graph_;
   std::shared_ptr<domi::ModelTaskDef> task_;  /*lint !e148*/
@@ -91,9 +93,9 @@ class GeModel : public AttrHolder {
   Buffer weights_buffer_;  /*lint !e148*/
 
   std::string name_;
-  uint32_t version_ = {0};
+  uint32_t version_ = {0U};
   std::string platform_version_;
-  uint8_t platform_type_ = {0};
+  uint8_t platform_type_ = {0U};
   uint32_t model_id_ = INVALID_MODEL_ID;
   std::map<uint32_t, uint64_t> model_id_to_session_id_map_;
 };

ge/ge_runtime/CMakeLists.txt

@@ -16,6 +16,7 @@ set(GE_SRC_LIST
     "task/label_goto_task.cc"
     "task/label_set_task.cc"
     "task/label_switch_task.cc"
+    "task/label_manager.cc"
 )
 
 add_library(ge_runtime SHARED ${GE_SRC_LIST})

ge/ge_runtime/task/hccl_task.cc

@@ -53,15 +53,7 @@ HcclTask::HcclTask(const ModelContext &model_context, const std::shared_ptr<Hccl
   }
 }
 
-HcclTask::~HcclTask() {
-  if (workspace_mem_ != nullptr) {
-    rtError_t rt_ret = rtFree(workspace_mem_);
-    if (rt_ret != RT_ERROR_NONE) {
-      GELOGE(RT_FAILED, "rtFree workspace_mem_ failed! ret: 0x%X.", rt_ret);
-    }
-    workspace_mem_ = nullptr;
-  }
-}
+HcclTask::~HcclTask() {}
 
 bool HcclTask::Distribute() {
   // Ops kernel info store
@@ -80,11 +72,7 @@ bool HcclTask::Distribute() {
   SetSecondaryStream();
 
   if (task_info_->workspace_size() > 0) {
-    rtError_t rt_ret = rtMalloc(&workspace_mem_, task_info_->workspace_size(), RT_MEMORYINFO_HBM);
-    if (rt_ret != RT_ERROR_NONE) {
-      GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
-      return false;
-    }
+    workspace_mem_ = task_info_->workspace_addr();
   }
 
   GELOGI("HcclTaskInfo Distribute Start. begin to call function LoadTask in hccl.");

ge/ge_runtime/task/label_goto_task.cc

@@ -16,33 +16,46 @@
 
 #include "ge_runtime/task/label_goto_task.h"
 #include "ge_runtime/task/task_factory.h"
-#include "framework/common/util.h"
 
 namespace ge {
 namespace model_runner {
 LabelGotoTask::LabelGotoTask(const ModelContext &model_context, const std::shared_ptr<LabelGotoTaskInfo> &task_info)
-    : TaskRepeater<LabelGotoTaskInfo>(model_context, task_info), task_info_(task_info) {
+    : TaskRepeater<LabelGotoTaskInfo>(model_context, task_info),
+      task_info_(task_info),
+      stream_(nullptr),
+      index_value_(nullptr) {
   if (task_info_ == nullptr) {
     GELOGW("task_info_ is null!");
     return;
   }
   auto stream_list = model_context.stream_list();
   auto label_list = model_context.label_list();
+  rt_model_handle_ = model_context.rt_model_handle();
   uint32_t stream_id = task_info->stream_id();
-  uint32_t label_id = task_info->label_id();
+  label_id_ = task_info->label_id();
   GELOGI("Stream list size:%zu, stream id:%u.", stream_list.size(), stream_id);
-  GELOGI("Label list size:%zu, label id:%u.", label_list.size(), label_id);
-  if (stream_id >= stream_list.size() || label_id >= label_list.size()) {
+  GELOGI("Label list size:%zu, label id:%u.", label_list.size(), label_id_);
+  if (stream_id >= stream_list.size() || label_id_ >= label_list.size()) {
     GELOGW("Stream/Label id invalid.");
     return;
   }
   stream_ = stream_list[stream_id];
-  label_ = label_list[label_id];
+  label_manager_ = LabelManager::GetInstance();
+  if (label_manager_ == nullptr) {
+    GELOGW("Get label manager instance failed.");
+    return;
+  }
+  label_info_ = label_manager_->GetLabelInfo(rt_model_handle_, {label_id_}, label_list);
 }
 
 LabelGotoTask::~LabelGotoTask() {
-  GE_FREE_RT_LOG(label_info_);
-  GE_FREE_RT_LOG(index_value_);
+  if (index_value_ != nullptr) {
+    rtError_t rt_ret = rtFree(index_value_);
+    if (rt_ret != RT_ERROR_NONE) {
+      GELOGE(RT_FAILED, "rtFree index_value_ failed! ret: 0x%X.", rt_ret);
+    }
+    index_value_ = nullptr;
+  }
 }
 
 bool LabelGotoTask::Distribute() {
@@ -94,21 +107,34 @@ bool LabelGotoTask::CheckParamValid() {
     return false;
   }
 
-  if (label_ == nullptr) {
-    GELOGE(PARAM_INVALID, "label is null!");
+  if (label_info_ == nullptr) {
+    GELOGE(PARAM_INVALID, "label info is null!");
     return false;
   }
 
-  if (label_info_ != nullptr) {
-    GELOGE(PARAM_INVALID, "label_info_ has dirty data.");
-    return false;
+  if (index_value_ == nullptr) {
+    rtError_t rt_ret = rtMalloc(&index_value_, sizeof(uint64_t), RT_MEMORY_HBM);
+    if (rt_ret != RT_ERROR_NONE) {
+      GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
+      return false;
+    }
+
+    uint64_t index = 0;
+    rt_ret = rtMemcpy(index_value_, sizeof(uint64_t), &index, sizeof(index), RT_MEMCPY_HOST_TO_DEVICE);
+    if (rt_ret != RT_ERROR_NONE) {
+      GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
+      return false;
+    }
   }
 
-  if (index_value_ != nullptr) {
-    GELOGE(PARAM_INVALID, "index_value_ has dirty data.");
+  void *label_info = label_info_->GetLabelInfo();
+  rtError_t rt_ret = rtLabelSwitchByIndex(index_value_, 1, label_info, stream_);
+  if (rt_ret != RT_ERROR_NONE) {
+    GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
     return false;
   }
 
+  GELOGI("DistributeTask end.");
   return true;
 }
 

ge/ge_runtime/task/label_goto_task.h

@@ -18,7 +18,11 @@
 #define GE_GE_RUNTIME_TASK_LABEL_GOTO_TASK_H_
 
 #include <memory>
+#include <vector>
+#include <map>
+#include <mutex>
 #include "ge_runtime/task/task.h"
+#include "ge_runtime/task/label_manager.h"
 
 namespace ge {
 namespace model_runner {
@@ -31,13 +35,13 @@ class LabelGotoTask : public TaskRepeater<LabelGotoTaskInfo> {
   bool Distribute() override;
 
  private:
-  bool CheckParamValid();
-
   std::shared_ptr<LabelGotoTaskInfo> task_info_;
-  void *stream_{nullptr};
-  void *label_{nullptr};
-  void *label_info_{nullptr};
-  void *index_value_{nullptr};
+  void *stream_;
+  std::shared_ptr<LabelGuard> label_info_;
+  void *index_value_;
+  uint32_t label_id_;
+  rtModel_t rt_model_handle_;
+  std::shared_ptr<LabelManager> label_manager_;
 };
 }  // namespace model_runner
 }  // namespace ge

ge/ge_runtime/task/label_manager.cc

@@ -0,0 +1,119 @@
+/**
+ * Copyright 2021 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#include "ge_runtime/task/label_manager.h"
+#include <algorithm>
+#include <string>
+#include "runtime/mem.h"
+#include "runtime/rt_model.h"
+#include "common/ge_inner_error_codes.h"
+#include "framework/common/debug/ge_log.h"
+
+namespace ge {
+namespace model_runner {
+std::weak_ptr<LabelManager> LabelManager::instance_;
+std::mutex LabelManager::instance_mutex_;
+
+template <class T>
+static std::string GetVectorString(const std::vector<T> &vec) {
+  std::string ret;
+  for (size_t i = 0; i < vec.size(); ++i) {
+    if (i != 0) {
+      ret.push_back(',');
+    }
+    ret += std::to_string(vec[i]);
+  }
+  return ret;
+}
+
+LabelGuard::~LabelGuard() {
+  void *label_info = GetLabelInfo();
+  if (label_info != nullptr) {
+    rtError_t rt_ret = rtFree(label_info);
+    if (rt_ret != RT_ERROR_NONE) {
+      GELOGE(RT_FAILED, "rtFree label_info failed! ret: 0x%X.", rt_ret);
+    }
+  }
+}
+
+std::shared_ptr<LabelManager> LabelManager::GetInstance() {
+  std::lock_guard<std::mutex> lock(instance_mutex_);
+  auto instance = instance_.lock();
+  if (instance != nullptr) {
+    return instance;
+  }
+
+  instance = std::make_shared<LabelManager>();
+  instance_ = instance;
+  return instance;
+}
+
+std::shared_ptr<LabelGuard> LabelManager::GetLabelInfo(rtModel_t model, const std::vector<uint32_t> &label_ids,
+                                                       const std::vector<void *> &all_label) {
+  std::lock_guard<std::mutex> lock(model_info_mapping_mutex_);
+  rtError_t rt_ret;
+  auto model_iter = model_info_mapping_.find(model);
+  if (model_iter == model_info_mapping_.end()) {
+    model_info_mapping_.emplace(model, std::map<std::string, std::weak_ptr<LabelGuard>>());
+    model_iter = model_info_mapping_.find(model);
+  }
+
+  std::string label_id_str = GetVectorString(label_ids);
+  auto &label_map = model_iter->second;
+  auto label_iter = label_map.find(label_id_str);
+  if (label_iter != label_map.end()) {
+    auto label_guard = label_iter->second.lock();
+    if (label_guard != nullptr) {
+      GELOGI("model %p find same label id %s.", model, label_id_str.c_str());
+      return label_guard;
+    }
+  }
+
+  GELOGI("Alloc label id %s for model %p.", label_id_str.c_str(), model);
+  void *label_info;
+  std::vector<void *> label_list;
+  bool status = true;
+  std::transform(label_ids.begin(), label_ids.end(), std::back_inserter(label_list),
+                 [&all_label, &status](uint32_t idx) -> void * {
+                   if (idx >= all_label.size()) {
+                     GELOGE(PARAM_INVALID, "Invalid label id %u, all label list size %zu.", idx, all_label.size());
+                     status = false;
+                     return nullptr;
+                   }
+                   return all_label[idx];
+                 });
+  if (!status) {
+    GELOGE(PARAM_INVALID, "Get label info failed.");
+    return nullptr;
+  }
+  uint32_t label_info_size = sizeof(rtLabelDevInfo) * label_list.size();
+  rt_ret = rtMalloc(&label_info, label_info_size, RT_MEMORY_HBM);
+  if (rt_ret != RT_ERROR_NONE) {
+    GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
+    return nullptr;
+  }
+
+  rt_ret = rtLabelListCpy(label_list.data(), label_list.size(), label_info, label_info_size);
+  if (rt_ret != RT_ERROR_NONE) {
+    GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
+    return nullptr;
+  }
+
+  auto label_guard = std::make_shared<LabelGuard>(label_info);
+  label_map.emplace(label_id_str, label_guard);
+  return label_guard;
+}
+}  // namespace model_runner
+}  // namespace ge

ge/ge_runtime/task/label_manager.h

@@ -0,0 +1,54 @@
+/**
+ * Copyright 2021 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#ifndef GE_GE_RUNTIME_TASK_LABEL_MANAGER_H_
+#define GE_GE_RUNTIME_TASK_LABEL_MANAGER_H_
+
+#include <vector>
+#include <memory>
+#include <mutex>
+#include <map>
+#include <runtime/base.h>
+
+namespace ge {
+namespace model_runner {
+class LabelGuard {
+ public:
+  explicit LabelGuard(void *label_info) : label_info_(reinterpret_cast<uintptr_t>(label_info)) {}
+  ~LabelGuard();
+  void *GetLabelInfo() { return reinterpret_cast<void *>(label_info_); }
+
+ private:
+  uintptr_t label_info_;
+};
+
+class LabelManager {
+ public:
+  static std::shared_ptr<LabelManager> GetInstance();
+  std::shared_ptr<LabelGuard> GetLabelInfo(rtModel_t model, const std::vector<uint32_t> &label_ids,
+                                           const std::vector<void *> &all_label);
+
+ private:
+  std::mutex model_info_mapping_mutex_;
+  std::map<rtModel_t, std::map<std::string, std::weak_ptr<LabelGuard>>> model_info_mapping_;
+
+  static std::weak_ptr<LabelManager> instance_;
+  static std::mutex instance_mutex_;
+};
+
+
+}  // namespace model_runner
+}  // namespace ge
+#endif  // GE_GE_RUNTIME_TASK_LABEL_MANAGER_H_

ge/ge_runtime/task/label_switch_task.cc

@@ -24,14 +24,14 @@ LabelSwitchTask::LabelSwitchTask(const ModelContext &model_context,
     : TaskRepeater<LabelSwitchTaskInfo>(model_context, task_info),
       task_info_(task_info),
       stream_(nullptr),
-      all_label_resource_(),
       label_info_(nullptr) {
   if (task_info_ == nullptr) {
     GELOGW("task_info_ is null!");
     return;
   }
 
-  all_label_resource_ = model_context.label_list();
+  rt_model_handle_ = model_context.rt_model_handle();
+  auto all_label_resource = model_context.label_list();
   auto stream_list = model_context.stream_list();
   uint32_t stream_id = task_info->stream_id();
   GELOGI("Stream list size:%zu, stream id:%u.", stream_list.size(), stream_id);
@@ -40,18 +40,16 @@ LabelSwitchTask::LabelSwitchTask(const ModelContext &model_context,
     return;
   }
   stream_ = stream_list[stream_id];
-}
-
-LabelSwitchTask::~LabelSwitchTask() {
-  if (label_info_ != nullptr) {
-    rtError_t rt_ret = rtFree(label_info_);
-    if (rt_ret != RT_ERROR_NONE) {
-      GELOGE(RT_FAILED, "rtFree fwkOpBuf failed! ret: 0x%X.", rt_ret);
-    }
-    label_info_ = nullptr;
+  label_manager_ = LabelManager::GetInstance();
+  if (label_manager_ == nullptr) {
+    GELOGW("Get label manager instance failed.");
+    return;
   }
+  label_info_ = label_manager_->GetLabelInfo(rt_model_handle_, task_info_->label_list(), all_label_resource);
 }
 
+LabelSwitchTask::~LabelSwitchTask() {}
+
 bool LabelSwitchTask::Distribute() {
   GELOGI("LabelSwitchTask Distribute start.");
   if (!CheckParamValid()) {
@@ -117,8 +115,8 @@ bool LabelSwitchTask::CheckParamValid() {
     return false;
   }
 
-  if (label_info_ != nullptr) {
-    GELOGE(PARAM_INVALID, "label_info_ has dirty data.");
+  if (label_info_ == nullptr) {
+    GELOGE(PARAM_INVALID, "CopyLabelList failed, label info is null.");
     return false;
   }
 
@@ -126,6 +124,5 @@ bool LabelSwitchTask::CheckParamValid() {
 }
 
 REGISTER_TASK(TaskInfoType::LABEL_SWITCH, LabelSwitchTask, LabelSwitchTaskInfo);
-
 }  // namespace model_runner
 }  // namespace ge

ge/ge_runtime/task/label_switch_task.h

@@ -19,6 +19,7 @@
 
 #include <memory>
 #include "ge_runtime/task/task.h"
+#include "ge_runtime/task/label_manager.h"
 
 namespace ge {
 namespace model_runner {
@@ -35,8 +36,9 @@ class LabelSwitchTask : public TaskRepeater<LabelSwitchTaskInfo> {
 
   std::shared_ptr<LabelSwitchTaskInfo> task_info_;
   void *stream_;
-  std::vector<void *> all_label_resource_;
-  void *label_info_;
+  rtModel_t rt_model_handle_;
+  std::shared_ptr<LabelGuard> label_info_;
+  std::shared_ptr<LabelManager> label_manager_;
 };
 }  // namespace model_runner
 }  // namespace ge

inc/external/acl/acl.h

@@ -0,0 +1,82 @@
+/**
+ * Copyright 2019-2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef INC_EXTERNAL_ACL_ACL_H_
+#define INC_EXTERNAL_ACL_ACL_H_
+
+#include "acl_rt.h"
+#include "acl_op.h"
+#include "acl_mdl.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// Current version is 1.2.0
+#define ACL_MAJOR_VERSION 1
+#define ACL_MINOR_VERSION 2
+#define ACL_PATCH_VERSION 0
+
+/**
+ * @ingroup AscendCL
+ * @brief acl initialize
+ *
+ * @par Restriction
+ * The aclInit interface can be called only once in a process
+ * @param configPath [IN]    the config path,it can be NULL
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclInit(const char *configPath);
+
+/**
+ * @ingroup AscendCL
+ * @brief acl finalize
+ *
+ * @par Restriction
+ * Need to call aclFinalize before the process exits.
+ * After calling aclFinalize,the services cannot continue to be used normally.
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclFinalize();
+
+/**
+ * @ingroup AscendCL
+ * @brief query ACL interface version
+ *
+ * @param majorVersion[OUT] ACL interface major version
+ * @param minorVersion[OUT] ACL interface minor version
+ * @param patchVersion[OUT] ACL interface patch version
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclrtGetVersion(int32_t *majorVersion, int32_t *minorVersion, int32_t *patchVersion);
+
+/**
+ * @ingroup AscendCL
+ * @brief get recent error message
+ *
+ * @retval null for failed
+ * @retval OtherValues success
+ */
+ACL_FUNC_VISIBILITY const char *aclGetRecentErrMsg();
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // INC_EXTERNAL_ACL_ACL_H_

inc/external/acl/acl_base.h

@@ -0,0 +1,656 @@
+/**
+ * Copyright 2019-2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef INC_EXTERNAL_ACL_ACL_BASE_H_
+#define INC_EXTERNAL_ACL_ACL_BASE_H_
+
+#include <stdint.h>
+#include <stddef.h>
+#include "error_codes/rt_error_codes.h"
+#include "error_codes/ge_error_codes.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#if defined(_MSC_VER)
+#ifdef FUNC_VISIBILITY
+#define ACL_FUNC_VISIBILITY _declspec(dllexport)
+#else
+#define ACL_FUNC_VISIBILITY
+#endif
+#else
+#ifdef FUNC_VISIBILITY
+#define ACL_FUNC_VISIBILITY __attribute__((visibility("default")))
+#else
+#define ACL_FUNC_VISIBILITY
+#endif
+#endif
+
+#ifdef __GNUC__
+#define ACL_DEPRECATED __attribute__((deprecated))
+#define ACL_DEPRECATED_MESSAGE(message) __attribute__((deprecated(message)))
+#elif defined(_MSC_VER)
+#define ACL_DEPRECATED __declspec(deprecated)
+#define ACL_DEPRECATED_MESSAGE(message) __declspec(deprecated(message))
+#else
+#define ACL_DEPRECATED
+#define ACL_DEPRECATED_MESSAGE(message)
+#endif
+
+typedef void *aclrtStream;
+typedef void *aclrtEvent;
+typedef void *aclrtContext;
+typedef int aclError;
+typedef uint16_t aclFloat16;
+typedef struct aclDataBuffer aclDataBuffer;
+typedef struct aclTensorDesc aclTensorDesc;
+
+static const int ACL_ERROR_NONE = 0;
+static const int ACL_SUCCESS = 0;
+
+static const int ACL_ERROR_INVALID_PARAM = 100000;
+static const int ACL_ERROR_UNINITIALIZE = 100001;
+static const int ACL_ERROR_REPEAT_INITIALIZE = 100002;
+static const int ACL_ERROR_INVALID_FILE = 100003;
+static const int ACL_ERROR_WRITE_FILE = 100004;
+static const int ACL_ERROR_INVALID_FILE_SIZE = 100005;
+static const int ACL_ERROR_PARSE_FILE = 100006;
+static const int ACL_ERROR_FILE_MISSING_ATTR = 100007;
+static const int ACL_ERROR_FILE_ATTR_INVALID = 100008;
+static const int ACL_ERROR_INVALID_DUMP_CONFIG = 100009;
+static const int ACL_ERROR_INVALID_PROFILING_CONFIG = 100010;
+static const int ACL_ERROR_INVALID_MODEL_ID = 100011;
+static const int ACL_ERROR_DESERIALIZE_MODEL = 100012;
+static const int ACL_ERROR_PARSE_MODEL = 100013;
+static const int ACL_ERROR_READ_MODEL_FAILURE = 100014;
+static const int ACL_ERROR_MODEL_SIZE_INVALID = 100015;
+static const int ACL_ERROR_MODEL_MISSING_ATTR = 100016;
+static const int ACL_ERROR_MODEL_INPUT_NOT_MATCH = 100017;
+static const int ACL_ERROR_MODEL_OUTPUT_NOT_MATCH = 100018;
+static const int ACL_ERROR_MODEL_NOT_DYNAMIC = 100019;
+static const int ACL_ERROR_OP_TYPE_NOT_MATCH = 100020;
+static const int ACL_ERROR_OP_INPUT_NOT_MATCH = 100021;
+static const int ACL_ERROR_OP_OUTPUT_NOT_MATCH = 100022;
+static const int ACL_ERROR_OP_ATTR_NOT_MATCH = 100023;
+static const int ACL_ERROR_OP_NOT_FOUND = 100024;
+static const int ACL_ERROR_OP_LOAD_FAILED = 100025;
+static const int ACL_ERROR_UNSUPPORTED_DATA_TYPE = 100026;
+static const int ACL_ERROR_FORMAT_NOT_MATCH = 100027;
+static const int ACL_ERROR_BIN_SELECTOR_NOT_REGISTERED = 100028;
+static const int ACL_ERROR_KERNEL_NOT_FOUND = 100029;
+static const int ACL_ERROR_BIN_SELECTOR_ALREADY_REGISTERED = 100030;
+static const int ACL_ERROR_KERNEL_ALREADY_REGISTERED = 100031;
+static const int ACL_ERROR_INVALID_QUEUE_ID = 100032;
+static const int ACL_ERROR_REPEAT_SUBSCRIBE = 100033;
+static const int ACL_ERROR_STREAM_NOT_SUBSCRIBE = 100034;
+static const int ACL_ERROR_THREAD_NOT_SUBSCRIBE = 100035;
+static const int ACL_ERROR_WAIT_CALLBACK_TIMEOUT = 100036;
+static const int ACL_ERROR_REPEAT_FINALIZE = 100037;
+static const int ACL_ERROR_NOT_STATIC_AIPP = 100038;
+static const int ACL_ERROR_COMPILING_STUB_MODE = 100039;
+static const int ACL_ERROR_GROUP_NOT_SET = 100040;
+static const int ACL_ERROR_GROUP_NOT_CREATE = 100041;
+static const int ACL_ERROR_PROF_ALREADY_RUN = 100042;
+static const int ACL_ERROR_PROF_NOT_RUN = 100043;
+static const int ACL_ERROR_DUMP_ALREADY_RUN = 100044;
+static const int ACL_ERROR_DUMP_NOT_RUN = 100045;
+static const int ACL_ERROR_PROF_REPEAT_SUBSCRIBE = 148046;
+static const int ACL_ERROR_PROF_API_CONFLICT = 148047;
+static const int ACL_ERROR_INVALID_MAX_OPQUEUE_NUM_CONFIG = 148048;
+static const int ACL_ERROR_INVALID_OPP_PATH = 148049;
+static const int ACL_ERROR_OP_UNSUPPORTED_DYNAMIC = 148050;
+static const int ACL_ERROR_RELATIVE_RESOURCE_NOT_CLEARED = 148051;
+static const int ACL_ERROR_UNSUPPORTED_JPEG = 148052;
+
+static const int ACL_ERROR_BAD_ALLOC = 200000;
+static const int ACL_ERROR_API_NOT_SUPPORT = 200001;
+static const int ACL_ERROR_INVALID_DEVICE = 200002;
+static const int ACL_ERROR_MEMORY_ADDRESS_UNALIGNED = 200003;
+static const int ACL_ERROR_RESOURCE_NOT_MATCH = 200004;
+static const int ACL_ERROR_INVALID_RESOURCE_HANDLE = 200005;
+static const int ACL_ERROR_FEATURE_UNSUPPORTED = 200006;
+static const int ACL_ERROR_PROF_MODULES_UNSUPPORTED = 200007;
+
+static const int ACL_ERROR_STORAGE_OVER_LIMIT = 300000;
+
+static const int ACL_ERROR_INTERNAL_ERROR = 500000;
+static const int ACL_ERROR_FAILURE = 500001;
+static const int ACL_ERROR_GE_FAILURE = 500002;
+static const int ACL_ERROR_RT_FAILURE = 500003;
+static const int ACL_ERROR_DRV_FAILURE = 500004;
+static const int ACL_ERROR_PROFILING_FAILURE = 500005;
+
+#define ACL_TENSOR_SHAPE_RANGE_NUM 2
+#define ACL_TENSOR_VALUE_RANGE_NUM 2
+#define ACL_UNKNOWN_RANK 0xFFFFFFFFFFFFFFFE
+
+typedef enum {
+  ACL_DT_UNDEFINED = -1,
+  ACL_FLOAT = 0,
+  ACL_FLOAT16 = 1,
+  ACL_INT8 = 2,
+  ACL_INT32 = 3,
+  ACL_UINT8 = 4,
+  ACL_INT16 = 6,
+  ACL_UINT16 = 7,
+  ACL_UINT32 = 8,
+  ACL_INT64 = 9,
+  ACL_UINT64 = 10,
+  ACL_DOUBLE = 11,
+  ACL_BOOL = 12,
+  ACL_STRING = 13,
+  ACL_COMPLEX64 = 16,
+  ACL_COMPLEX128 = 17,
+  ACL_BF16 = 27
+} aclDataType;
+
+typedef enum {
+  ACL_FORMAT_UNDEFINED = -1,
+  ACL_FORMAT_NCHW = 0,
+  ACL_FORMAT_NHWC = 1,
+  ACL_FORMAT_ND = 2,
+  ACL_FORMAT_NC1HWC0 = 3,
+  ACL_FORMAT_FRACTAL_Z = 4,
+  ACL_FORMAT_NC1HWC0_C04 = 12,
+  ACL_FORMAT_HWCN = 16,
+  ACL_FORMAT_NDHWC = 27,
+  ACL_FORMAT_FRACTAL_NZ = 29,
+  ACL_FORMAT_NCDHW = 30,
+  ACL_FORMAT_NDC1HWC0 = 32,
+  ACL_FRACTAL_Z_3D = 33
+} aclFormat;
+
+typedef enum {
+  ACL_DEBUG = 0,
+  ACL_INFO = 1,
+  ACL_WARNING = 2,
+  ACL_ERROR = 3,
+} aclLogLevel;
+
+typedef enum { ACL_MEMTYPE_DEVICE = 0, ACL_MEMTYPE_HOST = 1, ACL_MEMTYPE_HOST_COMPILE_INDEPENDENT = 2 } aclMemType;
+
+/**
+ * @ingroup AscendCL
+ * @brief Converts data of type aclFloat16 to data of type float
+ *
+ * @param value [IN]   Data to be converted
+ *
+ * @retval Transformed data
+ */
+ACL_FUNC_VISIBILITY float aclFloat16ToFloat(aclFloat16 value);
+
+/**
+ * @ingroup AscendCL
+ * @brief Converts data of type float to data of type aclFloat16
+ *
+ * @param value [IN]   Data to be converted
+ *
+ * @retval Transformed data
+ */
+ACL_FUNC_VISIBILITY aclFloat16 aclFloatToFloat16(float value);
+
+/**
+ * @ingroup AscendCL
+ * @brief create data of aclDataBuffer
+ *
+ * @param data [IN]    pointer to data
+ * @li Need to be managed by the user,
+ *  call aclrtMalloc interface to apply for memory,
+ *  call aclrtFree interface to release memory
+ *
+ * @param size [IN]    size of data in bytes
+ *
+ * @retval pointer to created instance. nullptr if run out of memory
+ *
+ * @see aclrtMalloc | aclrtFree
+ */
+ACL_FUNC_VISIBILITY aclDataBuffer *aclCreateDataBuffer(void *data, size_t size);
+
+/**
+ * @ingroup AscendCL
+ * @brief destroy data of aclDataBuffer
+ *
+ * @par Function
+ *  Only the aclDataBuffer type data is destroyed here.
+ *  The memory of the data passed in when the aclDataDataBuffer interface
+ *  is called to create aclDataBuffer type data must be released by the user
+ *
+ * @param  dataBuffer [IN]   pointer to the aclDataBuffer
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclCreateDataBuffer
+ */
+ACL_FUNC_VISIBILITY aclError aclDestroyDataBuffer(const aclDataBuffer *dataBuffer);
+
+/**
+ * @ingroup AscendCL
+ * @brief update new data of aclDataBuffer
+ *
+ * @param dataBuffer [OUT]    pointer to aclDataBuffer
+ * @li The old data need to be released by the user, otherwise it may occur memory leak leakage
+ *  call aclGetDataBufferAddr interface to get old data address
+ *  call aclrtFree interface to release memory
+ *
+ * @param data [IN]    pointer to new data
+ * @li Need to be managed by the user,
+ *  call aclrtMalloc interface to apply for memory,
+ *  call aclrtFree interface to release memory
+ *
+ * @param size [IN]    size of data in bytes
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclrtMalloc | aclrtFree | aclGetDataBufferAddr
+ */
+ACL_FUNC_VISIBILITY aclError aclUpdateDataBuffer(aclDataBuffer *dataBuffer, void *data, size_t size);
+
+/**
+ * @ingroup AscendCL
+ * @brief get data address from aclDataBuffer
+ *
+ * @param dataBuffer [IN]    pointer to the data of aclDataBuffer
+ *
+ * @retval data address
+ */
+ACL_FUNC_VISIBILITY void *aclGetDataBufferAddr(const aclDataBuffer *dataBuffer);
+
+/**
+ * @ingroup AscendCL
+ * @brief get data size of aclDataBuffer
+ *
+ * @param  dataBuffer [IN]    pointer to the data of aclDataBuffer
+ *
+ * @retval data size
+ */
+ACL_DEPRECATED_MESSAGE("aclGetDataBufferSize is deprecated, use aclGetDataBufferSizeV2 instead")
+ACL_FUNC_VISIBILITY uint32_t aclGetDataBufferSize(const aclDataBuffer *dataBuffer);
+
+/**
+ * @ingroup AscendCL
+ * @brief get data size of aclDataBuffer to replace aclGetDataBufferSize
+ *
+ * @param  dataBuffer [IN]    pointer to the data of aclDataBuffer
+ *
+ * @retval data size
+ */
+ACL_FUNC_VISIBILITY size_t aclGetDataBufferSizeV2(const aclDataBuffer *dataBuffer);
+
+/**
+ * @ingroup AscendCL
+ * @brief get size of aclDataType
+ *
+ * @param  dataType [IN]    aclDataType data the size to get
+ *
+ * @retval size of the aclDataType
+ */
+ACL_FUNC_VISIBILITY size_t aclDataTypeSize(aclDataType dataType);
+
+// interfaces of tensor desc
+/**
+ * @ingroup AscendCL
+ * @brief create data aclTensorDesc
+ *
+ * @param  dataType [IN]    Data types described by tensor
+ * @param  numDims [IN]     the number of dimensions of the shape
+ * @param  dims [IN]        the size of the specified dimension
+ * @param  format [IN]      tensor format
+ *
+ * @retval aclTensorDesc pointer.
+ * @retval nullptr if param is invalid or run out of memory
+ */
+ACL_FUNC_VISIBILITY aclTensorDesc *aclCreateTensorDesc(aclDataType dataType, int numDims, const int64_t *dims,
+                                                       aclFormat format);
+
+/**
+ * @ingroup AscendCL
+ * @brief destroy data aclTensorDesc
+ *
+ * @param desc [IN]     pointer to the data of aclTensorDesc to destroy
+ */
+ACL_FUNC_VISIBILITY void aclDestroyTensorDesc(const aclTensorDesc *desc);
+
+/**
+ * @ingroup AscendCL
+ * @brief set tensor shape range for aclTensorDesc
+ *
+ * @param  desc [OUT]     pointer to the data of aclTensorDesc
+ * @param  dimsCount [IN]     the number of dimensions of the shape
+ * @param  dimsRange [IN]     the range of dimensions of the shape
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclSetTensorShapeRange(aclTensorDesc *desc, size_t dimsCount,
+                                                    int64_t dimsRange[][ACL_TENSOR_SHAPE_RANGE_NUM]);
+
+/**
+ * @ingroup AscendCL
+ * @brief set value range for aclTensorDesc
+ *
+ * @param  desc [OUT]     pointer to the data of aclTensorDesc
+ * @param  valueCount [IN]     the number of value
+ * @param  valueRange [IN]     the range of value
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclSetTensorValueRange(aclTensorDesc *desc, size_t valueCount,
+                                                    int64_t valueRange[][ACL_TENSOR_VALUE_RANGE_NUM]);
+/**
+ * @ingroup AscendCL
+ * @brief get data type specified by the tensor description
+ *
+ * @param desc [IN]        pointer to the instance of aclTensorDesc
+ *
+ * @retval data type specified by the tensor description.
+ * @retval ACL_DT_UNDEFINED if description is null
+ */
+ACL_FUNC_VISIBILITY aclDataType aclGetTensorDescType(const aclTensorDesc *desc);
+
+/**
+ * @ingroup AscendCL
+ * @brief get data format specified by the tensor description
+ *
+ * @param  desc [IN]        pointer to the instance of aclTensorDesc
+ *
+ * @retval data format specified by the tensor description.
+ * @retval ACL_FORMAT_UNDEFINED if description is null
+ */
+ACL_FUNC_VISIBILITY aclFormat aclGetTensorDescFormat(const aclTensorDesc *desc);
+
+/**
+ * @ingroup AscendCL
+ * @brief get tensor size specified by the tensor description
+ *
+ * @param  desc [IN]        pointer to the instance of aclTensorDesc
+ *
+ * @retval data size specified by the tensor description.
+ * @retval 0 if description is null
+ */
+ACL_FUNC_VISIBILITY size_t aclGetTensorDescSize(const aclTensorDesc *desc);
+
+/**
+ * @ingroup AscendCL
+ * @brief get element count specified by the tensor description
+ *
+ * @param  desc [IN]        pointer to the instance of aclTensorDesc
+ *
+ * @retval element count specified by the tensor description.
+ * @retval 0 if description is null
+ */
+ACL_FUNC_VISIBILITY size_t aclGetTensorDescElementCount(const aclTensorDesc *desc);
+
+/**
+ * @ingroup AscendCL
+ * @brief get number of dims specified by the tensor description
+ *
+ * @param  desc [IN]        pointer to the instance of aclTensorDesc
+ *
+ * @retval number of dims specified by the tensor description.
+ * @retval 0 if description is null
+ * @retval ACL_UNKNOWN_RANK if the tensor dim is -2
+ */
+ACL_FUNC_VISIBILITY size_t aclGetTensorDescNumDims(const aclTensorDesc *desc);
+
+/**
+ * @ingroup AscendCL
+ * @brief Get the size of the specified dim in the tensor description
+ *
+ * @param  desc [IN]        pointer to the instance of aclTensorDesc
+ * @param  index [IN]       index of dims, start from 0.
+ *
+ * @retval dim specified by the tensor description and index.
+ * @retval -1 if description or index is invalid
+ */
+ACL_DEPRECATED_MESSAGE("aclGetTensorDescDim is deprecated, use aclGetTensorDescDimV2 instead")
+ACL_FUNC_VISIBILITY int64_t aclGetTensorDescDim(const aclTensorDesc *desc, size_t index);
+
+/**
+ * @ingroup AscendCL
+ * @brief Get the size of the specified dim in the tensor description
+ *
+ * @param  desc [IN]        pointer to the instance of aclTensorDesc
+ * @param  index [IN]       index of dims, start from 0.
+ * @param  dimSize [OUT]    size of the specified dim.
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclGetTensorDescDimV2(const aclTensorDesc *desc, size_t index, int64_t *dimSize);
+
+/**
+ * @ingroup AscendCL
+ * @brief Get the range of the specified dim in the tensor description
+ *
+ * @param  desc [IN]        pointer to the instance of aclTensorDesc
+ * @param  index [IN]       index of dims, start from 0.
+ * @param  dimRangeNum [IN]     number of dimRange.
+ * @param  dimRange [OUT]       range of the specified dim.
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclGetTensorDescDimRange(const aclTensorDesc *desc, size_t index, size_t dimRangeNum,
+                                                      int64_t *dimRange);
+
+/**
+ * @ingroup AscendCL
+ * @brief set tensor description name
+ *
+ * @param desc [OUT]       pointer to the instance of aclTensorDesc
+ * @param name [IN]        tensor description name
+ */
+ACL_FUNC_VISIBILITY void aclSetTensorDescName(aclTensorDesc *desc, const char *name);
+
+/**
+ * @ingroup AscendCL
+ * @brief get tensor description name
+ *
+ * @param  desc [IN]        pointer to the instance of aclTensorDesc
+ *
+ * @retval tensor description name.
+ * @retval empty string if description is null
+ */
+ACL_FUNC_VISIBILITY const char *aclGetTensorDescName(aclTensorDesc *desc);
+
+/**
+ * @ingroup AscendCL
+ * @brief Convert the format in the source aclTensorDesc according to
+ * the specified dstFormat to generate a new target aclTensorDesc.
+ * The format in the source aclTensorDesc remains unchanged.
+ *
+ * @param  srcDesc [IN]     pointer to the source tensor desc
+ * @param  dstFormat [IN]   destination format
+ * @param  dstDesc [OUT]    pointer to the pointer to the destination tensor desc
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclTransTensorDescFormat(const aclTensorDesc *srcDesc, aclFormat dstFormat,
+                                                      aclTensorDesc **dstDesc);
+
+/**
+ * @ingroup AscendCL
+ * @brief Set the storage format specified by the tensor description
+ *
+ * @param  desc [OUT]     pointer to the instance of aclTensorDesc
+ * @param  format [IN]    the storage format
+ *
+ * @retval ACL_SUCCESS    The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_DEPRECATED_MESSAGE("aclSetTensorStorageFormat is deprecated, use aclSetTensorFormat instead")
+ACL_FUNC_VISIBILITY aclError aclSetTensorStorageFormat(aclTensorDesc *desc, aclFormat format);
+
+/**
+ * @ingroup AscendCL
+ * @brief Set the storage shape specified by the tensor description
+ *
+ * @param  desc [OUT]      pointer to the instance of aclTensorDesc
+ * @param  numDims [IN]    the number of dimensions of the shape
+ * @param  dims [IN]       the size of the specified dimension
+ *
+ * @retval ACL_SUCCESS     The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_DEPRECATED_MESSAGE("aclSetTensorStorageShape is deprecated, use aclSetTensorShape instead")
+ACL_FUNC_VISIBILITY aclError aclSetTensorStorageShape(aclTensorDesc *desc, int numDims, const int64_t *dims);
+
+/**
+ * @ingroup AscendCL
+ * @brief Set the format specified by the tensor description
+ *
+ * @param  desc [OUT]     pointer to the instance of aclTensorDesc
+ * @param  format [IN]    the storage format
+ *
+ * @retval ACL_SUCCESS    The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclSetTensorFormat(aclTensorDesc *desc, aclFormat format);
+
+/**
+ * @ingroup AscendCL
+ * @brief Set the shape specified by the tensor description
+ *
+ * @param  desc [OUT]      pointer to the instance of aclTensorDesc
+ * @param  numDims [IN]    the number of dimensions of the shape
+ * @param  dims [IN]       the size of the specified dimension
+ *
+ * @retval ACL_SUCCESS     The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclSetTensorShape(aclTensorDesc *desc, int numDims, const int64_t *dims);
+
+/**
+ * @ingroup AscendCL
+ * @brief Set the original format specified by the tensor description
+ *
+ * @param  desc [OUT]     pointer to the instance of aclTensorDesc
+ * @param  format [IN]    the storage format
+ *
+ * @retval ACL_SUCCESS    The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclSetTensorOriginFormat(aclTensorDesc *desc, aclFormat format);
+
+/**
+ * @ingroup AscendCL
+ * @brief Set the original shape specified by the tensor description
+ *
+ * @param  desc [OUT]      pointer to the instance of aclTensorDesc
+ * @param  numDims [IN]    the number of dimensions of the shape
+ * @param  dims [IN]       the size of the specified dimension
+ *
+ * @retval ACL_SUCCESS     The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclSetTensorOriginShape(aclTensorDesc *desc, int numDims, const int64_t *dims);
+
+/**
+ * @ingroup AscendCL
+ * @brief get op description info
+ *
+ * @param desc [IN]     pointer to tensor description
+ * @param index [IN]    index of tensor
+ *
+ * @retval null for failed.
+ * @retval OtherValues success.
+ */
+ACL_FUNC_VISIBILITY aclTensorDesc *aclGetTensorDescByIndex(aclTensorDesc *desc, size_t index);
+
+/**
+ * @ingroup AscendCL
+ * @brief get address of tensor
+ *
+ * @param desc [IN]    pointer to tensor description
+ *
+ * @retval null for failed
+ * @retval OtherValues success
+ */
+ACL_FUNC_VISIBILITY void *aclGetTensorDescAddress(const aclTensorDesc *desc);
+
+/**
+ * @ingroup AscendCL
+ * @brief Set the dynamic input name specified by the tensor description
+ *
+ * @param  desc [OUT]      pointer to the instance of aclTensorDesc
+ * @param  dynamicInputName [IN]       pointer to the dynamic input name
+ *
+ * @retval ACL_SUCCESS     The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclSetTensorDynamicInput(aclTensorDesc *desc, const char *dynamicInputName);
+
+/**
+ * @ingroup AscendCL
+ * @brief Set const data specified by the tensor description
+ *
+ * @param  desc [OUT]      pointer to the instance of aclTensorDesc
+ * @param  dataBuffer [IN]       pointer to the const databuffer
+ * @param  length [IN]       the length of const databuffer
+ *
+ * @retval ACL_SUCCESS     The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclSetTensorConst(aclTensorDesc *desc, void *dataBuffer, size_t length);
+
+/**
+ * @ingroup AscendCL
+ * @brief Set tensor memory type specified by the tensor description
+ *
+ * @param  desc [OUT]      pointer to the instance of aclTensorDesc
+ * @param  memType [IN]       ACL_MEMTYPE_DEVICE means device, ACL_MEMTYPE_HOST or
+ * ACL_MEMTYPE_HOST_COMPILE_INDEPENDENT means host
+ *
+ * @retval ACL_SUCCESS     The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclSetTensorPlaceMent(aclTensorDesc *desc, aclMemType memType);
+
+/**
+ * @ingroup AscendCL
+ * @brief an interface for users to output  APP logs
+ *
+ * @param logLevel [IN]    the level of current log
+ * @param func [IN]        the function where the log is located
+ * @param file [IN]        the file where the log is located
+ * @param line [IN]        Number of source lines where the log is located
+ * @param fmt [IN]         the format of current log
+ * @param ... [IN]         the value of current log
+ */
+ACL_FUNC_VISIBILITY void aclAppLog(aclLogLevel logLevel, const char *func, const char *file, uint32_t line,
+                                   const char *fmt, ...);
+
+/**
+ * @ingroup AscendCL
+ * @brief get soc name
+ *
+ * @retval null for failed
+ * @retval OtherValues success
+ */
+ACL_FUNC_VISIBILITY const char *aclrtGetSocName();
+
+#define ACL_APP_LOG(level, fmt, ...) aclAppLog(level, __FUNCTION__, __FILE__, __LINE__, fmt, ##__VA_ARGS__)
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // INC_EXTERNAL_ACL_ACL_BASE_H_

inc/external/acl/acl_op.h

@@ -0,0 +1,532 @@
+/**
+ * Copyright 2019-2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef INC_EXTERNAL_ACL_ACL_OP_H_
+#define INC_EXTERNAL_ACL_ACL_OP_H_
+
+#include "acl_base.h"
+#include "acl_rt.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef struct aclopHandle aclopHandle;
+typedef struct aclopAttr aclopAttr;
+typedef struct aclopKernelDesc aclopKernelDesc;
+
+typedef void (*aclDataDeallocator)(void *data, size_t length);
+
+static const int ACL_COMPILE_FLAG_BIN_SELECTOR = 1;
+
+typedef enum aclEngineType {
+  ACL_ENGINE_SYS,
+  ACL_ENGINE_AICORE,
+  ACL_ENGINE_VECTOR,
+} aclopEngineType;
+
+/**
+ * @ingroup AscendCL
+ * @brief Set base directory that contains single op models
+ *
+ * @par Restriction
+ * The aclopSetModelDir interface can be called only once in a process.
+ * @param  modelDir [IN]   path of the directory
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetModelDir(const char *modelDir);
+
+/**
+ * @ingroup AscendCL
+ * @brief load single op models from memory
+ *
+ * @par Restriction
+ * The aclopLoad interface can be called more than one times in a process.
+ * @param model [IN]        address of single op models
+ * @param modelSize [IN]    size of single op models
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopLoad(const void *model, size_t modelSize);
+
+/**
+ * @ingroup AscendCL
+ * @brief create data of type aclopAttr
+ *
+ * @retval pointer to created instance.
+ * @retval nullptr if run out of memory
+ */
+ACL_FUNC_VISIBILITY aclopAttr *aclopCreateAttr();
+
+/**
+ * @ingroup AscendCL
+ * @brief destroy data of typ aclopAttr
+ *
+ * @param attr [IN]   pointer to the instance of aclopAttr
+ */
+ACL_FUNC_VISIBILITY void aclopDestroyAttr(const aclopAttr *attr);
+
+/**
+ * @ingroup AscendCL
+ * @brief set an attribute. the type of the attribute is bool
+ *
+ * @param attr [OUT]       pointer to the instance of aclopAttr
+ * @param attrName [IN]    attribute name
+ * @param attrValue [IN]   attribute value
+ *                         false if attrValue is 0, true otherwise.
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetAttrBool(aclopAttr *attr, const char *attrName, uint8_t attrValue);
+
+/**
+ * @ingroup AscendCL
+ * @brief set an attribute. the type of the attribute is int64_t
+ *
+ * @param attr [OUT]       pointer to the instance of aclopAttr
+ * @param attrName [IN]    attribute name
+ * @param attrValue [IN]   attribute value
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetAttrInt(aclopAttr *attr, const char *attrName, int64_t attrValue);
+
+/**
+ * @ingroup AscendCL
+ * @brief set an attribute. the type of the attribute is float
+ *
+ * @param attr [OUT]       pointer to the instance of aclopAttr
+ * @param attrName [IN]    attribute name
+ * @param attrValue [IN]   attribute value
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetAttrFloat(aclopAttr *attr, const char *attrName, float attrValue);
+
+/**
+ * @ingroup AscendCL
+ * @brief set an attribute. the type of the attribute is string
+ *
+ * @param attr [OUT]       pointer to the instance of aclopAttr
+ * @param attrName [IN]    attribute name
+ * @param attrValue [IN]   attribute value
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetAttrString(aclopAttr *attr, const char *attrName, const char *attrValue);
+
+/**
+ * @ingroup AscendCL
+ * @brief set an attribute. the type of the attribute is aclDataType
+ *
+ * @param attr [OUT]       pointer to the instance of aclopAttr
+ * @param attrName [IN]    attribute name
+ * @param attrValue [IN]   attribute value
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetAttrDataType(aclopAttr *attr, const char *attrName, aclDataType attrValue);
+
+/**
+ * @ingroup AscendCL
+ * @brief set an attribute. the type of the attribute is list of aclDataType
+ *
+ * @param attr [OUT]       pointer to the instance of aclopAttr
+ * @param attrName [IN]    attribute name
+ * @param numValues [IN]   number of values. false if attrValue is 0, true otherwise.
+ * @param values [IN]      pointer to values
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetAttrListDataType(aclopAttr *attr, const char *attrName, int numValues,
+                                                      const aclDataType values[]);
+
+/**
+ * @ingroup AscendCL
+ * @brief set an attribute. the type of the attribute is list of bools
+ *
+ * @param attr [OUT]       pointer to the instance of aclopAttr
+ * @param attrName [IN]    attribute name
+ * @param numValues [IN]   number of values. false if attrValue is 0, true otherwise.
+ * @param values [IN]      pointer to values
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetAttrListBool(aclopAttr *attr, const char *attrName, int numValues,
+                                                  const uint8_t *values);
+
+/**
+ * @ingroup AscendCL
+ * @brief set an attribute. the type of the attribute is list of ints
+ *
+ * @param attr [OUT]       pointer to the instance of aclopAttr
+ * @param attrName [IN]    attribute name
+ * @param numValues [IN]   number of values
+ * @param values [IN]      pointer to values
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetAttrListInt(aclopAttr *attr, const char *attrName, int numValues,
+                                                 const int64_t *values);
+
+/**
+ * @ingroup AscendCL
+ * @brief set an attribute. the type of the attribute is list of floats
+ *
+ * @param attr [OUT]       pointer to the instance of aclopAttr
+ * @param attrName [IN]    attribute name
+ * @param numValues [IN]   number of values
+ * @param values [IN]      pointer to values
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetAttrListFloat(aclopAttr *attr, const char *attrName, int numValues,
+                                                   const float *values);
+
+/**
+ * @ingroup AscendCL
+ * @brief set an attribute. the type of the attribute is list of strings
+ *
+ * @param attr [OUT]       pointer to the instance of aclopAttr
+ * @param attrName [IN]    attribute name
+ * @param numValues [IN]   number of values
+ * @param values [IN]      pointer to values
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetAttrListString(aclopAttr *attr, const char *attrName, int numValues,
+                                                    const char **values);
+
+/**
+ * @ingroup AscendCL
+ * @brief set an attribute. the type of the attribute is list of list of ints
+ *
+ * @param attr [OUT]       pointer to the instance of aclopAttr
+ * @param attrName [IN]    attribute name
+ * @param numLists [IN]    number of lists
+ * @param numValues [IN]   pointer to number of values of each list
+ * @param values [IN]      pointer to values
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetAttrListListInt(aclopAttr *attr, const char *attrName, int numLists,
+                                                     const int *numValues, const int64_t *const values[]);
+
+/**
+ * @ingroup AscendCL
+ * @brief Load and execute the specified operator asynchronously
+ *
+ * @par Restriction
+ * @li The input and output organization of each operator is different,
+ * and the application needs to organize the operator strictly
+ * according to the operator input and output parameters when calling.
+ * @li When the user calls aclopExecute,
+ * the ACL finds the corresponding task according to the optype,
+ * the description of the input tesnsor,
+ * the description of the output tesnsor, and attr, and issues the execution.
+ *
+ * @param opType [IN]      type of op
+ * @param numInputs [IN]   number of inputs
+ * @param inputDesc [IN]   pointer to array of input tensor descriptions
+ * @param inputs [IN]      pointer to array of input buffers
+ * @param numOutputs [IN]  number of outputs
+ * @param outputDesc [IN]  pointer to array of output tensor descriptions
+ * @param outputs [OUT]    pointer to array of output buffers
+ * @param attr [IN]        pointer to instance of aclopAttr.
+ *                         may pass nullptr if the op has no attribute
+ * @param stream [IN]      stream
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_DEPRECATED_MESSAGE("aclopExecute is deprecated, use aclopExecuteV2 instead")
+ACL_FUNC_VISIBILITY aclError aclopExecute(const char *opType, int numInputs, const aclTensorDesc *const inputDesc[],
+                                          const aclDataBuffer *const inputs[], int numOutputs,
+                                          const aclTensorDesc *const outputDesc[], aclDataBuffer *const outputs[],
+                                          const aclopAttr *attr, aclrtStream stream);
+
+/**
+ * @ingroup AscendCL
+ * @brief Load and execute the specified operator
+ *        The difference with aclopExecute is that aclopExecuteV2 will refresh outputDesc
+ *
+ * @par Restriction
+ * @li The input and output organization of each operator is different,
+ * and the application needs to organize the operator strictly
+ * according to the operator input and output parameters when calling.
+ * @li When the user calls aclopExecuteV2,
+ * the ACL finds the corresponding task according to the optype,
+ * the description of the input tesnsor,
+ * the description of the output tesnsor, and attr, and issues the execution.
+ *
+ * @param opType [IN]      type of op
+ * @param numInputs [IN]   number of inputs
+ * @param inputDesc [IN]   pointer to array of input tensor descriptions
+ * @param inputs [IN]      pointer to array of input buffers
+ * @param numOutputs [IN]  number of outputs
+ * @param outputDesc [IN|OUT]  pointer to array of output tensor descriptions
+ * @param outputs [OUT]    pointer to array of output buffers
+ * @param attr [IN]        pointer to instance of aclopAttr.
+ *                         may pass nullptr if the op has no attribute
+ * @param stream [IN]      stream
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopExecuteV2(const char *opType, int numInputs, aclTensorDesc *inputDesc[],
+                                            aclDataBuffer *inputs[], int numOutputs, aclTensorDesc *outputDesc[],
+                                            aclDataBuffer *outputs[], aclopAttr *attr, aclrtStream stream);
+
+/**
+ * @ingroup AscendCL
+ * @brief create a instance of aclopHandle.
+ *
+ * @param opType [IN]      type of op
+ * @param numInputs [IN]   number of inputs
+ * @param inputDesc [IN]   pointer to array of input tensor descriptions
+ * @param numOutputs [IN]  number of outputs
+ * @param outputDesc [IN]  pointer to array of output tensor descriptions
+ * @param opAttr [IN]      pointer to instance of aclopAttr.
+ *                         may pass nullptr if the op has no attribute
+ * @param handle [OUT]     pointer to the pointer to the handle
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopCreateHandle(const char *opType, int numInputs,
+                                               const aclTensorDesc *const inputDesc[], int numOutputs,
+                                               const aclTensorDesc *const outputDesc[], const aclopAttr *opAttr,
+                                               aclopHandle **handle);
+
+/**
+ * @ingroup AscendCL
+ * @brief destroy aclopHandle instance
+ *
+ * @param handle [IN]   pointer to the instance of aclopHandle
+ */
+ACL_FUNC_VISIBILITY void aclopDestroyHandle(aclopHandle *handle);
+
+/**
+ * @ingroup AscendCL
+ * @brief execute an op with the handle.
+ *        can save op model matching cost compared with aclopExecute
+ *
+ * @param handle [IN]      pointer to the instance of aclopHandle.
+ *                         The aclopCreateHandle interface has been called
+ *                         in advance to create aclopHandle type data.
+ * @param numInputs [IN]   number of inputs
+ * @param inputs [IN]      pointer to array of input buffers.
+ *                         The aclCreateDataBuffer interface has been called
+ *                         in advance to create aclDataBuffer type data.
+ * @param numOutputs [IN]  number of outputs
+ * @param outputs [OUT]    pointer to array of output buffers
+ * @param stream [IN]      stream
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclopCreateHandle | aclCreateDataBuffer
+ */
+ACL_FUNC_VISIBILITY aclError aclopExecWithHandle(aclopHandle *handle, int numInputs,
+                                                 const aclDataBuffer *const inputs[], int numOutputs,
+                                                 aclDataBuffer *const outputs[], aclrtStream stream);
+
+/**
+ * @ingroup AscendCL
+ * @brief cast data type
+ *
+ * @param srcDesc [IN]     source tensor desc
+ * @param srcBuffer [IN]   source tensor buffer
+ * @param dstDesc [IN]     destination tensor desc
+ * @param dstBuffer [OUT]  destination tensor buffer
+ * @param truncate [IN]    do not truncate if value is 0, truncate otherwise
+ * @param stream [IN]      stream
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopCast(const aclTensorDesc *srcDesc, const aclDataBuffer *srcBuffer,
+                                       const aclTensorDesc *dstDesc, aclDataBuffer *dstBuffer, uint8_t truncate,
+                                       aclrtStream stream);
+
+/**
+ * @ingroup AscendCL
+ * @brief create a handle for casting datatype
+ *
+ * @param srcDesc [IN]    source tensor desc
+ * @param dstDesc [IN]    destination tensor desc
+ * @param truncate [IN]   do not truncate if value is 0, truncate otherwise
+ * @param handle [OUT]    pointer to the pointer to the handle
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopCreateHandleForCast(aclTensorDesc *srcDesc, aclTensorDesc *dstDesc, uint8_t truncate,
+                                                      aclopHandle **handle);
+
+/**
+ * @ingroup AscendCL
+ * @brief create kernel
+ *
+ * @param opType [IN]           op type
+ * @param kernelId [IN]         kernel id
+ * @param kernelName [IN]       kernel name
+ * @param binData [IN]          kernel bin data
+ * @param binSize [IN]          kernel bin size
+ * @param enginetype [IN]       enigne type
+ * @param deallocator [IN]      callback function for deallocating bin data,
+ *                              null if bin data to be deallocated by caller
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclopCompile
+ */
+ACL_FUNC_VISIBILITY aclError aclopCreateKernel(const char *opType, const char *kernelId, const char *kernelName,
+                                               void *binData, int binSize, aclopEngineType enginetype,
+                                               aclDataDeallocator deallocator);
+
+/**
+ * @ingroup AscendCL
+ * @brief create kernel
+ *
+ * @param numInputs [IN]            number of inputs
+ * @param inputDesc [IN]            pointer to array of input tensor descriptions
+ * @param numOutputs [IN]           number of outputs
+ * @param outputDesc [IN]           pointer to array of output tensor descriptions
+ * @param opAttr [IN]               pointer to instance of aclopAttr
+ * @param aclopKernelDesc [IN]      pointer to instance of aclopKernelDesc
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+typedef aclError (*aclopCompileFunc)(int numInputs, const aclTensorDesc *const inputDesc[], int numOutputs,
+                                     const aclTensorDesc *const outputDesc[], const aclopAttr *opAttr,
+                                     aclopKernelDesc *aclopKernelDesc);
+
+/**
+ * @ingroup AscendCL
+ * @brief register compile function
+ *
+ * @param opType [IN]         op type
+ * @param func [IN]           compile function
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclopUnregisterCompileFunc
+ */
+ACL_FUNC_VISIBILITY aclError aclopRegisterCompileFunc(const char *opType, aclopCompileFunc func);
+
+/**
+ * @ingroup AscendCL
+ * @brief unregister compile function
+ *
+ * @param opType [IN]         op type
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopUnregisterCompileFunc(const char *opType);
+
+/**
+ * @ingroup AscendCL
+ * @brief set kernel args
+ *
+ * @param kernelDesc [IN]               pointer to instance of aclopKernelDesc
+ * @param kernelId [IN]                 kernel id
+ * @param blockDim [IN]                 block dim
+ * @param args [IN]                     args
+ * @param argSize [IN]                  size in bytes of args
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetKernelArgs(aclopKernelDesc *kernelDesc, const char *kernelId, uint32_t blockDim,
+                                                const void *args, uint32_t argSize);
+
+/**
+ * @ingroup AscendCL
+ * @brief set workspace sizes
+ *
+ * @param kernelDesc [IN]               pointer to instance of aclopKernelDesc
+ * @param numWorkspaces [IN]            number of workspaces
+ * @param workspaceSizes [IN]           pointer to array of sizes of workspaces
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetKernelWorkspaceSizes(aclopKernelDesc *kernelDesc, int numWorkspaces,
+                                                          size_t *workspaceSizes);
+
+/**
+ * @ingroup AscendCL
+ * @brief compile op with dynamic shape
+ *
+ * @param opType [IN]       op type
+ * @param numInputs [IN]    number of inputs
+ * @param inputDesc [IN]    pointer to array of input tensor descriptions
+ * @param numOutputs [IN]   number of outputs
+ * @param outputDesc [IN]   pointer to array of output tensor descriptions
+ * @param attr [IN]         pointer to instance of aclopAttr.
+ *                          may pass nullptr if the op has no attribute
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopUpdateParams(const char *opType, int numInputs,
+                                               const aclTensorDesc *const inputDesc[], int numOutputs,
+                                               const aclTensorDesc *const outputDesc[], const aclopAttr *attr);
+
+/**
+ * @ingroup AscendCL
+ * @brief inferShape the specified operator synchronously
+ *
+ * @param opType [IN]       type of op
+ * @param numInputs [IN]    number of inputs
+ * @param inputDesc [IN]    pointer to array of input tensor descriptions
+ * @param inputs [IN]       pointer to array of input buffers
+ * @param numOutputs [IN]   number of outputs
+ * @param outputDesc [OUT]  pointer to array of output tensor descriptions
+ * @param attr [IN]         pointer to instance of aclopAttr.
+ *                          may pass nullptr if the op has no attribute
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopInferShape(const char *opType, int numInputs, aclTensorDesc *inputDesc[],
+                                             aclDataBuffer *inputs[], int numOutputs, aclTensorDesc *outputDesc[],
+                                             aclopAttr *attr);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // INC_EXTERNAL_ACL_ACL_OP_H_

inc/external/acl/acl_op_compiler.h

@@ -0,0 +1,200 @@
+/**
+ * Copyright 2019-2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef INC_EXTERNAL_ACL_ACL_OP_COMPILER_H_
+#define INC_EXTERNAL_ACL_ACL_OP_COMPILER_H_
+
+#include "acl_base.h"
+#include "acl_op.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef enum aclCompileType { ACL_COMPILE_SYS, ACL_COMPILE_UNREGISTERED } aclopCompileType;
+
+typedef enum {
+  ACL_PRECISION_MODE,
+  ACL_AICORE_NUM,
+  ACL_AUTO_TUNE_MODE,
+  ACL_OP_SELECT_IMPL_MODE,
+  ACL_OPTYPELIST_FOR_IMPLMODE,
+  ACL_OP_DEBUG_LEVEL,
+  ACL_DEBUG_DIR,
+  ACL_OP_COMPILER_CACHE_MODE,
+  ACL_OP_COMPILER_CACHE_DIR,
+  ACL_OP_PERFORMANCE_MODE
+} aclCompileOpt;
+
+typedef enum aclCompileFlag { ACL_OP_COMPILE_DEFAULT, ACL_OP_COMPILE_FUZZ } aclOpCompileFlag;
+
+typedef struct aclGraphDumpOption aclGraphDumpOption;
+
+/**
+ * @ingroup AscendCL
+ * @brief compile op
+ *
+ * @param opType [IN]           op type
+ * @param numInputs [IN]        number of inputs
+ * @param inputDesc [IN]        pointer to array of input tensor descriptions
+ * @param numOutputs [IN]       number of outputs
+ * @param outputDesc [IN]       pointer to array of output tensor descriptions
+ * @param attr [IN]           pointer to instance of aclopAttr.
+ *                              may pass nullptr if the op has no attribute
+ * @param engineType [IN]       engine type
+ * @param compileFlag [IN]      compile flag
+ * @param opPath [IN]           path of op
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopCompile(const char *opType, int numInputs, const aclTensorDesc *const inputDesc[],
+                                          int numOutputs, const aclTensorDesc *const outputDesc[],
+                                          const aclopAttr *attr, aclopEngineType engineType,
+                                          aclopCompileType compileFlag, const char *opPath);
+
+/**
+ * @ingroup AscendCL
+ * @brief compile and execute op
+ *
+ * @param opType [IN]           op type
+ * @param numInputs [IN]        number of inputs
+ * @param inputDesc [IN]        pointer to array of input tensor descriptions
+ * @param inputs [IN]           pointer to array of input buffers
+ * @param numOutputs [IN]       number of outputs
+ * @param outputDesc [IN]       pointer to array of output tensor descriptions
+ * @param outputs [IN]          pointer to array of outputs buffers
+ * @param attr [IN]             pointer to instance of aclopAttr.
+ *                              may pass nullptr if the op has no attribute
+ * @param engineType [IN]       engine type
+ * @param compileFlag [IN]      compile flag
+ * @param opPath [IN]           path of op
+ * @param stream [IN]           stream handle
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopCompileAndExecute(
+    const char *opType, int numInputs, const aclTensorDesc *const inputDesc[], const aclDataBuffer *const inputs[],
+    int numOutputs, const aclTensorDesc *const outputDesc[], aclDataBuffer *const outputs[], const aclopAttr *attr,
+    aclopEngineType engineType, aclopCompileType compileFlag, const char *opPath, aclrtStream stream);
+
+/**
+ * @ingroup AscendCL
+ * @brief compile and execute op
+ *
+ * @param opType [IN]           op type
+ * @param numInputs [IN]        number of inputs
+ * @param inputDesc [IN]        pointer to array of input tensor descriptions
+ * @param inputs [IN]           pointer to array of input buffers
+ * @param numOutputs [IN]       number of outputs
+ * @param outputDesc [IN|OUT]   pointer to array of output tensor descriptions
+ * @param outputs [IN]          pointer to array of outputs buffers
+ * @param attr [IN]             pointer to instance of aclopAttr.
+ *                              may pass nullptr if the op has no attribute
+ * @param engineType [IN]       engine type
+ * @param compileFlag [IN]      compile flag
+ * @param opPath [IN]           path of op
+ * @param stream [IN]           stream handle
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopCompileAndExecuteV2(const char *opType, int numInputs, aclTensorDesc *inputDesc[],
+                                                      aclDataBuffer *inputs[], int numOutputs,
+                                                      aclTensorDesc *outputDesc[], aclDataBuffer *outputs[],
+                                                      aclopAttr *attr, aclopEngineType engineType,
+                                                      aclopCompileType compileFlag, const char *opPath,
+                                                      aclrtStream stream);
+
+/**
+ * @ingroup AscendCL
+ * @brief set compile option
+ *
+ * @param aclCompileOpt [IN]      compile option
+ * @param value [IN]              pointer for the option value
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclSetCompileopt(aclCompileOpt opt, const char *value);
+
+/**
+ * @ingroup AscendCL
+ * @brief set compile flag
+ *
+ * @param flag [IN]    compile flag, ACL_OP_COMPILE_DEFAULT means compile with default mode
+ *                     ACL_OP_COMPILE_FUZZ means compile with fuzz mode
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetCompileFlag(aclOpCompileFlag flag);
+
+/**
+ * @ingroup AscendCL
+ * @brief generate graph and dump
+ *
+ * @param opType [IN]           op type
+ * @param numInputs [IN]        number of inputs
+ * @param inputDesc [IN]        pointer to array of input tensor descriptions
+ * @param inputs [IN]           pointer to array of input buffers
+ * @param numOutputs [IN]       number of outputs
+ * @param outputDesc [IN]       pointer to array of output tensor descriptions
+ * @param outputs [IN]          pointer to array of outputs buffers
+ * @param attr [IN]             pointer to instance of aclopAttr.
+ *                              may pass nullptr if the op has no attribute
+ * @param engineType [IN]       engine type
+ * @param graphDumpPath [IN]    dump path, if the suffix is ".txt", it means file path, else it means directory path
+ * @param graphDumpOpt [IN]     dump option, nullptr is supported
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclGenGraphAndDumpForOp(
+    const char *opType, int numInputs, const aclTensorDesc *const inputDesc[], const aclDataBuffer *const inputs[],
+    int numOutputs, const aclTensorDesc *const outputDesc[], aclDataBuffer *const outputs[], const aclopAttr *attr,
+    aclopEngineType engineType, const char *graphDumpPath, const aclGraphDumpOption *graphDumpOpt);
+
+/**
+ * @ingroup AscendCL
+ * @brief Create the graph dump option
+ *
+ * @retval null for failed
+ * @retval OtherValues success
+ *
+ * @see aclDestroyGraphDumpOpt
+ */
+ACL_FUNC_VISIBILITY aclGraphDumpOption *aclCreateGraphDumpOpt();
+
+/**
+ * @ingroup AscendCL
+ * @brief Destroy graph dump option
+ *
+ * @param graphDumpOpt [IN]  pointer to the graph dump option
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclCreateGraphDumpOpt
+ */
+ACL_FUNC_VISIBILITY aclError aclDestroyGraphDumpOpt(const aclGraphDumpOption *graphDumpOpt);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // INC_EXTERNAL_ACL_ACL_OP_COMPILER_H_

inc/external/acl/acl_prof.h

@@ -0,0 +1,485 @@
+/**
+ * Copyright 2019-2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef INC_EXTERNAL_ACL_PROF_H_
+#define INC_EXTERNAL_ACL_PROF_H_
+
+#if (defined(_WIN32) || defined(_WIN64) || defined(_MSC_VER))
+#define MSVP_PROF_API __declspec(dllexport)
+#else
+#define MSVP_PROF_API __attribute__((visibility("default")))
+#endif
+
+#include "acl_base.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define ACL_PROF_ACL_API 0x0001ULL
+#define ACL_PROF_TASK_TIME 0x0002ULL
+#define ACL_PROF_AICORE_METRICS 0x0004ULL
+#define ACL_PROF_AICPU 0x0008ULL
+#define ACL_PROF_L2CACHE 0x0010ULL
+#define ACL_PROF_HCCL_TRACE 0x0020ULL
+#define ACL_PROF_TRAINING_TRACE 0x0040ULL
+#define ACL_PROF_MSPROFTX 0x0080ULL
+#define ACL_PROF_RUNTIME_API 0x0100ULL
+
+/**
+ * @deprecated please use aclprofGetOpTypeLen and aclprofGetOpTNameLen instead
+ */
+#define ACL_PROF_MAX_OP_NAME_LEN 257
+#define ACL_PROF_MAX_OP_TYPE_LEN 65
+
+typedef enum {
+  ACL_AICORE_ARITHMETIC_UTILIZATION = 0,
+  ACL_AICORE_PIPE_UTILIZATION = 1,
+  ACL_AICORE_MEMORY_BANDWIDTH = 2,
+  ACL_AICORE_L0B_AND_WIDTH = 3,
+  ACL_AICORE_RESOURCE_CONFLICT_RATIO = 4,
+  ACL_AICORE_MEMORY_UB = 5,
+  ACL_AICORE_NONE = 0xFF
+} aclprofAicoreMetrics;
+
+typedef enum {
+  ACL_STEP_START = 0,  // step  start
+  ACL_STEP_END = 1     // step  end
+} aclprofStepTag;
+
+typedef enum {
+  ACL_SUBSCRIBE_OP = 0,
+  ACL_SUBSCRIBE_SUBGRAPH = 1,
+  ACL_SUBSCRIBE_OP_THREAD = 2,
+  ACL_SUBSCRIBE_NONE
+} aclprofSubscribeOpFlag;
+
+typedef enum { ACL_SUBSCRIBE_ATTRI_THREADID = 0, ACL_SUBSCRIBE_ATTRI_NONE } aclprofSubscribeOpAttri;
+
+typedef struct aclprofConfig aclprofConfig;
+typedef struct aclprofStopConfig aclprofStopConfig;
+typedef struct aclprofAicoreEvents aclprofAicoreEvents;
+typedef struct aclprofSubscribeConfig aclprofSubscribeConfig;
+typedef struct aclprofStepInfo aclprofStepInfo;
+
+/**
+ * @ingroup AscendCL
+ * @brief profiling initialize
+ *
+ * @param  profilerResultPath [IN]  path of profiling result
+ * @param  length [IN]              length of profilerResultPath
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclprofFinalize
+ */
+MSVP_PROF_API aclError aclprofInit(const char *profilerResultPath, size_t length);
+
+/**
+ * @ingroup AscendCL
+ * @brief profiling finalize
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclprofInit
+ */
+MSVP_PROF_API aclError aclprofFinalize();
+
+/**
+ * @ingroup AscendCL
+ * @brief Start profiling modules by profilerConfig
+ *
+ * @param  profilerConfig [IN]  config of profiling
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclprofStop
+ */
+MSVP_PROF_API aclError aclprofStart(const aclprofConfig *profilerConfig);
+
+/**
+ * @ingroup AscendCL
+ * @brief Create data of type aclprofConfig
+ *
+ * @param  deviceIdList [IN]      list of device id
+ * @param  deviceNums [IN]        number of devices
+ * @param  aicoreMetrics [IN]     type of aicore metrics
+ * @param  aicoreEvents [IN]      pointer to aicore events, only support NULL now
+ * @param  dataTypeConfig [IN]    config modules need profiling
+ *
+ * @retval the aclprofConfig pointer
+ *
+ * @see aclprofDestroyConfig
+ */
+MSVP_PROF_API aclprofConfig *aclprofCreateConfig(uint32_t *deviceIdList, uint32_t deviceNums,
+                                                 aclprofAicoreMetrics aicoreMetrics, aclprofAicoreEvents *aicoreEvents,
+                                                 uint64_t dataTypeConfig);
+
+/**
+ * @ingroup AscendCL
+ * @brief Destroy data of type aclprofConfig
+ *
+ * @param  profilerConfig [IN]  config of profiling
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclprofCreateConfig
+ */
+MSVP_PROF_API aclError aclprofDestroyConfig(const aclprofConfig *profilerConfig);
+
+/**
+ * @ingroup AscendCL
+ * @brief stop profiling modules by stopProfilingConfig
+ *
+ * @param  profilerConfig [IN]  pointer to stop config of profiling
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclprofStart
+ */
+MSVP_PROF_API aclError aclprofStop(const aclprofConfig *profilerConfig);
+
+/**
+ * @ingroup AscendCL
+ * @brief subscribe profiling data of model
+ *
+ * @param  modelId [IN]              the model id subscribed
+ * @param  profSubscribeConfig [IN]  pointer to config of model subscribe
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclprofModelUnSubscribe
+ */
+MSVP_PROF_API aclError aclprofModelSubscribe(uint32_t modelId, const aclprofSubscribeConfig *profSubscribeConfig);
+
+/**
+ * @ingroup AscendCL
+ * @brief unsubscribe profiling data of model
+ *
+ * @param  modelId [IN]  the model id unsubscribed
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclprofModelSubscribe
+ */
+MSVP_PROF_API aclError aclprofModelUnSubscribe(uint32_t modelId);
+
+/**
+ * @ingroup AscendCL
+ * @brief create subscribe config
+ *
+ * @param  timeInfoSwitch [IN] switch whether get time info from model
+ * @param  aicoreMetrics [IN]  aicore metrics
+ * @param  fd [IN]             pointer to write pipe
+ *
+ * @retval the aclprofSubscribeConfig pointer
+ *
+ * @see aclprofDestroySubscribeConfig
+ */
+MSVP_PROF_API aclprofSubscribeConfig *aclprofCreateSubscribeConfig(int8_t timeInfoSwitch,
+                                                                   aclprofAicoreMetrics aicoreMetrics, void *fd);
+
+/**
+ * @ingroup AscendCL
+ * @brief destroy subscribe config
+ *
+ * @param  profSubscribeConfig [IN]  subscribe config
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclprofCreateSubscribeConfig
+ */
+MSVP_PROF_API aclError aclprofDestroySubscribeConfig(const aclprofSubscribeConfig *profSubscribeConfig);
+
+/**
+ * @ingroup AscendCL
+ * @brief create subscribe config
+ *
+ * @param  opDescSize [OUT]  size of op desc
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+MSVP_PROF_API aclError aclprofGetOpDescSize(size_t *opDescSize);
+
+/**
+ * @ingroup AscendCL
+ * @brief get op number from subscription data
+ *
+ * @param  opInfo [IN]     pointer to subscription data
+ * @param  opInfoLen [IN]  memory size of subscription data
+ * @param  opNumber [OUT]  op number of subscription data
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+MSVP_PROF_API aclError aclprofGetOpNum(const void *opInfo, size_t opInfoLen, uint32_t *opNumber);
+
+/**
+ * @ingroup AscendCL
+ * @brief get length op type from subscription data
+ *
+ * @param  opInfo [IN]      pointer to subscription data
+ * @param  opInfoLen [IN]   memory size of subscription data
+ * @param  index [IN]       index of op array in opInfo
+ * @param  opTypeLen [OUT]  actual length of op type string
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+MSVP_PROF_API aclError aclprofGetOpTypeLen(const void *opInfo, size_t opInfoLen, uint32_t index, size_t *opTypeLen);
+
+/**
+ * @ingroup AscendCL
+ * @brief get op type from subscription data
+ *
+ * @param  opInfo [IN]      pointer to subscription data
+ * @param  opInfoLen [IN]   memory size of subscription data
+ * @param  index [IN]       index of op array in opInfo
+ * @param  opType [OUT]     obtained op type string
+ * @param  opTypeLen [IN]   obtained length of op type string
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+MSVP_PROF_API aclError aclprofGetOpType(const void *opInfo, size_t opInfoLen, uint32_t index, char *opType,
+                                        size_t opTypeLen);
+
+/**
+ * @ingroup AscendCL
+ * @brief get length op name from subscription data
+ *
+ * @param  opInfo [IN]      pointer to subscription data
+ * @param  opInfoLen [IN]   memory size of subscription data
+ * @param  index [IN]       index of op array in opInfo
+ * @param  opNameLen [OUT]  actual length of op name string
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+MSVP_PROF_API aclError aclprofGetOpNameLen(const void *opInfo, size_t opInfoLen, uint32_t index, size_t *opNameLen);
+
+/**
+ * @ingroup AscendCL
+ * @brief get op type from subscription data
+ *
+ * @param  opInfo [IN]      pointer to subscription data
+ * @param  opInfoLen [IN]   memory size of subscription data
+ * @param  index [IN]       index of op array in opInfo
+ * @param  opName [OUT]     obtained op name string
+ * @param  opNameLen [IN]   obtained length of op name string
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+MSVP_PROF_API aclError aclprofGetOpName(const void *opInfo, size_t opInfoLen, uint32_t index, char *opName,
+                                        size_t opNameLen);
+
+/**
+ * @ingroup AscendCL
+ * @brief get start time of specified op from subscription data
+ *
+ * @param  opInfo [IN]     pointer to subscription data
+ * @param  opInfoLen [IN]  memory size of subscription data
+ * @param  index [IN]      index of op array in opInfo
+ *
+ * @retval start time(us) of specified op with timestamp
+ * @retval 0 for failed
+ */
+MSVP_PROF_API uint64_t aclprofGetOpStart(const void *opInfo, size_t opInfoLen, uint32_t index);
+
+/**
+ * @ingroup AscendCL
+ * @brief get end time of specified op from subscription data
+ *
+ * @param  opInfo [IN]     pointer to subscription data
+ * @param  opInfoLen [IN]  memory size of subscription data
+ * @param  index [IN]      index of op array in opInfo
+ *
+ * @retval end time(us) of specified op with timestamp
+ * @retval 0 for failed
+ */
+MSVP_PROF_API uint64_t aclprofGetOpEnd(const void *opInfo, size_t opInfoLen, uint32_t index);
+
+/**
+ * @ingroup AscendCL
+ * @brief get excution time of specified op from subscription data
+ *
+ * @param  opInfo [IN]     pointer to subscription data
+ * @param  opInfoLen [IN]  memory size of subscription data
+ * @param  index [IN]      index of op array in opInfo
+ *
+ * @retval execution time(us) of specified op with timestamp
+ * @retval 0 for failed
+ */
+MSVP_PROF_API uint64_t aclprofGetOpDuration(const void *opInfo, size_t opInfoLen, uint32_t index);
+
+/**
+ * @ingroup AscendCL
+ * @brief get model id from subscription data
+ *
+ * @param  opInfo [IN]     pointer to subscription data
+ * @param  opInfoLen [IN]  memory size of subscription data
+ *
+ * @retval model id of subscription data
+ * @retval 0 for failed
+ */
+MSVP_PROF_API size_t aclprofGetModelId(const void *opInfo, size_t opInfoLen, uint32_t index);
+
+/**
+ * @ingroup AscendCL
+ * @brief get op flag from subscription data
+ *
+ * @param  opInfo [IN]     pointer to subscription data
+ * @param  opInfoLen [IN]  memory size of subscription data
+ * @param  index [IN]      index of op array in opInfo
+ *
+ * @retval op flag
+ * @retval ACL_SUBSCRIBE_NONE for failed
+ */
+MSVP_PROF_API aclprofSubscribeOpFlag aclprofGetOpFlag(const void *opInfo, size_t opInfoLen, uint32_t index);
+
+/**
+ * @ingroup AscendCL
+ * @brief get op flag from subscription data
+ *
+ * @param  opInfo [IN]     pointer to subscription data
+ * @param  opInfoLen [IN]  memory size of subscription data
+ * @param  index [IN]      index of op array in opInfo
+ * @param  attri [IN]      attribute of op
+ *
+ * @retval op flag
+ * @retval NULL for failed
+ */
+MSVP_PROF_API const char *aclprofGetOpAttriValue(const void *opInfo, size_t opInfoLen, uint32_t index,
+                                                 aclprofSubscribeOpAttri attri);
+
+/**
+ * @ingroup AscendCL
+ * @brief
+ *
+ * @param  stepInfo [IN]     pointer to stepInfo data
+ * @param  aclprofstepTag [IN] start or end flag
+ * @param  stream [IN] steam info
+ *
+ * @retval 0 for failed
+ */
+MSVP_PROF_API aclError aclprofGetStepTimestamp(aclprofStepInfo *stepInfo, aclprofStepTag tag, aclrtStream stream);
+
+/**
+ * @ingroup AscendCL
+ * @brief create pointer to aclprofStepInfo data
+ *
+ *
+ * @retval aclprofStepInfo pointer
+ */
+MSVP_PROF_API aclprofStepInfo *aclprofCreateStepInfo();
+
+/**
+ * @ingroup AscendCL
+ * @brief destroy aclprofStepInfo pointer
+ *
+ *
+ * @retval void
+ */
+MSVP_PROF_API void aclprofDestroyStepInfo(aclprofStepInfo *stepinfo);
+
+/**
+ * @ingroup AscendCL
+ * @brief create pointer to aclprofstamp
+ *
+ *
+ * @retval aclprofStamp pointer
+ */
+MSVP_PROF_API void *aclprofCreateStamp();
+
+/**
+ * @ingroup AscendCL
+ * @brief destory stamp pointer
+ *
+ *
+ * @retval void
+ */
+MSVP_PROF_API void aclprofDestroyStamp(void *stamp);
+
+/**
+ * @ingroup AscendCL
+ * @brief Record push timestamp
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+MSVP_PROF_API aclError aclprofPush(void *stamp);
+
+/**
+ * @ingroup AscendCL
+ * @brief Record pop timestamp
+ *
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+MSVP_PROF_API aclError aclprofPop();
+
+/**
+ * @ingroup AscendCL
+ * @brief Record range start timestamp
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+MSVP_PROF_API aclError aclprofRangeStart(void *stamp, uint32_t *rangeId);
+
+/**
+ * @ingroup AscendCL
+ * @brief Record range end timestamp
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+MSVP_PROF_API aclError aclprofRangeStop(uint32_t rangeId);
+
+/**
+ * @ingroup AscendCL
+ * @brief set message to stamp
+ *
+ *
+ * @retval void
+ */
+MSVP_PROF_API aclError aclprofSetStampTraceMessage(void *stamp, const char *msg, uint32_t msgLen);
+
+/**
+ * @ingroup AscendCL
+ * @brief Record mark timestamp
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+MSVP_PROF_API aclError aclprofMark(void *stamp);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // INC_EXTERNAL_ACL_PROF_H_

inc/external/acl/acl_tdt.h

@@ -0,0 +1,316 @@
+/**
+ * Copyright 2019-2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef INC_EXTERNAL_ACL_ACL_TDT_H_
+#define INC_EXTERNAL_ACL_ACL_TDT_H_
+
+#include "acl/acl_base.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+enum acltdtTensorType {
+  ACL_TENSOR_DATA_UNDEFINED = -1,
+  ACL_TENSOR_DATA_TENSOR,
+  ACL_TENSOR_DATA_END_OF_SEQUENCE,
+  ACL_TENSOR_DATA_ABNORMAL
+};
+
+typedef struct acltdtDataItem acltdtDataItem;
+typedef struct acltdtDataset acltdtDataset;
+typedef struct acltdtChannelHandle acltdtChannelHandle;
+
+/**
+ * @ingroup AscendCL
+ * @brief Get tensor type from item
+ *
+ * @param dataItem [IN] pointer to the data item
+ *
+ * @retval Tensor type.
+ * @retval ACL_DT_UNDEFINED if dataItem is null
+ */
+ACL_FUNC_VISIBILITY acltdtTensorType acltdtGetTensorTypeFromItem(const acltdtDataItem *dataItem);
+
+/**
+ * @ingroup AscendCL
+ * @brief Get data type from item
+ *
+ * @param dataItem [IN] pointer to the data item
+ *
+ * @retval Data type.
+ * @retval ACL_DT_UNDEFINED if dataItem is null
+ */
+ACL_FUNC_VISIBILITY aclDataType acltdtGetDataTypeFromItem(const acltdtDataItem *dataItem);
+
+/**
+ * @ingroup AscendCL
+ * @brief Get data address from item
+ *
+ * @param dataItem [IN] pointer to data item
+ *
+ * @retval null for failed
+ * @retval OtherValues success
+ */
+ACL_FUNC_VISIBILITY void *acltdtGetDataAddrFromItem(const acltdtDataItem *dataItem);
+
+/**
+ * @ingroup AscendCL
+ * @brief Get data size from item
+ *
+ * @param dataItem [IN] pointer to data item
+ *
+ * @retval 0 for failed
+ * @retval OtherValues success
+ */
+ACL_FUNC_VISIBILITY size_t acltdtGetDataSizeFromItem(const acltdtDataItem *dataItem);
+
+/**
+ * @ingroup AscendCL
+ * @brief Get dim's number from item
+ *
+ * @param dataItem [IN] pointer to data item
+ *
+ * @retval 0 for failed
+ * @retval OtherValues success
+ */
+ACL_FUNC_VISIBILITY size_t acltdtGetDimNumFromItem(const acltdtDataItem *dataItem);
+
+/**
+ * @ingroup AscendCL
+ * @brief Get dims from item
+ *
+ * @param  dataItem [IN]      the struct of data item
+ * @param  dims [IN|OUT]      pointer to the dims of dataTtem
+ * @param  dimNum [IN]        the size of the dims
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError acltdtGetDimsFromItem(const acltdtDataItem *dataItem, int64_t *dims, size_t dimNum);
+
+/**
+ * @ingroup AscendCL
+ * @brief Create the struct of data item
+ *
+ * @param tdtType [IN]  Tdt tensor type
+ * @param dims [IN]     pointer of tdtDataItem's dims
+ * @param dimNum [IN]   Dim number
+ * @param dataType [IN] Data type
+ * @param data [IN]     Data pointer
+ * @param size [IN]     Data size
+ *
+ * @retval null for failed
+ * @retval OtherValues success
+ *
+ * @see acltdtDestroyDataItem
+ */
+ACL_FUNC_VISIBILITY acltdtDataItem *acltdtCreateDataItem(acltdtTensorType tdtType, const int64_t *dims, size_t dimNum,
+                                                         aclDataType dataType, void *data, size_t size);
+
+/**
+ * @ingroup AscendCL
+ * @brief Destroy the struct of data item
+ *
+ * @param dataItem [IN]  pointer to the data item
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see acltdtCreateDataItem
+ */
+ACL_FUNC_VISIBILITY aclError acltdtDestroyDataItem(acltdtDataItem *dataItem);
+
+/**
+ * @ingroup AscendCL
+ * @brief Create the tdt dataset
+ *
+ * @retval null for failed
+ * @retval OtherValues success
+ *
+ * @see acltdtDestroyDataset
+ */
+ACL_FUNC_VISIBILITY acltdtDataset *acltdtCreateDataset();
+
+/**
+ * @ingroup AscendCL
+ * @brief Destroy the tdt dataset
+ *
+ * @param dataset [IN]  pointer to the dataset
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see acltdtCreateDataset
+ */
+ACL_FUNC_VISIBILITY aclError acltdtDestroyDataset(acltdtDataset *dataset);
+
+/**
+ * @ingroup AscendCL
+ * @brief Get the data item
+ *
+ * @param dataset [IN] pointer to the dataset
+ * @param index [IN]   index of the dataset
+ *
+ * @retval null for failed
+ * @retval OtherValues success
+ *
+ * @see acltdtAddDataItem
+ */
+ACL_FUNC_VISIBILITY acltdtDataItem *acltdtGetDataItem(const acltdtDataset *dataset, size_t index);
+
+/**
+ * @ingroup AscendCL
+ * @brief Get the data item
+ *
+ * @param dataset [OUT] pointer to the dataset
+ * @param dataItem [IN] pointer to the data item
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see acltdtGetDataItem
+ */
+ACL_FUNC_VISIBILITY aclError acltdtAddDataItem(acltdtDataset *dataset, acltdtDataItem *dataItem);
+
+/**
+ * @ingroup AscendCL
+ * @brief Get the size of dataset
+ *
+ * @param dataset [IN]  pointer to the dataset
+ *
+ * @retval 0 for failed
+ * @retval OtherValues success
+ */
+ACL_FUNC_VISIBILITY size_t acltdtGetDatasetSize(const acltdtDataset *dataset);
+
+/**
+ * @ingroup AscendCL
+ * @brief Get the name of dataset
+ *
+ * @param  dataset [IN]      pointer to the dataset
+ *
+ * @retval null for failed
+ * @retval OtherValues success
+ */
+ACL_FUNC_VISIBILITY const char *acltdtGetDatasetName(const acltdtDataset *dataset);
+
+/**
+ * @ingroup AscendCL
+ * @brief Stop the channel
+ *
+ * @param handle [IN]  pointer to the channel handle
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see acltdtCreateChannel | acltdtDestroyChannel
+ */
+ACL_FUNC_VISIBILITY aclError acltdtStopChannel(acltdtChannelHandle *handle);
+
+/**
+ * @ingroup AscendCL
+ * @brief Create the channel
+ *
+ * @param deviceId [IN]  the device id
+ * @param name [IN]      the name of channel
+ *
+ * @retval null for failed
+ * @retval OtherValues success
+ *
+ * @see acltdtStopChannel | acltdtDestroyChannel
+ */
+ACL_FUNC_VISIBILITY acltdtChannelHandle *acltdtCreateChannel(uint32_t deviceId, const char *name);
+
+/**
+ * @ingroup AscendCL
+ * @brief Create the channel with max size
+ *
+ * @param deviceId [IN]  the device id
+ * @param name [IN]      the name of channel
+ * @param capacity [IN]   the capacity of channel
+ *
+ * @retval null for failed
+ * @retval OtherValues success
+ *
+ * @see acltdtDestroyChannel
+ */
+ACL_FUNC_VISIBILITY acltdtChannelHandle *acltdtCreateChannelWithCapacity(uint32_t deviceId, const char *name,
+                                                                         size_t capacity);
+
+/**
+ * @ingroup AscendCL
+ * @brief Destroy the channel
+ *
+ * @param handle [IN]  pointer to the channel handle
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see acltdtCreateChannel | acltdtStopChannel
+ */
+ACL_FUNC_VISIBILITY aclError acltdtDestroyChannel(acltdtChannelHandle *handle);
+
+/**
+ * @ingroup AscendCL
+ * @brief Send tensor to device
+ *
+ * @param handle [IN]  pointer to the channel handle
+ * @param dataset [IN] pointer to the dataset
+ * @param timeout [IN] to be reserved, now it must be -1
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see acltdtReceiveTensor
+ */
+ACL_FUNC_VISIBILITY aclError acltdtSendTensor(const acltdtChannelHandle *handle, const acltdtDataset *dataset,
+                                              int32_t timeout);
+
+/**
+ * @ingroup AscendCL
+ * @brief Receive tensor from device
+ *
+ * @param handle [IN]      pointer to the channel handle
+ * @param dataset [OUT]    pointer to the dataset
+ * @param timeout [IN]     to be reserved, now it must be -1
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see acltdtSendTensor
+ */
+ACL_FUNC_VISIBILITY aclError acltdtReceiveTensor(const acltdtChannelHandle *handle, acltdtDataset *dataset,
+                                                 int32_t timeout);
+
+/**
+ * @ingroup AscendCL
+ * @brief query the size of the channel
+ *
+ * @param handle [IN]      pointer to the channel handle
+ * @param size [OUT]       current size of this channel
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ */
+ACL_FUNC_VISIBILITY aclError acltdtQueryChannelSize(const acltdtChannelHandle *handle, size_t *size);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // INC_EXTERNAL_ACL_ACL_TDT_H_

inc/external/acl/acl_tdt_queue.h

@@ -0,0 +1,476 @@
+/**
+ * Copyright 2019-2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef INC_EXTERNAL_ACL_ACL_TDT_QUEUE_H_
+#define INC_EXTERNAL_ACL_ACL_TDT_QUEUE_H_
+
+#include "acl/acl_base.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define ACL_TDT_QUEUE_PERMISSION_MANAGE 1
+#define ACL_TDT_QUEUE_PERMISSION_DEQUEUE 2
+#define ACL_TDT_QUEUE_PERMISSION_ENQUEUE 4
+
+typedef void *acltdtBuf;
+typedef struct tagMemQueueAttr acltdtQueueAttr;
+typedef struct acltdtQueueRouteList acltdtQueueRouteList;
+typedef struct acltdtQueueRouteQueryInfo acltdtQueueRouteQueryInfo;
+typedef struct acltdtQueueRoute acltdtQueueRoute;
+
+typedef enum { ACL_TDT_QUEUE_NAME_PTR = 0, ACL_TDT_QUEUE_DEPTH_UINT32 } acltdtQueueAttrType;
+
+typedef enum {
+  ACL_TDT_QUEUE_ROUTE_SRC_UINT32 = 0,
+  ACL_TDT_QUEUE_ROUTE_DST_UINT32,
+  ACL_TDT_QUEUE_ROUTE_STATUS_INT32
+} acltdtQueueRouteParamType;
+
+typedef enum {
+  ACL_TDT_QUEUE_ROUTE_QUERY_SRC = 0,
+  ACL_TDT_QUEUE_ROUTE_QUERY_DST,
+  ACL_TDT_QUEUE_ROUTE_QUERY_SRC_AND_DST
+} acltdtQueueRouteQueryMode;
+
+typedef enum {
+  ACL_TDT_QUEUE_ROUTE_QUERY_MODE_ENUM = 0,
+  ACL_TDT_QUEUE_ROUTE_QUERY_SRC_ID_UINT32,
+  ACL_TDT_QUEUE_ROUTE_QUERY_DST_ID_UINT32
+} acltdtQueueRouteQueryInfoParamType;
+
+/**
+ * @ingroup AscendCL
+ * @brief create queue
+ *
+ * @param attr [IN] pointer to the queue attr
+ * @param qid [OUT] pointer to the qid
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see acltdtDestroyQueue
+ */
+ACL_FUNC_VISIBILITY aclError acltdtCreateQueue(const acltdtQueueAttr *attr, uint32_t *qid);
+
+/**
+ * @ingroup AscendCL
+ * @brief destroy queue
+ *
+ * @param qid [IN] qid which to be destroyed
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see acltdtCreateQueue
+ */
+ACL_FUNC_VISIBILITY aclError acltdtDestroyQueue(uint32_t qid);
+
+/**
+ * @ingroup AscendCL
+ * @brief enqueue function
+ *
+ * @param qid [IN] qid
+ * @param buf [IN] acltdtBuf
+ * @param timeout [IN] timeout, -1 means blocking
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see acltdtDequeue
+ */
+ACL_FUNC_VISIBILITY aclError acltdtEnqueue(uint32_t qid, acltdtBuf buf, int32_t timeout);
+
+/**
+ * @ingroup AscendCL
+ * @brief dequeue function
+ *
+ * @param qid [IN] qid
+ * @param buf [OUT] pointer to the acltdtBuf
+ * @param timeout [IN] timeout, -1 means blocking
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see acltdtEnqueue
+ */
+ACL_FUNC_VISIBILITY aclError acltdtDequeue(uint32_t qid, acltdtBuf *buf, int32_t timeout);
+
+/**
+ * @ingroup AscendCL
+ * @brief enqueue function
+ *
+ * @param qid [IN] qid
+ * @param data [IN] the pointer to data buf
+ * @param dataSize [IN] the size of data buf
+ * @param userData [IN] the pointer to user data buf
+ * @param userDataSize [IN] the size of user data buf
+ * @param timeout [IN] timeout, -1 means blocking
+ * @param rsv [IN] reserved param
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see acltdtDequeueData
+ */
+ACL_FUNC_VISIBILITY aclError acltdtEnqueueData(uint32_t qid, const void *data, size_t dataSize, const void *userData,
+                                               size_t userDataSize, int32_t timeout, uint32_t rsv);
+
+/**
+ * @ingroup AscendCL
+ * @brief dequeue function
+ *
+ * @param qid [IN] qid
+ * @param data [IN|OUT] the pointer to data buf
+ * @param dataSize [IN] the size of data buf
+ * @param retDataSize [OUT] the return size of data buf
+ * @param userData [IN|OUT] the pointer to user data buf
+ * @param userDataSize [IN] the size of user data buf
+ * @param timeout [IN] timeout, -1 means blocking
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see acltdtEnqueueData
+ */
+ACL_FUNC_VISIBILITY aclError acltdtDequeueData(uint32_t qid, void *data, size_t dataSize, size_t *retDataSize,
+                                               void *userData, size_t userDataSize, int32_t timeout);
+
+/**
+ * @ingroup AscendCL
+ * @brief grant queue to other process
+ *
+ * @param qid [IN] qid
+ * @param pid [IN] pid of dst process
+ * @param permission [IN] permission of queue
+ * @param timeout [IN] timeout, -1 means blocking
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see ACL_TDT_QUEUE_PERMISSION_MANAGE | ACL_TDT_QUEUE_PERMISSION_DEQUEUE | ACL_TDT_QUEUE_PERMISSION_ENQUEUE
+ */
+ACL_FUNC_VISIBILITY aclError acltdtGrantQueue(uint32_t qid, int32_t pid, uint32_t permission, int32_t timeout);
+
+/**
+ * @ingroup AscendCL
+ * @brief attach queue in current process
+ *
+ * @param qid [IN] qid
+ * @param timeout [IN] timeout, -1 means blocking
+ * @param permission [OUT] permission of queue
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see acltdtGrantQueue
+ */
+ACL_FUNC_VISIBILITY aclError acltdtAttachQueue(uint32_t qid, int32_t timeout, uint32_t *permission);
+
+/**
+ * @ingroup AscendCL
+ * @brief bind queue routes
+ *
+ * @param qRouteList [IN|OUT] pointer to the route list
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError acltdtBindQueueRoutes(acltdtQueueRouteList *qRouteList);
+
+/**
+ * @ingroup AscendCL
+ * @brief unbind queue routes
+ *
+ * @param qRouteList [IN|OUT] pointer to the route list
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError acltdtUnbindQueueRoutes(acltdtQueueRouteList *qRouteList);
+
+/**
+ * @ingroup AscendCL
+ * @brief query queue routes according to query mode
+ *
+ * @param queryInfo [IN] pointer to the queue route query info
+ * @param qRouteList [IN|OUT] pointer to the route list
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError acltdtQueryQueueRoutes(const acltdtQueueRouteQueryInfo *queryInfo,
+                                                    acltdtQueueRouteList *qRouteList);
+
+/**
+ * @ingroup AscendCL
+ * @brief alloc acltdtBuf
+ *
+ * @param size [IN] alloc buf size
+ * @param type [IN] reserved parameters, need to set zero currently
+ * @param buf [OUT] pointer to the acltdtBuf
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see acltdtFreeBuf
+ */
+ACL_FUNC_VISIBILITY aclError acltdtAllocBuf(size_t size, uint32_t type, acltdtBuf *buf);
+
+/**
+ * @ingroup AscendCL
+ * @brief free acltdtBuf
+ *
+ * @param buf [IN] pointer to the acltdtBuf
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see acltdtAllocBuf
+ */
+ACL_FUNC_VISIBILITY aclError acltdtFreeBuf(acltdtBuf buf);
+
+/**
+ * @ingroup AscendCL
+ * @brief get data buf address
+ *
+ * @param buf [IN] acltdtBuf
+ * @param dataPtr [OUT] pointer to the data ptr which is acquired from acltdtBuf
+ * @param size [OUT] pointer to the size
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see acltdtAllocBuf
+ */
+ACL_FUNC_VISIBILITY aclError acltdtGetBufData(const acltdtBuf buf, void **dataPtr, size_t *size);
+
+/**
+ * @ingroup AscendCL
+ * @brief Create the queue attr
+ *
+ * @retval null for failed
+ * @retval OtherValues success
+ *
+ * @see acltdtDestroyQueueAttr
+ */
+ACL_FUNC_VISIBILITY acltdtQueueAttr *acltdtCreateQueueAttr();
+
+/**
+ * @ingroup AscendCL
+ * @brief Destroy the queue attr
+ *
+ * @param attr [IN]  pointer to the queue attr
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see acltdtCreateQueueAttr
+ */
+ACL_FUNC_VISIBILITY aclError acltdtDestroyQueueAttr(const acltdtQueueAttr *attr);
+
+/**
+ * @ingroup AscendCL
+ * @brief Set parameter for queue attr
+ *
+ * @param attr [IN|OUT] pointer to the queue attr
+ * @param type [IN]    parameter type
+ * @param len [IN]       parameter length
+ * @param param [IN]        pointer to parameter value
+ *
+ * @retval ACL_SUCCESS for success, other for failure
+ *
+ * @see acltdtCreateQueueAttr
+ */
+ACL_FUNC_VISIBILITY aclError acltdtSetQueueAttr(acltdtQueueAttr *attr, acltdtQueueAttrType type, size_t len,
+                                                const void *param);
+
+/**
+ * @ingroup AscendCL
+ *
+ * @brief Get parameter for queue attr.
+ *
+ * @param attr [IN]   pointer to the queue attr
+ * @param type [IN]     parameter type
+ * @param len [IN]        parameter length
+ * @param paramRetSize [OUT] pointer to parameter real length
+ * @param param [OUT]        pointer to parameter value
+ *
+ * @retval ACL_SUCCESS for success, other for failure
+ *
+ * @see acltdtCreateQueueAttr
+ */
+ACL_FUNC_VISIBILITY aclError acltdtGetQueueAttr(const acltdtQueueAttr *attr, acltdtQueueAttrType type, size_t len,
+                                                size_t *paramRetSize, void *param);
+
+/**
+ * @ingroup AscendCL
+ * @brief Create the queue route
+ *
+ * @param srcId [IN]   src id of queue route
+ * @param dstId [IN]   dst id of queue route
+ *
+ * @retval null for failed
+ * @retval OtherValues success
+ *
+ * @see acltdtDestroyQueueRoute
+ */
+ACL_FUNC_VISIBILITY acltdtQueueRoute *acltdtCreateQueueRoute(uint32_t srcId, uint32_t dstId);
+
+/**
+ * @ingroup AscendCL
+ * @brief Destroy the queue attr
+ *
+ * @param route [IN]  pointer to the queue route
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see acltdtCreateQueueRoute
+ */
+ACL_FUNC_VISIBILITY aclError acltdtDestroyQueueRoute(const acltdtQueueRoute *route);
+
+/**
+ * @ingroup AscendCL
+ *
+ * @brief Get parameter for queue route.
+ *
+ * @param route [IN]   pointer to the queue route
+ * @param type [IN]     parameter type
+ * @param len [IN]        parameter length
+ * @param paramRetSize [OUT] pointer to parameter real length
+ * @param param [OUT]        pointer to parameter value
+ *
+ * @retval ACL_SUCCESS for success, other for failure
+ *
+ * @see acltdtCreateQueueRoute
+ */
+ACL_FUNC_VISIBILITY aclError acltdtGetQueueRouteParam(const acltdtQueueRoute *route, acltdtQueueRouteParamType type,
+                                                      size_t len, size_t *paramRetSize, void *param);
+
+/**
+ * @ingroup AscendCL
+ * @brief Create the queue route list
+ *
+ * @retval null for failed
+ * @retval OtherValues success
+ *
+ * @see acltdtDestroyQueueRouteList
+ */
+ACL_FUNC_VISIBILITY acltdtQueueRouteList *acltdtCreateQueueRouteList();
+
+/**
+ * @ingroup AscendCL
+ * @brief Destroy the queue route list
+ *
+ * @param routeList [IN]  pointer to the queue route list
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see acltdtCreateQueueRouteList
+ */
+ACL_FUNC_VISIBILITY aclError acltdtDestroyQueueRouteList(const acltdtQueueRouteList *routeList);
+
+/**
+ * @ingroup AscendCL
+ * @brief add queue route to the route list
+ *
+ * @param routeList [IN|OUT]  pointer to the queue route list
+ * @param route [IN]  pointer to the queue route
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see acltdtCreateQueueRouteList | acltdtCreateQueueRoute
+ *
+ */
+ACL_FUNC_VISIBILITY aclError acltdtAddQueueRoute(acltdtQueueRouteList *routeList, const acltdtQueueRoute *route);
+
+/**
+ * @ingroup AscendCL
+ * @brief get queue route from route list
+ *
+ * @param routeList [IN]  pointer to the queue route list
+ * @param index [IN]  index of queue route in route list
+ * @param route [IN|OUT]  pointer to the queue route
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see acltdtCreateQueueRouteList | acltdtCreateQueueRoute
+ *
+ */
+ACL_FUNC_VISIBILITY aclError acltdtGetQueueRoute(const acltdtQueueRouteList *routeList, size_t index,
+                                                 acltdtQueueRoute *route);
+
+/**
+ * @ingroup AscendCL
+ * @brief get queue route num from route list
+ *
+ * @param routeList [IN]  pointer to the queue route list
+ *
+ * @retval the number of queue route
+ *
+ */
+ACL_FUNC_VISIBILITY size_t acltdtGetQueueRouteNum(const acltdtQueueRouteList *routeList);
+
+/**
+ * @ingroup AscendCL
+ * @brief Create the queue route query info
+ *
+ * @retval null for failed
+ * @retval OtherValues success
+ *
+ * @see acltdtDestroyQueueRouteQueryInfo
+ */
+ACL_FUNC_VISIBILITY acltdtQueueRouteQueryInfo *acltdtCreateQueueRouteQueryInfo();
+
+/**
+ * @ingroup AscendCL
+ * @brief Destroy the queue route query info
+ *
+ * @param info [IN]  pointer to the queue route info
+ *
+ * @retval ACL_SUCCESS  The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see acltdtCreateQueueRouteQueryInfo
+ *
+ */
+ACL_FUNC_VISIBILITY aclError acltdtDestroyQueueRouteQueryInfo(const acltdtQueueRouteQueryInfo *info);
+
+/**
+ * @ingroup AscendCL
+ * @brief Set parameter for queue route info
+ *
+ * @param attr [IN|OUT] pointer to the queue route info
+ * @param type [IN]    parameter type
+ * @param len [IN]       parameter length
+ * @param param [IN]        pointer to parameter value
+ *
+ * @retval ACL_SUCCESS for success, other for failure
+ *
+ * @see acltdtCreateQueueRouteQueryInfo
+ */
+ACL_FUNC_VISIBILITY aclError acltdtSetQueueRouteQueryInfo(acltdtQueueRouteQueryInfo *param,
+                                                          acltdtQueueRouteQueryInfoParamType type, size_t len,
+                                                          const void *value);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // INC_EXTERNAL_ACL_ACL_TDT_QUEUE_H_

inc/external/acl/error_codes/ge_error_codes.h

@@ -0,0 +1,77 @@
+/**
+ * Copyright 2019-2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef INC_EXTERNAL_GE_GE_ERROR_CODES_H_
+#define INC_EXTERNAL_GE_GE_ERROR_CODES_H_
+
+#if defined(_MSC_VER)
+#ifdef FUNC_VISIBILITY
+#define GE_FUNC_VISIBILITY _declspec(dllexport)
+#else
+#define GE_FUNC_VISIBILITY
+#endif
+#else
+#ifdef FUNC_VISIBILITY
+#define GE_FUNC_VISIBILITY __attribute__((visibility("default")))
+#else
+#define GE_FUNC_VISIBILITY
+#endif
+#endif
+
+#include <stddef.h>
+#include <stdint.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+static const uint32_t ACL_ERROR_GE_PARAM_INVALID = 145000U;
+static const uint32_t ACL_ERROR_GE_EXEC_NOT_INIT = 145001U;
+static const uint32_t ACL_ERROR_GE_EXEC_MODEL_PATH_INVALID = 145002U;
+static const uint32_t ACL_ERROR_GE_EXEC_MODEL_ID_INVALID = 145003U;
+static const uint32_t ACL_ERROR_GE_EXEC_MODEL_DATA_SIZE_INVALID = 145006U;
+static const uint32_t ACL_ERROR_GE_EXEC_MODEL_ADDR_INVALID = 145007U;
+static const uint32_t ACL_ERROR_GE_EXEC_MODEL_QUEUE_ID_INVALID = 145008U;
+static const uint32_t ACL_ERROR_GE_EXEC_LOAD_MODEL_REPEATED = 145009U;
+static const uint32_t ACL_ERROR_GE_DYNAMIC_INPUT_ADDR_INVALID = 145011U;
+static const uint32_t ACL_ERROR_GE_DYNAMIC_INPUT_LENGTH_INVALID = 145012U;
+static const uint32_t ACL_ERROR_GE_DYNAMIC_BATCH_SIZE_INVALID = 145013U;
+static const uint32_t ACL_ERROR_GE_AIPP_BATCH_EMPTY = 145014U;
+static const uint32_t ACL_ERROR_GE_AIPP_NOT_EXIST = 145015U;
+static const uint32_t ACL_ERROR_GE_AIPP_MODE_INVALID = 145016U;
+static const uint32_t ACL_ERROR_GE_OP_TASK_TYPE_INVALID = 145017U;
+static const uint32_t ACL_ERROR_GE_OP_KERNEL_TYPE_INVALID = 145018U;
+static const uint32_t ACL_ERROR_GE_PLGMGR_PATH_INVALID = 145019U;
+static const uint32_t ACL_ERROR_GE_FORMAT_INVALID = 145020U;
+static const uint32_t ACL_ERROR_GE_SHAPE_INVALID = 145021U;
+static const uint32_t ACL_ERROR_GE_DATATYPE_INVALID = 145022U;
+static const uint32_t ACL_ERROR_GE_MEMORY_ALLOCATION = 245000U;
+static const uint32_t ACL_ERROR_GE_MEMORY_OPERATE_FAILED = 245001U;
+static const uint32_t ACL_ERROR_GE_DEVICE_MEMORY_OPERATE_FAILED = 245002U;
+static const uint32_t ACL_ERROR_GE_INTERNAL_ERROR = 545000U;
+static const uint32_t ACL_ERROR_GE_LOAD_MODEL = 545001U;
+static const uint32_t ACL_ERROR_GE_EXEC_LOAD_MODEL_PARTITION_FAILED = 545002U;
+static const uint32_t ACL_ERROR_GE_EXEC_LOAD_WEIGHT_PARTITION_FAILED = 545003U;
+static const uint32_t ACL_ERROR_GE_EXEC_LOAD_TASK_PARTITION_FAILED = 545004U;
+static const uint32_t ACL_ERROR_GE_EXEC_LOAD_KERNEL_PARTITION_FAILED = 545005U;
+static const uint32_t ACL_ERROR_GE_EXEC_RELEASE_MODEL_DATA = 545006U;
+static const uint32_t ACL_ERROR_GE_COMMAND_HANDLE = 545007U;
+static const uint32_t ACL_ERROR_GE_GET_TENSOR_INFO = 545008U;
+static const uint32_t ACL_ERROR_GE_UNLOAD_MODEL = 545009U;
+
+#ifdef __cplusplus
+}  // namespace ge
+#endif
+#endif  // INC_EXTERNAL_GE_GE_ERROR_CODES_H_

inc/external/acl/error_codes/rt_error_codes.h

@@ -0,0 +1,123 @@
+/**
+ * Copyright 2019-2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef __INC_EXTERNEL_RT_ERROR_CODES_H__
+#define __INC_EXTERNEL_RT_ERROR_CODES_H__
+
+#include <stddef.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+static const int32_t ACL_RT_SUCCESS = 0;  // success
+
+static const int32_t ACL_ERROR_RT_PARAM_INVALID = 107000;             // param invalid
+static const int32_t ACL_ERROR_RT_INVALID_DEVICEID = 107001;          // invalid device id
+static const int32_t ACL_ERROR_RT_CONTEXT_NULL = 107002;              // current context null
+static const int32_t ACL_ERROR_RT_STREAM_CONTEXT = 107003;            // stream not in current context
+static const int32_t ACL_ERROR_RT_MODEL_CONTEXT = 107004;             // model not in current context
+static const int32_t ACL_ERROR_RT_STREAM_MODEL = 107005;              // stream not in model
+static const int32_t ACL_ERROR_RT_EVENT_TIMESTAMP_INVALID = 107006;   // event timestamp invalid
+static const int32_t ACL_ERROR_RT_EVENT_TIMESTAMP_REVERSAL = 107007;  // event timestamp reversal
+static const int32_t ACL_ERROR_RT_ADDR_UNALIGNED = 107008;            // memory address unaligned
+static const int32_t ACL_ERROR_RT_FILE_OPEN = 107009;                 // open file failed
+static const int32_t ACL_ERROR_RT_FILE_WRITE = 107010;                // write file failed
+static const int32_t ACL_ERROR_RT_STREAM_SUBSCRIBE = 107011;          // error subscribe stream
+static const int32_t ACL_ERROR_RT_THREAD_SUBSCRIBE = 107012;          // error subscribe thread
+static const int32_t ACL_ERROR_RT_GROUP_NOT_SET = 107013;             // group not set
+static const int32_t ACL_ERROR_RT_GROUP_NOT_CREATE = 107014;          // group not create
+static const int32_t ACL_ERROR_RT_STREAM_NO_CB_REG = 107015;          // callback not register to stream
+static const int32_t ACL_ERROR_RT_INVALID_MEMORY_TYPE = 107016;       // invalid memory type
+static const int32_t ACL_ERROR_RT_INVALID_HANDLE = 107017;            // invalid handle
+static const int32_t ACL_ERROR_RT_INVALID_MALLOC_TYPE = 107018;       // invalid malloc type
+static const int32_t ACL_ERROR_RT_WAIT_TIMEOUT = 107019;              // wait timeout
+static const int32_t ACL_ERROR_RT_TASK_TIMEOUT = 107020;              // task timeout
+
+static const int32_t ACL_ERROR_RT_FEATURE_NOT_SUPPORT = 207000;  // feature not support
+static const int32_t ACL_ERROR_RT_MEMORY_ALLOCATION = 207001;    // memory allocation error
+static const int32_t ACL_ERROR_RT_MEMORY_FREE = 207002;          // memory free error
+static const int32_t ACL_ERROR_RT_AICORE_OVER_FLOW = 207003;     // aicore over flow
+static const int32_t ACL_ERROR_RT_NO_DEVICE = 207004;            // no device
+static const int32_t ACL_ERROR_RT_RESOURCE_ALLOC_FAIL = 207005;  // resource alloc fail
+static const int32_t ACL_ERROR_RT_NO_PERMISSION = 207006;        // no permission
+static const int32_t ACL_ERROR_RT_NO_EVENT_RESOURCE = 207007;    // no event resource
+static const int32_t ACL_ERROR_RT_NO_STREAM_RESOURCE = 207008;   // no stream resource
+static const int32_t ACL_ERROR_RT_NO_NOTIFY_RESOURCE = 207009;   // no notify resource
+static const int32_t ACL_ERROR_RT_NO_MODEL_RESOURCE = 207010;    // no model resource
+static const int32_t ACL_ERROR_RT_NO_CDQ_RESOURCE = 207011;      // no cdq resource
+static const int32_t ACL_ERROR_RT_OVER_LIMIT = 207012;           // over limit
+static const int32_t ACL_ERROR_RT_QUEUE_EMPTY = 207013;          // queue is empty
+static const int32_t ACL_ERROR_RT_QUEUE_FULL = 207014;           // queue is full
+static const int32_t ACL_ERROR_RT_REPEATED_INIT = 207015;        // repeated init
+static const int32_t ACL_ERROR_RT_AIVEC_OVER_FLOW = 207016;      // aivec over flow
+static const int32_t ACL_ERROR_RT_OVER_FLOW = 207017;            // common over flow
+
+static const int32_t ACL_ERROR_RT_INTERNAL_ERROR = 507000;                   // runtime internal error
+static const int32_t ACL_ERROR_RT_TS_ERROR = 507001;                         // ts internel error
+static const int32_t ACL_ERROR_RT_STREAM_TASK_FULL = 507002;                 // task full in stream
+static const int32_t ACL_ERROR_RT_STREAM_TASK_EMPTY = 507003;                // task empty in stream
+static const int32_t ACL_ERROR_RT_STREAM_NOT_COMPLETE = 507004;              // stream not complete
+static const int32_t ACL_ERROR_RT_END_OF_SEQUENCE = 507005;                  // end of sequence
+static const int32_t ACL_ERROR_RT_EVENT_NOT_COMPLETE = 507006;               // event not complete
+static const int32_t ACL_ERROR_RT_CONTEXT_RELEASE_ERROR = 507007;            // context release error
+static const int32_t ACL_ERROR_RT_SOC_VERSION = 507008;                      // soc version error
+static const int32_t ACL_ERROR_RT_TASK_TYPE_NOT_SUPPORT = 507009;            // task type not support
+static const int32_t ACL_ERROR_RT_LOST_HEARTBEAT = 507010;                   // ts lost heartbeat
+static const int32_t ACL_ERROR_RT_MODEL_EXECUTE = 507011;                    // model execute failed
+static const int32_t ACL_ERROR_RT_REPORT_TIMEOUT = 507012;                   // report timeout
+static const int32_t ACL_ERROR_RT_SYS_DMA = 507013;                          // sys dma error
+static const int32_t ACL_ERROR_RT_AICORE_TIMEOUT = 507014;                   // aicore timeout
+static const int32_t ACL_ERROR_RT_AICORE_EXCEPTION = 507015;                 // aicore exception
+static const int32_t ACL_ERROR_RT_AICORE_TRAP_EXCEPTION = 507016;            // aicore trap exception
+static const int32_t ACL_ERROR_RT_AICPU_TIMEOUT = 507017;                    // aicpu timeout
+static const int32_t ACL_ERROR_RT_AICPU_EXCEPTION = 507018;                  // aicpu exception
+static const int32_t ACL_ERROR_RT_AICPU_DATADUMP_RSP_ERR = 507019;           // aicpu datadump response error
+static const int32_t ACL_ERROR_RT_AICPU_MODEL_RSP_ERR = 507020;              // aicpu model operate response error
+static const int32_t ACL_ERROR_RT_PROFILING_ERROR = 507021;                  // profiling error
+static const int32_t ACL_ERROR_RT_IPC_ERROR = 507022;                        // ipc error
+static const int32_t ACL_ERROR_RT_MODEL_ABORT_NORMAL = 507023;               // model abort normal
+static const int32_t ACL_ERROR_RT_KERNEL_UNREGISTERING = 507024;             // kernel unregistering
+static const int32_t ACL_ERROR_RT_RINGBUFFER_NOT_INIT = 507025;              // ringbuffer not init
+static const int32_t ACL_ERROR_RT_RINGBUFFER_NO_DATA = 507026;               // ringbuffer no data
+static const int32_t ACL_ERROR_RT_KERNEL_LOOKUP = 507027;                    // kernel lookup error
+static const int32_t ACL_ERROR_RT_KERNEL_DUPLICATE = 507028;                 // kernel register duplicate
+static const int32_t ACL_ERROR_RT_DEBUG_REGISTER_FAIL = 507029;              // debug register failed
+static const int32_t ACL_ERROR_RT_DEBUG_UNREGISTER_FAIL = 507030;            // debug unregister failed
+static const int32_t ACL_ERROR_RT_LABEL_CONTEXT = 507031;                    // label not in current context
+static const int32_t ACL_ERROR_RT_PROGRAM_USE_OUT = 507032;                  // program register num use out
+static const int32_t ACL_ERROR_RT_DEV_SETUP_ERROR = 507033;                  // device setup error
+static const int32_t ACL_ERROR_RT_VECTOR_CORE_TIMEOUT = 507034;              // vector core timeout
+static const int32_t ACL_ERROR_RT_VECTOR_CORE_EXCEPTION = 507035;            // vector core exception
+static const int32_t ACL_ERROR_RT_VECTOR_CORE_TRAP_EXCEPTION = 507036;       // vector core trap exception
+static const int32_t ACL_ERROR_RT_CDQ_BATCH_ABNORMAL = 507037;               // cdq alloc batch abnormal
+static const int32_t ACL_ERROR_RT_DIE_MODE_CHANGE_ERROR = 507038;            // can not change die mode
+static const int32_t ACL_ERROR_RT_DIE_SET_ERROR = 507039;                    // single die mode can not set die
+static const int32_t ACL_ERROR_RT_INVALID_DIEID = 507040;                    // invalid die id
+static const int32_t ACL_ERROR_RT_DIE_MODE_NOT_SET = 507041;                 // die mode not set
+static const int32_t ACL_ERROR_RT_AICORE_TRAP_READ_OVERFLOW = 507042;        // aic trap read overflow
+static const int32_t ACL_ERROR_RT_AICORE_TRAP_WRITE_OVERFLOW = 507043;       // aic trap write overflow
+static const int32_t ACL_ERROR_RT_VECTOR_CORE_TRAP_READ_OVERFLOW = 507044;   // aiv trap read overflow
+static const int32_t ACL_ERROR_RT_VECTOR_CORE_TRAP_WRITE_OVERFLOW = 507045;  // aiv trap write overflow
+
+static const int32_t ACL_ERROR_RT_DRV_INTERNAL_ERROR = 507899;    // drv internal error
+static const int32_t ACL_ERROR_RT_AICPU_INTERNAL_ERROR = 507900;  // aicpu internal error
+static const int32_t ACL_ERROR_RT_SOCKET_CLOSE = 507901;          // hdc disconnect
+
+#ifdef __cplusplus
+}
+#endif
+#endif  // __INC_EXTERNEL_RT_ERROR_CODES_H__

inc/external/acl/ops/acl_cblas.h

@@ -0,0 +1,334 @@
+/**
+ * Copyright 2019-2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef INC_EXTERNAL_ACL_OPS_ACL_CBLAS_H_
+#define INC_EXTERNAL_ACL_OPS_ACL_CBLAS_H_
+
+#include "acl/acl.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef enum aclTransType { ACL_TRANS_N, ACL_TRANS_T, ACL_TRANS_NZ, ACL_TRANS_NZ_T } aclTransType;
+
+typedef enum aclComputeType { ACL_COMPUTE_HIGH_PRECISION, ACL_COMPUTE_LOW_PRECISION } aclComputeType;
+
+/**
+ * @ingroup AscendCL
+ * @brief perform the matrix-vector multiplication
+ *
+ * @param transA [IN]      transpose type of matrix A
+ * @param m [IN]           number of rows of matrix A
+ * @param n [IN]           number of columns of matrix A
+ * @param alpha [IN]       pointer to scalar used for multiplication.
+ *                         of same type as dataTypeC
+ * @param a [IN]           pointer to matrix A
+ * @param lda [IN]         leading dimension used to store the matrix A
+ * @param dataTypeA [IN]   datatype of matrix A
+ * @param x [IN]           pointer to vector x
+ * @param incx [IN]        stride between consecutive elements of vector x
+ * @param dataTypeX [IN]   datatype of vector x
+ * @param beta [IN]        pointer to scalar used for multiplication.
+ *                         of same type as dataTypeC If beta == 0,
+ *                         then y does not have to be a valid input
+ * @param y [IN|OUT]       pointer to vector y
+ * @param incy [IN]        stride between consecutive elements of vector y
+ * @param dataTypeY [IN]   datatype of vector y
+ * @param type [IN]        computation type
+ * @param stream [IN]      stream
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclblasGemvEx(aclTransType transA, int m, int n, const void *alpha, const void *a, int lda,
+                                           aclDataType dataTypeA, const void *x, int incx, aclDataType dataTypeX,
+                                           const void *beta, void *y, int incy, aclDataType dataTypeY,
+                                           aclComputeType type, aclrtStream stream);
+
+/**
+ * @ingroup AscendCL
+ * @brief create a handle for performing the matrix-vector multiplication
+ *
+ * @param transA [IN]      transpose type of matrix A
+ * @param m [IN]           number of rows of matrix A
+ * @param n [IN]           number of columns of matrix A
+ * @param dataTypeA [IN]   datatype of matrix A
+ * @param dataTypeX [IN]   datatype of vector x
+ * @param dataTypeY [IN]   datatype of vector y
+ * @param type [IN]        computation type
+ * @param handle [OUT]     pointer to the pointer to the handle
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclblasCreateHandleForGemvEx(aclTransType transA, int m, int n, aclDataType dataTypeA,
+                                                          aclDataType dataTypeX, aclDataType dataTypeY,
+                                                          aclComputeType type, aclopHandle **handle);
+
+/**
+ * @ingroup AscendCL
+ * @brief perform the matrix-vector multiplication
+ *
+ * @param transA [IN]      transpose type of matrix A
+ * @param m [IN]           number of rows of matrix A
+ * @param n [IN]           number of columns of matrix A
+ * @param alpha [IN]       pointer to scalar used for multiplication
+ * @param a [IN]           pointer to matrix A
+ * @param lda [IN]         leading dimension used to store the matrix A
+ * @param x [IN]           pointer to vector x
+ * @param incx [IN]        stride between consecutive elements of vector x
+ * @param beta [IN]        pointer to scalar used for multiplication.
+ *                         If beta value == 0,
+ *                         then y does not have to be a valid input
+ * @param y [IN|OUT]       pointer to vector y
+ * @param incy [IN]        stride between consecutive elements of vector y
+ * @param type [IN]        computation type
+ * @param stream [IN]      stream
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclblasHgemv(aclTransType transA, int m, int n, const aclFloat16 *alpha,
+                                          const aclFloat16 *a, int lda, const aclFloat16 *x, int incx,
+                                          const aclFloat16 *beta, aclFloat16 *y, int incy, aclComputeType type,
+                                          aclrtStream stream);
+
+/**
+ * @ingroup AscendCL
+ * @brief create a handle for performing the matrix-vector multiplication
+ *
+ * @param transA [IN]      transpose type of matrix A
+ * @param m [IN]           number of rows of matrix A
+ * @param n [IN]           number of columns of matrix A
+ * @param type [IN]        computation type
+ * @param handle [OUT]     pointer to the pointer to the handle
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclblasCreateHandleForHgemv(aclTransType transA, int m, int n, aclComputeType type,
+                                                         aclopHandle **handle);
+
+/**
+ * @ingroup AscendCL
+ * @brief perform the matrix-vector multiplication
+ *
+ * @param transA [IN]      transpose type of matrix A
+ * @param m [IN]           number of rows of matrix A
+ * @param n [IN]           number of columns of matrix A
+ * @param alpha [IN]       pointer to scalar used for multiplication
+ * @param a [IN]           pointer to matrix A
+ * @param lda [IN]         leading dimension used to store the matrix A
+ * @param x [IN]           pointer to vector x
+ * @param incx [IN]        stride between consecutive elements of vector x
+ * @param beta [IN]        pointer to scalar used for multiplication.
+ *                         If beta value == 0,
+ *                         then y does not have to be a valid input
+ * @param y [IN|OUT]       pointer to vector y
+ * @param incy [IN]        stride between consecutive elements of vector y
+ * @param type [IN]        computation type
+ * @param stream [IN]      stream
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclblasS8gemv(aclTransType transA, int m, int n, const int32_t *alpha, const int8_t *a,
+                                           int lda, const int8_t *x, int incx, const int32_t *beta, int32_t *y,
+                                           int incy, aclComputeType type, aclrtStream stream);
+
+/**
+ * @ingroup AscendCL
+ * @brief create a handle for performing the matrix-vector multiplication
+ *
+ * @param transA [IN]      transpose type of matrix A
+ * @param m [IN]           number of rows of matrix A
+ * @param n [IN]           number of columns of matrix A
+ * @param handle [OUT]     pointer to the pointer to the handle
+ * @param type [IN]        computation type
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclblasCreateHandleForS8gemv(aclTransType transA, int m, int n, aclComputeType type,
+                                                          aclopHandle **handle);
+
+/**
+ * @ingroup AscendCL
+ * @brief perform the matrix-matrix multiplication
+ *
+ * @param transA [IN]      transpose type of matrix A
+ * @param transB [IN]      transpose type of matrix B
+ * @param transC [IN]      transpose type of matrix C
+ * @param m [IN]           number of rows of matrix A and matrix C
+ * @param n [IN]           number of columns of matrix B and matrix C
+ * @param k [IN]           number of columns of matrix A and rows of matrix B
+ * @param alpha [IN]       pointer to scalar used for multiplication. of same type as dataTypeC
+ * @param matrixA [IN]     pointer to matrix A
+ * @param lda [IN]         leading dimension array used to store  matrix A
+ * @param dataTypeA [IN]   datatype of matrix A
+ * @param matrixB [IN]     pointer to matrix B
+ * @param ldb [IN]         leading dimension array used to store  matrix B
+ * @param dataTypeB [IN]   datatype of matrix B
+ * @param beta [IN]        pointer to scalar used for multiplication.
+ *                         of same type as dataTypeC If beta == 0,
+ *                         then matrixC does not have to be a valid input
+ * @param matrixC [IN|OUT] pointer to matrix C
+ * @param ldc [IN]         leading dimension array used to store  matrix C
+ * @param dataTypeC [IN]   datatype of matrix C
+ * @param type [IN]        computation type
+ * @param stream [IN]      stream
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclblasGemmEx(aclTransType transA, aclTransType transB, aclTransType transC, int m, int n,
+                                           int k, const void *alpha, const void *matrixA, int lda,
+                                           aclDataType dataTypeA, const void *matrixB, int ldb, aclDataType dataTypeB,
+                                           const void *beta, void *matrixC, int ldc, aclDataType dataTypeC,
+                                           aclComputeType type, aclrtStream stream);
+
+/**
+ * @ingroup AscendCL
+ * @brief create a handle for performing the matrix-matrix multiplication
+ *
+ * @param transA [IN]      transpose type of matrix A
+ * @param transB [IN]      transpose type of matrix B
+ * @param transC [IN]      transpose type of matrix C
+ * @param m [IN]           number of rows of matrix A and matrix C
+ * @param n [IN]           number of columns of matrix B and matrix C
+ * @param k [IN]           number of columns of matrix A and rows of matrix B
+ * @param dataTypeA [IN]   datatype of matrix A
+ * @param dataTypeB [IN]   datatype of matrix B
+ * @param dataTypeC [IN]   datatype of matrix C
+ * @param type [IN]        computation type
+ * @param handle [OUT]     pointer to the pointer to the handle
+ * @param type [IN]        computation type
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclblasCreateHandleForGemmEx(aclTransType transA, aclTransType transB, aclTransType transC,
+                                                          int m, int n, int k, aclDataType dataTypeA,
+                                                          aclDataType dataTypeB, aclDataType dataTypeC,
+                                                          aclComputeType type, aclopHandle **handle);
+
+/**
+ * @ingroup AscendCL
+ * @brief perform the matrix-matrix multiplication
+ *
+ * @param transA [IN]      transpose type of matrix A
+ * @param transB [IN]      transpose type of matrix B
+ * @param transC [IN]      transpose type of matrix C
+ * @param m [IN]           number of rows of matrix A and matrix C
+ * @param n [IN]           number of columns of matrix B and matrix C
+ * @param k [IN]           number of columns of matrix A and rows of matrix B
+ * @param alpha [IN]       pointer to scalar used for multiplication
+ * @param matrixA [IN]     pointer to matrix A
+ * @param lda [IN]         leading dimension used to store the matrix A
+ * @param matrixB [IN]     pointer to matrix B
+ * @param ldb [IN]         leading dimension used to store the matrix B
+ * @param beta [IN]        pointer to scalar used for multiplication.
+ *                         If beta value == 0,
+ *                         then matrixC does not have to be a valid input
+ * @param matrixC [IN|OUT] pointer to matrix C
+ * @param ldc [IN]         leading dimension used to store the matrix C
+ * @param type [IN]        computation type
+ * @param stream [IN]      stream
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclblasHgemm(aclTransType transA, aclTransType transB, aclTransType transC, int m, int n,
+                                          int k, const aclFloat16 *alpha, const aclFloat16 *matrixA, int lda,
+                                          const aclFloat16 *matrixB, int ldb, const aclFloat16 *beta,
+                                          aclFloat16 *matrixC, int ldc, aclComputeType type, aclrtStream stream);
+
+/**
+ * @ingroup AscendCL
+ * @brief create a handle for performing the matrix-matrix multiplication
+ *
+ * @param transA [IN]      transpose type of matrix A
+ * @param transB [IN]      transpose type of matrix B
+ * @param transC [IN]      transpose type of matrix C
+ * @param m [IN]           number of rows of matrix A and matrix C
+ * @param n [IN]           number of columns of matrix B and matrix C
+ * @param k [IN]           number of columns of matrix A and rows of matrix B
+ * @param type [IN]        computation type
+ * @param handle [OUT]     pointer to the pointer to the handle
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclblasCreateHandleForHgemm(aclTransType transA, aclTransType transB, aclTransType transC,
+                                                         int m, int n, int k, aclComputeType type,
+                                                         aclopHandle **handle);
+
+/**
+ * @ingroup AscendCL
+ * @brief perform the matrix-matrix multiplication
+ *
+ * @param transA [IN]      transpose type of matrix A
+ * @param transB [IN]      transpose type of matrix B
+ * @param transC [IN]      transpose type of matrix C
+ * @param m [IN]           number of rows of matrix A and matrix C
+ * @param n [IN]           number of columns of matrix B and matrix C
+ * @param k [IN]           number of columns of matrix A and rows of matrix B
+ * @param alpha [IN]       pointer to scalar used for multiplication
+ * @param matrixA [IN]     pointer to matrix A
+ * @param lda [IN]         leading dimension used to store the matrix A
+ * @param matrixB [IN]     pointer to matrix B
+ * @param ldb [IN]         leading dimension used to store the matrix B
+ * @param beta [IN]        pointer to scalar used for multiplication.
+ *                         If beta value == 0,
+ *                         then matrixC does not have to be a valid input
+ * @param matrixC [IN|OUT] pointer to matrix C
+ * @param ldc [IN]         leading dimension used to store the matrix C
+ * @param type [IN]        computation type
+ * @param stream [IN]      stream
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclblasS8gemm(aclTransType transA, aclTransType transB, aclTransType transC, int m, int n,
+                                           int k, const int32_t *alpha, const int8_t *matrixA, int lda,
+                                           const int8_t *matrixB, int ldb, const int32_t *beta, int32_t *matrixC,
+                                           int ldc, aclComputeType type, aclrtStream stream);
+
+/**
+ * @ingroup AscendCL
+ * @brief create a handle for performing the matrix-matrix multiplication
+ *
+ * @param transA [IN]      transpose type of matrix A
+ * @param transB [IN]      transpose type of matrix B
+ * @param transC [IN]      transpose type of matrix C
+ * @param m [IN]           number of rows of matrix A and matrix C
+ * @param n [IN]           number of columns of matrix B and matrix C
+ * @param k [IN]           number of columns of matrix A and rows of matrix B
+ * @param type [IN]        computation type
+ * @param handle [OUT]     pointer to the pointer to the handle
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclblasCreateHandleForS8gemm(aclTransType transA, aclTransType transB, aclTransType transC,
+                                                          int m, int n, int k, aclComputeType type,
+                                                          aclopHandle **handle);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // INC_EXTERNAL_ACL_OPS_ACL_CBLAS_H_

inc/external/acl/ops/acl_fv.h

@@ -0,0 +1,348 @@
+/**
+ * Copyright 2019-2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef INC_EXTERNAL_ACL_OPS_ACL_RETR_H_
+#define INC_EXTERNAL_ACL_OPS_ACL_RETR_H_
+
+#include "acl/acl.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef struct aclfvInitPara aclfvInitPara;
+typedef struct aclfvFeatureInfo aclfvFeatureInfo;
+typedef struct aclfvRepoRange aclfvRepoRange;
+typedef struct aclfvQueryTable aclfvQueryTable;
+typedef struct aclfvSearchInput aclfvSearchInput;
+typedef struct aclfvSearchResult aclfvSearchResult;
+
+// search operation type
+enum aclfvSearchType {
+  SEARCH_1_N,  // 1:N operation type
+  SEARCH_N_M   // N:M operation type
+};
+
+/**
+ * @ingroup AscendCL
+ * @brief Create fv init param.
+ *
+ * @param fsNum [IN]  The feature num
+ *
+ * @retval null for failed.
+ * @retval OtherValues success.
+ */
+ACL_FUNC_VISIBILITY aclfvInitPara *aclfvCreateInitPara(uint64_t fsNum);
+
+/**
+ * @ingroup AscendCL
+ * @brief Destroy fv init param.
+ *
+ * @par Function
+ * Can only destroy fv init param information created
+ * through aclfvCreateInitPara interface.
+ *
+ * @param initPara [IN]   fv init param.
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclfvCreateInitPara
+ */
+ACL_FUNC_VISIBILITY aclError aclfvDestroyInitPara(aclfvInitPara *initPara);
+
+/**
+ * @ingroup AscendCL
+ * @brief set value for maxTopNumFor1N which in fv init param.
+ *
+ * @param initPara [IN|OUT]     fv init param.
+ * @param maxTopNumFor1N [IN]   maxTopNumFor1N value for init param.
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclfvSet1NTopNum(aclfvInitPara *initPara, uint32_t maxTopNumFor1N);
+
+/**
+ * @ingroup AscendCL
+ * @brief set value for maxTopNumForNM which in fv init param.
+ *
+ * @param initPara [IN|OUT]        fv init param.
+ * @param maxTopNumForNM [IN]   maxTopNumForNM value for init param.
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclfvSetNMTopNum(aclfvInitPara *initPara, uint32_t maxTopNumForNM);
+
+/**
+ * @ingroup AscendCL
+ * @brief Create fv feature info.
+ *
+ * @param id0 [IN]     The first level library id0
+ * @param id1 [IN]     Secondary library id1
+ * @param offset [IN]  The offset of the first feature in the library
+ * @param featureLen [IN]       Single feature length
+ * @param featureCount [IN]     Single feature count
+ * @param featureData [IN]      Feature value list
+ * @param featureDataLen [IN]   Feature value list length
+ *
+ * @retval null for failed.
+ * @retval OtherValues success.
+ */
+ACL_FUNC_VISIBILITY aclfvFeatureInfo *aclfvCreateFeatureInfo(uint32_t id0, uint32_t id1, uint32_t offset,
+                                                             uint32_t featureLen, uint32_t featureCount,
+                                                             uint8_t *featureData, uint32_t featureDataLen);
+
+/**
+ * @ingroup AscendCL
+ * @brief Destroy fv feature info.
+ *
+ * @par Function
+ * Can only destroy fv feature info information created
+ * through aclfvCreateFeatureInfo interface.
+ *
+ * @param featureInfo [IN]     fv feature info.
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclfvCreateFeatureInfo
+ */
+ACL_FUNC_VISIBILITY aclError aclfvDestroyFeatureInfo(aclfvFeatureInfo *featureInfo);
+
+/**
+ * @ingroup AscendCL
+ * @brief Create fv repo range.
+ *
+ * @param id0Min [IN]  id0 start value
+ * @param id0Min [IN]  id0 max
+ * @param id1Min [IN]  id0 start value
+ * @param id1Max [IN]  id1 max
+ *
+ * @retval null for failed. OtherValues success
+ */
+ACL_FUNC_VISIBILITY aclfvRepoRange *aclfvCreateRepoRange(uint32_t id0Min, uint32_t id0Max, uint32_t id1Min,
+                                                         uint32_t id1Max);
+
+/**
+ * @ingroup AscendCL
+ * @brief Destroy fv repo range.
+ *
+ * @par Function
+ * Can only destroy fv repo range information created
+ * through aclfvCreateRepoRange interface.
+ *
+ * @param repoRange [IN]     fv repo range.
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclfvCreateRepoRange
+ */
+ACL_FUNC_VISIBILITY aclError aclfvDestroyRepoRange(aclfvRepoRange *repoRange);
+
+/**
+ * @ingroup AscendCL
+ * @brief Create query table.
+ *
+ * @param queryCnt [IN]   Number of tables, the maximum number is 6
+ * @param tableLen [IN]   Single table length, table length is 32KB
+ * @param tableData [IN]  Feature value list
+ * @param tableDataLen [IN]   The length of memory requested by the featureData pointer
+ *
+ * @retval null for failed. OtherValues success
+ */
+ACL_FUNC_VISIBILITY aclfvQueryTable *aclfvCreateQueryTable(uint32_t queryCnt, uint32_t tableLen, uint8_t *tableData,
+                                                           uint32_t tableDataLen);
+
+/**
+ * @ingroup AscendCL
+ * @brief Destroy query table.
+ *
+ * @par Function
+ * Can only destroy query table information created
+ * through aclfvCreateQueryTable interface.
+ *
+ * @param queryTable [IN]     query table.
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclfvCreateQueryTable
+ */
+ACL_FUNC_VISIBILITY aclError aclfvDestroyQueryTable(aclfvQueryTable *queryTable);
+
+/**
+ * @ingroup AscendCL
+ * @brief Create search input.
+ *
+ * @param queryTable [IN]  query table
+ * @param repoRange [IN]   query repo range
+ * @param topk [IN]  query topk
+ *
+ * @retval null for failed. OtherValues success
+ */
+ACL_FUNC_VISIBILITY aclfvSearchInput *aclfvCreateSearchInput(aclfvQueryTable *queryTable, aclfvRepoRange *repoRange,
+                                                             uint32_t topk);
+
+/**
+ * @ingroup AscendCL
+ * @brief Destroy search input.
+ *
+ * @par Function
+ * Can only destroy search input information created
+ * through aclfvCreateSearchInput interface.
+ *
+ * @param searchInput [IN]     search input.
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclfvCreateSearchInput
+ */
+ACL_FUNC_VISIBILITY aclError aclfvDestroySearchInput(aclfvSearchInput *searchInput);
+
+/**
+ * @ingroup AscendCL
+ * @brief Create search result.
+ *
+ * @param queryCnt [IN]   Retrieve the number of features
+ * @param resultNum [IN]  The number of search results for each feature, the number is queryCnt
+ * @param resultNumDataLen [IN]  resultNum memory length
+ * @param id0 [IN]  Level 1 library id0
+ * @param id1 [IN]  Secondary library id1
+ * @param resultOffset [IN]   The offset of the bottom library corresponding
+ * to each feature retrieval result, total length topK * queryCnt
+ * @param resultDistance [IN]  Distance, total length topK * queryCnt
+ * @param dataLen [IN]  The memory size requested by
+ * id0\id1\reslutOffset\resultDistance
+ *
+ * @retval null for failed. OtherValues success
+ */
+ACL_FUNC_VISIBILITY aclfvSearchResult *aclfvCreateSearchResult(uint32_t queryCnt, uint32_t *resultNum,
+                                                               uint32_t resultNumDataLen, uint32_t *id0, uint32_t *id1,
+                                                               uint32_t *resultOffset, float *resultDistance,
+                                                               uint32_t dataLen);
+
+/**
+ * @ingroup AscendCL
+ * @brief Destroy search result.
+ *
+ * @par Function
+ * Can only destroy search result information created
+ * through aclfvCreateSearchResult interface.
+ *
+ * @param searchResult [IN]     search result.
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclfvCreateSearchResult
+ */
+ACL_FUNC_VISIBILITY aclError aclfvDestroySearchResult(aclfvSearchResult *searchResult);
+
+/**
+ * @ingroup AscendCL
+ * @brief fv IP initialize.
+ *
+ * @param initPara [IN]     fv init param.
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure.
+ */
+ACL_FUNC_VISIBILITY aclError aclfvInit(aclfvInitPara *initPara);
+
+/**
+ * @ingroup AscendCL
+ * @brief release fv resources.
+ *
+ * @par Function
+ * Can only release fv resources created
+ * through aclfvInit interface.
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure.
+ *
+ * @see aclfvInit
+ */
+ACL_FUNC_VISIBILITY aclError aclfvRelease();
+
+/**
+ * @ingroup AscendCL
+ * @brief fv repo add.
+ *
+ * @param type [IN]          repo add type
+ * @param featureInfo [IN]   add feature information
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure.
+ */
+ACL_FUNC_VISIBILITY aclError aclfvRepoAdd(aclfvSearchType type, aclfvFeatureInfo *featureInfo);
+
+/**
+ * @ingroup AscendCL
+ * @brief fv repo del.
+ *
+ * @param type [IN]       repo delete type
+ * @param repoRange [IN]  repo range information
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure.
+ */
+ACL_FUNC_VISIBILITY aclError aclfvRepoDel(aclfvSearchType type, aclfvRepoRange *repoRange);
+
+/**
+ * @ingroup AscendCL
+ * @brief fv accurate del.
+ *
+ * @param featureInfo [IN]   accurate delete feature information
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure.
+ */
+ACL_FUNC_VISIBILITY aclError aclfvDel(aclfvFeatureInfo *featureInfo);
+
+/**
+ * @ingroup AscendCL
+ * @brief fv accurate modify.
+ *
+ * @param featureInfo [IN]  accurate modify feature information
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure.
+ */
+ACL_FUNC_VISIBILITY aclError aclfvModify(aclfvFeatureInfo *featureInfo);
+
+/**
+ * @ingroup AscendCL
+ * @brief fv search.
+ *
+ * @param type [IN]  search type
+ * @param searchInput [IN]    search input
+ * @param searchRst [OUT]     search result
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure.
+ */
+ACL_FUNC_VISIBILITY aclError aclfvSearch(aclfvSearchType type, aclfvSearchInput *searchInput,
+                                         aclfvSearchResult *searchRst);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // INC_EXTERNAL_ACL_OPS_ACL_RETR_H_

inc/external/ge/ge_api.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright 2019-2022 Huawei Technologies Co., Ltd
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -57,22 +57,22 @@ class GE_FUNC_VISIBILITY Session {
 
   ///
   /// @ingroup client
-  /// @brief add a graph with a specific graphId
-  /// @param [in] graphId graph id
+  /// @brief add a graph with a specific graph id
+  /// @param [in] graph_id graph id
   /// @return Status result of function
   ///
-  Status AddGraph(uint32_t graphId, const Graph &graph);
+  Status AddGraph(uint32_t graph_id, const Graph &graph);
 
   ///
   /// @ingroup client
-  /// @brief add a graph with a specific graphId and graphOptions
+  /// @brief add a graph with a specific graph id and graphOptions
   /// @param [in] graphId graph id
   /// @param [in] graph the graph
   /// @param [in] options graph options
   /// @return Status result of function
   ///
   ATTRIBUTED_DEPRECATED(Status AddGraph(uint32_t, const Graph &, const std::map<AscendString, AscendString> &))
-  Status AddGraph(uint32_t graphId, const Graph &graph, const std::map<std::string, std::string> &options);
+  Status AddGraph(uint32_t graph_id, const Graph &graph, const std::map<std::string, std::string> &options);
 
   ///
   /// @ingroup client
@@ -82,7 +82,7 @@ class GE_FUNC_VISIBILITY Session {
   /// @param [in] options graph options
   /// @return Status result of function
   ///
-  Status AddGraph(uint32_t graphId, const Graph &graph, const std::map<AscendString, AscendString> &options);
+  Status AddGraph(uint32_t graph_id, const Graph &graph, const std::map<AscendString, AscendString> &options);
 
   ///
   /// @ingroup client
@@ -106,10 +106,10 @@ class GE_FUNC_VISIBILITY Session {
   ///
   /// @ingroup ge_graph
   /// @brief remove a graph of the session with specific session id
-  /// @param [in] graphId graph id
+  /// @param [in] graph_d graph id
   /// @return Status result of function
   ///
-  Status RemoveGraph(uint32_t graphId);
+  Status RemoveGraph(uint32_t graph_id);
 
   ///
   /// @ingroup ge_graph
@@ -119,7 +119,7 @@ class GE_FUNC_VISIBILITY Session {
   /// @param [out] outputs output data
   /// @return Status result of function
   ///
-  Status RunGraph(uint32_t graphId, const std::vector<Tensor> &inputs, std::vector<Tensor> &outputs);
+  Status RunGraph(uint32_t graph_id, const std::vector<Tensor> &inputs, std::vector<Tensor> &outputs);
 
   ///
   /// @ingroup ge_graph
@@ -140,9 +140,9 @@ class GE_FUNC_VISIBILITY Session {
   /// @param [in] inputs: input data
   /// @return Status result of function
   ///
-  Status BuildGraph(uint32_t graphId, const std::vector<InputTensorInfo> &inputs);
+  Status BuildGraph(uint32_t graph_id, const std::vector<InputTensorInfo> &inputs);
 
-  Status BuildGraph(uint32_t graphId, const std::vector<ge::Tensor> &inputs);  /*lint !e148*/
+  Status BuildGraph(uint32_t graph_id, const std::vector<ge::Tensor> &inputs); /*lint !e148*/
 
   ///
   /// @ingroup ge_graph
@@ -154,7 +154,7 @@ class GE_FUNC_VISIBILITY Session {
   ///                        Please ensure that the implementation of the function is trusted.
   /// @return Status result of function
   ///
-  Status RunGraphAsync(uint32_t graphId, const std::vector<ge::Tensor> &inputs, RunAsyncCallback callback);
+  Status RunGraphAsync(uint32_t graph_id, const std::vector<ge::Tensor> &inputs, RunAsyncCallback callback);
 
   ///
   /// @ingroup ge_graph
@@ -189,7 +189,7 @@ class GE_FUNC_VISIBILITY Session {
 
   Status RegisterCallBackFunc(const char *key, const session::pCallBackFunc &callback);
 
-  bool IsGraphNeedRebuild(uint32_t graphId);
+  bool IsGraphNeedRebuild(uint32_t graph_id);
 
  private:
   uint64_t sessionId_;

inc/external/ge/ge_api_error_codes.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright 2019-2022 Huawei Technologies Co., Ltd
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -20,14 +20,27 @@
 #include <map>
 #include <string>
 #include "ge_error_codes.h"
+#include "ge_api_types.h"
 
-namespace ge {
 #ifdef __GNUC__
 #define ATTRIBUTED_DEPRECATED(replacement) __attribute__((deprecated("Please use " #replacement " instead.")))
 #else
 #define ATTRIBUTED_DEPRECATED(replacement) __declspec(deprecated("Please use " #replacement " instead."))
 #endif
 
+// Code compose(4 byte), runtime: 2 bit,  type: 2 bit,   level: 3 bit,  sysid: 8 bit, modid: 5 bit, value: 12 bit
+#define GE_ERRORNO(runtime, type, level, sysid, modid, name, value, desc)                                \
+  constexpr ge::Status name = (static_cast<uint32_t>(0xFFU & (static_cast<uint32_t>(runtime))) << 30U) | \
+                              (static_cast<uint32_t>(0xFFU & (static_cast<uint32_t>(type))) << 28U) |    \
+                              (static_cast<uint32_t>(0xFFU & (static_cast<uint32_t>(level))) << 25U) |   \
+                              (static_cast<uint32_t>(0xFFU & (static_cast<uint32_t>(sysid))) << 17U) |   \
+                              (static_cast<uint32_t>(0xFFU & (static_cast<uint32_t>(modid))) << 12U) |   \
+                              (static_cast<uint32_t>(0x0FFFU) & (static_cast<uint32_t>(value)));         \
+  const ErrorNoRegisterar g_errorno_##name((name), (desc));
+
+#define GE_ERRORNO_EXTERNAL(name, desc) const ErrorNoRegisterar g_errorno_##name((name), (desc));
+
+namespace ge {
 class GE_FUNC_VISIBILITY StatusFactory {
  public:
   static StatusFactory *Instance() {
@@ -35,7 +48,7 @@ class GE_FUNC_VISIBILITY StatusFactory {
     return &instance;
   }
 
-  void RegisterErrorNo(uint32_t err, const std::string &desc) {
+  void RegisterErrorNo(const uint32_t err, const std::string &desc) {
     // Avoid repeated addition
     if (err_desc_.find(err) != err_desc_.end()) {
       return;
@@ -43,19 +56,19 @@ class GE_FUNC_VISIBILITY StatusFactory {
     err_desc_[err] = desc;
   }
 
-  void RegisterErrorNo(uint32_t err, const char *desc) {
+  void RegisterErrorNo(const uint32_t err, const char *const desc) {
     if (desc == nullptr) {
       return;
     }
-    std::string error_desc = desc;
+    const std::string error_desc = desc;
     if (err_desc_.find(err) != err_desc_.end()) {
       return;
     }
     err_desc_[err] = error_desc;
   }
 
-  std::string GetErrDesc(uint32_t err) {
-    auto iter_find = err_desc_.find(err);
+  std::string GetErrDesc(const uint32_t err) {
+    const std::map<uint32_t, std::string>::const_iterator iter_find = err_desc_.find(err);
     if (iter_find == err_desc_.end()) {
       return "";
     }
@@ -72,61 +85,18 @@ class GE_FUNC_VISIBILITY StatusFactory {
 
 class GE_FUNC_VISIBILITY ErrorNoRegisterar {
  public:
-  ErrorNoRegisterar(uint32_t err, const std::string &desc) { StatusFactory::Instance()->RegisterErrorNo(err, desc); }
-  ErrorNoRegisterar(uint32_t err, const char *desc) { StatusFactory::Instance()->RegisterErrorNo(err, desc); }
+  ErrorNoRegisterar(const uint32_t err, const std::string &desc) noexcept {
+    StatusFactory::Instance()->RegisterErrorNo(err, desc);
+  }
+  ErrorNoRegisterar(const uint32_t err, const char *const desc) noexcept {
+    StatusFactory::Instance()->RegisterErrorNo(err, desc);
+  }
   ~ErrorNoRegisterar() {}
 };
 
-// Code compose(4 byte), runtime: 2 bit,  type: 2 bit,   level: 3 bit,  sysid: 8 bit, modid: 5 bit, value: 12 bit
-#define GE_ERRORNO(runtime, type, level, sysid, modid, name, value, desc)                              \
-  constexpr ge::Status name =                                                                          \
-    ((0xFF & (static_cast<uint8_t>(runtime))) << 30) | ((0xFF & (static_cast<uint8_t>(type))) << 28) | \
-    ((0xFF & (static_cast<uint8_t>(level))) << 25) | ((0xFF & (static_cast<uint8_t>(sysid))) << 17) |  \
-    ((0xFF & (static_cast<uint8_t>(modid))) << 12) | (0x0FFF & (static_cast<uint16_t>(value)));        \
-  const ErrorNoRegisterar g_##name##_errorno(name, desc);
-
-#define GE_ERRORNO_EXTERNAL(name, desc) const ErrorNoRegisterar g_##name##_errorno(name, desc);
-
-using Status = uint32_t;
-
 // General error code
 GE_ERRORNO(0, 0, 0, 0, 0, SUCCESS, 0, "success");
-GE_ERRORNO(0b11, 0b11, 0b111, 0xFF, 0b11111, FAILED, 0xFFF, "failed"); /*lint !e401*/
-
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_PARAM_INVALID, "Parameter invalid.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_EXEC_NOT_INIT, "GE executor not initialized yet.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_EXEC_MODEL_PATH_INVALID, "Model file path invalid.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_EXEC_MODEL_ID_INVALID, "Model id invalid.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_EXEC_MODEL_DATA_SIZE_INVALID, "Data size of model invalid.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_EXEC_MODEL_ADDR_INVALID, "Model addr invalid.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_EXEC_MODEL_QUEUE_ID_INVALID, "Queue id of model invalid.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_EXEC_LOAD_MODEL_REPEATED, "The model loaded repeatedly.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_DYNAMIC_INPUT_ADDR_INVALID, "Dynamic input addr invalid.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_DYNAMIC_INPUT_LENGTH_INVALID, "Dynamic input size invalid.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_DYNAMIC_BATCH_SIZE_INVALID, "Dynamic batch size invalid.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_AIPP_BATCH_EMPTY, "AIPP batch parameter empty.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_AIPP_NOT_EXIST, "AIPP parameter not exist.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_AIPP_MODE_INVALID, "AIPP mode invalid.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_OP_TASK_TYPE_INVALID, "Task type invalid.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_OP_KERNEL_TYPE_INVALID, "Kernel type invalid.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_PLGMGR_PATH_INVALID, "Plugin path is invalid.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_FORMAT_INVALID, "Format is invalid.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_SHAPE_INVALID, "Shape is invalid.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_DATATYPE_INVALID, "Datatype is invalid.");
-
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_MEMORY_ALLOCATION, "Memory allocation error.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_MEMORY_OPERATE_FAILED, "Failed to operate memory.");
-
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_INTERNAL_ERROR, "Internal error.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_LOAD_MODEL, "Load model error.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_EXEC_LOAD_MODEL_PARTITION_FAILED, "Failed to load model partition.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_EXEC_LOAD_WEIGHT_PARTITION_FAILED, "Failed to load weight partition.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_EXEC_LOAD_TASK_PARTITION_FAILED, "Failed to load task partition.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_EXEC_LOAD_KERNEL_PARTITION_FAILED, "Failed to load op kernel partition.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_EXEC_RELEASE_MODEL_DATA, "Failed to release the model data.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_COMMAND_HANDLE, "Command handle error.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_GET_TENSOR_INFO, "Get tensor info error.");
-GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_UNLOAD_MODEL, "Load model error.");
+GE_ERRORNO(0b11, 0b11, 0b111, 0xFFU, 0b11111, FAILED, 0xFFFU, "failed"); /*lint !e401*/
 
 }  // namespace ge
 

inc/external/ge/ge_api_types.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright 2019-2022 Huawei Technologies Co., Ltd
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -24,96 +24,111 @@
 #include <functional>
 #include <memory>
 #include "graph/tensor.h"
+#include "graph/types.h"
 
 namespace ge {
 // Option key: graph run mode
-const char *const OPTION_GRAPH_RUN_MODE = "ge.graphRunMode";
+const char_t *const OPTION_GRAPH_RUN_MODE = "ge.graphRunMode";
+const char_t *const OPTION_DEVICE_TYPE = "ge.deviceType";
 
 // Option key: ome init
-const char *const OPTION_EXEC_SESSION_ID = "ge.exec.sessionId";
-const char *const OPTION_EXEC_DEVICE_ID = "ge.exec.deviceId";
-const char *const OPTION_EXEC_JOB_ID = "ge.exec.jobId";
-const char *const OPTION_EXEC_IS_USEHCOM = "ge.exec.isUseHcom";
-const char *const OPTION_EXEC_IS_USEHVD = "ge.exec.isUseHvd";
-const char *const OPTION_EXEC_RANK_ID = "ge.exec.rankId";
-const char *const OPTION_EXEC_POD_NAME = "ge.exec.podName";
-const char *const OPTION_EXEC_DEPLOY_MODE = "ge.exec.deployMode";
-const char *const OPTION_EXEC_RANK_TABLE_FILE = "ge.exec.rankTableFile";
-const char *const GE_AICPU_FLAG = "ge.aicpuFlag";
-const char *const OPTION_EXEC_EXTERN_PLUGIN_PATH = "ge.soLoadPath";
+const char_t *const OPTION_EXEC_SESSION_ID = "ge.exec.sessionId";
+const char_t *const OPTION_EXEC_DEVICE_ID = "ge.exec.deviceId";
+const char_t *const OPTION_EXEC_JOB_ID = "ge.exec.jobId";
+const char_t *const OPTION_EXEC_IS_USEHCOM = "ge.exec.isUseHcom";
+const char_t *const OPTION_EXEC_IS_USEHVD = "ge.exec.isUseHvd";
+const char_t *const OPTION_EXEC_RANK_ID = "ge.exec.rankId";
+const char_t *const OPTION_EXEC_POD_NAME = "ge.exec.podName";
+const char_t *const OPTION_EXEC_DEPLOY_MODE = "ge.exec.deployMode";
+const char_t *const OPTION_EXEC_RANK_TABLE_FILE = "ge.exec.rankTableFile";
+const char_t *const GE_AICPU_FLAG = "ge.aicpuFlag";
+const char_t *const OPTION_EXEC_EXTERN_PLUGIN_PATH = "ge.soLoadPath";
 // Dump flag and para
-const char *const OPTION_EXEC_ENABLE_DUMP = "ge.exec.enableDump";
-const char *const OPTION_EXEC_DUMP_PATH = "ge.exec.dumpPath";
-const char *const OPTION_EXEC_DUMP_STEP = "ge.exec.dumpStep";
-const char *const OPTION_EXEC_DUMP_MODE = "ge.exec.dumpMode";
-const char *const OPTION_EXEC_ENABLE_DUMP_DEBUG = "ge.exec.enableDumpDebug";
-const char *const OPTION_EXEC_DUMP_DEBUG_MODE = "ge.exec.dumpDebugMode";
-const char *const OPTION_EXEC_ENABLE_INCRE_BUILD = "ge.exec.enableIncreBuild";
-const char *const OPTION_EXEC_INCRE_BUILD_CACHE_PATH = "ge.exec.increBuildCachePath";
-const char *const OPTION_EXEC_ENABLE_EXCEPTION_DUMP = "ge.exec.enable_exception_dump";
-const char *const OPTION_EXEC_ENABLE_SCOPE_FUSION_PASSES = "ge.exec.enableScopeFusionPasses";
-const char *const OPTION_EXEC_PROFILING_FPPONIT_OPTIONS = "ge.exec.profilingFpPointOptions";
-const char *const OPTION_EXEC_PROFILING_BPPONIT_OPTIONS = "ge.exec.profilingBpPointOptions";
+const char_t *const OPTION_EXEC_ENABLE_DUMP = "ge.exec.enableDump";
+const char_t *const OPTION_EXEC_DUMP_PATH = "ge.exec.dumpPath";
+const char_t *const OPTION_EXEC_DUMP_STEP = "ge.exec.dumpStep";
+const char_t *const OPTION_EXEC_DUMP_MODE = "ge.exec.dumpMode";
+const char_t *const OPTION_EXEC_ENABLE_DUMP_DEBUG = "ge.exec.enableDumpDebug";
+const char_t *const OPTION_EXEC_DUMP_DEBUG_MODE = "ge.exec.dumpDebugMode";
+const char_t *const OPTION_EXEC_ENABLE_INCRE_BUILD = "ge.exec.enableIncreBuild";
+const char_t *const OPTION_EXEC_INCRE_BUILD_CACHE_PATH = "ge.exec.increBuildCachePath";
+const char_t *const OPTION_EXEC_ENABLE_EXCEPTION_DUMP = "ge.exec.enable_exception_dump";
+const char_t *const OPTION_EXEC_ENABLE_SCOPE_FUSION_PASSES = "ge.exec.enableScopeFusionPasses";
+const char_t *const OPTION_EXEC_PROFILING_FPPONIT_OPTIONS = "ge.exec.profilingFpPointOptions";
+const char_t *const OPTION_EXEC_PROFILING_BPPONIT_OPTIONS = "ge.exec.profilingBpPointOptions";
 // profiling flag
-const char *const OPTION_EXEC_PROFILING_MODE = "ge.exec.profilingMode";
-const char *const OPTION_EXEC_PROFILING_OPTIONS = "ge.exec.profilingOptions";
+const char_t *const OPTION_EXEC_PROFILING_MODE = "ge.exec.profilingMode";
+const char_t *const OPTION_EXEC_PROFILING_OPTIONS = "ge.exec.profilingOptions";
 // Hccl flag, if ge.exec.hcclFlag =1, it means load plugin for opskernel, else:ge.exec.hcclFlag =0
-const char *const OPTION_EXEC_HCCL_FLAG = "ge.exec.hcclFlag";
-const char *const OPTION_EXEC_ATOMIC_FLAG = "ge.exec.enable_atomic";
-const char *const OPTION_EXEC_DISABLE_REUSED_MEMORY = "ge.exec.disableReuseMemory";
-const char *const OPTION_EXEC_ENABLE_TAILING_OPTIMIZATION = "ge.exec.isTailingOptimization";
+const char_t *const OPTION_EXEC_HCCL_FLAG = "ge.exec.hcclFlag";
+const char_t *const OPTION_EXEC_ATOMIC_FLAG = "ge.exec.enable_atomic";
+const char_t *const OPTION_EXEC_DISABLE_REUSED_MEMORY = "ge.exec.disableReuseMemory";
+const char_t *const OPTION_EXEC_ENABLE_TAILING_OPTIMIZATION = "ge.exec.isTailingOptimization";
 // Dynamic input flag. ge.exec.dynamicInput=1, means enable dynaimc input,
 // ge.exec.dynamicGraphExecuteMode, dynamic_execute[default]
-const char *const OPTION_EXEC_DYNAMIC_INPUT = "ge.exec.dynamicInput";
-const char *const OPTION_EXEC_DYNAMIC_EXECUTE_MODE = "ge.exec.dynamicGraphExecuteMode";
-const char *const OPTION_EXEC_DATA_INPUTS_SHAPE_RANGE = "ge.exec.dataInputsShapeRange";
+const char_t *const OPTION_EXEC_DYNAMIC_INPUT = "ge.exec.dynamicInput";
+const char_t *const OPTION_EXEC_DYNAMIC_EXECUTE_MODE = "ge.exec.dynamicGraphExecuteMode";
+const char_t *const OPTION_EXEC_DATA_INPUTS_SHAPE_RANGE = "ge.exec.dataInputsShapeRange";
+const char_t *const OPTION_EXEC_ENABLE_COPY_OUTPUT_ADDR = "ge.exec.enableCopyOutputAddr";
+const char_t *const OPTION_EXEC_GRAPH_EXEC_TIMEOUT = "ge.exec.graphExecTimeout";
 
 // Option key: memory init
-const char *const GRAPH_MEMORY_MAX_SIZE = "ge.graphMemoryMaxSize";
-const char *const VARIABLE_MEMORY_MAX_SIZE = "ge.variableMemoryMaxSize";
+const char_t *const GRAPH_MEMORY_MAX_SIZE = "ge.graphMemoryMaxSize";
+const char_t *const VARIABLE_MEMORY_MAX_SIZE = "ge.variableMemoryMaxSize";
+const char_t *const OPTION_EXEC_REUSE_ZERO_COPY_MEMORY = "ge.exec.reuseZeroCopyMemory";
+
+const std::string ATOMIC_CLEAN_POLICY = "ge.exec.atomicCleanPolicy";
+
+const char_t *const OPTION_EXEC_LOGICAL_DEVICE_CLUSTER_DEPLOY_MODE = "ge.exec.logicalDeviceClusterDeployMode";
+const char_t *const OPTION_EXEC_LOGICAL_DEVICE_ID = "ge.exec.logicalDeviceId";
+
 namespace configure_option {
-const char *const STREAM_NUM = "ge.streamNum";
-const char *const HEAD_STREAM = "ge.headStream";
-const char *const PERF_LEVEL = "ge.perfLevel";
-const char *const ENCRYPT_MODE = "ge.encryptMode";
-const char *const EK_FILE = "ge.ekFile";
-const char *const CERT_FILE = "ge.certFile";
-const char *const HW_KEY_FILE = "ge.hwKeyFile";
-const char *const PRIVATE_KEY_FILE = "ge.privateKeyFile";
-const char *const FRAMEWORK_TYPE = "ge.frameworkType";
-const char *const CALIBRATION_CONF_FILE = "ge.calibrationConfFile";
-const char *const INSERT_OP_FILE = "ge.insertOpFile";
-const char *const OUTPUT_NODE_NAME = "ge.outputNodeName";
-const char *const COMPRESS_FLAG = "ge.compressFlag";
-const char *const PRECISION_MODE = "ge.exec.precision_mode";
-const char *const SINGLE_OP_FLAG = "ge.exec.single_op";
-const char *const TRAIN_FLAG = "ge.trainFlag";
-const char *const RUN_FLAG = "ge.runFlag";
-const char *const LOCAL_FMKOP_FLAG = "ge.enabledLocalFmkop";
-const char *const TBE_PLUGIN_PATH_FLAG = "ge.TBE_plugin_path";
-const char *const DDK_VERSION_FLAG = "ge.DDK_version";
-const char *const GE_FE_FLAG = "ge.feFlag";
-const char *const STREAM_MAX_PARALLEL_NUM = "ge.streamMaxParallelNum";
-const char *const OUTPUT_DATATYPE = "ge.outputDatatype";
-const char *const OP_SELECT_IMPL_MODE = "ge.opSelectImplmode";
-const char *const OPTYPELIST_FOR_IMPLMODE = "ge.optypelistForImplmode";
-const char *const HCOM_PARALLEL = "ge.hcomParallel";
-const char *const AUTO_TUNE_MODE = "ge.autoTuneMode";
-const char *const SOC_VERSION = "ge.socVersion";
-const char *const CORE_TYPE = "ge.engineType";
-const char *const AICORE_NUM = "ge.aicoreNum";
-const char *const L1_FUSION = "ge.l1Fusion";
-const char *const BUFFER_OPTIMIZE = "ge.bufferOptimize";
-const char *const ENABLE_SMALL_CHANNEL = "ge.enableSmallChannel";
-const char *const ENABLE_COMPRESS_WEIGHT = "ge.enableCompressWeight";
-const char *const FUSION_SWITCH_FILE = "ge.fusionSwitchFile";
-const char *const SAVE_ORIGINAL_MODEL = "ge.saveOriginalModel";
-const char *const ORIGINAL_MODEL_FILE = "ge.originalModelFile";
-const char *const INPUT_FP16_NODES = "ge.INPUT_NODES_SET_FP16";
-const char *const OP_DEBUG_LEVEL = "ge.opDebugLevel";
-const char *const PERFORMANCE_MODE = "ge.performance_mode";
-const char *const MODIFY_MIXLIST = "ge.exec.modify_mixlist";
-const char *const OP_PRECISION_MODE = "ge.exec.op_precision_mode";
+const char_t *const STREAM_NUM = "ge.streamNum";
+const char_t *const HEAD_STREAM = "ge.headStream";
+const char_t *const PERF_LEVEL = "ge.perfLevel";
+const char_t *const ENCRYPT_MODE = "ge.encryptMode";
+const char_t *const EK_FILE = "ge.ekFile";
+const char_t *const CERT_FILE = "ge.certFile";
+const char_t *const HW_KEY_FILE = "ge.hwKeyFile";
+const char_t *const PRIVATE_KEY_FILE = "ge.privateKeyFile";
+const char_t *const FRAMEWORK_TYPE = "ge.frameworkType";
+const char_t *const CALIBRATION_CONF_FILE = "ge.calibrationConfFile";
+const char_t *const INSERT_OP_FILE = "ge.insertOpFile";
+const char_t *const OUTPUT_NODE_NAME = "ge.outputNodeName";
+const char_t *const COMPRESS_FLAG = "ge.compressFlag";
+const char_t *const PRECISION_MODE = "ge.exec.precision_mode";
+const char_t *const SINGLE_OP_FLAG = "ge.exec.single_op";
+const char_t *const TRAIN_FLAG = "ge.trainFlag";
+const char_t *const RUN_FLAG = "ge.runFlag";
+const char_t *const LOCAL_FMKOP_FLAG = "ge.enabledLocalFmkop";
+const char_t *const TBE_PLUGIN_PATH_FLAG = "ge.TBE_plugin_path";
+const char_t *const DDK_VERSION_FLAG = "ge.DDK_version";
+const char_t *const GE_FE_FLAG = "ge.feFlag";
+const char_t *const STREAM_MAX_PARALLEL_NUM = "ge.streamMaxParallelNum";
+const char_t *const OUTPUT_DATATYPE = "ge.outputDatatype";
+const char_t *const OP_SELECT_IMPL_MODE = "ge.opSelectImplmode";
+const char_t *const OPTYPELIST_FOR_IMPLMODE = "ge.optypelistForImplmode";
+const char_t *const HCOM_PARALLEL = "ge.hcomParallel";
+const char_t *const AUTO_TUNE_MODE = "ge.autoTuneMode";
+const char_t *const SOC_VERSION = "ge.socVersion";
+const char_t *const VIRTUAL_TYPE = "ge.virtual_type";
+const char_t *const CORE_TYPE = "ge.engineType";
+const char_t *const AICORE_NUM = "ge.aicoreNum";
+const char_t *const L1_FUSION = "ge.l1Fusion";
+const char_t *const BUFFER_OPTIMIZE = "ge.bufferOptimize";
+const char_t *const ENABLE_SMALL_CHANNEL = "ge.enableSmallChannel";
+const char_t *const ENABLE_COMPRESS_WEIGHT = "ge.enableCompressWeight";
+const char_t *const FUSION_SWITCH_FILE = "ge.fusionSwitchFile";
+const char_t *const SAVE_ORIGINAL_MODEL = "ge.saveOriginalModel";
+const char_t *const ORIGINAL_MODEL_FILE = "ge.originalModelFile";
+const char_t *const INPUT_FP16_NODES = "ge.INPUT_NODES_SET_FP16";
+const char_t *const OP_DEBUG_LEVEL = "ge.opDebugLevel";
+const char_t *const PERFORMANCE_MODE = "ge.performance_mode";
+const char_t *const SHAPE_GENERALIZED_BUILD_MODE = "ge.shape_generalized_build_mode";
+const char_t *const MODIFY_MIXLIST = "ge.exec.modify_mixlist";
+const char_t *const OP_PRECISION_MODE = "ge.exec.op_precision_mode";
+const char_t *const CUSTOMIZE_DTYPES = "ge.customizeDtypes";
+const char_t *const COMPRESSION_OPTIMIZE_CONF = "ge.compressionOptimizeConf";
 }  // namespace configure_option
 // Configure stream num by Session constructor options param,
 // its value should be int32_t type, default value is "1"
@@ -223,7 +238,7 @@ const std::string OPTYPELIST_FOR_IMPLMODE = "ge.optypelistForImplmode";
 const std::string HCOM_PARALLEL = "ge.hcomParallel";
 
 // configure whether to use dynamic batch size
-const char *const kDynamicBatchSize = "ge.dynamicBatchSize";
+const char_t *const kDynamicBatchSize = "ge.dynamicBatchSize";
 
 // configure threshold of fusion data size for communication op
 const std::string FUSION_TENSOR_SIZE = "ge.fusionTensorSize";
@@ -232,10 +247,10 @@ const std::string INPUT_SHAPE = "ge.inputShape";
 
 const std::string DYNAMIC_NODE_TYPE = "ge.dynamicNodeType";
 // configure whether to use dynamic image size
-const char *const kDynamicImageSize = "ge.dynamicImageSize";
+const char_t *const kDynamicImageSize = "ge.dynamicImageSize";
 
 // Configure whether to use dynamic dims
-const char *const kDynamicDims = "ge.dynamicDims";
+const char_t *const kDynamicDims = "ge.dynamicDims";
 
 // Configure auto tune mode, this option only take effect while AUTO_TUNE_FLAG is Y,
 // example: GA|RL, support configure multiple, split by |
@@ -244,9 +259,16 @@ const std::string AUTO_TUNE_MODE = "ge.autoTuneMode";
 // Configure soc version , example: "Ascend310"
 const std::string SOC_VERSION = "ge.socVersion";
 
+// configure whether to enable virtualization,
+// its value should be "0" or "1", default value is "0"
+const std::string VIRTUAL_TYPE = "ge.virtual_type";
+
 // Configure core type "VectorEngine", default value is "AIcoreEngine"
 const std::string CORE_TYPE = "ge.engineType";
 
+// Configure graph exclude one or more engines
+const std::string EXCLUDE_ENGINES = "ge.exec.exclude_engines";
+
 // Configure AICORE NUM
 const std::string AICORE_NUM = "ge.aicoreNum";
 
@@ -262,38 +284,51 @@ const std::string ENABLE_SMALL_CHANNEL = "ge.enableSmallChannel";
 // Configure Compress Weight flag
 const std::string ENABLE_COMPRESS_WEIGHT = "ge.enableCompressWeight";
 
+// Configure Sparse Matrix Weight flag
+const std::string ENABLE_SPARSE_MATRIX_WEIGHT = "ge.enableSparseMatrixWeight";
+
 // Configure fusion switch file path
 const std::string FUSION_SWITCH_FILE = "ge.fusionSwitchFile";
 
+// Configure compression optimize file path
+const std::string COMPRESSION_OPTIMIZE_CONF = "ge.compressionOptimizeConf";
+
+// Configure customize dtypes path
+const std::string CUSTOMIZE_DTYPES = "ge.customizeDtypes";
+
 // Save original model
 const std::string SAVE_ORIGINAL_MODEL = "ge.saveOriginalModel";
 
 // Save original model file name
 const std::string ORIGINAL_MODEL_FILE = "ge.originalModelFile";
 
-const char *const OPTION_GE_MAX_DUMP_FILE_NUM = "ge.maxDumpFileNum";
-const char *const OPTION_GE_MAX_DUMP_FILE_SIZE = "ge.maxDumpFileSize";
-const char *const OPTION_GE_MAX_DUMP_OP_NUM = "ge.maxDumpOpNum";
+const char_t *const OPTION_GE_MAX_DUMP_FILE_NUM = "ge.maxDumpFileNum";
+const char_t *const OPTION_GE_MAX_DUMP_FILE_SIZE = "ge.maxDumpFileSize";
+const char_t *const OPTION_GE_MAX_DUMP_OP_NUM = "ge.maxDumpOpNum";
 
 // Configure for print op pass
 // Its value should be "0" or "1", default value is "1"
-const char *const ENABLE_PRINT_OP_PASS = "ge.enablePrintOpPass";
+const char_t *const ENABLE_PRINT_OP_PASS = "ge.enablePrintOpPass";
 
 // Configure operator compilation path
 // Its value should be file path, default value is "./"
-const char *const DEBUG_DIR = "ge.debugDir";
+const char_t *const DEBUG_DIR = "ge.debugDir";
+
+// Configure switch for op status check such as overflow
+// Its value should be true of flase
+const char_t *const STATUS_CHECK = "ge.status_check";
 
 // Configure operator compiler cache path
 // Its value should be file path, default value is "./"
-const char *const OP_COMPILER_CACHE_DIR = "ge.op_compiler_cache_dir";
+const char_t *const OP_COMPILER_CACHE_DIR = "ge.op_compiler_cache_dir";
 
 // Configure operator compiler cache mode
 // Its value should be "disable", "enable" or "force", default value is "disable"
-const char *const OP_COMPILER_CACHE_MODE = "ge.op_compiler_cache_mode";
+const char_t *const OP_COMPILER_CACHE_MODE = "ge.op_compiler_cache_mode";
 
 // Configure whether to use single stream.
 // Its value should be "true" or "false", default value is "false"
-const char *const ENABLE_SINGLE_STREAM = "ge.enableSingleStream";
+const char_t *const ENABLE_SINGLE_STREAM = "ge.enableSingleStream";
 
 // Configure input fp16 nodes
 const std::string INPUT_FP16_NODES = "ge.INPUT_NODES_SET_FP16";
@@ -318,17 +353,28 @@ const std::string OP_BANK_UPDATE_FLAG = "ge.op_bank_update";
 const std::string HCOM_MULTI_MODE = "ge.hcomMultiMode";
 
 // atc and ir option
-const char *const INPUT_SHAPE_RANGE = "input_shape_range";
+const char_t *const INPUT_SHAPE_RANGE = "input_shape_range";
 
 // Configure express high compile performance or high execute performance
 // normal: no need to compile, used saved .o files directly
 // high: need to recompile, high execute performance mode
 const std::string PERFORMANCE_MODE = "ge.performance_mode";
 
+// For selecting the mode of shape generalization when build graph.
+// shape_generalized: Shape will be generalized during graph build.
+// shape_precise: Shape will not be generalized, use precise shape.
+const std::string SHAPE_GENERALIZED_BUILD_MODE = "ge.shape_generalized_build_mode";
+
 const std::string MODIFY_MIXLIST = "ge.exec.modify_mixlist";
 
 const std::string OP_PRECISION_MODE = "ge.exec.op_precision_mode";
 
+const std::string OP_WAIT_TIMEOUT = "ge.exec.opWaitTimeout";
+
+const std::string OP_EXECUTE_TIMEOUT = "ge.exec.opExecuteTimeout";
+
+const char_t *const FILE_CONSTANT_PATH = "ge.exec.value_bins";
+
 // Graph run mode
 enum GraphRunMode { PREDICTION = 0, TRAIN };
 
@@ -345,7 +391,7 @@ struct OutputTensorInfo {
   std::vector<int64_t> dims;        // shape description
   std::unique_ptr<uint8_t[]> data;  // tensor data
   int64_t length;                   // tensor length
-  OutputTensorInfo() : data_type(0), dims({}), data(nullptr), length(0) {}
+  OutputTensorInfo() : data_type(0U), dims({}), data(nullptr), length(0) {}
   OutputTensorInfo(OutputTensorInfo &&out)
       : data_type(out.data_type), dims(out.dims), data(std::move(out.data)), length(out.length) {}
 
@@ -367,48 +413,53 @@ using RunAsyncCallback = std::function<void(Status, std::vector<ge::Tensor> &)>;
 
 // for ir build
 namespace ir_option {
-static const char *const INPUT_FORMAT = "input_format";
-static const char *const INPUT_SHAPE = "input_shape";
-static const char *const INPUT_SHAPE_RANGE = ge::INPUT_SHAPE_RANGE;
-static const char *const OP_NAME_MAP = "op_name_map";
-static const char *const IS_DYNAMIC_INPUT = "is_dynamic_input";
-static const char *const IS_INPUT_ADJUST_HW_LAYOUT = "is_input_adjust_hw_layout";
-static const char *const IS_OUTPUT_ADJUST_HW_LAYOUT = "is_output_adjust_hw_layout";
-static const char *const ENABLE_SCOPE_FUSION_PASSES = "enable_scope_fusion_passes";
-static const char *const OUTPUT = "output";
-static const char *const DYNAMIC_BATCH_SIZE = kDynamicBatchSize;
-static const char *const DYNAMIC_IMAGE_SIZE = kDynamicImageSize;
-static const char *const DYNAMIC_DIMS = kDynamicDims;
-static const char *const INSERT_OP_FILE = ge::INSERT_OP_FILE.c_str();
-static const char *const PRECISION_MODE = ge::PRECISION_MODE.c_str();
-static const char *const TUNE_DEVICE_IDS = ge::TUNE_DEVICE_IDS.c_str();
-static const char *const EXEC_DISABLE_REUSED_MEMORY = ge::OPTION_EXEC_DISABLE_REUSED_MEMORY;
-static const char *const AUTO_TUNE_MODE = ge::AUTO_TUNE_MODE.c_str();
-static const char *const CORE_TYPE = ge::CORE_TYPE.c_str();
-static const char *const SOC_VERSION = ge::SOC_VERSION.c_str();
-static const char *const ENABLE_SINGLE_STREAM = ge::ENABLE_SINGLE_STREAM;
-static const char *const AICORE_NUM = ge::AICORE_NUM.c_str();
-static const char *const FUSION_SWITCH_FILE = ge::FUSION_SWITCH_FILE.c_str();
-static const char *const ENABLE_SMALL_CHANNEL = ge::ENABLE_SMALL_CHANNEL.c_str();
-static const char *const OP_SELECT_IMPL_MODE = ge::OP_SELECT_IMPL_MODE.c_str();
-static const char *const OUTPUT_TYPE = ge::OUTPUT_DATATYPE.c_str();
-static const char *const BUFFER_OPTIMIZE = ge::BUFFER_OPTIMIZE.c_str();
-static const char *const ENABLE_COMPRESS_WEIGHT = ge::ENABLE_COMPRESS_WEIGHT.c_str();
-static const char *const COMPRESS_WEIGHT_CONF = "compress_weight_conf";
-static const char *const OUT_NODES = ge::OUTPUT_NODE_NAME.c_str();
-static const char *const INPUT_FP16_NODES = ge::INPUT_FP16_NODES.c_str();
-static const char *const LOG_LEVEL = "log";
-static const char *const OPTYPELIST_FOR_IMPLMODE = ge::OPTYPELIST_FOR_IMPLMODE.c_str();
-static const char *const DEBUG_DIR = ge::DEBUG_DIR;
-static const char *const OP_COMPILER_CACHE_DIR = ge::OP_COMPILER_CACHE_DIR;
-static const char *const OP_COMPILER_CACHE_MODE = ge::OP_COMPILER_CACHE_MODE;
-static const char *const MDL_BANK_PATH = ge::MDL_BANK_PATH_FLAG.c_str();
-static const char *const OP_BANK_PATH = ge::OP_BANK_PATH_FLAG.c_str();
-static const char *const OP_BANK_UPDATE = ge::OP_BANK_UPDATE_FLAG.c_str();
-static const char *const OP_DEBUG_LEVEL = ge::OP_DEBUG_LEVEL.c_str();
-static const char *const PERFORMANCE_MODE = ge::PERFORMANCE_MODE.c_str();
-static const char *const MODIFY_MIXLIST = ge::MODIFY_MIXLIST.c_str();
-static const char *const OP_PRECISION_MODE = ge::OP_PRECISION_MODE.c_str();
+static const char_t *const INPUT_FORMAT = "input_format";
+static const char_t *const INPUT_SHAPE = "input_shape";
+static const char_t *const INPUT_SHAPE_RANGE = ge::INPUT_SHAPE_RANGE;
+static const char_t *const OP_NAME_MAP = "op_name_map";
+static const char_t *const IS_DYNAMIC_INPUT = "is_dynamic_input";
+static const char_t *const IS_INPUT_ADJUST_HW_LAYOUT = "is_input_adjust_hw_layout";
+static const char_t *const IS_OUTPUT_ADJUST_HW_LAYOUT = "is_output_adjust_hw_layout";
+static const char_t *const ENABLE_SCOPE_FUSION_PASSES = "enable_scope_fusion_passes";
+static const char_t *const OUTPUT = "output";
+static const char_t *const DYNAMIC_BATCH_SIZE = kDynamicBatchSize;
+static const char_t *const DYNAMIC_IMAGE_SIZE = kDynamicImageSize;
+static const char_t *const DYNAMIC_DIMS = kDynamicDims;
+static const char_t *const INSERT_OP_FILE = ge::INSERT_OP_FILE.c_str();
+static const char_t *const PRECISION_MODE = ge::PRECISION_MODE.c_str();
+static const char_t *const TUNE_DEVICE_IDS = ge::TUNE_DEVICE_IDS.c_str();
+static const char_t *const EXEC_DISABLE_REUSED_MEMORY = ge::OPTION_EXEC_DISABLE_REUSED_MEMORY;
+static const char_t *const AUTO_TUNE_MODE = ge::AUTO_TUNE_MODE.c_str();
+static const char_t *const CORE_TYPE = ge::CORE_TYPE.c_str();
+static const char_t *const SOC_VERSION = ge::SOC_VERSION.c_str();
+static const char_t *const VIRTUAL_TYPE = ge::VIRTUAL_TYPE.c_str();
+static const char_t *const ENABLE_SINGLE_STREAM = ge::ENABLE_SINGLE_STREAM;
+static const char_t *const AICORE_NUM = ge::AICORE_NUM.c_str();
+static const char_t *const FUSION_SWITCH_FILE = ge::FUSION_SWITCH_FILE.c_str();
+static const char_t *const ENABLE_SMALL_CHANNEL = ge::ENABLE_SMALL_CHANNEL.c_str();
+static const char_t *const OP_SELECT_IMPL_MODE = ge::OP_SELECT_IMPL_MODE.c_str();
+static const char_t *const OUTPUT_TYPE = ge::OUTPUT_DATATYPE.c_str();
+static const char_t *const BUFFER_OPTIMIZE = ge::BUFFER_OPTIMIZE.c_str();
+static const char_t *const ENABLE_COMPRESS_WEIGHT = ge::ENABLE_COMPRESS_WEIGHT.c_str();
+static const char_t *const SPARSITY = ge::ENABLE_SPARSE_MATRIX_WEIGHT.c_str();
+static const char_t *const COMPRESS_WEIGHT_CONF = "compress_weight_conf";
+static const char_t *const OUT_NODES = ge::OUTPUT_NODE_NAME.c_str();
+static const char_t *const INPUT_FP16_NODES = ge::INPUT_FP16_NODES.c_str();
+static const char_t *const LOG_LEVEL = "log";
+static const char_t *const OPTYPELIST_FOR_IMPLMODE = ge::OPTYPELIST_FOR_IMPLMODE.c_str();
+static const char_t *const DEBUG_DIR = ge::DEBUG_DIR;
+static const char_t *const OP_COMPILER_CACHE_DIR = ge::OP_COMPILER_CACHE_DIR;
+static const char_t *const OP_COMPILER_CACHE_MODE = ge::OP_COMPILER_CACHE_MODE;
+static const char_t *const MDL_BANK_PATH = ge::MDL_BANK_PATH_FLAG.c_str();
+static const char_t *const OP_BANK_PATH = ge::OP_BANK_PATH_FLAG.c_str();
+static const char_t *const OP_BANK_UPDATE = ge::OP_BANK_UPDATE_FLAG.c_str();
+static const char_t *const OP_DEBUG_LEVEL = ge::OP_DEBUG_LEVEL.c_str();
+static const char_t *const PERFORMANCE_MODE = ge::PERFORMANCE_MODE.c_str();
+static const char_t *const SHAPE_GENERALIZED_BUILD_MODE = ge::SHAPE_GENERALIZED_BUILD_MODE.c_str();
+static const char_t *const MODIFY_MIXLIST = ge::MODIFY_MIXLIST.c_str();
+static const char_t *const OP_PRECISION_MODE = ge::OP_PRECISION_MODE.c_str();
+static const char_t *const CUSTOMIZE_DTYPES = "ge.customizeDtypes";
+static const char_t *const COMPRESSION_OPTIMIZE_CONF = "ge.compressionOptimizeConf";
 
 // for interface: aclgrphBuildModel
 #ifdef __GNUC__
@@ -437,19 +488,23 @@ const std::set<std::string> ir_builder_suppported_options = {INPUT_FORMAT,
                                                              OP_BANK_PATH,
                                                              OP_BANK_UPDATE,
                                                              PERFORMANCE_MODE,
-                                                             MODIFY_MIXLIST};
+                                                             SHAPE_GENERALIZED_BUILD_MODE,
+                                                             MODIFY_MIXLIST,
+                                                             CUSTOMIZE_DTYPES};
 
 // for interface: aclgrphParse
 const std::set<std::string> ir_parser_suppported_options = {
-  INPUT_FP16_NODES, IS_INPUT_ADJUST_HW_LAYOUT, IS_OUTPUT_ADJUST_HW_LAYOUT, OUTPUT,
-  OUT_NODES,        ENABLE_SCOPE_FUSION_PASSES};
+    INPUT_FP16_NODES, IS_INPUT_ADJUST_HW_LAYOUT, IS_OUTPUT_ADJUST_HW_LAYOUT, OUTPUT,
+    OUT_NODES,        ENABLE_SCOPE_FUSION_PASSES};
 
 // for interface: aclgrphBuildInitialize
 const std::set<std::string> global_options = {CORE_TYPE,
                                               SOC_VERSION,
+                                              VIRTUAL_TYPE,
                                               BUFFER_OPTIMIZE,
                                               ENABLE_COMPRESS_WEIGHT,
                                               COMPRESS_WEIGHT_CONF,
+                                              SPARSITY,
                                               PRECISION_MODE,
                                               TUNE_DEVICE_IDS,
                                               EXEC_DISABLE_REUSED_MEMORY,
@@ -464,7 +519,8 @@ const std::set<std::string> global_options = {CORE_TYPE,
                                               DEBUG_DIR,
                                               OP_COMPILER_CACHE_DIR,
                                               OP_COMPILER_CACHE_MODE,
-                                              MODIFY_MIXLIST};
+                                              MODIFY_MIXLIST,
+                                              COMPRESSION_OPTIMIZE_CONF};
 #endif
 }  // namespace ir_option
 }  // namespace ge

inc/external/ge/ge_error_codes.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright 2019-2022 Huawei Technologies Co., Ltd
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -32,42 +32,44 @@
 #endif
 
 #include <stddef.h>
+#include <stdint.h>
 
 #ifdef __cplusplus
 extern "C" {
 #endif
-static const uint32_t ACL_ERROR_GE_PARAM_INVALID = 145000;
-static const uint32_t ACL_ERROR_GE_EXEC_NOT_INIT = 145001;
-static const uint32_t ACL_ERROR_GE_EXEC_MODEL_PATH_INVALID = 145002;
-static const uint32_t ACL_ERROR_GE_EXEC_MODEL_ID_INVALID = 145003;
-static const uint32_t ACL_ERROR_GE_EXEC_MODEL_DATA_SIZE_INVALID = 145006;
-static const uint32_t ACL_ERROR_GE_EXEC_MODEL_ADDR_INVALID = 145007;
-static const uint32_t ACL_ERROR_GE_EXEC_MODEL_QUEUE_ID_INVALID = 145008;
-static const uint32_t ACL_ERROR_GE_EXEC_LOAD_MODEL_REPEATED = 145009;
-static const uint32_t ACL_ERROR_GE_DYNAMIC_INPUT_ADDR_INVALID = 145011;
-static const uint32_t ACL_ERROR_GE_DYNAMIC_INPUT_LENGTH_INVALID = 145012;
-static const uint32_t ACL_ERROR_GE_DYNAMIC_BATCH_SIZE_INVALID = 145013;
-static const uint32_t ACL_ERROR_GE_AIPP_BATCH_EMPTY = 145014;
-static const uint32_t ACL_ERROR_GE_AIPP_NOT_EXIST = 145015;
-static const uint32_t ACL_ERROR_GE_AIPP_MODE_INVALID = 145016;
-static const uint32_t ACL_ERROR_GE_OP_TASK_TYPE_INVALID = 145017;
-static const uint32_t ACL_ERROR_GE_OP_KERNEL_TYPE_INVALID = 145018;
-static const uint32_t ACL_ERROR_GE_PLGMGR_PATH_INVALID = 145019;
-static const uint32_t ACL_ERROR_GE_FORMAT_INVALID = 145020;
-static const uint32_t ACL_ERROR_GE_SHAPE_INVALID = 145021;
-static const uint32_t ACL_ERROR_GE_DATATYPE_INVALID = 145022;
-static const uint32_t ACL_ERROR_GE_MEMORY_ALLOCATION = 245000;
-static const uint32_t ACL_ERROR_GE_MEMORY_OPERATE_FAILED = 245001;
-static const uint32_t ACL_ERROR_GE_INTERNAL_ERROR = 545000;
-static const uint32_t ACL_ERROR_GE_LOAD_MODEL = 545001;
-static const uint32_t ACL_ERROR_GE_EXEC_LOAD_MODEL_PARTITION_FAILED = 545002;
-static const uint32_t ACL_ERROR_GE_EXEC_LOAD_WEIGHT_PARTITION_FAILED = 545003;
-static const uint32_t ACL_ERROR_GE_EXEC_LOAD_TASK_PARTITION_FAILED = 545004;
-static const uint32_t ACL_ERROR_GE_EXEC_LOAD_KERNEL_PARTITION_FAILED = 545005;
-static const uint32_t ACL_ERROR_GE_EXEC_RELEASE_MODEL_DATA = 545006;
-static const uint32_t ACL_ERROR_GE_COMMAND_HANDLE = 545007;
-static const uint32_t ACL_ERROR_GE_GET_TENSOR_INFO = 545008;
-static const uint32_t ACL_ERROR_GE_UNLOAD_MODEL = 545009;
+static const uint32_t ACL_ERROR_GE_PARAM_INVALID = 145000U;
+static const uint32_t ACL_ERROR_GE_EXEC_NOT_INIT = 145001U;
+static const uint32_t ACL_ERROR_GE_EXEC_MODEL_PATH_INVALID = 145002U;
+static const uint32_t ACL_ERROR_GE_EXEC_MODEL_ID_INVALID = 145003U;
+static const uint32_t ACL_ERROR_GE_EXEC_MODEL_DATA_SIZE_INVALID = 145006U;
+static const uint32_t ACL_ERROR_GE_EXEC_MODEL_ADDR_INVALID = 145007U;
+static const uint32_t ACL_ERROR_GE_EXEC_MODEL_QUEUE_ID_INVALID = 145008U;
+static const uint32_t ACL_ERROR_GE_EXEC_LOAD_MODEL_REPEATED = 145009U;
+static const uint32_t ACL_ERROR_GE_DYNAMIC_INPUT_ADDR_INVALID = 145011U;
+static const uint32_t ACL_ERROR_GE_DYNAMIC_INPUT_LENGTH_INVALID = 145012U;
+static const uint32_t ACL_ERROR_GE_DYNAMIC_BATCH_SIZE_INVALID = 145013U;
+static const uint32_t ACL_ERROR_GE_AIPP_BATCH_EMPTY = 145014U;
+static const uint32_t ACL_ERROR_GE_AIPP_NOT_EXIST = 145015U;
+static const uint32_t ACL_ERROR_GE_AIPP_MODE_INVALID = 145016U;
+static const uint32_t ACL_ERROR_GE_OP_TASK_TYPE_INVALID = 145017U;
+static const uint32_t ACL_ERROR_GE_OP_KERNEL_TYPE_INVALID = 145018U;
+static const uint32_t ACL_ERROR_GE_PLGMGR_PATH_INVALID = 145019U;
+static const uint32_t ACL_ERROR_GE_FORMAT_INVALID = 145020U;
+static const uint32_t ACL_ERROR_GE_SHAPE_INVALID = 145021U;
+static const uint32_t ACL_ERROR_GE_DATATYPE_INVALID = 145022U;
+static const uint32_t ACL_ERROR_GE_MEMORY_ALLOCATION = 245000U;
+static const uint32_t ACL_ERROR_GE_MEMORY_OPERATE_FAILED = 245001U;
+static const uint32_t ACL_ERROR_GE_DEVICE_MEMORY_OPERATE_FAILED = 245002U;
+static const uint32_t ACL_ERROR_GE_INTERNAL_ERROR = 545000U;
+static const uint32_t ACL_ERROR_GE_LOAD_MODEL = 545001U;
+static const uint32_t ACL_ERROR_GE_EXEC_LOAD_MODEL_PARTITION_FAILED = 545002U;
+static const uint32_t ACL_ERROR_GE_EXEC_LOAD_WEIGHT_PARTITION_FAILED = 545003U;
+static const uint32_t ACL_ERROR_GE_EXEC_LOAD_TASK_PARTITION_FAILED = 545004U;
+static const uint32_t ACL_ERROR_GE_EXEC_LOAD_KERNEL_PARTITION_FAILED = 545005U;
+static const uint32_t ACL_ERROR_GE_EXEC_RELEASE_MODEL_DATA = 545006U;
+static const uint32_t ACL_ERROR_GE_COMMAND_HANDLE = 545007U;
+static const uint32_t ACL_ERROR_GE_GET_TENSOR_INFO = 545008U;
+static const uint32_t ACL_ERROR_GE_UNLOAD_MODEL = 545009U;
 
 #ifdef __cplusplus
 }  // namespace ge

inc/external/ge/ge_ir_build.h

@@ -1,18 +1,18 @@
 /**
-* Copyright 2020 Huawei Technologies Co., Ltd
-
-* Licensed under the Apache License, Version 2.0 (the "License");
-* you may not use this file except in compliance with the License.
-* You may obtain a copy of the License at
-
-* http://www.apache.org/licenses/LICENSE-2.0
-
-* Unless required by applicable law or agreed to in writing, software
-* distributed under the License is distributed on an "AS IS" BASIS,
-* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-* See the License for the specific language governing permissions and
-* limitations under the License.
-*/
+ * Copyright 2019-2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
 
 #ifndef INC_EXTERNAL_GE_IR_BUILD_H_
 #define INC_EXTERNAL_GE_IR_BUILD_H_
@@ -36,14 +36,10 @@
 #include <memory>
 #include "graph/graph.h"
 #include "graph/ge_error_codes.h"
-
-namespace {
-const int IR_MAJOR_VERSION = 1;
-const int IR_MINOR_VERSION = 0;
-const int IR_PATCH_VERSION = 0;
-}  // namespace
-
 namespace ge {
+const int32_t IR_MAJOR_VERSION = 1;
+const int32_t IR_MINOR_VERSION = 0;
+const int32_t IR_PATCH_VERSION = 0;
 
 struct ModelBufferData {
   std::shared_ptr<uint8_t> data = nullptr;
@@ -102,10 +98,10 @@ GE_FUNC_VISIBILITY graphStatus aclgrphBuildModel(const ge::Graph &graph,
  * @retval GRAPH_SUCCESS The function is successfully executed.
  * @retval OtherValues Failure
  */
-ATTRIBUTED_DEPRECATED(GE_FUNC_VISIBILITY graphStatus aclgrphSaveModel(const char *, const ModelBufferData &))
-GE_FUNC_VISIBILITY graphStatus aclgrphSaveModel(const string &output_file, const ModelBufferData &model);
+ATTRIBUTED_DEPRECATED(GE_FUNC_VISIBILITY graphStatus aclgrphSaveModel(const char_t *, const ModelBufferData &))
+GE_FUNC_VISIBILITY graphStatus aclgrphSaveModel(const std::string &output_file, const ModelBufferData &model);
 
-GE_FUNC_VISIBILITY graphStatus aclgrphSaveModel(const char *output_file, const ModelBufferData &model);
+GE_FUNC_VISIBILITY graphStatus aclgrphSaveModel(const char_t *output_file, const ModelBufferData &model);
 
 /**
  * @ingroup AscendCL
@@ -117,7 +113,8 @@ GE_FUNC_VISIBILITY graphStatus aclgrphSaveModel(const char *output_file, const M
  * @retval GRAPH_SUCCESS The function is successfully executed.
  * @retval OtherValues Failure
  */
-GE_FUNC_VISIBILITY graphStatus aclgrphGetIRVersion(int *major_version, int *minor_version, int *patch_version);
+GE_FUNC_VISIBILITY graphStatus aclgrphGetIRVersion(int32_t *major_version, int32_t *minor_version,
+                                                   int32_t *patch_version);
 
 /**
  * @ingroup AscendCL
@@ -129,7 +126,7 @@ GE_FUNC_VISIBILITY graphStatus aclgrphGetIRVersion(int *major_version, int *mino
  * @retval GRAPH_SUCCESS The function is successfully executed.
  * @retval OtherValues Failure
  */
-GE_FUNC_VISIBILITY graphStatus aclgrphDumpGraph(const ge::Graph &graph, const char *file, const size_t len);
+GE_FUNC_VISIBILITY graphStatus aclgrphDumpGraph(const ge::Graph &graph, const char_t *file, const size_t len);
 
 /**
  * @ingroup AscendCL
@@ -153,7 +150,7 @@ GE_FUNC_VISIBILITY graphStatus aclgrphGenerateForOp(const AscendString &op_type,
  * @param cfg_path   [IN] the config file path
  * @return graphStatus
  */
-GE_FUNC_VISIBILITY graphStatus aclgrphSetOpAttr(Graph &graph, aclgrphAttrType attr_type, const char *cfg_path);
+GE_FUNC_VISIBILITY graphStatus aclgrphSetOpAttr(Graph &graph, aclgrphAttrType attr_type, const char_t *cfg_path);
 
 };      // namespace ge
 #endif  // INC_EXTERNAL_GE_IR_BUILD_H_

inc/external/hccl/hccl.h

@@ -0,0 +1,210 @@
+/**
+ * Copyright 2019-2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+/**
+ * @file hccl.h
+ * @brief HCCL API
+ */
+
+#ifndef HCCL_H_
+#define HCCL_H_
+
+#include <hccl/hccl_types.h>
+#include <acl/acl.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif  // __cplusplus
+
+/**
+ * @brief Initialize HCCL.
+ *
+ * @param clusterInfo A string identifying the cluster info file path, include file name.
+ * @param rank A integer identifying the identify for the rank.
+ * @param comm A pointer identifying the initialized communication resource.
+ * @return HcclResult
+ * @see HcclCommDestroy()
+ */
+extern HcclResult HcclCommInitClusterInfo(const char *clusterInfo, uint32_t rank, HcclComm *comm);
+
+/**
+ * @brief Get hccl root info.
+ *
+ * @param rootInfo A pointer identifying the hccl root info.
+ * @return HcclResult
+ */
+extern HcclResult HcclGetRootInfo(HcclRootInfo *rootInfo);
+
+/**
+ * @brief Initialize HCCL with root info.
+ *
+ * @param nRanks A integer identifying the rank size of the cluster.
+ * @param rootInfo A struct identifying the hccl root info.
+ * @param rank A integer identifying the identify for the rank.
+ * @param comm A pointer identifying the initialized communication resource.
+ * @return HcclResult
+ * @see HcclCommDestroy()
+ */
+extern HcclResult HcclCommInitRootInfo(uint32_t nRanks, const HcclRootInfo *rootInfo, uint32_t rank, HcclComm *comm);
+
+/**
+ * @brief AllReduce operator.
+ *
+ * @param sendBuf A pointer identifying the input data address of the operator.
+ * @param recvBuf A pointer identifying the output data address of the operator.
+ * @param count An integer(u64) identifying the number of the output data.
+ * @param dataType The data type of the operator, must be one of the following types: int8, int16, int32, float16,
+ * float32.
+ * @param op The reduction type of the operator, must be one of the following types: sum, min, max, prod.
+ * @param comm A pointer identifying the communication resource based on.
+ * @param stream A pointer identifying the stream information.
+ * @return HcclResult
+ */
+extern HcclResult HcclAllReduce(void *sendBuf, void *recvBuf, uint64_t count, HcclDataType dataType, HcclReduceOp op,
+                                HcclComm comm, aclrtStream stream);
+
+/**
+ * @brief Broadcast operator.
+ *
+ * @param buf A pointer identifying the data address of the operator.
+ * @param count An integer(u64) identifying the number of the data.
+ * @param dataType The data type of the operator, must be one of the following types: int8, int32, float16, float32.
+ * @param root An integer(u32) identifying the the root rank in the operator.
+ * @param comm A pointer identifying the communication resource based on
+ * @param stream A pointer identifying the stream information.
+ * @return HcclResult
+ */
+extern HcclResult HcclBroadcast(void *buf, uint64_t count, HcclDataType dataType, uint32_t root, HcclComm comm,
+                                aclrtStream stream);
+
+/**
+ * @brief ReduceScatter operator.
+ *
+ * @param sendBuf A pointer identifying the input data address of the operator.
+ * @param recvBuf A pointer identifying the output data address of the operator.
+ * @param recvCount An integer(u64) identifying the number of the output data.
+ * @param dataType The data type of the operator, must be one of the following types: int8, int32, float16, float32.
+ * @param op The reduction type of the operator, must be one of the following types: sum, min, max, prod.
+ * @param comm A pointer identifying the communication resource based on.
+ * @param stream A pointer identifying the stream information.
+ * @return HcclResult
+ */
+extern HcclResult HcclReduceScatter(void *sendBuf, void *recvBuf, uint64_t recvCount, HcclDataType dataType,
+                                    HcclReduceOp op, HcclComm comm, aclrtStream stream);
+
+/**
+ * @brief AllGather operator.
+ *
+ * @param sendBuf A pointer identifying the input data address of the operator.
+ * @param recvBuf A pointer identifying the output data address of the operator.
+ * @param sendCount An integer(u64) identifying the number of the input data.
+ * @param dataType The data type of the operator, must be one of the following types: int8, int32, float16, float32.
+ * @param comm A pointer identifying the communication resource based on.
+ * @param stream A pointer identifying the stream information.
+ * @return HcclResult
+ */
+extern HcclResult HcclAllGather(void *sendBuf, void *recvBuf, uint64_t sendCount, HcclDataType dataType, HcclComm comm,
+                                aclrtStream stream);
+/**
+ * @brief Get the rank size of this comm.
+ *
+ * @param comm A pointer identifying the communication resource based on.
+ * @param rankSize  A pointer identifying the rank size.
+ * @return HcclResult
+ */
+extern HcclResult HcclGetRankSize(HcclComm comm, uint32_t *rankSize);
+
+/**
+ * @brief Get the rank id of this comm.
+ *
+ * @param comm A pointer identifying the communication resource based on.
+ * @param rankSize  A pointer identifying the rank id.
+ * @return HcclResult
+ */
+extern HcclResult HcclGetRankId(HcclComm comm, uint32_t *rank);
+/**
+ * @brief Barrier operator.
+ *
+ * @param comm A pointer identifying the communication resource based on.
+ * @param stream A pointer identifying the stream information.
+ * @return HcclResult
+ */
+extern HcclResult HcclBarrier(HcclComm comm, aclrtStream stream);
+
+/**
+ * @brief Send operator.
+ *
+ * @param sendBuff A pointer identifying the input data address of the operator.
+ * @param count An integer(u64) identifying the number of the send data.
+ * @param dataType The data type of the operator, must be one of the following types: int8, int32, float16, float32.
+ * @param destRank An integer identifying the destination rank.
+ * @param comm A pointer identifying the communication resource based on.
+ * @param stream A pointer identifying the stream information.
+ * @return HcclResult
+ */
+extern HcclResult HcclSend(void *sendBuf, uint64_t count, HcclDataType dataType, uint32_t destRank, HcclComm comm,
+                           aclrtStream stream);
+/**
+ * @brief Recv operator.
+ *
+ * @param recvBuff A pointer identifying the output data address of the operator.
+ * @param count An integer(u64) identifying the number of the receive data.
+ * @param dataType The data type of the operator, must be one of the following types: int8, int32, float16, float32.
+ * @param srcRank An integer identifying the source rank.
+ * @param comm A pointer identifying the communication resource based on.
+ * @param stream A pointer identifying the stream information.
+ * @return HcclResult
+ */
+extern HcclResult HcclRecv(void *recvBuf, uint64_t count, HcclDataType dataType, uint32_t srcRank, HcclComm comm,
+                           aclrtStream stream);
+
+/**
+ * @brief AlltoAllV operator.
+ *
+ * @param sendBuff A pointer identifying the input data address of the operator.
+ * @param sendCounts Integer array, where entry i specifies the number of elements to send to rank i.
+ * @param sdispls Integer array, where entry i specifies the displacement (offset from sendbuf, in units of sendtype)
+ * from which to send data to rank i.
+ * @param sendType Datatype of send buffer elements, must be one of the following types: int8, int32, int64, uint64,
+ * float16, float32.
+ * @param recvBuf A pointer identifying the output data address of the operator.
+ * @param recvCounts Integer array, where entry j specifies the number of elements to receive from rank j.
+ * @param rdispls Integer array, where entry j specifies the displacement (offset from recvbuf, in units of recvtype) to
+ * which data from rank j should be written.
+ * @param recvType Datatype of receive buffer elements, must be one of the following types: int8, int32, int64, uint64,
+ * float16, float32.
+ * @param comm A pointer identifying the communication resource based on.
+ * @param stream A pointer identifying the stream information.
+ * @return HcclResult
+ */
+
+extern HcclResult HcclAlltoAllV(const void *sendBuf, const void *sendCounts, const void *sdispls, HcclDataType sendType,
+                                const void *recvBuf, const void *recvCounts, const void *rdispls, HcclDataType recvType,
+                                HcclComm comm, aclrtStream stream);
+
+/**
+ * @brief Destroy HCCL comm
+ *
+ * @param comm A pointer identifying the communication resource targetting
+ * @return HcclResult
+ * @see HcclCommInitClusterInfo()
+ */
+extern HcclResult HcclCommDestroy(HcclComm comm);
+
+#ifdef __cplusplus
+}
+#endif  // __cplusplus
+#endif  // HCCL_H_

inc/external/hccl/hccl_types.h

@@ -0,0 +1,101 @@
+/**
+ * Copyright 2019-2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+/**
+ * @file hccl_types.h
+ * @brief HCCL data type definition
+ *
+ */
+
+#ifndef HCCL_TYPES_H_
+#define HCCL_TYPES_H_
+
+#include <stdint.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif  // __cplusplus
+
+/**
+ * @brief HCCL functions return value definition
+ */
+typedef enum {
+  HCCL_SUCCESS = 0,              /**< success */
+  HCCL_E_PARA = 1,               /**< parameter error */
+  HCCL_E_PTR = 2,                /**< empty pointer */
+  HCCL_E_MEMORY = 3,             /**< memory error */
+  HCCL_E_INTERNAL = 4,           /**< internal error */
+  HCCL_E_NOT_SUPPORT = 5,        /**< not support feature */
+  HCCL_E_NOT_FOUND = 6,          /**< not found specific resource */
+  HCCL_E_UNAVAIL = 7,            /**< resource unavailable */
+  HCCL_E_SYSCALL = 8,            /**< call system interface error */
+  HCCL_E_TIMEOUT = 9,            /**< timeout */
+  HCCL_E_OPEN_FILE_FAILURE = 10, /**< open file fail */
+  HCCL_E_TCP_CONNECT = 11,       /**< tcp connect fail */
+  HCCL_E_ROCE_CONNECT = 12,      /**< roce connect fail */
+  HCCL_E_TCP_TRANSFER = 13,      /**< tcp transfer fail */
+  HCCL_E_ROCE_TRANSFER = 14,     /**< roce transfer fail */
+  HCCL_E_RUNTIME = 15,           /**< call runtime api fail */
+  HCCL_E_DRV = 16,               /**< call driver api fail */
+  HCCL_E_PROFILING = 17,         /**< call profiling api fail */
+  HCCL_E_CCE = 18,               /**< call cce api fail */
+  HCCL_E_NETWORK = 19,           /**< call network api fail */
+  HCCL_E_AGAIN = 20,             /**< try again */
+  HCCL_E_RESERVED                /**< reserved */
+} HcclResult;
+
+/**
+ * @brief handle to HCCL communicator
+ */
+typedef void *HcclComm;
+
+/**
+ * @brief HCCL Reduction opperation
+ */
+typedef enum {
+  HCCL_REDUCE_SUM = 0,  /**< sum */
+  HCCL_REDUCE_PROD = 1, /**< prod */
+  HCCL_REDUCE_MAX = 2,  /**< max */
+  HCCL_REDUCE_MIN = 3,  /**< min */
+  HCCL_REDUCE_RESERVED  /**< reserved */
+} HcclReduceOp;
+
+/**
+ * @brief HCCL data type
+ */
+typedef enum {
+  HCCL_DATA_TYPE_INT8 = 0,   /**< int8 */
+  HCCL_DATA_TYPE_INT16 = 1,  /**< int16 */
+  HCCL_DATA_TYPE_INT32 = 2,  /**< int32 */
+  HCCL_DATA_TYPE_FP16 = 3,   /**< fp16 */
+  HCCL_DATA_TYPE_FP32 = 4,   /**< fp32 */
+  HCCL_DATA_TYPE_INT64 = 5,  /**< int64 */
+  HCCL_DATA_TYPE_UINT64 = 6, /**< uint64 */
+  HCCL_DATA_TYPE_RESERVED    /**< reserved */
+} HcclDataType;
+
+const uint32_t HCCL_ROOT_INFO_BYTES = 4108;  // 4108: root info length
+/**
+ * @brief HCCL root info
+ */
+typedef struct HcclRootInfoDef {
+  char internal[HCCL_ROOT_INFO_BYTES];
+} HcclRootInfo;
+
+#ifdef __cplusplus
+}
+#endif  // __cplusplus
+#endif  // HCCL_TYPES_H_

inc/external/runtime/rt_error_codes.h

@@ -0,0 +1,123 @@
+/**
+ * Copyright 2019-2022 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef __INC_EXTERNEL_RT_ERROR_CODES_H__
+#define __INC_EXTERNEL_RT_ERROR_CODES_H__
+
+#include <stddef.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+static const int32_t ACL_RT_SUCCESS = 0;  // success
+
+static const int32_t ACL_ERROR_RT_PARAM_INVALID = 107000;             // param invalid
+static const int32_t ACL_ERROR_RT_INVALID_DEVICEID = 107001;          // invalid device id
+static const int32_t ACL_ERROR_RT_CONTEXT_NULL = 107002;              // current context null
+static const int32_t ACL_ERROR_RT_STREAM_CONTEXT = 107003;            // stream not in current context
+static const int32_t ACL_ERROR_RT_MODEL_CONTEXT = 107004;             // model not in current context
+static const int32_t ACL_ERROR_RT_STREAM_MODEL = 107005;              // stream not in model
+static const int32_t ACL_ERROR_RT_EVENT_TIMESTAMP_INVALID = 107006;   // event timestamp invalid
+static const int32_t ACL_ERROR_RT_EVENT_TIMESTAMP_REVERSAL = 107007;  // event timestamp reversal
+static const int32_t ACL_ERROR_RT_ADDR_UNALIGNED = 107008;            // memory address unaligned
+static const int32_t ACL_ERROR_RT_FILE_OPEN = 107009;                 // open file failed
+static const int32_t ACL_ERROR_RT_FILE_WRITE = 107010;                // write file failed
+static const int32_t ACL_ERROR_RT_STREAM_SUBSCRIBE = 107011;          // error subscribe stream
+static const int32_t ACL_ERROR_RT_THREAD_SUBSCRIBE = 107012;          // error subscribe thread
+static const int32_t ACL_ERROR_RT_GROUP_NOT_SET = 107013;             // group not set
+static const int32_t ACL_ERROR_RT_GROUP_NOT_CREATE = 107014;          // group not create
+static const int32_t ACL_ERROR_RT_STREAM_NO_CB_REG = 107015;          // callback not register to stream
+static const int32_t ACL_ERROR_RT_INVALID_MEMORY_TYPE = 107016;       // invalid memory type
+static const int32_t ACL_ERROR_RT_INVALID_HANDLE = 107017;            // invalid handle
+static const int32_t ACL_ERROR_RT_INVALID_MALLOC_TYPE = 107018;       // invalid malloc type
+static const int32_t ACL_ERROR_RT_WAIT_TIMEOUT = 107019;              // wait timeout
+static const int32_t ACL_ERROR_RT_TASK_TIMEOUT = 107020;              // task timeout
+
+static const int32_t ACL_ERROR_RT_FEATURE_NOT_SUPPORT = 207000;  // feature not support
+static const int32_t ACL_ERROR_RT_MEMORY_ALLOCATION = 207001;    // memory allocation error
+static const int32_t ACL_ERROR_RT_MEMORY_FREE = 207002;          // memory free error
+static const int32_t ACL_ERROR_RT_AICORE_OVER_FLOW = 207003;     // aicore over flow
+static const int32_t ACL_ERROR_RT_NO_DEVICE = 207004;            // no device
+static const int32_t ACL_ERROR_RT_RESOURCE_ALLOC_FAIL = 207005;  // resource alloc fail
+static const int32_t ACL_ERROR_RT_NO_PERMISSION = 207006;        // no permission
+static const int32_t ACL_ERROR_RT_NO_EVENT_RESOURCE = 207007;    // no event resource
+static const int32_t ACL_ERROR_RT_NO_STREAM_RESOURCE = 207008;   // no stream resource
+static const int32_t ACL_ERROR_RT_NO_NOTIFY_RESOURCE = 207009;   // no notify resource
+static const int32_t ACL_ERROR_RT_NO_MODEL_RESOURCE = 207010;    // no model resource
+static const int32_t ACL_ERROR_RT_NO_CDQ_RESOURCE = 207011;      // no cdq resource
+static const int32_t ACL_ERROR_RT_OVER_LIMIT = 207012;           // over limit
+static const int32_t ACL_ERROR_RT_QUEUE_EMPTY = 207013;          // queue is empty
+static const int32_t ACL_ERROR_RT_QUEUE_FULL = 207014;           // queue is full
+static const int32_t ACL_ERROR_RT_REPEATED_INIT = 207015;        // repeated init
+static const int32_t ACL_ERROR_RT_AIVEC_OVER_FLOW = 207016;      // aivec over flow
+static const int32_t ACL_ERROR_RT_OVER_FLOW = 207017;            // common over flow
+
+static const int32_t ACL_ERROR_RT_INTERNAL_ERROR = 507000;                   // runtime internal error
+static const int32_t ACL_ERROR_RT_TS_ERROR = 507001;                         // ts internel error
+static const int32_t ACL_ERROR_RT_STREAM_TASK_FULL = 507002;                 // task full in stream
+static const int32_t ACL_ERROR_RT_STREAM_TASK_EMPTY = 507003;                // task empty in stream
+static const int32_t ACL_ERROR_RT_STREAM_NOT_COMPLETE = 507004;              // stream not complete
+static const int32_t ACL_ERROR_RT_END_OF_SEQUENCE = 507005;                  // end of sequence
+static const int32_t ACL_ERROR_RT_EVENT_NOT_COMPLETE = 507006;               // event not complete
+static const int32_t ACL_ERROR_RT_CONTEXT_RELEASE_ERROR = 507007;            // context release error
+static const int32_t ACL_ERROR_RT_SOC_VERSION = 507008;                      // soc version error
+static const int32_t ACL_ERROR_RT_TASK_TYPE_NOT_SUPPORT = 507009;            // task type not support
+static const int32_t ACL_ERROR_RT_LOST_HEARTBEAT = 507010;                   // ts lost heartbeat
+static const int32_t ACL_ERROR_RT_MODEL_EXECUTE = 507011;                    // model execute failed
+static const int32_t ACL_ERROR_RT_REPORT_TIMEOUT = 507012;                   // report timeout
+static const int32_t ACL_ERROR_RT_SYS_DMA = 507013;                          // sys dma error
+static const int32_t ACL_ERROR_RT_AICORE_TIMEOUT = 507014;                   // aicore timeout
+static const int32_t ACL_ERROR_RT_AICORE_EXCEPTION = 507015;                 // aicore exception
+static const int32_t ACL_ERROR_RT_AICORE_TRAP_EXCEPTION = 507016;            // aicore trap exception
+static const int32_t ACL_ERROR_RT_AICPU_TIMEOUT = 507017;                    // aicpu timeout
+static const int32_t ACL_ERROR_RT_AICPU_EXCEPTION = 507018;                  // aicpu exception
+static const int32_t ACL_ERROR_RT_AICPU_DATADUMP_RSP_ERR = 507019;           // aicpu datadump response error
+static const int32_t ACL_ERROR_RT_AICPU_MODEL_RSP_ERR = 507020;              // aicpu model operate response error
+static const int32_t ACL_ERROR_RT_PROFILING_ERROR = 507021;                  // profiling error
+static const int32_t ACL_ERROR_RT_IPC_ERROR = 507022;                        // ipc error
+static const int32_t ACL_ERROR_RT_MODEL_ABORT_NORMAL = 507023;               // model abort normal
+static const int32_t ACL_ERROR_RT_KERNEL_UNREGISTERING = 507024;             // kernel unregistering
+static const int32_t ACL_ERROR_RT_RINGBUFFER_NOT_INIT = 507025;              // ringbuffer not init
+static const int32_t ACL_ERROR_RT_RINGBUFFER_NO_DATA = 507026;               // ringbuffer no data
+static const int32_t ACL_ERROR_RT_KERNEL_LOOKUP = 507027;                    // kernel lookup error
+static const int32_t ACL_ERROR_RT_KERNEL_DUPLICATE = 507028;                 // kernel register duplicate
+static const int32_t ACL_ERROR_RT_DEBUG_REGISTER_FAIL = 507029;              // debug register failed
+static const int32_t ACL_ERROR_RT_DEBUG_UNREGISTER_FAIL = 507030;            // debug unregister failed
+static const int32_t ACL_ERROR_RT_LABEL_CONTEXT = 507031;                    // label not in current context
+static const int32_t ACL_ERROR_RT_PROGRAM_USE_OUT = 507032;                  // program register num use out
+static const int32_t ACL_ERROR_RT_DEV_SETUP_ERROR = 507033;                  // device setup error
+static const int32_t ACL_ERROR_RT_VECTOR_CORE_TIMEOUT = 507034;              // vector core timeout
+static const int32_t ACL_ERROR_RT_VECTOR_CORE_EXCEPTION = 507035;            // vector core exception
+static const int32_t ACL_ERROR_RT_VECTOR_CORE_TRAP_EXCEPTION = 507036;       // vector core trap exception
+static const int32_t ACL_ERROR_RT_CDQ_BATCH_ABNORMAL = 507037;               // cdq alloc batch abnormal
+static const int32_t ACL_ERROR_RT_DIE_MODE_CHANGE_ERROR = 507038;            // can not change die mode
+static const int32_t ACL_ERROR_RT_DIE_SET_ERROR = 507039;                    // single die mode can not set die
+static const int32_t ACL_ERROR_RT_INVALID_DIEID = 507040;                    // invalid die id
+static const int32_t ACL_ERROR_RT_DIE_MODE_NOT_SET = 507041;                 // die mode not set
+static const int32_t ACL_ERROR_RT_AICORE_TRAP_READ_OVERFLOW = 507042;        // aic trap read overflow
+static const int32_t ACL_ERROR_RT_AICORE_TRAP_WRITE_OVERFLOW = 507043;       // aic trap write overflow
+static const int32_t ACL_ERROR_RT_VECTOR_CORE_TRAP_READ_OVERFLOW = 507044;   // aiv trap read overflow
+static const int32_t ACL_ERROR_RT_VECTOR_CORE_TRAP_WRITE_OVERFLOW = 507045;  // aiv trap write overflow
+
+static const int32_t ACL_ERROR_RT_DRV_INTERNAL_ERROR = 507899;    // drv internal error
+static const int32_t ACL_ERROR_RT_AICPU_INTERNAL_ERROR = 507900;  // aicpu internal error
+static const int32_t ACL_ERROR_RT_SOCKET_CLOSE = 507901;          // hdc disconnect
+
+#ifdef __cplusplus
+}
+#endif
+#endif  // __INC_EXTERNEL_RT_ERROR_CODES_H__

inc/framework/common/aicpu_op.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.

inc/framework/common/debug/ge_log.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -17,6 +17,7 @@
 #ifndef INC_FRAMEWORK_COMMON_DEBUG_GE_LOG_H_
 #define INC_FRAMEWORK_COMMON_DEBUG_GE_LOG_H_
 
+#include <cinttypes>
 #include <cstdint>
 
 #include "framework/common/ge_inner_error_codes.h"
@@ -33,7 +34,7 @@
 extern "C" {
 #endif
 
-#define GE_MODULE_NAME static_cast<int>(GE)
+#define GE_MODULE_NAME static_cast<int32_t>(GE)
 
 // trace status of log
 enum TraceStatus { TRACE_INIT = 0, TRACE_RUNNING, TRACE_WAITING, TRACE_STOP };
@@ -42,60 +43,79 @@ class GE_FUNC_VISIBILITY GeLog {
  public:
   static uint64_t GetTid() {
 #ifdef __GNUC__
-    uint64_t tid = static_cast<uint64_t>(syscall(__NR_gettid));
+    const uint64_t tid = static_cast<uint64_t>(syscall(__NR_gettid));
 #else
-    uint64_t tid = static_cast<uint64_t>(GetCurrentThreadId());
+    const uint64_t tid = static_cast<uint64_t>(GetCurrentThreadId());
 #endif
     return tid;
   }
 };
 
-inline bool IsLogEnable(int module_name, int log_level) {
-  int32_t enable = CheckLogLevel(module_name, log_level);
+inline bool IsLogEnable(const int32_t module_name, const int32_t log_level) {
+  const int32_t enable = CheckLogLevel(module_name, log_level);
   // 1:enable, 0:disable
   return (enable == 1);
 }
 
-#define GELOGE(ERROR_CODE, fmt, ...)                                                                         \
-  dlog_error(GE_MODULE_NAME, "%lu %s: ErrorNo: %d(%s) %s" fmt, GeLog::GetTid(), __FUNCTION__, ERROR_CODE,    \
-             ((GE_GET_ERRORNO_STR(ERROR_CODE)).c_str()), ErrorManager::GetInstance().GetLogHeader().c_str(), \
-             ##__VA_ARGS__)
-#define GELOGW(fmt, ...)                      \
-  if (IsLogEnable(GE_MODULE_NAME, DLOG_WARN)) \
-  dlog_warn(GE_MODULE_NAME, "%lu %s:" fmt, GeLog::GetTid(), __FUNCTION__, ##__VA_ARGS__)
-#define GELOGI(fmt, ...)                      \
-  if (IsLogEnable(GE_MODULE_NAME, DLOG_INFO)) \
-  dlog_info(GE_MODULE_NAME, "%lu %s:" fmt, GeLog::GetTid(), __FUNCTION__, ##__VA_ARGS__)
-#define GELOGD(fmt, ...)                       \
-  if (IsLogEnable(GE_MODULE_NAME, DLOG_DEBUG)) \
-  dlog_debug(GE_MODULE_NAME, "%lu %s:" fmt, GeLog::GetTid(), __FUNCTION__, ##__VA_ARGS__)
-
-#define GEEVENT(fmt, ...) dlog_event(GE_MODULE_NAME, "%lu %s:" fmt, GeLog::GetTid(), __FUNCTION__, ##__VA_ARGS__)
-
-#define GELOGT(VALUE, fmt, ...)                                                                                    \
-  do {                                                                                                             \
-    TraceStatus stat = VALUE;                                                                                      \
-    const char *const TraceStatStr[] = {"INIT", "RUNNING", "WAITING", "STOP"};                                     \
-    int idx = static_cast<int>(stat);                                                                              \
-    char *k = const_cast<char *>("status");                                                                        \
-    char *v = const_cast<char *>(TraceStatStr[idx]);                                                               \
-    KeyValue kv = {k, v};                                                                                          \
-    DlogWithKV(static_cast<int>(GE_MODULE_NAME), DLOG_TRACE, &kv, 1, "%lu %s:" fmt, GeLog::GetTid(), __FUNCTION__, \
-               ##__VA_ARGS__);                                                                                     \
-  } while (0)
-
-#define GE_LOG_ERROR(MOD_NAME, ERROR_CODE, fmt, ...)                                                         \
-  dlog_error(MOD_NAME, "%lu %s: ErrorNo: %d(%s) %s" fmt, GeLog::GetTid(), __FUNCTION__, ERROR_CODE,          \
-             ((GE_GET_ERRORNO_STR(ERROR_CODE)).c_str()), ErrorManager::GetInstance().GetLogHeader().c_str(), \
-             ##__VA_ARGS__)
+#define GELOGE(ERROR_CODE, fmt, ...)                                                                   \
+  do {                                                                                                 \
+    dlog_error(GE_MODULE_NAME, "%" PRIu64 " %s: ErrorNo: %" PRIuLEAST8 "(%s) %s" fmt, GeLog::GetTid(), \
+               &__FUNCTION__[0], (ERROR_CODE), ((GE_GET_ERRORNO_STR(ERROR_CODE)).c_str()),             \
+               ErrorManager::GetInstance().GetLogHeader().c_str(), ##__VA_ARGS__);                     \
+  } while (false)
+
+#define GELOGW(fmt, ...)                                                                                  \
+  do {                                                                                                    \
+    if (IsLogEnable(GE_MODULE_NAME, DLOG_WARN)) {                                                         \
+      dlog_warn(GE_MODULE_NAME, "%" PRIu64 " %s:" fmt, GeLog::GetTid(), &__FUNCTION__[0], ##__VA_ARGS__); \
+    }                                                                                                     \
+  } while (false)
+
+#define GELOGI(fmt, ...)                                                                                  \
+  do {                                                                                                    \
+    if (IsLogEnable(GE_MODULE_NAME, DLOG_INFO)) {                                                         \
+      dlog_info(GE_MODULE_NAME, "%" PRIu64 " %s:" fmt, GeLog::GetTid(), &__FUNCTION__[0], ##__VA_ARGS__); \
+    }                                                                                                     \
+  } while (false)
+
+#define GELOGD(fmt, ...)                                                                                   \
+  do {                                                                                                     \
+    if (IsLogEnable(GE_MODULE_NAME, DLOG_DEBUG)) {                                                         \
+      dlog_debug(GE_MODULE_NAME, "%" PRIu64 " %s:" fmt, GeLog::GetTid(), &__FUNCTION__[0], ##__VA_ARGS__); \
+    }                                                                                                      \
+  } while (false)
+
+#define GEEVENT(fmt, ...)                                                                                \
+  do {                                                                                                   \
+    dlog_event(GE_MODULE_NAME, "%" PRIu64 " %s:" fmt, GeLog::GetTid(), &__FUNCTION__[0], ##__VA_ARGS__); \
+  } while (false)
+
+#define GELOGT(VALUE, fmt, ...)                                                                              \
+  do {                                                                                                       \
+    TraceStatus stat = (VALUE);                                                                              \
+    const char_t *const TraceStatStr[] = {"INIT", "RUNNING", "WAITING", "STOP"};                             \
+    const int32_t idx = static_cast<int32_t>(stat);                                                          \
+    char_t *k = const_cast<char_t *>("status");                                                              \
+    char_t *v = const_cast<char_t *>(TraceStatStr[idx]);                                                     \
+    KeyValue kv = {k, v};                                                                                    \
+    DlogWithKV(GE_MODULE_NAME, DLOG_TRACE, &kv, 1, "%" PRIu64 " %s:" fmt, GeLog::GetTid(), &__FUNCTION__[0], \
+               ##__VA_ARGS__);                                                                               \
+  } while (false)
+
+#define GE_LOG_ERROR(MOD_NAME, ERROR_CODE, fmt, ...)                                                                 \
+  do {                                                                                                               \
+    dlog_error((MOD_NAME), "%" PRIu64 " %s: ErrorNo: %" PRIuLEAST8 "(%s) %s" fmt, GeLog::GetTid(), &__FUNCTION__[0], \
+               (ERROR_CODE), ((GE_GET_ERRORNO_STR(ERROR_CODE)).c_str()),                                             \
+               ErrorManager::GetInstance().GetLogHeader().c_str(), ##__VA_ARGS__);                                   \
+  } while (false)
 
 // print memory when it is greater than 1KB.
-#define GE_PRINT_DYNAMIC_MEMORY(FUNC, PURPOSE, SIZE)                                                        \
-  do {                                                                                                      \
-    if ((SIZE) > 1024) {                                                                                    \
-      GELOGI("MallocMemory, func=%s, size=%zu, purpose=%s", (#FUNC), static_cast<size_t>(SIZE), (PURPOSE)); \
-    }                                                                                                       \
-  } while (0);
+#define GE_PRINT_DYNAMIC_MEMORY(FUNC, PURPOSE, SIZE)                                                                \
+  do {                                                                                                              \
+    if (static_cast<size_t>(SIZE) > 1024UL) {                                                                       \
+      GELOGI("MallocMemory, func=%s, size=%" PRIu64 ", purpose=%s", (#FUNC), static_cast<size_t>(SIZE), (PURPOSE)); \
+    }                                                                                                               \
+  } while (false)
 #ifdef __cplusplus
 }
 #endif

inc/framework/common/debug/log.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -19,17 +19,14 @@
 
 #include <string>
 #include <sstream>
-#include <securec.h>
 
-#include "runtime/rt.h"
-#include "common/string_util.h"
-#include "common/util.h"
-#include "common/util/error_manager/error_manager.h"
+#include "framework/common/string_util.h"
+#include "framework/common/util.h"
 #include "framework/common/debug/ge_log.h"
-#include "ge/ge_api_error_codes.h"
+#include "external/ge/ge_api_error_codes.h"
 
 #if !defined(__ANDROID__) && !defined(ANDROID)
-#define DOMI_LOGE(fmt, ...) GE_LOG_ERROR(GE_MODULE_NAME, ge::FAILED, fmt, ##__VA_ARGS__)
+#define DOMI_LOGE(fmt, ...) GE_LOG_ERROR(GE_MODULE_NAME, (ge::FAILED), fmt, ##__VA_ARGS__)
 #else
 #include <android/log.h>
 #if defined(BUILD_VERSION_PERF)
@@ -52,196 +49,116 @@
     GELOGW(__VA_ARGS__);           \
   }
 
-#define GE_LOGE_IF(condition, ...) \
-  if ((condition)) {               \
-    DOMI_LOGE(__VA_ARGS__);        \
+#define GE_LOGE_IF(condition, ...)     \
+  if ((condition)) {                   \
+    GELOGE((ge::FAILED), __VA_ARGS__); \
   }
 
 // If expr is not SUCCESS, print the log and return the same value
-#define GE_CHK_STATUS_RET(expr, ...)   \
-  do {                                 \
-    const ge::Status _status = (expr); \
-    if (_status != ge::SUCCESS) {      \
-      DOMI_LOGE(__VA_ARGS__);          \
-      return _status;                  \
-    }                                  \
-  } while (0);
+#define GE_CHK_STATUS_RET(expr, ...)       \
+  do {                                     \
+    const ge::Status _chk_status = (expr); \
+    if (_chk_status != ge::SUCCESS) {      \
+      GELOGE((ge::FAILED), __VA_ARGS__);   \
+      return _chk_status;                  \
+    }                                      \
+  } while (false)
 
 // If expr is not SUCCESS, print the log and do not execute return
-#define GE_CHK_STATUS(expr, ...)       \
-  do {                                 \
-    const ge::Status _status = (expr); \
-    if (_status != ge::SUCCESS) {      \
-      DOMI_LOGE(__VA_ARGS__);          \
-    }                                  \
-  } while (0);
+#define GE_CHK_STATUS(expr, ...)           \
+  do {                                     \
+    const ge::Status _chk_status = (expr); \
+    if (_chk_status != ge::SUCCESS) {      \
+      GELOGE(_chk_status, __VA_ARGS__);    \
+    }                                      \
+  } while (false)
 
 // If expr is not SUCCESS, return the same value
-#define GE_CHK_STATUS_RET_NOLOG(expr)  \
-  do {                                 \
-    const ge::Status _status = (expr); \
-    if (_status != ge::SUCCESS) {      \
-      return _status;                  \
-    }                                  \
-  } while (0);
+#define GE_CHK_STATUS_RET_NOLOG(expr)      \
+  do {                                     \
+    const ge::Status _chk_status = (expr); \
+    if (_chk_status != ge::SUCCESS) {      \
+      return _chk_status;                  \
+    }                                      \
+  } while (false)
 
 // If expr is not GRAPH_SUCCESS, print the log and return FAILED
 #define GE_CHK_GRAPH_STATUS_RET(expr, ...)                  \
   do {                                                      \
     if ((expr) != ge::GRAPH_SUCCESS) {                      \
       REPORT_CALL_ERROR("E19999", "Operator graph failed"); \
-      DOMI_LOGE(__VA_ARGS__);                               \
-      return FAILED;                                        \
+      GELOGE(ge::FAILED, __VA_ARGS__);                      \
+      return (ge::FAILED);                                  \
     }                                                       \
-  } while (0);
+  } while (false)
 
 // If expr is not SUCCESS, print the log and execute a custom statement
-#define GE_CHK_STATUS_EXEC(expr, exec_expr, ...)                  \
-  do {                                                            \
-    const ge::Status _status = (expr);                            \
-    GE_CHK_BOOL_EXEC(_status == SUCCESS, exec_expr, __VA_ARGS__); \
-  } while (0);
+#define GE_CHK_STATUS_EXEC(expr, exec_expr, ...)                      \
+  do {                                                                \
+    const ge::Status _chk_status = (expr);                            \
+    GE_CHK_BOOL_EXEC(_chk_status == SUCCESS, exec_expr, __VA_ARGS__); \
+  } while (false)
 
 // If expr is not true, print the log and return the specified status
 #define GE_CHK_BOOL_RET_STATUS(expr, _status, ...) \
   do {                                             \
-    bool b = (expr);                               \
+    const bool b = (expr);                         \
     if (!b) {                                      \
       REPORT_INNER_ERROR("E19999", __VA_ARGS__);   \
-      GELOGE(_status, __VA_ARGS__);                \
-      return _status;                              \
+      GELOGE((_status), __VA_ARGS__);              \
+      return (_status);                            \
     }                                              \
-  } while (0);
+  } while (false)
 
 // If expr is not true, print the log and return the specified status
 #define GE_CHK_BOOL_RET_STATUS_NOLOG(expr, _status, ...) \
   do {                                                   \
-    bool b = (expr);                                     \
+    const bool b = (expr);                               \
     if (!b) {                                            \
-      return _status;                                    \
+      return (_status);                                  \
     }                                                    \
-  } while (0);
+  } while (false)
 
 // If expr is not true, print the log and execute a custom statement
 #define GE_CHK_BOOL_EXEC(expr, exec_expr, ...) \
   {                                            \
-    bool b = (expr);                           \
+    const bool b = (expr);                     \
     if (!b) {                                  \
-      DOMI_LOGE(__VA_ARGS__);                  \
+      GELOGE(ge::FAILED, __VA_ARGS__);         \
       exec_expr;                               \
     }                                          \
   }
 
-// If expr is not true, print the log and execute a custom statement
-#define GE_CHK_BOOL_EXEC_WARN(expr, exec_expr, ...) \
-  {                                                 \
-    bool b = (expr);                                \
-    if (!b) {                                       \
-      GELOGW(__VA_ARGS__);                          \
-      exec_expr;                                    \
-    }                                               \
-  }
-// If expr is not true, print the log and execute a custom statement
-#define GE_CHK_BOOL_EXEC_INFO(expr, exec_expr, ...) \
-  {                                                 \
-    bool b = (expr);                                \
-    if (!b) {                                       \
-      GELOGI(__VA_ARGS__);                          \
-      exec_expr;                                    \
-    }                                               \
-  }
-
-// If expr is not true, print the log and execute a custom statement
-#define GE_CHK_BOOL_TRUE_EXEC_INFO(expr, exec_expr, ...) \
-  {                                                      \
-    bool b = (expr);                                     \
-    if (b) {                                             \
-      GELOGI(__VA_ARGS__);                               \
-      exec_expr;                                         \
-    }                                                    \
-  }
-
-// If expr is true, print logs and execute custom statements
-#define GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(expr, exec_expr, ...) \
-  {                                                          \
-    bool b = (expr);                                         \
-    if (b) {                                                 \
-      DOMI_LOGE(__VA_ARGS__);                                \
-      exec_expr;                                             \
-    }                                                        \
-  }
-// If expr is true, print the Information log and execute a custom statement
-#define GE_CHK_TRUE_EXEC_INFO(expr, exec_expr, ...) \
-  {                                                 \
-    bool b = (expr);                                \
-    if (b) {                                        \
-      GELOGI(__VA_ARGS__);                          \
-      exec_expr;                                    \
-    }                                               \
-  }
-
-// If expr is not SUCCESS, print the log and execute the expression + return
-#define GE_CHK_BOOL_TRUE_RET_VOID(expr, exec_expr, ...) \
-  {                                                     \
-    bool b = (expr);                                    \
-    if (b) {                                            \
-      DOMI_LOGE(__VA_ARGS__);                           \
-      exec_expr;                                        \
-      return;                                           \
-    }                                                   \
-  }
-
-// If expr is not SUCCESS, print the log and execute the expression + return _status
-#define GE_CHK_BOOL_TRUE_EXEC_RET_STATUS(expr, _status, exec_expr, ...) \
-  {                                                                     \
-    bool b = (expr);                                                    \
-    if (b) {                                                            \
-      REPORT_INNER_ERROR("E19999", __VA_ARGS__);                        \
-      DOMI_LOGE(__VA_ARGS__);                                           \
-      exec_expr;                                                        \
-      return _status;                                                   \
-    }                                                                   \
-  }
-
-// If expr is not true, execute a custom statement
-#define GE_CHK_BOOL_EXEC_NOLOG(expr, exec_expr) \
-  {                                             \
-    bool b = (expr);                            \
-    if (!b) {                                   \
-      exec_expr;                                \
-    }                                           \
-  }
-
 // -----------------runtime related macro definitions-------------------------------
 // If expr is not RT_ERROR_NONE, print the log
-#define GE_CHK_RT(expr)                                    \
-  do {                                                     \
-    rtError_t _rt_ret = (expr);                            \
-    if (_rt_ret != RT_ERROR_NONE) {                        \
-      DOMI_LOGE("Call rt api failed, ret: 0x%X", _rt_ret); \
-    }                                                      \
-  } while (0);
+#define GE_CHK_RT(expr)                                                \
+  do {                                                                 \
+    const rtError_t _rt_err = (expr);                                  \
+    if (_rt_err != RT_ERROR_NONE) {                                    \
+      GELOGE(ge::RT_FAILED, "Call rt api failed, ret: 0x%X", _rt_err); \
+    }                                                                  \
+  } while (false)
 
 // If expr is not RT_ERROR_NONE, print the log and execute the exec_expr expression
-#define GE_CHK_RT_EXEC(expr, exec_expr)                    \
-  {                                                        \
-    rtError_t _rt_ret = (expr);                            \
-    if (_rt_ret != RT_ERROR_NONE) {                        \
-      DOMI_LOGE("Call rt api failed, ret: 0x%X", _rt_ret); \
-      exec_expr;                                           \
-    }                                                      \
-  }
+#define GE_CHK_RT_EXEC(expr, exec_expr)                                \
+  do {                                                                 \
+    const rtError_t _rt_ret = (expr);                                  \
+    if (_rt_ret != RT_ERROR_NONE) {                                    \
+      GELOGE(ge::RT_FAILED, "Call rt api failed, ret: 0x%X", _rt_ret); \
+      exec_expr;                                                       \
+    }                                                                  \
+  } while (false)
 
 // If expr is not RT_ERROR_NONE, print the log and return
 #define GE_CHK_RT_RET(expr)                                                   \
   do {                                                                        \
-    rtError_t _rt_ret = (expr);                                               \
+    const rtError_t _rt_ret = (expr);                                         \
     if (_rt_ret != RT_ERROR_NONE) {                                           \
       REPORT_CALL_ERROR("E19999", "Call %s fail, ret: 0x%X", #expr, _rt_ret); \
-      DOMI_LOGE("Call rt api failed, ret: 0x%X", _rt_ret);                    \
+      GELOGE(ge::RT_FAILED, "Call rt api failed, ret: 0x%X", _rt_ret);        \
       return RT_ERROR_TO_GE_STATUS(_rt_ret);                                  \
     }                                                                         \
-  } while (0);
+  } while (false)
 
 // If expr is true, execute exec_expr without printing logs
 #define GE_IF_BOOL_EXEC(expr, exec_expr) \
@@ -256,32 +173,32 @@
   try {                                        \
     exec_expr0;                                \
   } catch (const std::bad_alloc &) {           \
-    DOMI_LOGE("Make shared failed");           \
+    GELOGE(ge::FAILED, "Make shared failed");  \
     exec_expr1;                                \
   }
 
-#define GE_ERRORLOG_AND_ERRORMSG(_status, errormsg)    \
-  {                                                    \
-    GELOGE(_status, "[Check][InnerData]%s", errormsg); \
-    REPORT_INNER_ERROR("E19999", "%s", errormsg);      \
-  }
-
-#define GE_WARNINGLOG_AND_ERRORMSG(errormsg)                                           \
-  {                                                                                    \
-    GELOGW("%s", errormsg);                                                            \
-    ErrorManager::GetInstance().ATCReportErrMessage("E19021", {"reason"}, {errormsg}); \
-  }
-
-#define GE_CHK_LOG_AND_ERRORMSG(expr, _status, errormsg)                                 \
-  do {                                                                                   \
-    bool b = (expr);                                                                     \
-    if (!b) {                                                                            \
-      GELOGE(_status, "%s", errormsg);                                                   \
-      ErrorManager::GetInstance().ATCReportErrMessage("E19021", {"reason"}, {errormsg}); \
-      return _status;                                                                    \
-    }                                                                                    \
-  } while (0)
-
+#define GE_ERRORLOG_AND_ERRORMSG(_status, errormsg)        \
+  {                                                        \
+    GELOGE((_status), "[Check][InnerData]%s", (errormsg)); \
+    REPORT_INNER_ERROR("E19999", "%s", (errormsg));        \
+  }
+
+#define GE_WARNINGLOG_AND_ERRORMSG(errormsg)                                             \
+  {                                                                                      \
+    GELOGW("%s", (errormsg));                                                            \
+    ErrorManager::GetInstance().ATCReportErrMessage("E10052", {"reason"}, {(errormsg)}); \
+  }
+
+#define GE_CHK_LOG_AND_ERRORMSG(expr, _status, errormsg)                                   \
+  do {                                                                                     \
+    const bool b = (expr);                                                                 \
+    if (!b) {                                                                              \
+      GELOGE((_status), "%s", (errormsg));                                                 \
+      ErrorManager::GetInstance().ATCReportErrMessage("E10052", {"reason"}, {(errormsg)}); \
+      return (_status);                                                                    \
+    }                                                                                      \
+  } while (false)
+namespace ge {
 template <typename T>
 GE_FUNC_VISIBILITY std::string FmtToStr(const T &t) {
   std::string fmt;
@@ -290,5 +207,5 @@ GE_FUNC_VISIBILITY std::string FmtToStr(const T &t) {
   fmt = st.str();
   return fmt;
 }
-
+}  // namespace ge
 #endif  // INC_FRAMEWORK_COMMON_DEBUG_LOG_H_

inc/framework/common/file_constant_util.h

@@ -0,0 +1,51 @@
+/**
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef INC_FRAMEWORK_COMMON_FILE_CONSTANT_UTIL_H
+#define INC_FRAMEWORK_COMMON_FILE_CONSTANT_UTIL_H
+
+#include <map>
+#include <string>
+#include <vector>
+#include "ge/ge_api_error_codes.h"
+#include "nlohmann/json.hpp"
+#include "graph/op_desc.h"
+#include "graph/ge_tensor.h"
+
+namespace ge {
+struct FileConstantInfo {
+  std::string value_bin_file_id;
+  std::string value_bin_file_path;
+};
+
+struct OptionInfo {
+  std::vector<FileConstantInfo> info;
+};
+
+void from_json(const nlohmann::json &j, FileConstantInfo &info);
+
+void from_json(const nlohmann::json &j, OptionInfo &option_info);
+
+Status GetFilePathFromOption(std::map<std::string, std::string> &file_id_and_path_map);
+
+Status CopyOneWeightFromFile(const void *const curr_dev_ptr, const std::string &value, const size_t file_constant_size,
+                             size_t &left_size);
+
+Status GetFilePath(const OpDescPtr &op_desc, const std::map<std::string, std::string> &file_id_and_path_map,
+                   std::string &file_path);
+}  // namespace ge
+
+#endif  // INC_FRAMEWORK_COMMON_FILE_CONSTANT_UTIL_H

inc/framework/common/fmk_error_codes.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -20,14 +20,18 @@
 #if defined(_MSC_VER)
 #ifdef FUNC_VISIBILITY
 #define GE_FUNC_VISIBILITY _declspec(dllexport)
+#define GE_OBJECT_VISIBILITY
 #else
 #define GE_FUNC_VISIBILITY
+#define GE_OBJECT_VISIBILITY
 #endif
 #else
 #ifdef FUNC_VISIBILITY
 #define GE_FUNC_VISIBILITY __attribute__((visibility("default")))
+#define GE_OBJECT_VISIBILITY
 #else
 #define GE_FUNC_VISIBILITY
+#define GE_OBJECT_VISIBILITY __attribute__((visibility("hidden")))
 #endif
 #endif
 
@@ -42,23 +46,23 @@
 #define DECLARE_ERRORNO_OME(name, value) DECLARE_ERRORNO(SYSID_FWK, MODID_OME, name, value)
 #define DECLARE_ERRORNO_CALIBRATION(name, value) DECLARE_ERRORNO(SYSID_FWK, MODID_CALIBRATION, name, value)
 
-#define DEF_ERRORNO(name, desc) const ErrorNoRegisterar g_##name##_errorno(name, desc);
+#define DEF_ERRORNO(name, desc) const ErrorNoRegisterar g_##name##_errorno((name), (desc));
 
 // Interface for Obtaining Error Code Description
 #define GET_ERRORNO_STR(value) domi::StatusFactory::Instance()->GetErrDesc(value)
 
-const int MODID_OMG = 1;          // OMG module ID
-const int MODID_OME = 2;          // OME module ID
-const int MODID_CALIBRATION = 3;  // Calibration module ID
-
 namespace domi {
+constexpr int32_t MODID_OMG = 1;          // OMG module ID
+constexpr int32_t MODID_OME = 2;          // OME module ID
+constexpr int32_t MODID_CALIBRATION = 3;  // Calibration module ID
+
 class GE_FUNC_VISIBILITY StatusFactory {
  public:
   static StatusFactory *Instance();
 
-  void RegisterErrorNo(uint32_t err, const std::string &desc);
+  void RegisterErrorNo(const uint32_t err, const std::string &desc);
 
-  std::string GetErrDesc(uint32_t err);
+  std::string GetErrDesc(const uint32_t err);
 
  protected:
   StatusFactory() {}
@@ -70,7 +74,9 @@ class GE_FUNC_VISIBILITY StatusFactory {
 
 class GE_FUNC_VISIBILITY ErrorNoRegisterar {
  public:
-  ErrorNoRegisterar(uint32_t err, const std::string &desc) { StatusFactory::Instance()->RegisterErrorNo(err, desc); }
+  ErrorNoRegisterar(const uint32_t err, const std::string &desc) {
+    StatusFactory::Instance()->RegisterErrorNo(err, desc);
+  }
   ~ErrorNoRegisterar() {}
 };
 

inc/framework/common/fmk_types.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.

inc/framework/common/ge_compiler_options.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.

inc/framework/common/ge_format_util.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -19,7 +19,7 @@
 
 #include <vector>
 
-#include "common/ge_inner_error_codes.h"
+#include "framework/common/ge_inner_error_codes.h"
 #include "graph/tensor.h"
 
 namespace ge {
@@ -33,7 +33,7 @@ class GE_FUNC_VISIBILITY GeFormatUtil {
   /// @param  [out] dst_shape     destination shape
   /// @return Status
   ///
-  static Status TransShape(const TensorDesc &src_desc, Format dst_format, std::vector<int64_t> &dst_shape);
+  static Status TransShape(const TensorDesc &src_desc, const Format dst_format, std::vector<int64_t> &dst_shape);
 };
 }  // namespace ge
 

inc/framework/common/ge_inner_error_codes.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -22,17 +22,57 @@
 #include <string>
 #include "ge/ge_api_error_codes.h"
 
+// Each module defines error codes using the following macros, name can not be modified to (name)
+#define GE_ERRORNO_COMMON(name, value, desc)                                                                \
+  GE_ERRORNO(ge::InnLogRuntime::RT_HOST, ge::InnErrorCodeType::ERROR_CODE, ge::InnErrorLevel::COMMON_LEVEL, \
+             ge::InnSystemIdType::SYSID_GE, ge::InnSubModuleId::COMMON_MODULE, name, (value), (desc))
+#define GE_ERRORNO_CLIENT(name, value, desc)                                                                \
+  GE_ERRORNO(ge::InnLogRuntime::RT_HOST, ge::InnErrorCodeType::ERROR_CODE, ge::InnErrorLevel::COMMON_LEVEL, \
+             ge::InnSystemIdType::SYSID_GE, ge::InnSubModuleId::CLIENT_MODULE, name, (value), (desc))
+#define GE_ERRORNO_INIT(name, value, desc)                                                                  \
+  GE_ERRORNO(ge::InnLogRuntime::RT_HOST, ge::InnErrorCodeType::ERROR_CODE, ge::InnErrorLevel::COMMON_LEVEL, \
+             ge::InnSystemIdType::SYSID_GE, ge::InnSubModuleId::INIT_MODULE, name, (value), (desc))
+#define GE_ERRORNO_SESSION(name, value, desc)                                                               \
+  GE_ERRORNO(ge::InnLogRuntime::RT_HOST, ge::InnErrorCodeType::ERROR_CODE, ge::InnErrorLevel::COMMON_LEVEL, \
+             ge::InnSystemIdType::SYSID_GE, ge::InnSubModuleId::SESSION_MODULE, name, (value), (desc))
+#define GE_ERRORNO_GRAPH(name, value, desc)                                                                 \
+  GE_ERRORNO(ge::InnLogRuntime::RT_HOST, ge::InnErrorCodeType::ERROR_CODE, ge::InnErrorLevel::COMMON_LEVEL, \
+             ge::InnSystemIdType::SYSID_GE, ge::InnSubModuleId::GRAPH_MODULE, name, (value), (desc))
+#define GE_ERRORNO_ENGINE(name, value, desc)                                                                \
+  GE_ERRORNO(ge::InnLogRuntime::RT_HOST, ge::InnErrorCodeType::ERROR_CODE, ge::InnErrorLevel::COMMON_LEVEL, \
+             ge::InnSystemIdType::SYSID_GE, ge::InnSubModuleId::ENGINE_MODULE, name, (value), (desc))
+#define GE_ERRORNO_OPS(name, value, desc)                                                                   \
+  GE_ERRORNO(ge::InnLogRuntime::RT_HOST, ge::InnErrorCodeType::ERROR_CODE, ge::InnErrorLevel::COMMON_LEVEL, \
+             ge::InnSystemIdType::SYSID_GE, ge::InnSubModuleId::OPS_MODULE, name, (value), (desc))
+#define GE_ERRORNO_PLUGIN(name, value, desc)                                                                \
+  GE_ERRORNO(ge::InnLogRuntime::RT_HOST, ge::InnErrorCodeType::ERROR_CODE, ge::InnErrorLevel::COMMON_LEVEL, \
+             ge::InnSystemIdType::SYSID_GE, ge::InnSubModuleId::PLUGIN_MODULE, name, (value), (desc))
+#define GE_ERRORNO_RUNTIME(name, value, desc)                                                               \
+  GE_ERRORNO(ge::InnLogRuntime::RT_HOST, ge::InnErrorCodeType::ERROR_CODE, ge::InnErrorLevel::COMMON_LEVEL, \
+             ge::InnSystemIdType::SYSID_GE, ge::InnSubModuleId::RUNTIME_MODULE, name, (value), (desc))
+#define GE_ERRORNO_EXECUTOR(name, value, desc)                                                                \
+  GE_ERRORNO(ge::InnLogRuntime::RT_DEVICE, ge::InnErrorCodeType::ERROR_CODE, ge::InnErrorLevel::COMMON_LEVEL, \
+             ge::InnSystemIdType::SYSID_GE, ge::InnSubModuleId::EXECUTOR_MODULE, name, (value), (desc))
+#define GE_ERRORNO_GENERATOR(name, value, desc)                                                             \
+  GE_ERRORNO(ge::InnLogRuntime::RT_HOST, ge::InnErrorCodeType::ERROR_CODE, ge::InnErrorLevel::COMMON_LEVEL, \
+             ge::InnSystemIdType::SYSID_GE, ge::InnSubModuleId::GENERATOR_MODULE, name, (value), (desc))
+
+// Get error code description
+#define GE_GET_ERRORNO_STR(value) ge::StatusFactory::Instance()->GetErrDesc(value)
+
+#define RT_ERROR_TO_GE_STATUS(RT_ERROR) static_cast<Status>(RT_ERROR)
+
 namespace ge {
 // System ID
-enum SystemIdType { SYSID_GE = 8 };
+enum class InnSystemIdType { SYSID_GE = 8 };
 // Runtime location
-enum LogRuntime {
+enum class InnLogRuntime {
   RT_HOST = 0b01,
   RT_DEVICE = 0b10,
 };
 
 // Sub model
-enum SubModuleId {
+enum class InnSubModuleId {
   COMMON_MODULE = 0,
   CLIENT_MODULE = 1,
   INIT_MODULE = 2,
@@ -47,13 +87,13 @@ enum SubModuleId {
 };
 
 // Error code type
-enum ErrorCodeType {
+enum class InnErrorCodeType {
   ERROR_CODE = 0b01,
   EXCEPTION_CODE = 0b10,
 };
 
 // Error level
-enum ErrorLevel {
+enum class InnErrorLevel {
   COMMON_LEVEL = 0b000,
   SUGGESTION_LEVEL = 0b001,
   MINOR_LEVEL = 0b010,
@@ -61,33 +101,6 @@ enum ErrorLevel {
   CRITICAL_LEVEL = 0b100,
 };
 
-// Each module defines error codes using the following macros
-#define GE_ERRORNO_COMMON(name, value, desc) \
-  GE_ERRORNO(RT_HOST, ERROR_CODE, COMMON_LEVEL, SYSID_GE, COMMON_MODULE, name, value, desc)
-#define GE_ERRORNO_CLIENT(name, value, desc) \
-  GE_ERRORNO(RT_HOST, ERROR_CODE, COMMON_LEVEL, SYSID_GE, CLIENT_MODULE, name, value, desc)
-#define GE_ERRORNO_INIT(name, value, desc) \
-  GE_ERRORNO(RT_HOST, ERROR_CODE, COMMON_LEVEL, SYSID_GE, INIT_MODULE, name, value, desc)
-#define GE_ERRORNO_SESSION(name, value, desc) \
-  GE_ERRORNO(RT_HOST, ERROR_CODE, COMMON_LEVEL, SYSID_GE, SESSION_MODULE, name, value, desc)
-#define GE_ERRORNO_GRAPH(name, value, desc) \
-  GE_ERRORNO(RT_HOST, ERROR_CODE, COMMON_LEVEL, SYSID_GE, GRAPH_MODULE, name, value, desc)
-#define GE_ERRORNO_ENGINE(name, value, desc) \
-  GE_ERRORNO(RT_HOST, ERROR_CODE, COMMON_LEVEL, SYSID_GE, ENGINE_MODULE, name, value, desc)
-#define GE_ERRORNO_OPS(name, value, desc) \
-  GE_ERRORNO(RT_HOST, ERROR_CODE, COMMON_LEVEL, SYSID_GE, OPS_MODULE, name, value, desc)
-#define GE_ERRORNO_PLUGIN(name, value, desc) \
-  GE_ERRORNO(RT_HOST, ERROR_CODE, COMMON_LEVEL, SYSID_GE, PLUGIN_MODULE, name, value, desc)
-#define GE_ERRORNO_RUNTIME(name, value, desc) \
-  GE_ERRORNO(RT_HOST, ERROR_CODE, COMMON_LEVEL, SYSID_GE, RUNTIME_MODULE, name, value, desc)
-#define GE_ERRORNO_EXECUTOR(name, value, desc) \
-  GE_ERRORNO(RT_DEVICE, ERROR_CODE, COMMON_LEVEL, SYSID_GE, EXECUTOR_MODULE, name, value, desc)
-#define GE_ERRORNO_GENERATOR(name, value, desc) \
-  GE_ERRORNO(RT_HOST, ERROR_CODE, COMMON_LEVEL, SYSID_GE, GENERATOR_MODULE, name, value, desc)
-
-// Get error code description
-#define GE_GET_ERRORNO_STR(value) ge::StatusFactory::Instance()->GetErrDesc(value)
-
 // Common module error code definition
 GE_ERRORNO_COMMON(MEMALLOC_FAILED, 0, "Failed to allocate memory!");  // 1343225856
 GE_ERRORNO_COMMON(PARAM_INVALID, 1, "Parameter's invalid!");          // 1343225857
@@ -125,13 +138,13 @@ GE_ERRORNO_CLIENT(GE_CLI_GE_ALREADY_INITIALIZED, 10, "GE is already initialized.
 GE_ERRORNO_CLIENT(GE_CLI_GE_NOT_INITIALIZED, 11, "GE is not yet initialized or is finalized.");  // 1343229963
 
 // Init module error code definition
-GE_ERRORNO_INIT(GE_MULTI_INIT, 0, "Multiple initializations are not supported.");            // 1343234048
-GE_ERRORNO_INIT(GE_FINALIZE_NOT_INIT, 1, "Finalize is not allowed before initialization.");  // 1343234049
-GE_ERRORNO_INIT(GE_MULTI_FINALIZE, 2, "Multiple finalizations are not supported.");          // 1343234050
-GE_ERRORNO_INIT(GE_PROF_MULTI_INIT, 3, "Multiple profiling initializations are not supported.");          // 1343234051
-GE_ERRORNO_INIT(GE_PROF_NOT_INIT, 4, "Profing initializations have not been done.");          // 1343234052
+GE_ERRORNO_INIT(GE_MULTI_INIT, 0, "Multiple initializations are not supported.");                 // 1343234048
+GE_ERRORNO_INIT(GE_FINALIZE_NOT_INIT, 1, "Finalize is not allowed before initialization.");       // 1343234049
+GE_ERRORNO_INIT(GE_MULTI_FINALIZE, 2, "Multiple finalizations are not supported.");               // 1343234050
+GE_ERRORNO_INIT(GE_PROF_MULTI_INIT, 3, "Multiple profiling initializations are not supported.");  // 1343234051
+GE_ERRORNO_INIT(GE_PROF_NOT_INIT, 4, "Profing initializations have not been done.");              // 1343234052
 GE_ERRORNO_INIT(GE_PROF_MODE_CONFLICT, 5,
-                "Profiling command mode which is preferred is running, the api mode will not work.");   // 1343234053
+                "Profiling command mode which is preferred is running, the api mode will not work.");  // 1343234053
 
 // Session module error code definition
 GE_ERRORNO_SESSION(GE_SESS_INIT_FAILED, 0, "Failed to initialize session.");                          // 1343238144
@@ -216,8 +229,8 @@ GE_ERRORNO_ENGINE(GE_ENG_FINALIZE_FAILED, 1, "Engine finalize failed.");
 GE_ERRORNO_ENGINE(GE_ENG_MEMTYPE_ERROR, 2, "Memory type HBM is necessary when engine is in device");  // 1343246338
 
 // Optimize errocode
-GE_ERRORNO_GRAPH(TO_BE_DELETED, 63, "The node of the graph to be deleted.");          // 1343242303
-GE_ERRORNO_GRAPH(NOT_CHANGED, 64, "The node of the graph no changed.");               // 1343242304
+GE_ERRORNO_GRAPH(TO_BE_DELETED, 63, "The node of the graph to be deleted.");  // 1343242303
+GE_ERRORNO_GRAPH(NOT_CHANGED, 64, "The node of the graph no changed.");       // 1343242304
 
 // Ops module error code definition
 GE_ERRORNO_OPS(GE_OPS_KERNEL_STORE_INIT_FAILED, 0, "Failed to initialize OpsKernelInfoStore.");  // 1343250432
@@ -313,7 +326,6 @@ GE_ERRORNO_GENERATOR(GE_GENERATOR_GRAPH_MANAGER_BUILD_GRAPH_FAILED, 3, "Graph ma
 GE_ERRORNO_GENERATOR(GE_GENERATOR_GRAPH_MANAGER_FINALIZE_FAILED, 4, "Graph manager finalize failed.");
 GE_ERRORNO_GENERATOR(GE_GENERATOR_GRAPH_MANAGER_SAVE_MODEL_FAILED, 5, "Graph manager save model failed.");
 
-#define RT_ERROR_TO_GE_STATUS(RT_ERROR) static_cast<Status>(RT_ERROR)
 }  // namespace ge
 
 #endif  // INC_FRAMEWORK_COMMON_GE_INNER_ERROR_CODES_H_

inc/framework/common/ge_types.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -17,31 +17,39 @@
 #ifndef INC_FRAMEWORK_COMMON_GE_TYPES_H_
 #define INC_FRAMEWORK_COMMON_GE_TYPES_H_
 
-#include <stdint.h>
+#include <cstdint>
 
 #include <string>
 #include <vector>
 
 #include "framework/common/fmk_error_codes.h"
-#include "ge/ge_api_error_codes.h"
+#include "external/ge/ge_api_error_codes.h"
 #include "external/graph/types.h"
 #include "external/ge/ge_api_types.h"
 
 namespace ge {
 enum RuntimeType { HOST = 0, DEVICE = 1 };
 
-enum PerfLevel { GEN_TASK_WITH_FUSION = -1, GEN_TASK_WITHOUT_L2FUSION = 3, GEN_TASK_WITHOUT_FUSION = 4 };
+enum class PerfLevel : int32_t {
+  GEN_TASK_WITH_FUSION = -1,
+  GEN_TASK_WITHOUT_L2FUSION = 3,
+  GEN_TASK_WITHOUT_FUSION = 4
+};
 
 enum FrameworkType {
   CAFFE = 0,
   MINDSPORE = 1,
   TENSORFLOW = 3,
-  ANDROID_NN,
-  ONNX,
+  ANDROID_NN = 4,
+  ONNX = 5,
 };
 
+enum class GraphStage : int64_t { GRAPH_STAGE_FUZZ = 0, GRAPH_STAGE_RESERVED };
+
+const char_t *const kGraphDumpStage = "DumpStage";
+
 const std::map<std::string, std::string> kFwkTypeToStr = {
-  {"0", "Caffe"}, {"1", "MindSpore"}, {"3", "TensorFlow"}, {"4", "Android_NN"}, {"5", "Onnx"}};
+    {"0", "Caffe"}, {"1", "MindSpore"}, {"3", "TensorFlow"}, {"4", "Android_NN"}, {"5", "Onnx"}};
 
 enum OpEngineType {
   ENGINE_SYS = 0,  // default engine
@@ -53,28 +61,52 @@ enum OpEngineType {
 
 enum InputAippType { DATA_WITHOUT_AIPP = 0, DATA_WITH_STATIC_AIPP, DATA_WITH_DYNAMIC_AIPP, DYNAMIC_AIPP_NODE };
 
-const char *const GE_ENGINE_ATTR_MEM_TYPE_HBM = "HBM";
-const char *const GE_OPTION_EXEC_PLACEMENT = "ge.exec.placement";
+const char_t *const GE_ENGINE_ATTR_MEM_TYPE_HBM = "HBM";
+const char_t *const GE_OPTION_EXEC_PLACEMENT = "ge.exec.placement";
 
 // profiling data
+
 const std::string kTaskTypeAicore = "AI_CORE";
 const std::string kTaskTypeAicpu = "AI_CPU";
 const std::string kTaskTypeInvalid = "TASK_TYPE_INVALID";
+const std::string kTaskTypeFftsPlus = "FFTS_PLUS";
+const std::string kEngineNameVectorCore = "VectorEngine";
+
+const std::string kEngineNameHccl = "ops_kernel_info_hccl";
+const std::string kEngineNameRts = "DNN_VM_RTS_OP_STORE";
+const std::string kEngineNameHostCpu = "DNN_VM_HOST_CPU_OP_STORE";
+const std::string kEngineNameGeLocal = "DNN_VM_GE_LOCAL_OP_STORE";
+const std::string kEngineNameAiCpu = "aicpu_ascend_kernel";
+const std::string kEngineNameAiCpuTf = "aicpu_tf_kernel";
+const std::string kEngineNameAiCore = "AIcoreEngine";
+const std::string kAtomicOpType = "DynamicAtomicAddrClean";
+
+const std::string kShapeTypeStatic = "static";
+const std::string kShapeTypeDynamic = "dynamic";
+
+constexpr uint64_t kInferSessionId = 0U;
+constexpr uint64_t kReleaseFlag = 1U;
+constexpr uint32_t kInvalidModelId = 0xFFFFFFFFU;
+constexpr size_t kNumTaskWithAtomicAddrCleanTask = 2U;
+constexpr uint32_t INVALID_MODEL_ID = 0xFFFFFFFFUL;
 
 // dynamic execute mode
-const char *const kLazyRecompile = "lazy_recompile";
+const char_t *const kLazyRecompile = "lazy_recompile";
+
+constexpr size_t kMaxHostMemInputLen = 128U;  // 64 aligned
 
 // Data cache, including data address and length
 struct DataBuffer {
- public:
   void *data;       // Data address
   uint64_t length;  // Data length
   bool isDataSupportMemShare = false;
-  uint32_t placement = 0;
-  DataBuffer(void *dataIn, uint64_t len, bool isSupportMemShare, uint32_t placement = 0)
-      : data(dataIn), length(len), isDataSupportMemShare(isSupportMemShare), placement(placement) {}
+  uint32_t placement = 0U;
 
-  DataBuffer() : data(nullptr), length(0), isDataSupportMemShare(false) {}
+  DataBuffer(void *const data_in, const uint64_t data_len, const bool is_support_mem_share = false,
+             const uint32_t data_placement = 0U)
+      : data(data_in), length(data_len), isDataSupportMemShare(is_support_mem_share), placement(data_placement) {}
+
+  DataBuffer() : data(nullptr), length(0UL), isDataSupportMemShare(false), placement(0U) {}
 };
 
 ///
@@ -86,7 +118,7 @@ struct InputData {
   uint32_t timestamp;                        // Data creation time
   uint32_t timeout;                          // Processing timeout
   uint32_t model_id;                         // Model ID required for data processing
-  uint64_t request_id = 0;                   // Request ID
+  uint64_t request_id = 0UL;                 // Request ID
   std::vector<DataBuffer> blobs;             // Actual input data, currently only supports one input
   bool is_dynamic_batch = false;             // Whether is dynamic batch size scene, default:false
   std::string batch_label;                   // Gear used for current inference in dynamic batch scene
@@ -113,10 +145,10 @@ struct Command {
 
 // The definition of I/O shape description
 struct ShapeDescription {
-  int64_t num = 0;
-  int64_t channel = 0;
-  int64_t height = 0;
-  int64_t width = 0;
+  int64_t num = 0L;
+  int64_t channel = 0L;
+  int64_t height = 0L;
+  int64_t width = 0L;
   std::vector<int64_t> dims;
   std::vector<std::pair<int64_t, int64_t>> shape_ranges;
 };
@@ -186,14 +218,14 @@ struct AippConfigInfo {
   int32_t mean_chn_1;
   int32_t mean_chn_2;
   int32_t mean_chn_3;
-  float min_chn_0;
-  float min_chn_1;
-  float min_chn_2;
-  float min_chn_3;
-  float var_reci_chn_0;
-  float var_reci_chn_1;
-  float var_reci_chn_2;
-  float var_reci_chn_3;
+  float32_t min_chn_0;
+  float32_t min_chn_1;
+  float32_t min_chn_2;
+  float32_t min_chn_3;
+  float32_t var_reci_chn_0;
+  float32_t var_reci_chn_1;
+  float32_t var_reci_chn_2;
+  float32_t var_reci_chn_3;
   int8_t support_rotation;
   uint32_t related_input_rank;
   uint32_t max_src_image_size;
@@ -202,26 +234,42 @@ struct AippConfigInfo {
 // The structure of offline Modeldata
 struct ModelData {
   void *model_data = nullptr;  // Model binary data start addr
-  uint32_t model_len = 0;      // Model binary data length
+  uint32_t model_len = 0U;     // Model binary data length
   int32_t priority = 0;        // Model priority
   std::string key;             // Key path for encrypt model, Empty for unencrypt
   std::string om_name;         // om file name, used for data dump
 };
 
+struct ModelParam {
+  ModelParam() : priority(0), mem_base(0U), mem_size(0U), weight_base(0U), weight_size(0U) {}
+  ModelParam(const int32_t pri, const uintptr_t m_base, const size_t m_len, const uintptr_t w_base, const size_t w_len)
+      : priority(pri), mem_base(m_base), mem_size(m_len), weight_base(w_base), weight_size(w_len) {}
+  ~ModelParam() = default;
+
+  int32_t priority;
+  uintptr_t mem_base;
+  size_t mem_size;
+  uintptr_t weight_base;
+  size_t weight_size;
+};
+
 // The definition of Model information
 struct ModelInfo {
-  uint32_t version = 0;
+  uint32_t version = 0U;
   std::string name;
-  bool is_encrypt = 0;  //  0:unencrypt, 1:encrypt
+  bool is_encrypt = false;  //  0:unencrypt, 1:encrypt
   std::vector<ShapeDescription> input_desc;
   std::vector<ShapeDescription> output_desc;
-  uint8_t reserved[3] = {0};  // 3-byte reserved field
+  uint8_t reserved[3] = {0U};  // 3-byte reserved field
 };
 
 // Asynchronous callback interface, implemented by the caller
 class GE_FUNC_VISIBILITY ModelListener {
  public:
   virtual ~ModelListener() {}
+  ModelListener() = default;
+  ModelListener(const ModelListener &) = delete;
+  ModelListener &operator=(const ModelListener &) = delete;
   ///
   /// @brief Asynchronous callback interface
   /// @param [in] model_id   Model ID of the callback
@@ -230,6 +278,18 @@ class GE_FUNC_VISIBILITY ModelListener {
   ///
   virtual Status OnComputeDone(uint32_t model_id, uint32_t data_index, uint32_t result_code,
                                std::vector<ge::Tensor> &outputs) = 0;
+
+  virtual void SetCallback(const RunAsyncCallback &callback) {
+    (void)callback;
+  }
+
+  virtual uint32_t GetResultCode() {
+    return 0U;
+  };
+
+  virtual Status ResetResult() {
+    return SUCCESS;
+  };
 };
 
 // OMM configuration item
@@ -249,6 +309,7 @@ struct Options {
   int32_t physical_device_id;
   std::string profiling_mode;
   std::string profiling_options;
+  int32_t graphExecTimeout;
 };
 
 // Profiling info of task
@@ -268,13 +329,24 @@ struct TaskDescInfo {
   std::vector<Format> output_format;
   std::vector<std::vector<int64_t>> output_shape;
   std::vector<DataType> output_data_type;
+  uint32_t context_id = 0xFFFFFFFFUL;
 };
 
 struct OpDescInfo {
   std::string op_name;
   std::string op_type;
-  uint32_t task_id;
-  uint32_t stream_id;
+  uint32_t task_id = 0U;
+  uint32_t stream_id = 0U;
+  uint32_t imply_type = 0U;
+  uint32_t block_dim = 0U;
+  std::string op_file_path;
+  std::string dev_func;
+  std::string tvm_magic;
+  uint32_t tiling_key = 0U;
+  uintptr_t args = 0U;
+  std::string tiling_data;
+  std::string node_info;
+  std::vector<int64_t> workspace_bytes;
   std::vector<Format> input_format;
   std::vector<std::vector<int64_t>> input_shape;
   std::vector<DataType> input_data_type;

inc/framework/common/gflags_util.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.

inc/framework/common/helper/model_helper.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -20,11 +20,10 @@
 #include <memory>
 #include <string>
 
-#include "common/fmk_types.h"
-#include "common/helper/om_file_helper.h"
+#include "framework/common/helper/om_file_helper.h"
 #include "common/model/ge_model.h"
 #include "common/model/ge_root_model.h"
-#include "common/types.h"
+#include "framework/common/types.h"
 #include "graph/model.h"
 
 namespace ge {
@@ -34,77 +33,71 @@ class GE_FUNC_VISIBILITY ModelHelper {
   ~ModelHelper();
 
   Status SaveToOmModel(const GeModelPtr &ge_model, const SaveParam &save_param, const std::string &output_file,
-                       ge::ModelBufferData &model);
-  Status SaveToOmRootModel(const GeRootModelPtr &ge_root_model, const SaveParam &save_param, const string &output_file,
-                           ModelBufferData &model, bool is_unknown_shape);
-  Status SaveOriginalGraphToOmModel(const ge::Graph &graph, const std::string &output_file);
+                       ge::ModelBufferData &model) const;
+  Status SaveToOmRootModel(const GeRootModelPtr &ge_root_model, const SaveParam &save_param,
+                           const std::string &output_file, ModelBufferData &model, const bool is_unknown_shape) const;
+  Status SaveOriginalGraphToOmModel(const ge::Graph &graph, const std::string &output_file) const;
   Status LoadModel(const ge::ModelData &model_data);
   Status LoadRootModel(const ge::ModelData &model_data);
-  Status GetModelBufferData(ge::ModelBufferData &model);
-
-  const ModelFileHeader *GetFileHeader() const {
-    return file_header_;
-  }
+  static void SetModelToGeModel(const GeModelPtr &ge_model, Model &model);
 
   GeModelPtr GetGeModel();
   GeRootModelPtr GetGeRootModel();
-  void SetSaveMode(bool val) {
+  void SetSaveMode(const bool val) {
     is_offline_ = val;
   }
-  bool GetSaveMode(void) const {
-    return is_offline_;
-  }
+
   bool GetModelType() const {
     return is_unknown_shape_model_;
-  };
+  }
 
-  Status GetBaseNameFromFileName(const std::string &file_name, std::string &base_name);
-  Status GetModelNameFromMergedGraphName(const std::string &graph_name, std::string &model_name);
+  Status GetBaseNameFromFileName(const std::string &file_name, std::string &base_name) const;
+  Status GetModelNameFromMergedGraphName(const ComputeGraphPtr &compute_graph, std::string &model_name) const;
 
  private:
   bool is_assign_model_ = false;
   bool is_offline_ = true;
   bool is_unknown_shape_model_ = false;
   ModelFileHeader *file_header_ = nullptr;
-  // Encrypted model need delete temp model and unencrypted model need not delete model
-  uint8_t *model_addr_tmp_ = nullptr;
-  uint32_t model_len_tmp_ = 0;
   GeModelPtr model_;
   GeRootModelPtr root_model_;
 
-  ModelHelper(const ModelHelper &);
-  ModelHelper &operator=(const ModelHelper &);
-  Status GenerateGeModel(OmFileLoadHelper &om_load_helper);
-  Status GenerateGeRootModel(OmFileLoadHelper &om_load_helper);
-  Status LoadModelData(OmFileLoadHelper &om_load_helper);
-  void SetModelToGeModel(GeModelPtr &ge_model, Model &model);
-  Status LoadModelData(OmFileLoadHelper &om_load_helper, GeModelPtr &cur_model, size_t mode_index);
-  Status LoadWeights(OmFileLoadHelper &om_load_helper);
-  Status LoadWeights(OmFileLoadHelper &om_load_helper, GeModelPtr &cur_model, size_t mode_index);
-  Status LoadTask(OmFileLoadHelper &om_load_helper);
-  Status LoadTask(OmFileLoadHelper &om_load_helper, GeModelPtr &cur_model, size_t mode_index);
-  Status LoadTBEKernelStore(OmFileLoadHelper &om_load_helper);
-  Status LoadTBEKernelStore(OmFileLoadHelper &om_load_helper, GeModelPtr &cur_model, size_t mode_index);
-  Status LoadCustAICPUKernelStore(OmFileLoadHelper &om_load_helper);
-  Status LoadCustAICPUKernelStore(OmFileLoadHelper &om_load_helper, GeModelPtr &cur_model, size_t mode_index);
-  Status ReleaseLocalModelData() noexcept;
-  Status SaveModelPartition(std::shared_ptr<OmFileSaveHelper> &om_file_save_helper, ModelPartitionType type,
-                            const uint8_t *data, size_t size, size_t model_index);
+  ModelHelper(const ModelHelper &) = default;
+  ModelHelper &operator=(const ModelHelper &) = default;
+
+  bool IsPartitionedGraph(const GeModelPtr &cur_model) const;
+
+  Status GenerateGeModel(const OmFileLoadHelper &om_load_helper, GeModelPtr &cur_model, const size_t mode_index,
+                         const bool is_dyn_root);
+  Status GenerateGeRootModel(const OmFileLoadHelper &om_load_helper);
+
+  Status LoadModelData(const OmFileLoadHelper &om_load_helper, const GeModelPtr &cur_model,
+                       const size_t mode_index) const;
+  Status LoadWeights(const OmFileLoadHelper &om_load_helper, const GeModelPtr &cur_model,
+                     const size_t mode_index) const;
+  Status LoadTask(const OmFileLoadHelper &om_load_helper, const GeModelPtr &cur_model, const size_t mode_index) const;
+  Status LoadTBEKernelStore(const OmFileLoadHelper &om_load_helper, const GeModelPtr &cur_model,
+                            const size_t mode_index) const;
+  Status LoadCustAICPUKernelStore(const OmFileLoadHelper &om_load_helper, const GeModelPtr &cur_model,
+                                  const size_t mode_index) const;
+
+  Status SaveModelPartition(std::shared_ptr<OmFileSaveHelper> &om_file_save_helper, const ModelPartitionType type,
+                            const uint8_t *const data, const size_t size, const size_t model_index) const;
   Status SaveModelDef(shared_ptr<OmFileSaveHelper> &om_file_save_helper, const GeModelPtr &ge_model,
-                      Buffer &model_buffer, size_t model_index = 0);
-  Status SaveSizeToModelDef(const GeModelPtr &ge_model);
+                      Buffer &model_buffer, const size_t model_index = 0U) const;
+  Status SaveSizeToModelDef(const GeModelPtr &ge_model) const;
   Status SaveModelWeights(shared_ptr<OmFileSaveHelper> &om_file_save_helper, const GeModelPtr &ge_model,
-                          size_t model_index = 0);
+                          const size_t model_index = 0U) const;
   Status SaveModelTbeKernel(shared_ptr<OmFileSaveHelper> &om_file_save_helper, const GeModelPtr &ge_model,
-                            size_t model_index = 0);
+                            const size_t model_index = 0U) const;
   Status SaveModelCustAICPU(shared_ptr<OmFileSaveHelper> &om_file_save_helper, const GeModelPtr &ge_model,
-                            size_t model_index = 0);
+                            const size_t model_index = 0U) const;
   Status SaveModelTaskDef(shared_ptr<OmFileSaveHelper> &om_file_save_helper, const GeModelPtr &ge_model,
-                          Buffer &task_buffer, size_t model_index = 0);
+                          Buffer &task_buffer, const size_t model_index = 0U) const;
   Status SaveModelHeader(shared_ptr<OmFileSaveHelper> &om_file_save_helper, const GeModelPtr &ge_model,
-                         size_t model_num = 1);
+                         const size_t model_num = 1U) const;
   Status SaveAllModelPartiton(shared_ptr<OmFileSaveHelper> &om_file_save_helper, const GeModelPtr &ge_model,
-                              Buffer &model_buffer, Buffer &task_buffer, size_t model_index = 0);
+                              Buffer &model_buffer, Buffer &task_buffer, const size_t model_index = 0U) const;
 };
 }  // namespace ge
 #endif  // INC_FRAMEWORK_COMMON_HELPER_MODEL_HELPER_H_

inc/framework/common/helper/om_file_helper.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -21,25 +21,20 @@
 #include <vector>
 
 #include "external/ge/ge_ir_build.h"
-#include "framework/common/fmk_types.h"
 #include "framework/common/types.h"
 #include "framework/common/ge_types.h"
 
-using ProcParam = struct PROC_PARAM;
-using std::string;
-using std::vector;
-
 namespace ge {
 struct ModelPartition {
   ModelPartitionType type;
-  uint8_t *data = 0;
-  uint32_t size = 0;
+  const uint8_t *data = nullptr;
+  uint32_t size = 0U;
 };
 
 struct OmFileContext {
   std::vector<ModelPartition> partition_datas_;
-  std::vector<char> partition_table_;
-  uint32_t model_data_len_ = 0;
+  std::vector<char_t> partition_table_;
+  uint32_t model_data_len_ = 0U;
 };
 
 struct SaveParam {
@@ -53,57 +48,56 @@ struct SaveParam {
 
 class GE_FUNC_VISIBILITY OmFileLoadHelper {
  public:
-  Status Init(const ge::ModelData &model);
+  Status Init(const ModelData &model);
 
-  Status Init(uint8_t *model_data, const uint32_t model_data_size);
+  Status Init(uint8_t *const model_data, const uint32_t model_data_size);
 
-  Status Init(uint8_t *model_data, const uint32_t model_data_size, uint32_t model_num);
+  Status Init(uint8_t *const model_data, const uint32_t model_data_size, const uint32_t model_num);
 
-  Status GetModelPartition(ModelPartitionType type, ModelPartition &partition);
+  Status GetModelPartition(const ModelPartitionType type, ModelPartition &partition);
 
-  Status GetModelPartition(ModelPartitionType type, ModelPartition &partition, size_t model_index);
+  Status GetModelPartition(const ModelPartitionType type, ModelPartition &partition, const size_t model_index) const;
 
   OmFileContext context_;
 
-  vector<OmFileContext> model_contexts_;
+  std::vector<OmFileContext> model_contexts_;
 
  private:
-  Status CheckModelValid(const ge::ModelData &model) const;
-
-  Status LoadModelPartitionTable(uint8_t *model_data, const uint32_t model_data_size);
+  Status LoadModelPartitionTable(uint8_t *const model_data, const uint32_t model_data_size, const size_t model_index,
+                                 size_t &mem_offset);
 
-  Status LoadModelPartitionTable(uint8_t *model_data, const uint32_t model_data_size, uint32_t model_num);
+  Status LoadModelPartitionTable(uint8_t *const model_data, const uint32_t model_data_size, const uint32_t model_num);
 
   bool is_inited_{false};
 };
 
 class GE_FUNC_VISIBILITY OmFileSaveHelper {
  public:
-  ModelFileHeader &GetModelFileHeader() { return model_header_; }
+  ModelFileHeader &GetModelFileHeader() {
+    return model_header_;
+  }
 
-  uint32_t GetModelDataSize() const { return context_.model_data_len_; }
+  uint32_t GetModelDataSize() const;
 
   ModelPartitionTable *GetPartitionTable();
 
-  Status AddPartition(ModelPartition &partition);
+  Status AddPartition(const ModelPartition &partition);
 
-  Status AddPartition(ModelPartition &partition, size_t cur_index);
+  Status AddPartition(const ModelPartition &partition, const size_t cur_index);
 
-  const std::vector<ModelPartition> &GetModelPartitions() const;
+  Status SaveModel(const SaveParam &save_param, const char_t *const output_file, ModelBufferData &model,
+                   const bool is_offline = true);
 
-  Status SaveModel(const SaveParam &save_param, const char *target_file, ge::ModelBufferData &model,
-                   bool is_offline = true);
+  Status SaveModelToFile(const char_t *const output_file, ModelBufferData &model, const bool is_offline = true);
 
-  Status SaveModelToFile(const char *output_file, ge::ModelBufferData &model, bool is_offline = true);
+  ModelPartitionTable *GetPartitionTable(const size_t cur_ctx_index);
 
-  vector<OmFileContext> model_contexts_;
+  Status SaveRootModel(const SaveParam &save_param, const char_t *const output_file, ModelBufferData &model,
+                       const bool is_offline);
 
+ private:
   ModelFileHeader model_header_;
-  OmFileContext context_;
-
-  ModelPartitionTable *GetPartitionTable(size_t cur_ctx_index);
-
-  Status SaveRootModel(const SaveParam &save_param, const char *output_file, ModelBufferData &model, bool is_offline);
+  std::vector<OmFileContext> model_contexts_;
 };
 }  // namespace ge
 #endif  // INC_FRAMEWORK_COMMON_HELPER_OM_FILE_HELPER_H_

inc/framework/common/l2_cache_optimize.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -17,107 +17,24 @@
 #ifndef INC_FRAMEWORK_COMMON_L2_CACHE_OPTIMIZE_H_
 #define INC_FRAMEWORK_COMMON_L2_CACHE_OPTIMIZE_H_
 
-#include <stdint.h>
+#include <cstdint>
 
 #include <algorithm>
 #include <functional>
 #include <string>
 #include <vector>
 
-#include "common/types.h"
-#include "common/util.h"
+#include "framework/common/types.h"
+#include "framework/common/util.h"
 #include "graph/compute_graph.h"
 
-using std::vector;
-
 namespace ge {
 // Size of RC memory alignment, 2M
-constexpr size_t ALIGN_SIZE = 2097152;
-
-constexpr uint32_t RC_VALUE_DEFAULT = 1;
-constexpr uint32_t RC_VALUE_MAX = 32;
-
-// RC data type classification
-enum RCType {
-  RC_DEFAULT,      // Such as temporary workspace memory of operator, variable (including global and local variable)
-  RC_HCOM,         // Output of gradient aggregation, RC value should be set to 0
-  RC_L2LOSS,       // Parameter of L2 loss operator, RC value should be set to 0
-  RC_INPUTOUTPUT,  // Input and output tensor of operator, RC value is returned by FE calculation
-  RC_WEIGHTS,      // The weight, fp16, RC value used by FP/BP operator should be set to 1 or the actual access numbers
-  RC_DW,           // The gradient data DW and RC value output by BP operator
-                   // should be set to 1 or the actual access numbers
-  RC_ARGS          // Args of FlowTable, actual access numbers
-};
-
-enum MemType { INPUT_TENSOR, OUTPUT_TENSOR, WEIGHT, WORKSPACE };
-
-// Memory usage information < node, type, number >
-struct NodeInfo {
-  string nodeName;
-  MemType memType;
-  size_t index;
-};
-
-// Memory block RC value
-struct RCMemoryBlock {
-  RCType type;        // RC type
-  size_t blockSize;   // memory block size
-  size_t headOffset;  // Start offset from base address
-  size_t tailOffset;  // End offset from base address
-  uint32_t rcCount;   // RC value
-  NodeInfo nodeInfo;  // Input and output indexes of node objects to which RC belongs
-};
-
-// L2Cache optimizer
-class GE_FUNC_VISIBILITY L2CacheOptimize {
- public:
-  explicit L2CacheOptimize(ge::ComputeGraphPtr &graph);
-  ~L2CacheOptimize();
-
-  // Collect the information L2Cache Memory optimization
-  Status Gath();
-
- private:
-  ge::ComputeGraphPtr graph_;
-
-  // Save RC block information list
-  vector<RCMemoryBlock> weightRCs;
-  vector<RCMemoryBlock> opRCs;
-
-  // Extract RC information generated by FE from compiled graph
-  void RetirveRCinfo();
-
-  // Take the maximum common divisor of RC values for the duplicate address
-  void Merge(vector<RCMemoryBlock> &blocks);
-
-  // The RC information is aligned with the 2m address
-  void Align(vector<RCMemoryBlock> &blocks);
-
-  // Weight of l2loss operator, output of gradient aggregation output, RC value set to 0
-  void HandleOutputZeroRC(RCType type, ge::NodePtr node, vector<int64_t> &outputList, vector<RCMemoryBlock> &blocks);
-
-  // Processing operator input Tensor's RC
-  void HandOPInput(ge::NodePtr node, vector<int64_t> &inputList, vector<RCMemoryBlock> &blocks);
-
-  // Processing operator output Tensor's RC
-  void HandOPoutput(ge::NodePtr node, vector<int64_t> &outputList, vector<RCMemoryBlock> &blocks);
+constexpr size_t ALIGN_SIZE = 2097152U;
 
-  // maximum common divisor
-  uint32_t Measure(uint32_t x, uint32_t y) {
-    if (x == 0 || y == 0) return RC_VALUE_DEFAULT;
-    uint32_t z = y;
-    while (x % y != 0) {
-      z = x % y;
-      x = y;
-      y = z;
-    }
-    return z;
-  }
+constexpr uint32_t RC_VALUE_DEFAULT = 1U;
+constexpr uint32_t RC_VALUE_MAX = 32U;
 
-  bool Contain(const RCMemoryBlock &l_block, const RCMemoryBlock &r_block);
-  bool Cross(const RCMemoryBlock &l_block, const RCMemoryBlock &r_block);
-  bool Connect(const RCMemoryBlock &l_block, const RCMemoryBlock &r_block);
-};
 }  // namespace ge
 
 #endif  // INC_FRAMEWORK_COMMON_L2_CACHE_OPTIMIZE_H_

inc/framework/common/op/attr_value_util.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -34,141 +34,10 @@
 #include <google/protobuf/map.h>
 #include <unordered_map>
 #include <string>
-
 #include "graph/debug/ge_attr_define.h"
 #include "proto/om.pb.h"
 
-using domi::AttrDef;
-using domi::AttrDef_ListValue;
-using domi::ModelDef;
-using domi::NamedAttrs;
-using domi::OpDef;
-
 namespace ge {
-using AttrDefMap = ::google::protobuf::Map<::std::string, ::domi::AttrDef>;
-using AttrDefPair = ::google::protobuf::MapPair<std::string, domi::AttrDef>;
-
-GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, AttrDef &attr, OpDef *opdef);
-// DEFINE_ADD_ATTR_VALUE
-GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const std::string &value, AttrDefMap *attrs);
-GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const char *value, AttrDefMap *attrs);
-GE_FUNC_VISIBILITY void AddOpAttr(const char *key, const char *value, AttrDefMap *attrs);
-GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const uint32_t value, AttrDefMap *attrs);
-GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const int32_t value, AttrDefMap *attrs);
-GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const int64_t value, AttrDefMap *attrs);
-GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const float value, AttrDefMap *attrs);
-GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const double value, AttrDefMap *attrs);
-GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const bool value, AttrDefMap *attrs);
-
-GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const AttrDef_ListValue &value, AttrDefMap *attrs);
-
-// DEFINE_ADD_ATTR_VALUE
-GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const std::string &value, OpDef *opdef);
-GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const char *value, OpDef *opdef);
-GE_FUNC_VISIBILITY void AddOpAttr(const char *key, const char *value, OpDef *opdef);
-GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const uint32_t value, OpDef *opdef);
-GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const int32_t value, OpDef *opdef);
-GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const int64_t value, OpDef *opdef);
-GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const float value, OpDef *opdef);
-GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const double value, OpDef *opdef);
-GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const bool value, OpDef *opdef);
-
-GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const AttrDef_ListValue &value, OpDef *opdef);
-
-GE_FUNC_VISIBILITY void AddOpBytesAttr(const std::string &key, const void *value, size_t size, OpDef *opdef);
-
-// DEFINE_ADD_ATTR_VALUE_LIST
-GE_FUNC_VISIBILITY void AddOpAttrList(const std::string &key, const double value, AttrDefMap *attrs);
-GE_FUNC_VISIBILITY void AddOpAttrList(const std::string &key, const float value, AttrDefMap *attrs);
-GE_FUNC_VISIBILITY void AddOpAttrList(const std::string &key, const uint32_t value, AttrDefMap *attrs);
-GE_FUNC_VISIBILITY void AddOpAttrList(const std::string &key, const int32_t value, AttrDefMap *attrs);
-GE_FUNC_VISIBILITY void AddOpAttrList(const std::string &key, const std::string value, AttrDefMap *attrs);
-GE_FUNC_VISIBILITY void AddOpAttrList(const std::string &key, const double value, OpDef *opdef);
-GE_FUNC_VISIBILITY void AddOpAttrList(const std::string &key, const float value, OpDef *opdef);
-GE_FUNC_VISIBILITY void AddOpAttrList(const std::string &key, const uint32_t value, OpDef *opdef);
-GE_FUNC_VISIBILITY void AddOpAttrList(const std::string &key, const int32_t value, OpDef *opdef);
-GE_FUNC_VISIBILITY void AddOpAttrList(const std::string &key, const bool value, OpDef *opdef);
-GE_FUNC_VISIBILITY void AddOpAttrList(const std::string &key, const int64_t value, OpDef *opdef);
-
-GE_FUNC_VISIBILITY void AddOpAttrList(const std::string &key, const std::string &value, OpDef *opdef);
-
-GE_FUNC_VISIBILITY bool GetOpAttr(const std::string &key, std::string *value, const OpDef *opdef);
-GE_FUNC_VISIBILITY bool GetOpAttr(const std::string &key, int32_t *value, const OpDef *opdef);
-GE_FUNC_VISIBILITY bool GetOpAttr(const std::string &key, int64_t *value, const OpDef *opdef);
-GE_FUNC_VISIBILITY bool GetOpAttr(const std::string &key, uint32_t *value, const OpDef *opdef);
-GE_FUNC_VISIBILITY bool GetOpAttr(const std::string &key, float *value, const OpDef *opdef);
-GE_FUNC_VISIBILITY bool GetOpAttr(const std::string &key, double *value, const OpDef *opdef);
-GE_FUNC_VISIBILITY bool GetOpAttr(const std::string &key, bool *value, const OpDef *opdef);
-GE_FUNC_VISIBILITY bool GetOpAttr(const std::string &key, AttrDef_ListValue *value, const OpDef *opdef);
-
-GE_FUNC_VISIBILITY uint32_t GetOpAttrListSize(const std::string &key, std::string value, const OpDef *opdef);
-GE_FUNC_VISIBILITY uint32_t GetOpAttrListSize(const std::string &key, int32_t value, const OpDef *opdef);
-GE_FUNC_VISIBILITY uint32_t GetOpAttrListSize(const std::string &key, int64_t value, const OpDef *opdef);
-GE_FUNC_VISIBILITY uint32_t GetOpAttrListSize(const std::string &key, uint32_t value, const OpDef *opdef);
-GE_FUNC_VISIBILITY uint32_t GetOpAttrListSize(const std::string &key, float value, const OpDef *opdef);
-GE_FUNC_VISIBILITY uint32_t GetOpAttrListSize(const std::string &key, double value, const OpDef *opdef);
-GE_FUNC_VISIBILITY uint32_t GetOpAttrListSize(const std::string &key, bool value, const OpDef *opdef);
-
-GE_FUNC_VISIBILITY bool GetBytesAttr(const std::string &key, std::string *value, const OpDef *opdef);
-GE_FUNC_VISIBILITY bool GetBytesAttr(const std::string &key, std::string *value, const ModelDef *model_def);
-
-GE_FUNC_VISIBILITY void AddModelAttr(const std::string &key, const std::string &value, ModelDef *model_def);
-GE_FUNC_VISIBILITY void AddModelAttr(const std::string &key, const char *value, ModelDef *model_def);
-GE_FUNC_VISIBILITY void AddModelAttr(const char *key, const char *value, ModelDef *model_def);
-GE_FUNC_VISIBILITY void AddModelAttr(const std::string &key, const uint32_t value, ModelDef *model_def);
-GE_FUNC_VISIBILITY void AddModelAttr(const std::string &key, const int32_t value, ModelDef *model_def);
-GE_FUNC_VISIBILITY void AddModelAttr(const std::string &key, const int64_t value, ModelDef *model_def);
-GE_FUNC_VISIBILITY void AddModelAttr(const std::string &key, const float value, ModelDef *model_def);
-GE_FUNC_VISIBILITY void AddModelAttr(const std::string &key, const double value, ModelDef *model_def);
-GE_FUNC_VISIBILITY void AddModelAttr(const std::string &key, const bool value, ModelDef *model_def);
-GE_FUNC_VISIBILITY void AddModelAttr(const std::string &key, const void *value, size_t size, ModelDef *model_def);
-GE_FUNC_VISIBILITY void AddModelAttr(const std::string &key, const AttrDef_ListValue &value, ModelDef *model_def);
-
-GE_FUNC_VISIBILITY void AddModelAttrList(const std::string &key, const double value, ModelDef *model_def);
-GE_FUNC_VISIBILITY void AddModelAttrList(const std::string &key, const float value, ModelDef *model_def);
-GE_FUNC_VISIBILITY void AddModelAttrList(const std::string &key, const uint32_t value, ModelDef *model_def);
-GE_FUNC_VISIBILITY void AddModelAttrList(const std::string &key, const int32_t value, ModelDef *model_def);
-GE_FUNC_VISIBILITY void AddModelAttrList(const std::string &key, const std::string &value, ModelDef *model_def);
-
-GE_FUNC_VISIBILITY bool GetModelAttr(const std::string &key, std::string *value, const ModelDef *model_def);
-GE_FUNC_VISIBILITY bool GetModelAttr(const std::string &key, int32_t *value, const ModelDef *model_def);
-GE_FUNC_VISIBILITY bool GetModelAttr(const std::string &key, int64_t *value, const ModelDef *model_def);
-GE_FUNC_VISIBILITY bool GetModelAttr(const std::string &key, uint32_t *value, const ModelDef *model_def);
-GE_FUNC_VISIBILITY bool GetModelAttr(const std::string &key, float *value, const ModelDef *model_def);
-GE_FUNC_VISIBILITY bool GetModelAttr(const std::string &key, double *value, const ModelDef *model_def);
-GE_FUNC_VISIBILITY bool GetModelAttr(const std::string &key, bool *value, const ModelDef *model_def);
-GE_FUNC_VISIBILITY bool GetModelAttr(const std::string &key, AttrDef_ListValue *value, const ModelDef *model_def);
-
-GE_FUNC_VISIBILITY bool HasOpAttr(const OpDef *opdef, const std::string &attr_name);
-
-GE_FUNC_VISIBILITY void SetAttrDef(const std::string &value, AttrDef *out);
-GE_FUNC_VISIBILITY void SetAttrDef(const char *value, AttrDef *out);
-GE_FUNC_VISIBILITY void SetAttrDef(const uint32_t value, AttrDef *out);
-GE_FUNC_VISIBILITY void SetAttrDef(const int32_t value, AttrDef *out);
-GE_FUNC_VISIBILITY void SetAttrDef(const float value, AttrDef *out);
-GE_FUNC_VISIBILITY void SetAttrDef(const double value, AttrDef *out);
-GE_FUNC_VISIBILITY void SetAttrDef(const bool value, AttrDef *out);
-GE_FUNC_VISIBILITY void SetAttrList(const std::string &value, AttrDef *out);
-GE_FUNC_VISIBILITY void SetAttrList(const bool value, AttrDef *out);
-GE_FUNC_VISIBILITY void SetAttrList(const float value, AttrDef *out);
-GE_FUNC_VISIBILITY void SetAttrList(const double value, AttrDef *out);
-GE_FUNC_VISIBILITY void SetAttrList(const uint32_t value, AttrDef *out);
-
-GE_FUNC_VISIBILITY bool GetAttrDefValue(const std::string &key, std::string *value, const AttrDefMap &attr);
-GE_FUNC_VISIBILITY bool GetAttrDefValue(const std::string &key, int32_t *value, const AttrDefMap &attr);
-GE_FUNC_VISIBILITY bool GetAttrDefValue(const std::string &key, int64_t *value, const AttrDefMap &attr);
-GE_FUNC_VISIBILITY bool GetAttrDefValue(const std::string &key, uint32_t *value, const AttrDefMap &attr);
-GE_FUNC_VISIBILITY bool GetAttrDefValue(const std::string &key, float *value, const AttrDefMap &attr);
-GE_FUNC_VISIBILITY bool GetAttrDefValue(const std::string &key, double *value, const AttrDefMap &attr);
-GE_FUNC_VISIBILITY bool GetAttrDefValue(const std::string &key, bool *value, const AttrDefMap &attr);
-GE_FUNC_VISIBILITY bool GetAttrDefValue(const std::string &key, AttrDef_ListValue *value, const AttrDefMap &attr);
-GE_FUNC_VISIBILITY bool GetAttrDefValue(const std::string &key, NamedAttrs *&value, AttrDefMap *attr);
-GE_FUNC_VISIBILITY bool GetAttrDefValue(const std::string &key, const NamedAttrs *&value, const AttrDefMap &attr);
-
-GE_FUNC_VISIBILITY bool GetAttrDefListValue(const std::string &key, int idx, int32_t *value, const AttrDefMap &attr);
-GE_FUNC_VISIBILITY bool GetAttrDefListValue(const std::string &key, int idx, uint32_t *value, const AttrDefMap &attr);
-GE_FUNC_VISIBILITY bool GetAttrDefListValue(const std::string &key, int idx, float *value, const AttrDefMap &attr);
-GE_FUNC_VISIBILITY bool GetAttrDefListValue(const std::string &key, int idx, double *value, const AttrDefMap &attr);
+GE_FUNC_VISIBILITY void SetAttrDef(const std::string &value, domi::AttrDef *const out);
 }
-
-#endif  // INC_FRAMEWORK_COMMON_OP_ATTR_VALUE_UTIL_H_
+#endif  // INC_FRAMEWORK_COMMON_OP_ATTR_VALUE_UTIL_H_

inc/framework/common/op/ge_op_utils.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -20,10 +20,8 @@
 #include <memory>
 #include <vector>
 
-#include "common/op/attr_value_util.h"
-#include "register/register_types.h"
-#include "register/register_error_codes.h"
-#include "common/util.h"
+#include "framework/common/op/attr_value_util.h"
+#include "framework/common/util.h"
 #include "graph/attr_value.h"
 #include "graph/ge_tensor.h"
 #include "graph/node.h"
@@ -31,18 +29,16 @@
 #include "proto/insert_op.pb.h"
 
 namespace ge {
-using domi::Status;
 
 // Add Sub Mul
 GE_FUNC_VISIBILITY extern const uint32_t ADD_INPUT_NUM;
-GE_FUNC_VISIBILITY extern const uint32_t SUB_INPUT_NUM;
 GE_FUNC_VISIBILITY extern const uint32_t MUL_INPUT_NUM;
 
 // Permute
 GE_FUNC_VISIBILITY extern const int32_t PERMUTE_ORDER_NUM;
 
 // Ssd PriroBox
-GE_FUNC_VISIBILITY extern const double SSD_PRIORBOX_ASPECT_RATIO_VALUE;
+GE_FUNC_VISIBILITY extern const float64_t SSD_PRIORBOX_ASPECT_RATIO_VALUE;
 
 GE_FUNC_VISIBILITY extern const uint32_t STRIDEDSLICE_INPUT_NUM;
 
@@ -55,8 +51,8 @@ GE_FUNC_VISIBILITY extern const uint32_t SWITCH_DATA_INPUT;
 GE_FUNC_VISIBILITY extern const uint32_t SWITCH_PRED_INPUT;
 
 // Merge
-GE_FUNC_VISIBILITY extern const uint32_t MERGE_DATA_OUTPUT;
-GE_FUNC_VISIBILITY extern const uint32_t MERGE_INDEX_OUTPUT;
+GE_FUNC_VISIBILITY extern const int32_t MERGE_DATA_OUTPUT;
+GE_FUNC_VISIBILITY extern const int32_t MERGE_INDEX_OUTPUT;
 
 // FunctionOp
 GE_FUNC_VISIBILITY extern const uint32_t IF_COND_INPUT;
@@ -65,85 +61,36 @@ GE_FUNC_VISIBILITY extern const uint32_t FOR_LIMIT_INPUT;
 GE_FUNC_VISIBILITY extern const uint32_t FOR_DELTA_INPUT;
 GE_FUNC_VISIBILITY extern const uint32_t FOR_DATA_INPUT;
 
-GE_FUNC_VISIBILITY extern const int NORMAL_TENSOR_SIZE;
-
+GE_FUNC_VISIBILITY extern const int32_t NORMAL_TENSOR_SIZE;
+/*lint -e148*/
 class GE_FUNC_VISIBILITY OpUtils {
  public:
-  ///
-  /// @ingroup domi_ome
-  /// @brief Check whether check_value is in [min_enum_value, max_enum_value]
-  /// @return true Within
-  /// @return false out of range
-  //
-  static inline bool CheckEnumValid(int32_t check_value, int32_t min_enum_value, int32_t max_enum_value) {
-    return check_value < min_enum_value ? false : (check_value >= max_enum_value ? false : true);
-  }
-
-  ///
-  /// @ingroup domi_omg
-  /// @brief Determine whether to manually calculate the tensor size based on the values of format and dim
-  /// @param [in] format, Format information of the tensor
-  /// @param [in] real_dim_cnt, Tensor dim
-  /// @return true Manually calculate the size based on dim and datatype
-  /// @return false skip
-  ///
-  static bool IsComputDimsSize(const int32_t format, const uint32_t real_dim_cnt);
-
   ///
   /// @brief Extract AIPP parameters from AttrDefMap and splice them
   /// @param [in] aipp_attr attr of operator
   /// @param [out] aipp_params aipp parameters
   /// @return enum of tagCCAippInputFormat
   ///
-  static Status ConvertAippParams(const GeAttrValue::NamedAttrs &aipp_attr, domi::AippOpParams *aipp_params);
-  static Status TransferDim(const std::vector<int64_t> &dim, std::vector<int64_t> &dim_vector);
+
+  static Status ConvertAippParams(const GeAttrValue::NamedAttrs &aipp_attr, domi::AippOpParams &aipp_params);
   template <typename T>
-  static void SliceData(const std::vector<char *> &input, int64_t chunk_size, std::vector<char *> &output,
-                        int64_t begin, int64_t out_dim, int64_t stride);
+  static void SliceData(const std::vector<char_t *> &input, const int64_t chunk_size, std::vector<char_t *> &output,
+                        const int64_t begin, const int64_t out_dim, const int64_t stride);
   template <typename T>
-  static Status SetDataByDataType(size_t out_size, const std::vector<char *> &chunk_input,
-                                  const std::vector<char *> &chunk_output, GeTensor *output);
+  static Status SetDataByDataType(const size_t out_size, const std::vector<char_t *> &chunk_input,
+                                  const std::vector<char_t *> &chunk_output, GeTensor *const output);
   template <typename T>
-  static Status SetOutputSliceDataByDataType(void *data, int64_t data_size, const std::vector<int64_t> &input_dims,
-                                             const std::vector<int64_t> &begin, const std::vector<int64_t> &output_dims,
-                                             ge::GeTensor *output, const std::vector<int64_t> &stride);
-  static Status SetOutputSliceData(void *data, int64_t data_size, int32_t data_type, std::vector<int64_t> &input_dims,
-                                   std::vector<int64_t> &begin, std::vector<int64_t> &output_dims, ge::GeTensor *output,
-                                   std::vector<int64_t> &stride);
-
-  ///
-  /// @ingroup domi_omg
-  /// @brief Convert the convolutional weight data from [h, w, c, k] to [k, c, h, w]
-  /// @param [in] input Weight data in HWCK format
-  /// @param [in] H value of H dimension
-  /// @param [in] W value of W dimension
-  /// @param [in] C value of C dimension
-  /// @param [in] K value of K dimension
-  /// @param [out] output Data pointer after conversion. The format is KCHW.
-  ///
-  static void TransDataHWCK2KCHW(const void *input, int64_t H, int64_t W, int64_t C, int64_t K, void **output);
-  ///
-  /// @ingroup domi_omg
-  /// @brief Converts the convolutional weight data from [k, c, h, w] to [h, w, c, k].
-  /// @param [in] input Weight data in HWCK format
-  /// @param [in] K value of K dimension
-  /// @param [in] C value of C dimension
-  /// @param [in] H value of H dimension
-  /// @param [in] W value of W dimension
-  /// @param [out] output Data pointer after conversion. The format is HWCK
-  ///
-  static void TransDataKCHW2HWCK(const void *input, int64_t K, int64_t C, int64_t H, int64_t W, void *output);
-
-  static vector<ConstGeTensorPtr> GetWeights(const ge::Node &node);
-  static vector<ConstGeTensorPtr> GetWeights(ge::ConstNodePtr node);
-  static vector<GeTensorPtr> MutableWeights(const ge::Node &node);
-  static vector<GeTensorPtr> MutableWeights(const ge::NodePtr node);
-  static Status SetWeights(ge::Node &node, const vector<ge::GeTensorPtr> &weights);
-  static Status SetWeights(ge::NodePtr node, const vector<ge::GeTensorPtr> &weights);
-  static Status GetShapeDataFromConstTensor(const ConstGeTensorPtr &tensor, DataType type, std::vector<int64_t> &dims);
-
- private:
-  static uint32_t GetRealDimCnt(const GeTensorDesc &tensor_desc);
+  static Status SetOutputSliceDataByDataType(void *const data, const int64_t data_size,
+                                             const std::vector<int64_t> &input_dims, const std::vector<int64_t> &begin,
+                                             const std::vector<int64_t> &output_dims, ge::GeTensor *const output,
+                                             const std::vector<int64_t> &stride);
+  static Status SetOutputSliceData(void *const data, const int64_t data_size, const int32_t data_type,
+                                   const std::vector<int64_t> &input_dims, const std::vector<int64_t> &begin,
+                                   const std::vector<int64_t> &output_dims, GeTensor *const output,
+                                   const std::vector<int64_t> &stride);
+  static Status GetShapeDataFromConstTensor(const ConstGeTensorPtr &tensor, const DataType type,
+                                            std::vector<int64_t> &dims);
 };
+/*lint +e148*/
 }  // namespace ge
 #endif  // INC_FRAMEWORK_COMMON_OP_GE_OP_UTILS_H_

inc/framework/common/op/op_parser_util.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -17,9 +17,9 @@
 #ifndef INC_FRAMEWORK_COMMON_OP_OP_PARSER_UTIL_H_
 #define INC_FRAMEWORK_COMMON_OP_OP_PARSER_UTIL_H_
 
-#include <limits.h>
-#include <math.h>
-#include <stdint.h>
+#include <climits>
+#include <cmath>
+#include <cstdint>
 
 namespace ge {
 // general
@@ -30,7 +30,7 @@ const uint32_t NORMAL_OUTPUT_NUM = 1;
 const uint32_t NORMAL_WORKSPACE_NUM = 0;
 const int32_t NORMAL_1D_DIM_NUM = 1;
 const int32_t NORMAL_SCALE_DIM_NUM = 0;
-const int NORMAL_TENSOR_SIZE = 4;
+const int32_t NORMAL_TENSOR_SIZE = 4;
 const uint32_t DEFAULT_REAL_DIM_CNT = 4;
 
 // const
@@ -111,8 +111,8 @@ const int32_t ROIPOOLING_DEFAULT_SAMPLING_RATIO = -1;
 const int32_t DETECTIONOUTPUT_INPUT_SIZE = 3;
 const int32_t DETECTIONOUTPUT_OUTPUT_SIZE = 2;
 const int32_t DETECTIONOUTPUT_WORKSPACE_NUM = 1;
-const int DETECTIONOUTPUT_CLASS_NUM = 20;  // Number of background categories
-const int DETECTIONOUTPUT_NUM_CLASSES_DEFAULT_VALUE = 21;
+const int32_t DETECTIONOUTPUT_CLASS_NUM = 20;  // Number of background categories
+const int32_t DETECTIONOUTPUT_NUM_CLASSES_DEFAULT_VALUE = 21;
 const float DETECTIONOUTPUT_NMS_THRESHOLD_DEFAULT_VALUE = 0.3;
 const float DETECTIONOUTPUT_CONFIDENCE_THRESHOLD_DEFAULT_VALUE = 0.8;
 
@@ -128,8 +128,8 @@ const float PROPOSAL_SCALE_DIM_0_DEFAULT_VALUE = 8;
 const float PROPOSAL_SCALE_DIM_1_DEFAULT_VALUE = 16;
 const float PROPOSAL_SCALE_DIM_2_DEFAULT_VALUE = 32;
 const float PROPOSAL_MIN_SIZE_DEFAULT_VALUE = 16;
-const int PROPOSAL_PRE_NMS_TOPN_DEFAULT_VALUE = 6000;
-const int PROPOSAL_POST_NMS_TOPN_DEFAULT_VALUE = 304;
+const int32_t PROPOSAL_PRE_NMS_TOPN_DEFAULT_VALUE = 6000;
+const int32_t PROPOSAL_POST_NMS_TOPN_DEFAULT_VALUE = 304;
 const float PROPOSAL_NMS_THRESH_DEFAULT_VALUE = 0.7;
 const float PROPOSAL_FILTER_THRESH_DEFAULT_VALUE = 0;
 
@@ -150,7 +150,7 @@ const int32_t PERMUTE_WORKSPACE_NUM = 1;
 const int32_t PERMUTE_ORDER_NUM = 4;
 
 // Ssd normalize
-const int SSD_NORMALIZE_INPUT_SIZE = 1;
+const int32_t SSD_NORMALIZE_INPUT_SIZE = 1;
 const float SSD_NORMALIZE_EPS_DEFAULT_VALUE = 2e-7;
 
 // SsdPriroBox
@@ -163,9 +163,9 @@ const double SSD_PRIORBOX_VARIANCE_VALUE = 0.1;
 const double SSD_PRIORBOX_VARIANCE_SIZE_ONE = 1;
 const double SSD_PRIORBOX_VARIANCE_SIZE_FOUR = 4;
 const double SSD_PRIORBOX_ASPECT_RATIO_VALUE = 1.0;
-const int SSD_PRIOR_BOX_CODETYPE_CORNER_VALUE = 1;
-const int SSD_PRIOR_BOX_CODETYPE_CENTER_SIZE_VALUE = 2;
-const int SSD_PRIOR_BOX_CODETYPE_CORNER_SIZE_VALUE = 3;
+const int32_t SSD_PRIOR_BOX_CODETYPE_CORNER_VALUE = 1;
+const int32_t SSD_PRIOR_BOX_CODETYPE_CENTER_SIZE_VALUE = 2;
+const int32_t SSD_PRIOR_BOX_CODETYPE_CORNER_SIZE_VALUE = 3;
 
 // Ssd DetectionOutput
 const int32_t SSD_DETECTIONOUTPUT_INPUT_SIZE = 3;
@@ -205,8 +205,8 @@ const int32_t CHANNEL_AXPY_INPUT_DIM_SIZE = 4;
 const int32_t CHANNEL_AXPY_WORKSPACE_NUM = 1;
 
 // Psroi pooling
-const int PSROI_POOLING_INPUT_COUNT = 2;
-const int PSROI_POOLING_WORKSPACE_NUM = 1;
+const int32_t PSROI_POOLING_INPUT_COUNT = 2;
+const int32_t PSROI_POOLING_WORKSPACE_NUM = 1;
 
 // MaxPoolWithArgmax
 const uint32_t MAX_POOL_WITH_ARGMAX_OUTPUT_NUM = 2;
@@ -223,7 +223,7 @@ const int32_t ROIALIGN_DEFAULT_POOLED_W = 1;
 
 // Correlation
 const uint32_t CORRELATION_INPUT_NUM = 2;
-const int CORRELATION_WORKSPACE_NUM = 1;
+const int32_t CORRELATION_WORKSPACE_NUM = 1;
 
 // Detectionpostprocess
 const int32_t POSTPROCESS_INPUT_SIZE = 4;
@@ -394,15 +394,15 @@ const uint32_t ATTENTION_DECODER_WORKSPACE_NUM = 1;
 const uint32_t ATTENTION_DECODER_INPUT_DECODER_INPUTS = 0;
 const uint32_t ATTENTION_DECODER_INPUT_DECODER_INITIAL_HIDDEN = 1;
 
-const int ATTENTION_DECODER_ALGO_NORMAL = 0;
-const int ATTENTION_DECODER_SYMBOLS = 10000;
-const int ATTENTION_DECODER_EMBEDDING_SIZE = 128;
-const int ATTENTION_DECODER_ATTENTION_NUM_HIDDEN = 256;
-const int ATTENTION_DECODER_DECODER_NUM_HIDDEN = 128;
-const int ATTENTION_DECODER_DECODER_NUM_LAYERS = 2;
-const int ATTENTION_DECODER_RNN_UNBIDIRECTIONAL = 0;
-const int ATTENTION_DECODER_SEQLEN_VALUE = 57;
-const int ATTENTION_DECODER_GRU = 3;
+const int32_t ATTENTION_DECODER_ALGO_NORMAL = 0;
+const int32_t ATTENTION_DECODER_SYMBOLS = 10000;
+const int32_t ATTENTION_DECODER_EMBEDDING_SIZE = 128;
+const int32_t ATTENTION_DECODER_ATTENTION_NUM_HIDDEN = 256;
+const int32_t ATTENTION_DECODER_DECODER_NUM_HIDDEN = 128;
+const int32_t ATTENTION_DECODER_DECODER_NUM_LAYERS = 2;
+const int32_t ATTENTION_DECODER_RNN_UNBIDIRECTIONAL = 0;
+const int32_t ATTENTION_DECODER_SEQLEN_VALUE = 57;
+const int32_t ATTENTION_DECODER_GRU = 3;
 
 // Logicaland
 const int32_t LOGICAL_AND_INPUT_NUM = 2;

inc/framework/common/op_types.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -20,6 +20,8 @@
 #include <set>
 #include <string>
 
+#include "graph/types.h"
+
 namespace ge {
 class GE_FUNC_VISIBILITY OpTypeContainer {
  public:
@@ -29,11 +31,12 @@ class GE_FUNC_VISIBILITY OpTypeContainer {
   }
   ~OpTypeContainer() = default;
 
-  void Register(const std::string &op_type) { op_type_list_.insert(op_type); }
+  bool Register(const std::string &op_type) {
+    return op_type_list_.insert(op_type).second;
+  }
 
   bool IsExisting(const std::string &op_type) {
-    auto iter_find = op_type_list_.find(op_type);
-    return iter_find != op_type_list_.end();
+    return op_type_list_.find(op_type) != op_type_list_.end();
   }
 
  protected:
@@ -42,21 +45,14 @@ class GE_FUNC_VISIBILITY OpTypeContainer {
  private:
   std::set<std::string> op_type_list_;
 };
-
-class GE_FUNC_VISIBILITY OpTypeRegistrar {
- public:
-  explicit OpTypeRegistrar(const std::string &op_type) { OpTypeContainer::Instance()->Register(op_type); }
-  ~OpTypeRegistrar() {}
-};
+}  // namespace ge
 
 #define REGISTER_OPTYPE_DECLARE(var_name, str_name) \
-  FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const char *var_name;
+  FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const char_t *var_name;
 
 #define REGISTER_OPTYPE_DEFINE(var_name, str_name) \
-  const char *var_name = str_name;                 \
-  const OpTypeRegistrar g_##var_name##_reg(str_name);
-
-#define IS_OPTYPE_EXISTING(str_name) (OpTypeContainer::Instance()->IsExisting(str_name))
-}  // namespace ge
+  const char_t *var_name = str_name;               \
+  const bool g_##var_name##_reg = OpTypeContainer::Instance()->Register(str_name);
 
+#define IS_OPTYPE_EXISTING(str_name) (ge::OpTypeContainer::Instance()->IsExisting(str_name))
 #endif  // INC_FRAMEWORK_COMMON_OP_TYPES_H_

inc/framework/common/profiling/ge_profiling.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -17,39 +17,15 @@
 #ifndef INC_FRAMEWORK_COMMON_GE_PROFILING_H_
 #define INC_FRAMEWORK_COMMON_GE_PROFILING_H_
 
-#include "ge/ge_api_error_codes.h"
-#include "toolchain/prof_callback.h"
+#include "external/ge/ge_api_error_codes.h"
 #include "runtime/base.h"
 
-const int MAX_DEV_NUM = 64;
-
-enum ProfCommandHandleType {
-  kProfCommandhandleInit = 0,
-  kProfCommandhandleStart,
-  kProfCommandhandleStop,
-  kProfCommandhandleFinalize,
-  kProfCommandhandleModelSubscribe,
-  kProfCommandhandleModelUnsubscribe
-};
-
-struct ProfCommandHandleData {
-  uint64_t profSwitch;
-  uint32_t devNums;  // length of device id list
-  uint32_t devIdList[MAX_DEV_NUM];
-  uint32_t modelId;
-};
-
-GE_FUNC_VISIBILITY ge::Status RegProfCtrlCallback(MsprofCtrlCallback func);
-GE_FUNC_VISIBILITY ge::Status RegProfSetDeviceCallback(MsprofSetDeviceCallback func);
-GE_FUNC_VISIBILITY ge::Status RegProfReporterCallback(MsprofReporterCallback func);
-GE_FUNC_VISIBILITY ge::Status ProfCommandHandle(ProfCommandHandleType type, void *data, uint32_t len);
-
 ///
 /// @brief Output the profiling data of single operator in Pytorch, and does not support multithreading
 /// @return Status result
 ///
-GE_FUNC_VISIBILITY ge::Status ProfSetStepInfo(uint64_t index_id, uint16_t tag_id, rtStream_t stream);
+GE_FUNC_VISIBILITY ge::Status ProfSetStepInfo(const uint64_t index_id, const uint16_t tag_id, rtStream_t const stream);
 
-GE_FUNC_VISIBILITY ge::Status ProfGetDeviceFormGraphId(uint32_t graph_id, uint32_t &device_id);
+GE_FUNC_VISIBILITY ge::Status ProfGetDeviceFormGraphId(const uint32_t graph_id, uint32_t &device_id);
 
 #endif  // INC_FRAMEWORK_COMMON_GE_PROFILING_H_

inc/framework/common/profiling/ge_runner_profiling.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.

inc/framework/common/profiling_definitions.h

@@ -0,0 +1,220 @@
+/**
+ * Copyright 2021 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef AIR_CXX_PROFILING_DEFINITIONS_H
+#define AIR_CXX_PROFILING_DEFINITIONS_H
+#include <string>
+#include <iostream>
+#include <mutex>
+#include <unordered_map>
+#include "graph/profiler.h"
+#include "external/ge/ge_api_types.h"
+#include "toolchain/prof_callback.h"
+namespace ge {
+namespace profiling {
+enum {
+  kAclCompileAndExecute,
+  kAclMatchOpModel,
+  kAclMatchStaticOpModel,
+  kAclMatchDynamicOpModel,
+  kAclExecuteAsync,
+  kAclLoadSingleOp,
+  kAclBuildOpModel,
+  kInferShape,
+  kTiling,
+  kUpdateShape,
+  kConstPrepare,
+  kInitHybridExecuteArgs,
+  kInitInferShapeContext,
+  kDestroyInferShapeContext,
+  kResetSubgraphExecutor,
+  kCommitInferShapeTask,
+  kDeviceToHost,
+  kPrepareTask,
+  kLaunchTask,
+  kCommitTilingTask,
+  kAtomic,
+  kKernelLaunchPrepare,
+  kRtKernelLaunch,
+  kRtEventCreateRecord,
+  kRtEventSync,
+  kRtEventDestroy,
+  kRtStreamSync,
+  kOpExecute,
+  kModelExecute,
+  kAllocMem,
+  kCopyH2D,
+  kPrepareNode,
+  kWaitForPrepareDone,
+  kPropgateOutputs,
+  kOnNodeDoneCallback,
+  kValidateInputTensor,
+  kAfterExecuted,
+  kRtEventSychronize,
+  kInferShapeWaitDependShape,
+  kInferShapeWaitInputTensor,
+  kInferShapeCallInferFunc,
+  kInferShapePropgate,
+  // v2 control node
+  kSelectBranch,
+  kExecuteSubGraph,
+  kInitSubGraphExecutor,
+  // fuzz compile
+  kSelectBin,
+  kFindCompileCache,
+  kAddCompileCache,
+  kFuzzCompileOp,
+  kCalcRuningParam,
+  kGenTask,
+  kRegisterBin,
+
+  // FFTS Plus
+  kFftsPlusPreThread,
+  kFftsPlusNodeThread,
+  kFftsPlusInferShape,
+  kOpFftsCalculateV2,
+  kInitThreadRunInfo,
+  kFftsPlusGraphSchedule,
+  kKnownGetAddrAndPrefCnt,
+  kKernelGetAddrAndPrefCnt,
+  kUpdateAddrAndPrefCnt,
+  kInitOpRunInfo,
+  kGetAutoThreadParam,
+  kAllocateOutputs,
+  kAllocateWorkspaces,
+  kInitTaskAddrs,
+  kInitThreadRunParam,
+  kUpdateTaskAndCache,
+  kFftsPlusTaskLaunch,
+
+  // Add new definitions here
+  kProfilingIndexEnd
+};
+constexpr uint64_t kInvalidHashId = 0UL;
+
+class ProfilingContext {
+ public:
+  static bool IsDumpToStdEnabled();
+  static ProfilingContext &GetInstance();
+  ProfilingContext();
+  ~ProfilingContext();
+
+  /*
+   * 还有一种思路是`IsEnabled`只判断profiler_是否为空指针，不再设置单独的enabled标记位，这样可以少一个标记位。
+   * 但是这么做就意味着，profiler_实例在未使能profiling时，必须是空指针状态。
+   * 为了性能考虑，profiling机制在编译和加载时，就会调用`RegisterString`，向profiler_注册字符串，后续执行时，只会使用注册好的index了。
+   * 因此存在一种场景：编译时并未使能profiling（因为编译时间很长，使能profiling也无法真实反应执行时的耗时状态），
+   * 因此编译时注册字符串的动作并没有生效。在执行时，动态的打开了profiling，这种场景下，执行时无法拿到注册后字符串
+   */
+  bool IsEnabled() const noexcept {
+    return enabled_ && (profiler_ != nullptr);
+  }
+  void SetEnable() noexcept {
+    enabled_ = true;
+  }
+  void SetDisable() noexcept {
+    enabled_ = false;
+  }
+
+  void RecordCurrentThread(const int64_t element, const int64_t event, const EventType et,
+                           const std::chrono::time_point<std::chrono::system_clock> time_point) {
+    if (IsEnabled()) {
+      profiler_->RecordCurrentThread(element, event, et, time_point);
+    }
+  }
+
+  void RecordCurrentThread(const int64_t element, const int64_t event, const EventType et) {
+    RecordCurrentThread(element, event, et, std::chrono::system_clock::now());
+  }
+
+  const Profiler *GetProfiler() const {
+    return profiler_.get();
+  }
+
+  void Dump(std::ostream &out_stream) const {
+    if (IsEnabled()) {
+      profiler_->Dump(out_stream);
+    } else {
+      out_stream << "Profiling not enable, skip to dump" << std::endl;
+    }
+  }
+
+  void DumpToStdOut() const {
+    Dump(std::cout);
+  }
+
+  void Reset() {
+    if (IsEnabled()) {
+      profiler_->Reset();
+    }
+  }
+
+  int64_t RegisterString(const std::string &str);
+  int64_t RegisterStringHash(const uint64_t hash_id, const std::string &str);
+  void UpdateElementHashId();
+  static Status QueryHashId(const std::string &src_str, uint64_t &hash_id);
+  size_t GetRegisterStringNum() const {
+    return strings_to_index_.size();
+  }
+
+  void Init();
+
+ private:
+  void UpdateHashByStr(const std::string &str, const uint64_t hash);
+
+ private:
+  bool inited_;
+  bool enabled_;
+  int64_t str_index_;
+  std::unordered_map<std::string, int64_t> strings_to_index_;
+  std::mutex strings_to_index_mutex_;
+  std::unique_ptr<Profiler> profiler_;
+};
+
+class ScopeProfiler {
+ public:
+  ScopeProfiler(const int64_t element, const int64_t event) : element_(element), event_(event) {
+    if (ProfilingContext::GetInstance().IsEnabled()) {
+      start_trace_ = std::chrono::system_clock::now();
+    }
+  }
+  ~ScopeProfiler() {
+    if (ProfilingContext::GetInstance().IsEnabled()) {
+      ProfilingContext::GetInstance().RecordCurrentThread(element_, event_, EventType::kEventStart, start_trace_);
+      ProfilingContext::GetInstance().RecordCurrentThread(element_, event_, EventType::kEventEnd);
+    }
+  }
+  void SetElement(const int64_t element) {
+    element_ = element;
+  }
+
+ private:
+  std::chrono::time_point<std::chrono::system_clock> start_trace_;
+  int64_t element_;
+  int64_t event_;
+};
+}  // namespace profiling
+}  // namespace ge
+#define PROFILING_START(element, event)                                                  \
+  ge::profiling::ProfilingContext::GetInstance().RecordCurrentThread((element), (event), \
+                                                                     ge::profiling::EventType::kEventStart)
+#define PROFILING_END(element, event)                                                    \
+  ge::profiling::ProfilingContext::GetInstance().RecordCurrentThread((element), (event), \
+                                                                     ge::profiling::EventType::kEventEnd)
+#define PROFILING_SCOPE(element, event) ge::profiling::ScopeProfiler profiler((element), (event))
+#define PROFILING_SCOPE_CONST(element, event) const ge::profiling::ScopeProfiler profiler((element), (event))
+#define PROFILING_SCOPE_ELEMENT(element) profiler.SetElement((element))
+#endif  // AIR_CXX_PROFILING_DEFINITIONS_H

inc/framework/common/runtime_tensor_desc.h

@@ -0,0 +1,38 @@
+/**
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021-2022. All rights reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#ifndef INC_FRAMEWORK_COMMON_RUNTIME_TENSOR_DESC_H_
+#define INC_FRAMEWORK_COMMON_RUNTIME_TENSOR_DESC_H_
+
+#include <cstdint>
+
+namespace ge {
+constexpr int64_t kMaxDimSize = 32;
+
+#pragma pack(push, 1)
+struct RuntimeTensorDesc {
+  uint64_t data_addr;
+  int64_t data_offset_size;
+  int64_t dtype;
+  int64_t shape[kMaxDimSize + 1];           // shape:Dim_Num|DIM0|DIM1|...|DIM31
+  int64_t original_shape[kMaxDimSize + 1];  // original_shape:Dim_Num|DIM0|DIM1|...|DIM31
+  int64_t format;
+  int64_t sub_format;
+  uint8_t reserved[456];  // padding to 1024 bytes
+};
+#pragma pack(pop)
+}  // namespace ge
+
+#endif  // INC_FRAMEWORK_COMMON_RUNTIME_TENSOR_DESC_H_

inc/framework/common/scope_guard.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -18,14 +18,15 @@
 #define INC_FRAMEWORK_COMMON_SCOPE_GUARD_H_
 
 #include <functional>
-#include <iostream>
 
 /// Usage:
 /// Acquire Resource 1
 /// MAKE_GUARD([&] { Release Resource 1 })
 /// Acquire Resource 2
 // MAKE_GUARD([&] { Release Resource 2 })
-#define GE_MAKE_GUARD(var, callback) ScopeGuard make_guard_##var(callback)
+#define GE_MAKE_GUARD(var, callback) const ::ge::ScopeGuard const_guard_##var(callback)
+
+#define GE_DISMISSABLE_GUARD(var, callback) ::ge::ScopeGuard make_guard_##var(callback)
 #define GE_DISMISS_GUARD(var) make_guard_##var.Dismiss()
 
 namespace ge {
@@ -42,13 +43,16 @@ class GE_FUNC_VISIBILITY ScopeGuard {
       if (on_exit_scope_ != nullptr) {
         try {
           on_exit_scope_();
-        } catch (std::bad_function_call &e) { }
-          catch (...) { }
+        } catch (std::bad_function_call &) {
+        } catch (...) {
+        }
       }
     }
   }
 
-  void Dismiss() { dismissed_ = true; }
+  void Dismiss() {
+    dismissed_ = true;
+  }
 
  private:
   std::function<void()> on_exit_scope_;

inc/framework/common/string_util.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -39,24 +39,27 @@
 #include <sstream>
 #include <string>
 #include <vector>
+#include "graph/types.h"
 
 namespace ge {
 class GE_FUNC_VISIBILITY StringUtils {
  public:
   static std::string &Ltrim(std::string &s) {
 #if __cplusplus >= 201103L
-    (void)s.erase(s.begin(), std::find_if(s.begin(), s.end(), [](int c) { return !std::isspace(c); }));
+    (void)s.erase(s.begin(), std::find_if(s.begin(), s.end(), [](const int32_t c) { return std::isspace(c) == 0; }));
 #else
-    (void)s.erase(s.begin(), std::find_if(s.begin(), s.end(), std::not1(std::ptr_fun<int, int>(std::isspace))));
+    (void)s.erase(s.begin(), std::find_if(s.begin(), s.end(), std::not1(std::ptr_fun<int32_t, int32_t>(std::isspace))));
 #endif
     return s;
   }
   // lint -esym(551,*)
-  static std::string &Rtrim(std::string &s) {  /*lint !e618*/
+  static std::string &Rtrim(std::string &s) { /*lint !e618*/
 #if __cplusplus >= 201103L
-    (void)s.erase(s.begin(), std::find_if(s.begin(), s.end(), [](int c) { return !std::isspace(c); }));
+    (void)s.erase(std::find_if(s.rbegin(), s.rend(), [](const int32_t c) { return std::isspace(c) == 0; }).base(),
+                  s.end());
 #else
-    (void)s.erase(std::find_if(s.rbegin(), s.rend(), std::not1(std::ptr_fun<int, int>(std::isspace))).base(), s.end());
+    (void)s.erase(std::find_if(s.rbegin(), s.rend(), std::not1(std::ptr_fun<int32_t, int32_t>(std::isspace))).base(),
+                  s.end());
 #endif
     return s;
   }
@@ -67,7 +70,9 @@ class GE_FUNC_VISIBILITY StringUtils {
   ///  @param [in] string to be trimmed
   ///  @return string after trim
   ///
-  static std::string &Trim(std::string &s) { return Ltrim(Rtrim(s)); }
+  static std::string &Trim(std::string &s) {
+    return Ltrim(Rtrim(s));
+  }
 
   ///
   ///  @ingroup domi_common
@@ -76,8 +81,8 @@ class GE_FUNC_VISIBILITY StringUtils {
   ///  @param [in] delim  separator
   ///  @return string array after segmentation
   ///
-  static std::vector<std::string> Split(const std::string &str, char delim) {
-    std::vector<std::string> elems;
+  static std::vector<std::string, std::allocator<std::string>> Split(const std::string &str, const char_t delim) {
+    std::vector<std::string, std::allocator<std::string>> elems;
 
     if (str.empty()) {
       elems.emplace_back("");
@@ -91,8 +96,8 @@ class GE_FUNC_VISIBILITY StringUtils {
       elems.push_back(item);
     }
 
-    auto str_size = str.size();
-    if (str_size > 0 && str[str_size - 1] == delim) {
+    const auto str_size = str.size();
+    if ((str_size > 0U) && (str[str_size - 1U] == delim)) {
       elems.emplace_back("");
     }
 
@@ -104,13 +109,13 @@ class GE_FUNC_VISIBILITY StringUtils {
   ///  @param [in] s path name
   ///  @return file name
   ///
-  static std::string GetFileName(std::string &s) {
+  static std::string GetFileName(const std::string &s) {
     if (s.empty()) {
       return "";
     }
-    std::vector<std::string> files = StringUtils::Split(s, '/');
+    const std::vector<std::string> files = StringUtils::Split(s, '/');
 
-    return files.empty() ? "" : files[files.size() - 1];
+    return files.empty() ? "" : files[files.size() - 1U];
   }
   ///
   ///  @ingroup domi_common
@@ -122,12 +127,13 @@ class GE_FUNC_VISIBILITY StringUtils {
   ///  @return string after replacement
   ///
   static std::string ReplaceAll(std::string str, const std::string &old_value, const std::string &new_value) {
-    std::string::size_type cur_pos = 0;
-    std::string::size_type old_length = old_value.length();
-    std::string::size_type new_length = new_value.length();
+    std::string::size_type cur_pos = 0U;
+    const std::string::size_type old_length = old_value.length();
+    const std::string::size_type new_length = new_value.length();
     // cycle replace
     for (; cur_pos != std::string::npos; cur_pos += new_length) {
-      if ((cur_pos = str.find(old_value, cur_pos)) != std::string::npos) {
+      cur_pos = str.find(old_value, cur_pos);
+      if (cur_pos != std::string::npos) {
         (void)str.replace(cur_pos, old_length, new_value);
       } else {
         break;
@@ -145,7 +151,7 @@ class GE_FUNC_VISIBILITY StringUtils {
   ///  @return if the value is a prefix, true is returned. Otherwise, false is returned
   ///
   static bool StartWith(const std::string &str, const std::string str_x) {
-    return ((str.size() >= str_x.size()) && (str.compare(0, str_x.size(), str_x) == 0));
+    return ((str.size() >= str_x.size()) && (str.compare(0U, str_x.size(), str_x) == 0));
   }
 
   ///
@@ -156,14 +162,14 @@ class GE_FUNC_VISIBILITY StringUtils {
   ///  @param [in] ... format Filling Content
   ///  @return formatted string
   ///
-  static std::string FormatString(const char *format, ...) {
-    const uint32_t MAX_BUFFER_LEN = 1024;  // the stack memory plint check result must be less than 1024
+  static std::string FormatString(const char_t *const format, ...) {
+    const uint32_t MAX_BUFFER_LEN = 1024U;  // the stack memory plint check result must be less than 1024
     va_list args;
     va_start(args, format);
-    char buffer[MAX_BUFFER_LEN] = {0};
-    int32_t ret = vsnprintf_s(buffer, MAX_BUFFER_LEN, MAX_BUFFER_LEN - 1, format, args);
+    char_t buffer[MAX_BUFFER_LEN] = {};
+    const int32_t ret = vsnprintf_s(&buffer[0], MAX_BUFFER_LEN, MAX_BUFFER_LEN - 1U, format, args);
     va_end(args);
-    return ret > 0 ? buffer : "";
+    return (ret > 0) ? buffer : "";
   }
 };
 }  // namespace ge

inc/framework/common/taskdown_common.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -21,9 +21,9 @@
 
 namespace ge {
 
-const int CC_FUSION_OP_MAX = 32;
+const int32_t CC_FUSION_OP_MAX = 32;
 
-typedef enum tagCcStatus {
+enum class ccStatus_t {
   CC_STATUS_SUCCESS = 0,         /**< succ */
   CC_STATUS_NOT_INITIALIZED = 1, /**< not init */
   CC_STATUS_ALLOC_FAILED = 2,    /**< alloc mem failed */
@@ -33,10 +33,10 @@ typedef enum tagCcStatus {
   CC_STATUS_RUNTIME_ERROR = 6,   /**< runtime error */
   CC_STATUS_NOT_SUPPORTED = 7,   /**< unsupport error */
   CC_STATUS_INVALID_VALUE = 7,   /**< invalid value error for blas*/
-  CC_STATUS_RESERVED             /**< just for check */
-} ccStatus_t;
+  CC_STATUS_RESERVED = 8,        /**< just for check */
+};
 
-typedef enum tagccKernelType {
+enum class ccKernelType {
   CCE_AI_CORE = 0, /* cce aicore */
   CCE_AI_CPU = 1,  /* cce aicpu */
   TE = 2,          /* te operator*/
@@ -47,9 +47,9 @@ typedef enum tagccKernelType {
   CUST_AI_CPU = 7, /* custom aicpu*/
   HOST_CPU = 8,    /* host cpu */
   INVALID = 10000  /* unknown kernel type */
-} ccKernelType;
+};
 
-typedef struct tagOpContext {
+using ccOpContext = struct tagOpContext {
   ccKernelType kernelType;
   uint32_t opId;
   uint32_t kernelFuncId;
@@ -66,7 +66,28 @@ typedef struct tagOpContext {
   uint64_t genVariableBaseAddr;
   uint64_t genVariableBaseSize;
   uint64_t l2ctrlSize;
-} ccOpContext;
-}  // namespace ge
+};
+
+enum class tagOpTensorFormat { OP_TENSOR_FORMAT_NC1HWC0 = 0, OP_TENSOR_FORMAT_ND, OP_TENSOR_FORMAT_RESERVED };
 
+enum class tagOpDataType {
+  OP_DATA_FLOAT = 0,            /**< float type */
+  OP_DATA_HALF,                 /**< fp16 type */
+  OP_DATA_INT8,                 /**< int8 type */
+  OP_DATA_INT32,                /**< int32 type */
+  OP_DATA_UINT8,                /**< uint8 type */
+  OP_DATA_HALF_UINT16_PROPOSAL, /**< mixed type for proposal */
+  OP_DATA_RESERVED
+};
+
+// AICPU Tensor
+using ccAICPUTensor = struct tagOpTensor {
+  // real dim info
+  tagOpTensorFormat format;
+  tagOpDataType data_type;
+  int32_t dim_cnt;
+  int32_t mm;
+  int32_t dim[8];
+};
+}  // namespace ge
 #endif  // INC_FRAMEWORK_COMMON_TASKDOWN_COMMON_H_

inc/framework/common/types.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -17,28 +17,16 @@
 #ifndef INC_FRAMEWORK_COMMON_TYPES_H_
 #define INC_FRAMEWORK_COMMON_TYPES_H_
 
-#include <limits.h>
-#include <stdint.h>
-#include <algorithm>
+#include <cstdint>
 #include <map>
 #include <memory>
 #include <string>
-#include <utility>
-#include <vector>
 
 #include "framework/common/fmk_error_codes.h"
 #include "framework/common/fmk_types.h"
 #include "framework/common/op_types.h"
 #include "register/register_types.h"
 
-#if !defined(__ANDROID__) && !defined(ANDROID)
-#define DOMI_DYNAMIC_CAST static_cast
-#define DOMI_DYNAMIC_POINTER_CAST std::static_pointer_cast
-#else
-#define DOMI_DYNAMIC_CAST static_cast
-#define DOMI_DYNAMIC_POINTER_CAST std::static_pointer_cast
-#endif
-
 namespace ge {
 // dump
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string DUMP_MODEL;
@@ -51,54 +39,20 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string OP_DEB
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string OP_DEBUG_ATOMIC;
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string OP_DEBUG_ALL;
 
-// Supported public properties name
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string PROP_OME_START_TIME;  // Start time
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string PROP_OME_DUMP_PATH;   // Dump path
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string PROP_OME_LOG_PATH;    // Log path
-
 // Profile-related constants
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t CCE_PROFILE_ON;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t CCE_PROFILE_OFF;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string OME_PROFILE;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string CCE_PROFILE;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string RTS_PROFILE;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string PROFILER_JOBCTX;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string PROFILER_TARGET_PATH;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string RTS_PROFILE_PATH;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string PROFILE_STOP_KEY;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string PROFILE_STOP_VALUE;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::map<std::string, std::string> PROFILE_COMPONENT_MAP;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string PROFILE_CONFIG;
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string PROFILE_MODEL_ID;
 
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string MODEL_ATTR_TASKS;
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string MODEL_ATTR_TASK_GEN_BASE_ADDR;
+FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string MODEL_ATTR_TASK_GEN_HOST_BASE_ADDR;
+FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string MODEL_ATTR_HOST_MEMORY_SIZE;
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string MODEL_ATTR_TASK_GEN_WEIGHT_ADDR;
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string MODEL_ATTR_FUSION_MODEL_DEF;
-
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const int MODEL_MAX_SIZE;  // Max size of 2 GB minus 1 byte.
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint64_t FILE_HEADER_MAX_SIZE;  // Max size of 3 GB.
-
-#if !defined(__ANDROID__) && !defined(ANDROID)
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint64_t ALLOC_MEMORY_MAX_SIZE;  // Max size of 8 GB.
-#else
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint64_t ALLOC_MEMORY_MAX_SIZE;  // Max size of 512M.
-#endif
-
-template <typename K, typename V>
-static std::pair<V, K> flip_pair(const std::pair<K, V> &p) {
-  return std::pair<V, K>(p.second, p.first);
-}
-
-template <typename K, typename V>
-static std::map<V, K> flip_map(std::map<K, V> src) {
-  std::map<V, K> dst;
-  std::transform(src.begin(), src.end(), std::inserter(dst, dst.begin()), flip_pair<K, V>);
-  return dst;
-}
 
 REGISTER_OPTYPE_DECLARE(DATA, "Data");
 REGISTER_OPTYPE_DECLARE(AIPPDATA, "AippData");
+REGISTER_OPTYPE_DECLARE(QUEUE_DATA, "QueueData");
 REGISTER_OPTYPE_DECLARE(CONVOLUTION, "Convolution");
 REGISTER_OPTYPE_DECLARE(CORRELATION, "Correlation");
 REGISTER_OPTYPE_DECLARE(CORRELATIONV2, "Correlation_V2");
@@ -134,6 +88,7 @@ REGISTER_OPTYPE_DECLARE(DROPOUTDOMASKV3, "DropOutDoMaskV3");
 REGISTER_OPTYPE_DECLARE(DROPOUTDOMASKV3D, "DropOutDoMaskV3D");
 REGISTER_OPTYPE_DECLARE(SOFTMAXV2WITHDROPOUTDOMASKV3D, "SoftmaxV2WithDropOutDoMaskV3D");
 REGISTER_OPTYPE_DECLARE(DROPOUTGENMASK, "DropOutGenMask");
+REGISTER_OPTYPE_DECLARE(AXPYWITHSOFTMAXANDDROPOUTDOMASK, "AxpyWithSoftmaxAndDropOutDoMask");
 REGISTER_OPTYPE_DECLARE(CONCAT, "Concat");
 REGISTER_OPTYPE_DECLARE(ROIPOOLING, "ROIPooling");
 REGISTER_OPTYPE_DECLARE(PROPOSAL, "Proposal");
@@ -167,6 +122,10 @@ REGISTER_OPTYPE_DECLARE(SLICED, "SliceD");
 REGISTER_OPTYPE_DECLARE(FLOORDIV, "FloorDiv");
 REGISTER_OPTYPE_DECLARE(SQUEEZE, "Squeeze");
 REGISTER_OPTYPE_DECLARE(UNSQUEEZE, "Unsqueeze");
+REGISTER_OPTYPE_DECLARE(SQUEEZEV2, "SqueezeV2");
+REGISTER_OPTYPE_DECLARE(UNSQUEEZEV2, "UnsqueezeV2");
+REGISTER_OPTYPE_DECLARE(SQUEEZEV3, "SqueezeV3");
+REGISTER_OPTYPE_DECLARE(UNSQUEEZEV3, "UnsqueezeV3");
 REGISTER_OPTYPE_DECLARE(STRIDEDSLICE, "StridedSlice");
 REGISTER_OPTYPE_DECLARE(RANGE, "Range");
 REGISTER_OPTYPE_DECLARE(RPNPROPOSALS, "GenerateRpnProposals");
@@ -203,6 +162,7 @@ REGISTER_OPTYPE_DECLARE(_IF, "_If");
 REGISTER_OPTYPE_DECLARE(STATELESSIF, "StatelessIf");
 REGISTER_OPTYPE_DECLARE(IF, "If");
 REGISTER_OPTYPE_DECLARE(CASE, "Case");
+REGISTER_OPTYPE_DECLARE(STATELESSCASE, "StatelessCase");
 REGISTER_OPTYPE_DECLARE(_WHILE, "_While");
 REGISTER_OPTYPE_DECLARE(WHILE, "While");
 REGISTER_OPTYPE_DECLARE(STATELESSWHILE, "StatelessWhile");
@@ -339,10 +299,15 @@ REGISTER_OPTYPE_DECLARE(PLACEHOLDER, "PlaceHolder");
 REGISTER_OPTYPE_DECLARE(END, "End");
 REGISTER_OPTYPE_DECLARE(BASICLSTMCELL, "BasicLSTMCell");
 REGISTER_OPTYPE_DECLARE(GETNEXT, "GetNext");
+REGISTER_OPTYPE_DECLARE(ITERATOR, "Iterator");
+REGISTER_OPTYPE_DECLARE(ITERATORV2, "IteratorV2");
 REGISTER_OPTYPE_DECLARE(INITDATA, "InitData");
 REGISTER_OPTYPE_DECLARE(TRANSSHAPE, "TransShape")
 REGISTER_OPTYPE_DECLARE(REFIDENTITY, "RefIdentity");
 REGISTER_OPTYPE_DECLARE(BITCAST, "Bitcast");
+REGISTER_OPTYPE_DECLARE(GATHERSHAPES, "GatherShapes");
+REGISTER_OPTYPE_DECLARE(FLATTENV2, "FlattenV2");
+REGISTER_OPTYPE_DECLARE(FILECONSTANT, "FileConstant");
 
 // ANN dedicated operator
 REGISTER_OPTYPE_DECLARE(ANN_MEAN, "AnnMean");
@@ -460,6 +425,7 @@ REGISTER_OPTYPE_DECLARE(MODELEXIT, "ModelExit");
 REGISTER_OPTYPE_DECLARE(SEND, "Send");
 REGISTER_OPTYPE_DECLARE(RECV, "Recv");
 REGISTER_OPTYPE_DECLARE(ENDOFSEQUENCE, "EndOfSequence");
+REGISTER_OPTYPE_DECLARE(STARTOFSEQUENCE, "StartOfSequence");
 
 REGISTER_OPTYPE_DECLARE(LABELSET, "LabelSet");
 REGISTER_OPTYPE_DECLARE(LABELGOTO, "LabelGoto");
@@ -483,8 +449,6 @@ REGISTER_OPTYPE_DECLARE(ELU_GRAD, "EluGrad");
 REGISTER_OPTYPE_DECLARE(ADD_V2, "AddV2");
 REGISTER_OPTYPE_DECLARE(DATAFORMATDIMMAP, "DataFormatDimMap");
 REGISTER_OPTYPE_DECLARE(DATAFORMATVECPERMUTE, "DataFormatVecPermute");
-REGISTER_OPTYPE_DECLARE(BESSELI0e, "BesselI0e");
-REGISTER_OPTYPE_DECLARE(BESSELI1e, "BesselI1e");
 REGISTER_OPTYPE_DECLARE(DEQUANTIZE, "Dequantize");
 REGISTER_OPTYPE_DECLARE(APPLYADADELTA, "ApplyAdadelta");
 REGISTER_OPTYPE_DECLARE(APPLYADAGRAD, "ApplyAdagrad");
@@ -538,29 +502,11 @@ REGISTER_OPTYPE_DECLARE(GETDYNAMICDIMS, "GetDynamicDims");
 // profiling training trace node
 REGISTER_OPTYPE_DECLARE(PROFILINGTRAININGTRACE, "ProfilingTrainingTrace");
 
-enum InputMode { INPUT = 0, CONST_INPUT };
-
-// Definition of the processing status enum of the process module
-enum ModelProcessState {
-  INIT_STATE = 0,    // init status
-  WAIT_EVENT_STATE,  // Wait for the event status
-  IND_RSLT_STATE,    // The model execution result is being output to the high level
-  STOPPED_STATE,     // Model execution completed. The model enters this state after Model Manager::Stop
-  RESERVED_STATE,    // reserved
-};
-
-// Indicates the enun definition of the execution mode of the access module
-enum SysMode {
-  INFERENCE = 0,  // Normal, that is, Inference mode
-  DEBUG,          // Debug mode
-  TIME,           // Model execution time mode, including the execution time of each OP
-  STOP,           // STOP mode
-  RESET,          // RESET mode
-  PERFORMANCE,  // Impact of enabling the performance model: 1. The input data of the model is considered ready and does
-                // not need to be converted
-  ANDROID_DEBUG,  // Exports Android platform computing data
-  RESERVED,       // reserved
-};
+// Stack series
+REGISTER_OPTYPE_DECLARE(STACK, "Stack");
+REGISTER_OPTYPE_DECLARE(STACKPUSH, "StackPush");
+REGISTER_OPTYPE_DECLARE(STACKPOP, "StackPop");
+REGISTER_OPTYPE_DECLARE(STACKCLOSE, "StackClose");
 
 // @brief encryption type of the model file
 enum ModelEncryptType {
@@ -599,50 +545,22 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t MODEL_FIL
 ///
 /// @brief model name length
 ///
-static constexpr uint32_t MODEL_NAME_LENGTH = 32;
+constexpr uint32_t MODEL_NAME_LENGTH = 32U;
 
 ///
 /// @brief length of user-defined information
 ///
-static constexpr uint32_t USER_DEFINE_INFO_LENGTH = 32;
+constexpr uint32_t USER_DEFINE_INFO_LENGTH = 32U;
 
 ///
 /// @brief length of the model file signature
 ///
-static constexpr uint32_t MODEL_FILE_CHECKSUM_LENGTH = 64;
+constexpr uint32_t MODEL_FILE_CHECKSUM_LENGTH = 64U;
 
 ///
 /// @brief length of the reserved field in the model file header
 ///
-static constexpr uint32_t MODEL_FILE_RESERVED_LENGTH = 75;
-
-///
-/// @ingroup domi_omg
-/// @brief INPUT node type
-///
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string INPUT_TYPE;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string DUMMY_DATA;
-
-///
-/// @ingroup domi_omg
-/// @brief AIPP flag, indicating the aipp conv operator
-///
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string AIPP_CONV_FLAG;
-
-///
-/// @ingroup domi_omg
-/// @brief AIPP flag, indicating the aipp data operator
-///
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string AIPP_DATA_FLAG;
-
-// flag of the Data operator, indicating that the input will be input to the dynamic AIPP operator
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string INPUT_TO_DYNAMIC_AIPP;
-
-// records the W dimension of the model input corresponding to the dynamic AIPP
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string AIPP_RELATED_DATA_DIM_W;
-
-// H dimension of the model input corresponding to the dynamic AIPP
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string AIPP_RELATED_DATA_DIM_H;
+constexpr uint32_t MODEL_FILE_RESERVED_LENGTH = 75U;
 
 // DATA node type
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string DATA_TYPE;
@@ -655,10 +573,6 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string FRAMEW
 
 // DATA node type
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string ANN_DATA_TYPE;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string ANN_NETOUTPUT_TYPE;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string ANN_DEPTHCONV_TYPE;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string ANN_CONV_TYPE;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string ANN_FC_TYPE;
 // convolution node type
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string NODE_NAME_NET_OUTPUT;
 
@@ -667,161 +581,31 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string NODE_N
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string NODE_NAME_OP_DEBUG;
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string OP_TYPE_OP_DEBUG;
 
-// convolution node type
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string OP_TYPE_CONVOLUTION;
-// adds a convolutional node name for the hard AIPP
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string AIPP_CONV_OP_NAME;
 // delimiter of operator configuration items
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string OP_CONF_DELIMITER;
 
-// op attr name
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string ATTR_NAME_VALUE1;
-
-// op attr name, used to 6d_2_4d C channel
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string ATTR_NAME_INPUT_CVALUE;
-
-// op attr name
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string ATTR_NAME_VALUE1;
-
-// alpha default value
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const float ALPHA_DEFAULT_VALUE;
-
-// beta default value
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const float BETA_DEFAULT_VALUE;
-
-// coef default value
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const float COEF_DEFAULT_VALUE;
-
-// coef value of Relu6
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const float RELU6_COEF;
-
-// stride default value
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t STRIDE_DEFAULT_VALUE;
-
-// pad default value
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t PAD_DEFAULT_VALUE;
-
-// dilation default value
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const int DILATION_DEFAULT_VALUE;
-
-// kernel default value
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t KERNEL_DEFAULT_VALUE;
-
-// default conv Group Size
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t DEFAULT_CONV_GROUP;
-
-// default deconv adj
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t DEFAULT_DECONV_ADJ;
-
-// indicate num 1
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t NUM_ONE;
-
 // dim default size value
-static const int32_t DIM_DEFAULT_SIZE = 4;
-
-// the shape of c must be the mutiply of 16 for depthwise
-static const uint32_t DEPTHWISE_DIM_C_BASE_NUM = 16;
-
-// C1HWNCoC0 dim size
-static const int32_t DIM_C1HWNCoC0_SIZE = 6;
-// C1HWNCoC0 C0 value
-static const int C1HWCOC_C0_VALUE = 16;
-// spatial default dim size
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const int32_t SPATIAL_DIM_DEFAULT_SIZE;
+constexpr int32_t DIM_DEFAULT_SIZE = 4;
 
 // dim extension default value
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const int32_t DIM_DEFAULT_VALUE;
 
-// the first item in the weight list of opdef is filter
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const int32_t WEIGHT_FILTER_INDEX;
-
-// the second item in the weight list of opdef is bias.
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const int32_t WEIGHT_BIAS_INDEX;
-
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const int32_t TENSOR_ND_SUPPORT_SIZE;
-
 // default NCHW index
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t NCHW_DIM_N;
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t NCHW_DIM_C;
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t NCHW_DIM_H;
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t NCHW_DIM_W;
 
-// default C1HWNCoC0 index
-static const uint32_t C1HWNCoC0_DIM_C1 = 0;
-static const uint32_t C1HWNCoC0_DIM_H = 1;
-static const uint32_t C1HWNCoC0_DIM_W = 2;
-static const uint32_t C1HWNCoC0_DIM_N = 3;
-static const uint32_t C1HWNCoC0_DIM_Co = 4;
-static const uint32_t C1HWNCoC0_DIM_C0 = 5;
-
-// default KCHW index
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t KCHW_DIM_K;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t KCHW_DIM_C;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t KCHW_DIM_H;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t KCHW_DIM_W;
-
-// default HWCK index
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t HWCK_DIM_H;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t HWCK_DIM_W;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t HWCK_DIM_C;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t HWCK_DIM_K;
-
 // default NHWC index
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t NHWC_DIM_N;
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t NHWC_DIM_H;
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t NHWC_DIM_W;
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t NHWC_DIM_C;
 
-// default CHWN index
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t CHWN_DIM_N;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t CHWN_DIM_C;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t CHWN_DIM_H;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t CHWN_DIM_W;
-
-// default CHW index
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t CHW_DIM_C;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t CHW_DIM_H;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t CHW_DIM_W;
-
-// default HWC index
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t HWC_DIM_H;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t HWC_DIM_W;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t HWC_DIM_C;
-// default Pad index
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t PAD_H_HEAD;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t PAD_H_TAIL;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t PAD_W_HEAD;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t PAD_W_TAIL;
-
-// default window index
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t WINDOW_H;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t WINDOW_W;
-
-// default stride index
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t STRIDE_H;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t STRIDE_W;
-
-// default dilation index
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t DILATION_H;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t DILATION_W;
-
-// the num of XRBG channel
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t XRGB_CHN_NUM;
-
-// default tensor format
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const int DEFAULT_FORMAT;
-
-// default global pooling
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const bool DEFAULT_GLOBAL_POOLING;
-
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t MODEL_VERSION;  // model version 1.0
 
-// Number of inputs of the Eltwise operator
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const int ELTWISE_MIN_INPUT_SIZE;
-
 // flowctrl
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string NODE_NAME_STREAM_SWITCH;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string NODE_NAME_STREAM_ACTIVE;
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string NODE_NAME_FLOWCTRL_LOOP_PER_ITER;
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string NODE_NAME_FLOWCTRL_LOOP_COND;
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string NODE_NAME_FLOWCTRL_LOOP_INCREMENT;
@@ -833,41 +617,36 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t TRUE_STRE
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t STREAM_SWITCH_INPUT_NUM;
 
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string NODE_NAME_GLOBAL_STEP;
-FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string NODE_NAME_GLOBAL_STEP_ASSIGNADD;
 
-static const int PLATFORM_VERSION_LEN = 20;
+constexpr uint32_t PLATFORM_VERSION_LEN = 20U;
 
 // Definition of the file header of the model file
 struct ModelFileHeader {
-  uint32_t magic = MODEL_FILE_MAGIC_NUM;               // magic number of DOMI
-  uint32_t headsize = MODEL_FILE_HEAD_LEN;             // length of the model header. The value is fixed at 256
-  uint32_t version = MODEL_VERSION;                    // version 1.0
-  uint8_t checksum[MODEL_FILE_CHECKSUM_LENGTH] = {0};  // signature
-  uint32_t length = 0;  // Ciphertext length. In the non-encryption model, the length is the plaintext length.
-  uint8_t is_encrypt = ModelEncryptType::UNENCRYPTED;     // whether encrypted 0:not encrypt, 1:encrypt
-  uint8_t is_checksum = ModelCheckType::CHECK;            // whether to check the checksum
-  uint8_t modeltype = 0;                                  // 0：IR model 1：standard model 2: OM Tiny model
-  uint8_t genmode = 0;                                    // 0：offline generate 1：online generate
-  uint8_t name[MODEL_NAME_LENGTH] = {0};                  // Model name, which contains 32 characters
-  uint32_t ops = 0;                                       // Computing power (Kops)
-  uint8_t userdefineinfo[USER_DEFINE_INFO_LENGTH] = {0};  // User-defined information. The value contains 32 characters
-  uint32_t om_ir_version = 0;
-  uint32_t model_num = 0;
-  uint8_t platform_version[PLATFORM_VERSION_LEN] = {0};
-  uint8_t platform_type = {0};
-  uint8_t reserved[MODEL_FILE_RESERVED_LENGTH] = {0};  // Reserved field 75
+  uint32_t magic = MODEL_FILE_MAGIC_NUM;                // magic number of DOMI
+  uint32_t headsize = MODEL_FILE_HEAD_LEN;              // length of the model header. The value is fixed at 256
+  uint32_t version = MODEL_VERSION;                     // version 1.0
+  uint8_t checksum[MODEL_FILE_CHECKSUM_LENGTH] = {0U};  // signature
+  uint32_t length = 0U;  // Ciphertext length. In the non-encryption model, the length is the plaintext length.
+  // whether encrypted 0:not encrypt, 1:encrypt
+  uint8_t is_encrypt = static_cast<uint8_t>(ModelEncryptType::UNENCRYPTED);
+  uint8_t is_checksum = static_cast<uint8_t>(ModelCheckType::CHECK);  // whether to check the checksum
+  uint8_t modeltype = 0U;                                  // 0：IR model 1：standard model 2: OM Tiny model
+  uint8_t genmode = 0U;                                    // 0：offline generate 1：online generate
+  uint8_t name[MODEL_NAME_LENGTH] = {0U};                  // Model name, which contains 32 characters
+  uint32_t ops = 0U;                                       // Computing power (Kops)
+  uint8_t userdefineinfo[USER_DEFINE_INFO_LENGTH] = {0U};  // User-defined information. The value contains 32 characters
+  uint32_t om_ir_version = 0U;
+  uint32_t model_num = 0U;
+  uint8_t platform_version[PLATFORM_VERSION_LEN] = {0U};
+  uint8_t platform_type = {0U};
+  uint8_t reserved[MODEL_FILE_RESERVED_LENGTH] = {0U};  // Reserved field 75
 };
 
-static constexpr uint8_t TARGET_TYPE_LTTE_8BIT = 0;
-static constexpr uint8_t TARGET_TYPE_MINI_8BIT = 1;
-static constexpr uint8_t TARGET_TYPE_TINY_8BIT = 2;
-
-static constexpr int32_t PARTITION_TYPE_MODEL_DEF = 0;
-static constexpr int32_t PARTITION_TYPE_WEIGHTS = 1;
-static constexpr int32_t PARTITION_TYPE_TASK_INFO = 2;
+constexpr uint8_t TARGET_TYPE_LTTE_8BIT = 0U;
+constexpr uint8_t TARGET_TYPE_MINI_8BIT = 1U;
 
 // number of partitions in the current model
-static constexpr uint32_t PARTITION_SIZE = 5;
+constexpr uint32_t PARTITION_SIZE = 5U;
 
 enum ModelPartitionType { MODEL_DEF = 0, WEIGHTS_DATA, TASK_INFO, TBE_KERNELS, CUST_AICPU_KERNELS };
 
@@ -882,24 +661,9 @@ struct ModelPartitionTable {
   ModelPartitionMemInfo partition[0];
 };
 
-#define SIZE_OF_MODEL_PARTITION_TABLE(table) (sizeof(ModelPartitionTable) + sizeof(ModelPartitionMemInfo) * (table).num)
-
-static constexpr int32_t PTHREAD_CREAT_SUCCESS = 0;  // pthread_creat success
-
-// Filter format
-typedef enum tagDomiFilterFormat {
-  DOMI_FILTER_KCHW,  // KCHW
-  DOMI_FILTER_HWCK,  // HWCK
-  DOMI_FILTER_RESERVED
-} domiFilterFormat_t;
-
-// Const data trans type
-typedef enum tagDomiConstDataTransType {
-  DOMI_CONST_DATA_NOT_CHANGE = 0,  // No action is required
-  DOMI_CONST_DATA_TRANS_MATMUL,    // The const input to MatMul and needs to be transposed
-  DOMI_CONST_DATA_RESERVED
-} domiConstDataTransType_t;
-
+inline uint64_t SizeOfModelPartitionTable(const ModelPartitionTable &table) {
+  return sizeof(ModelPartitionTable) + (sizeof(ModelPartitionMemInfo) * static_cast<uint64_t>(table.num));
+}
 // mode of activation
 typedef enum tagDomiActivationMode {
   DOMI_ACTIVATION_SIGMOID = 0,   // sigmoid
@@ -919,190 +683,6 @@ typedef enum tagDomiActivationMode {
   DOMI_ACTIVATION_RESERVED
 } domiActivationMode_t;
 
-// mode of batchnorm
-typedef enum tagDomiBatchNormMode {
-  DOMI_BATCHNORM_PER_ACTIVATION = 0,  // bnScale, bnBias tensor dims are 1xCxHxW
-  DOMI_BATCHNORM_SPATIAL,             // bnScale, bnBias tensor dims are 1xCx1x1
-  DOMI_BATCHNORM_RESERVED
-} domiBatchNormMode_t;
-
-// eltwise mode
-typedef enum tagDomiEltwiseMode {
-  DOMI_ELTWISE_PROD = 0,  // prod
-  DOMI_ELTWISE_SUM,       // sum
-  DOMI_ELTWISE_MAX,       // max
-  DOMI_ELTWISE_RESERVED
-} domiEltwiseMode_t;
-
-// mode of padding
-typedef enum tagDomiPaddingMode {
-  DOMI_PADDING_CEIL = 0,      // Default padding mode
-  DOMI_PADDING_DIRECTASSIGN,  // Default padding mode: NOTSET
-  DOMI_PADDING_VALID,         // VALID padding mode
-  DOMI_PADDING_SAME,          // Padding values of 0 are always used
-  DOMI_PADDING_CEIL_NEW,      // Padding values of 0 are always used
-  DOMI_PADDING_VALID_NEW,     // Padding values of 0 are always used
-  DOMI_PADDING_SAME_NEW,      // Padding values of 0 are always used
-  DOMI_PADDING_RESERVED
-} domiPaddingMode_t;
-
-// algorithm of convolution forward
-typedef enum tagDomiConvolutionFwdAlgo {
-  DOMI_CONVOLUTION_FWD_ALGO_GEMM = 0,           // matrix gemm algo
-  DOMI_CONVOLUTION_FWD_ALGO_WINOGRAD,           // Winograd Transform algo
-  DOMI_CONVOLUTION_FWD_ALGO_GEMM_ACCU_FLOAT32,  // accumulate in L0c with FP32
-  DOMI_CONVOLUTION_FWD_ALGO_RESERVED
-} domiConvolutionFwdAlgo_t;
-
-typedef enum tagDomiFullConnectFwdAlgo {
-  DOMI_FULLCONNECT_FWD_ALGO_HALF = 0,  // accumulate in L0c with FP16
-  DOMI_FULLCONNECT_FWD_ALGO_FLOAT32    // accumulate in L0c with FP32
-} domiFullConnectFwdAlgo_t;
-
-typedef enum tagDomiPooingFwdAlgo {
-  DOMI_POOLING_FWD_ALGO_HALF = 0,  // accumulate in L0c with FP16
-  DOMI_POOLING_FWD_ALGO_FLOAT32    // accumulate in L0c with FP32
-} domiPooingFwdAlgo_t;
-
-// mode of convolution
-typedef enum tagDomiConvolutionMode {
-  DOMI_CONV_CONVOLUTION = 0,    // math convolution
-  DOMI_CONV_CROSS_CORRELATION,  // cross-correlation convolution
-  DOMI_CONV_DECONVOLUTION,      // deconvolution, also named transposed convolution
-  DOMI_CONV_MODE_DEPTHWISE,     // depthwise convolution
-  DOMI_CONV_MODE_RESERVED
-} domiConvolutionMode_t;
-
-// softmax mode
-typedef enum tagDomiSoftmaxMode {
-  DOMI_SOFTMAX_MODE_INSTANCE = 0,  // compute the softmax over all C, H, W for each N
-  DOMI_SOFTMAX_MODE_CHANNEL,       // compute the softmax over all C for each H, W, N
-  DOMI_SOFTMAX_MODE_HEIGHT,        // compute the softmax over all H for each N, C, W
-  DOMI_SOFTMAX_MODE_WIDTH,         // compute the softmax over all W for each N, C, H
-  DOMI_SOFTMAX_MODE_RESERVED
-} domiSoftmaxMode_t;
-
-// softmax algorithm
-typedef enum tagDomiSoftmaxAlgo {
-  DOMI_SOFTMAX_FAST = 0,  // straightforward implementation
-  DOMI_SOFTMAX_ACCURATE,  // subtract max from every point to avoid overflow
-  DOMI_SOFTMAX_LOG,       // perform the Log softmax operation to avoid overflow
-  DOMI_SOFTMAX_ACCURATE_FP32,
-  DOMI_SOFTMAX_RESERVED
-} domiSoftmaxAlgo_t;
-
-// algorithm of convolution backward
-typedef enum tagDomiConvolutionBwdAlgo {
-  DOMI_CONVOLUTION_BWD_ALGO_GEMM = 0,  // matrix gemm algo
-  DOMI_CONVOLUTION_BWD_ALGO_WINOGRAD,  // Winograd Transform algo
-  DOMI_CONVOLUTION_BWD_ALGO_RESERVED
-} domiConvolutionBwdAlgo_t;
-
-// mode of pooling
-typedef enum tagDomiPoolingMode {
-  DOMI_POOLING_MAX = 0,  // max pooling
-  DOMI_POOLING_AVG,      // average pooling
-  DOMI_POOLING_L2,       // L2 pooling
-  DOMI_POOLING_RESERVED
-} domiPoolingMode_t;
-
-// propagate Nan
-typedef enum tagDomiNanPropagation {
-  DOMI_NAN_NOT_PROPAGATE = 0,  // Nan numbers are not propagated
-  DOMI_NAN_PROPAGATE,          // Nan numbers are propagated
-  DOMI_NAN_PROPAGATE_RESERVED
-} domiNanPropagation_t;
-
-// mode of cropandresize
-typedef enum tagDomiCropAndResizeMode {
-  DOMI_RESIZE_METHOD_BILINEAR = 0,  // resize bilinear
-  DOMI_RESIZE_METHOD_NEAREST,       // resize nearest
-  DOMI_RESIZE_RESERVED
-} domiCropAndResizeMode_t;
-
-// yolo version
-typedef enum tagDomiYoloVersion { DOMI_YOLO_V2 = 1, DOMI_YOLO_V3, DOMI_YOLO_TRSERVED } domiYoloVersion_t;
-
-typedef enum tagDomiRNNScopePassType {
-  DOMI_STATIC_BIDIRECTIONAL_RNN_GENERAL_PASS = 0,
-  DOMI_DYNAMIC_BIDIRECTIONAL_RNN_GENERAL_PASS,
-  DOMI_DYNAMIC_BIDIRECTIONAL_RNN_BIDAF_PASS
-} domiRNNScopePassType;
-
-// RNNDataLayout
-typedef enum tagDomiRNNDataLayout {
-  DOMI_RNN_ND_TBX = 0,  // data[max_time,batch_size,Xt]
-  DOMI_RNN_ND_BTX,      // data[batch_size,max_time,Xt]
-  DOMI_RNN_5D_TX1BX,    // data[max_time,Xt,1,batch_size,Xt]
-  DOMI_RNN_5D_BX1TX,    // dataa[batch_size,Xt,1,max_time,Xt]
-  DOMI_RNN_4DTBX1,
-  DOMI_ENN_DL_RESERVED
-} domiRNNDataLayout_t;
-
-// RNNInputMode
-typedef enum tagDomiRNNInputMode { DOMI_RNN_LINEAR_INPUT = 0, DOMI_RNN_SKIP_INPUT } domiRNNInputMode_t;
-
-// RNNDirectionMode
-typedef enum tagDomiRNNDirectionMode { DOMI_RNN_UNIDIRECTIONAL = 0, DOMI_RNN_BIDIRECTIONAL } domiDirectionMode_t;
-
-typedef enum tagDomiPoolingCeilMode { DOMI_POOLING_FLOOR = 0, DOMI_POOLING_CEIL } domiPoolingCeilMode_t;
-
-// RNNMode
-typedef enum tagDomiRNNActivationMode {
-  DOMI_RNN_ACTIVATION_SIGMOID = 0,  // sigmoid
-  DOMI_RNN_ACTIVATION_TANH,         // tanh
-  DOMI_RNN_ACTIVATION_RELU,         // ReLU
-  DOMI_RNN_ACTIVATION_RELU1,        //  ReLU1
-  DOMI_RNN_ACTIVATION_RELU6,        //  ReLU6
-  DOMI_RNN_ACTIVATION_RESERVED
-} domiRNNActivationMode_t;
-
-typedef enum tagDomiRNNLSTMOutMode {
-  DOMI_RNN_LSTM_OUT_SEPARATE = 0,
-  DOMI_RNN_LSTM_OUT_CONCAT,
-  DOMI_RNN_LSTM_OUT_RESERVED
-} domiRNNLSTMOutPutMode_t;
-typedef enum tagDomiRNNLSTMStateOutMode {
-  DOMI_RNN_LSTM_STATE_OUT_SEPARATE = 0,
-  DOMI_RNN_LSTM_STATE_OUT_CONCAT_ALL,
-  DOMI_RNN_LSTM_STATE_OUT_RESERVED
-} domiRNNLSTMStateOutMode_t;
-
-typedef enum tagDomiRNNMode {
-  DOMI_RNN_RELU = 0,
-  DOMI_RNN_TANH,
-  DOMI_LSTM,
-  DOMI_GRU,
-  DOMI_RNN_MODE_RESERVED
-} domiRNNMode_t;
-
-typedef enum tagDomiResizeBilinearMode {
-  DOMI_RESIZE_OUTPUT_DIM_BY_ZOOM_FACTOR = 0,  // Output dimension specified by zoom factor
-  DOMI_RESIZE_OUTPUT_DIM_BY_SHRINK_FACTOR,    // specified by shrink factor
-  DOMI_RESIZE_OUTPUT_DIM_EXPLICIT,            // specified explicitly
-  DOMI_RESIZE_OUTPUT_DIM_RESERVED
-} domiResizeOutputDimMode_t;
-
-#pragma pack(1)  // single-byte alignment
-// DUMP file struct
-struct FileHeader {
-  int32_t Version;          // version
-  int32_t Output_Offset;    // output offset address
-  char Reserved[24] = {0};  // 24 bytes reserved
-};
-
-struct BasicInfo {
-  struct FileHeader header;  // file header
-  int32_t stream_id;         // stread id
-  uint64_t start_time;       // start time
-  uint64_t end_time;         // end time
-  uint32_t input_size;       // input memory size
-  uint32_t output_size;      // output memory size
-  uint32_t weight_size;      // weight Memory Size
-  uint32_t workspace_size;   // workspace
-  uint32_t total_size;       // total memory size
-};
-#pragma pack()  // Cancels single-byte alignment
 enum class MemorySizeCalcType { NORMAL = 0, ALWAYS_EMPTY };
 }  // namespace ge
 

inc/framework/common/util.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -14,33 +14,32 @@
  * limitations under the License.
  */
 
-#ifndef INC_FRAMEWORK_COMMON_UTIL_H_
-#define INC_FRAMEWORK_COMMON_UTIL_H_
+#ifndef AIR_INC_FRAMEWORK_COMMON_UTIL_H_
+#define AIR_INC_FRAMEWORK_COMMON_UTIL_H_
 
-#include <google/protobuf/text_format.h>
-#include <limits.h>
-#include <math.h>
+#include <cmath>
 #include <sstream>
 #include <string>
-#include <vector>
 
-#include "framework/common/debug/ge_log.h"
+#include <google/protobuf/text_format.h>
+#include "external/graph/types.h"
+#include "external/register/register.h"
 #include "framework/common/debug/log.h"
 #include "framework/common/scope_guard.h"
 #include "framework/common/ge_inner_error_codes.h"
-#include "mmpa/mmpa_api.h"
+#include "graph/detail/attributes_holder.h"
 
-#define GE_CHECK_POSITIVE_SIZE_RANGE(size)                    \
-  do {                                                        \
-    if (size <= 0) {                                          \
-      DOMI_LOGE("param[%s] is not a positive number", #size); \
-      return PARAM_INVALID;                                   \
-    }                                                         \
-  } while (0)
+#define GE_CHECK_POSITIVE_SIZE_RANGE(size)                             \
+  do {                                                                 \
+    if ((size) <= 0) {                                                 \
+      GELOGE(ge::FAILED, "param[%s] is not a positive number", #size); \
+      return PARAM_INVALID;                                            \
+    }                                                                  \
+  } while (false)
 
 #define CHECK_FALSE_EXEC(expr, exec_expr, ...) \
   {                                            \
-    bool b = (expr);                           \
+    const bool b = (expr);                     \
     if (!b) {                                  \
       exec_expr;                               \
     }                                          \
@@ -48,189 +47,244 @@
 
 // new ge marco
 // Encapsulate common resource releases
-#define GE_MAKE_GUARD_RTMEM(var)         \
-  GE_MAKE_GUARD(var, [&] {               \
-    if (var) GE_CHK_RT(rtFreeHost(var)); \
-  });
+#define GE_MAKE_GUARD_RTMEM(var)  \
+  GE_MAKE_GUARD(var, [&] {        \
+    if ((var) != nullptr) {       \
+      GE_CHK_RT(rtFreeHost(var)); \
+    }                             \
+  })
 
-#define GE_MAKE_GUARD_RTSTREAM(var)           \
-  GE_MAKE_GUARD(var, [&] {                    \
-    if (var) GE_CHK_RT(rtStreamDestroy(var)); \
-  });
+#define GE_MAKE_GUARD_RTSTREAM(var)    \
+  GE_MAKE_GUARD(var, [&] {             \
+    if ((var) != nullptr) {            \
+      GE_CHK_RT(rtStreamDestroy(var)); \
+    }                                  \
+  })
 
 // For propagating errors when calling a function.
-#define GE_RETURN_IF_ERROR(expr)         \
-  do {                                   \
-    const ::ge::Status _status = (expr); \
-    if (_status) return _status;         \
-  } while (0)
+#define GE_RETURN_IF_ERROR(expr)           \
+  do {                                     \
+    const ge::Status _chk_status = (expr); \
+    if (_chk_status != ge::SUCCESS) {      \
+      return _chk_status;                  \
+    }                                      \
+  } while (false)
 
 #define GE_RETURN_WITH_LOG_IF_ERROR(expr, ...) \
   do {                                         \
-    const ::ge::Status _status = (expr);       \
-    if (_status) {                             \
-      DOMI_LOGE(__VA_ARGS__);                  \
-      return _status;                          \
+    const ge::Status _chk_status = (expr);     \
+    if (_chk_status != ge::SUCCESS) {          \
+      GELOGE(ge::FAILED, __VA_ARGS__);         \
+      return _chk_status;                      \
     }                                          \
-  } while (0)
+  } while (false)
 
 // check whether the parameter is true. If it is, return FAILED and record the error log
 #define GE_RETURN_WITH_LOG_IF_TRUE(condition, ...) \
   do {                                             \
     if (condition) {                               \
-      DOMI_LOGE(__VA_ARGS__);                      \
+      GELOGE(ge::FAILED, __VA_ARGS__);             \
       return ge::FAILED;                           \
     }                                              \
-  } while (0)
+  } while (false)
 
 // Check if the parameter is false. If yes, return FAILED and record the error log
 #define GE_RETURN_WITH_LOG_IF_FALSE(condition, ...) \
   do {                                              \
-    bool _condition = (condition);                  \
+    const bool _condition = (condition);            \
     if (!_condition) {                              \
-      DOMI_LOGE(__VA_ARGS__);                       \
+      GELOGE(ge::FAILED, __VA_ARGS__);              \
       return ge::FAILED;                            \
     }                                               \
-  } while (0)
+  } while (false)
 
 // Checks whether the parameter is true. If so, returns PARAM_INVALID and records the error log
 #define GE_RT_PARAM_INVALID_WITH_LOG_IF_TRUE(condition, ...) \
   do {                                                       \
     if (condition) {                                         \
-      DOMI_LOGE(__VA_ARGS__);                                \
+      GELOGE(ge::FAILED, __VA_ARGS__);                       \
       return ge::PARAM_INVALID;                              \
     }                                                        \
-  } while (0)
+  } while (false)
 
 // Check if the parameter is false. If yes, return PARAM_INVALID and record the error log
 #define GE_RT_PARAM_INVALID_WITH_LOG_IF_FALSE(condition, ...) \
   do {                                                        \
-    bool _condition = (condition);                            \
+    const bool _condition = (condition);                      \
     if (!_condition) {                                        \
-      DOMI_LOGE(__VA_ARGS__);                                 \
+      GELOGE(ge::FAILED, __VA_ARGS__);                        \
       return ge::PARAM_INVALID;                               \
     }                                                         \
-  } while (0)
+  } while (false)
 
 // Check if the parameter is null. If yes, return PARAM_INVALID and record the error
-#define GE_CHECK_NOTNULL(val)                                                   \
-  do {                                                                          \
-    if (val == nullptr) {                                                       \
-      REPORT_INNER_ERROR("E19999", "Param:%s is nullptr, check invalid", #val); \
-      DOMI_LOGE("[Check][Param:%s]null is invalid.", #val);                     \
-      return ge::PARAM_INVALID;                                                 \
-    }                                                                           \
-  } while (0)
+#define GE_CHECK_NOTNULL(val, ...)                                                          \
+  do {                                                                                      \
+    if ((val) == nullptr) {                                                                 \
+      REPORT_INNER_ERROR("E19999", "Param:" #val " is nullptr, check invalid" __VA_ARGS__); \
+      GELOGE(ge::FAILED, "[Check][Param:" #val "]null is invalid" __VA_ARGS__);             \
+      return ge::PARAM_INVALID;                                                             \
+    }                                                                                       \
+  } while (false)
 
 // Check if the parameter is null. If yes, just return and record the error
-#define GE_CHECK_NOTNULL_JUST_RETURN(val)             \
-  do {                                                \
-    if (val == nullptr) {                             \
-      DOMI_LOGE("param[%s] must not be null.", #val); \
-      return;                                         \
-    }                                                 \
-  } while (0)
+#define GE_CHECK_NOTNULL_JUST_RETURN(val)                      \
+  do {                                                         \
+    if ((val) == nullptr) {                                    \
+      GELOGE(ge::FAILED, "param[%s] must not be null.", #val); \
+      return;                                                  \
+    }                                                          \
+  } while (false)
 
 // Check whether the parameter is null. If so, execute the exec_expr expression and record the error log
-#define GE_CHECK_NOTNULL_EXEC(val, exec_expr)         \
-  do {                                                \
-    if (val == nullptr) {                             \
-      DOMI_LOGE("param[%s] must not be null.", #val); \
-      exec_expr;                                      \
-    }                                                 \
-  } while (0)
+#define GE_CHECK_NOTNULL_EXEC(val, exec_expr)                  \
+  do {                                                         \
+    if ((val) == nullptr) {                                    \
+      GELOGE(ge::FAILED, "param[%s] must not be null.", #val); \
+      exec_expr;                                               \
+    }                                                          \
+  } while (false)
 
 // Check whether the parameter is null. If yes, return directly and record the error log
-#define GE_RT_VOID_CHECK_NOTNULL(val)                 \
-  do {                                                \
-    if (val == nullptr) {                             \
-      DOMI_LOGE("param[%s] must not be null.", #val); \
-      return;                                         \
-    }                                                 \
-  } while (0)
+#define GE_RT_VOID_CHECK_NOTNULL(val)                          \
+  do {                                                         \
+    if ((val) == nullptr) {                                    \
+      GELOGE(ge::FAILED, "param[%s] must not be null.", #val); \
+      return;                                                  \
+    }                                                          \
+  } while (false)
 
 // Check if the parameter is null. If yes, return false and record the error log
-#define GE_RT_FALSE_CHECK_NOTNULL(val)                \
-  do {                                                \
-    if (val == nullptr) {                             \
-      DOMI_LOGE("param[%s] must not be null.", #val); \
-      return false;                                   \
-    }                                                 \
-  } while (0)
+#define GE_RT_FALSE_CHECK_NOTNULL(val)                         \
+  do {                                                         \
+    if ((val) == nullptr) {                                    \
+      GELOGE(ge::FAILED, "param[%s] must not be null.", #val); \
+      return false;                                            \
+    }                                                          \
+  } while (false)
 
 // Check if the parameter is out of bounds
-#define GE_CHECK_SIZE(size)                          \
-  do {                                               \
-    if (size == 0) {                                 \
-      DOMI_LOGE("param[%s] is out of range", #size); \
-      return ge::PARAM_INVALID;                      \
-    }                                                \
-  } while (0)
+#define GE_CHECK_SIZE(size)                                   \
+  do {                                                        \
+    if ((size) == 0U) {                                       \
+      GELOGE(ge::FAILED, "param[%s] is out of range", #size); \
+      return ge::PARAM_INVALID;                               \
+    }                                                         \
+  } while (false)
 
 // Check if the value on the left is greater than or equal to the value on the right
-#define GE_CHECK_GE(lhs, rhs)                              \
-  do {                                                     \
-    if (lhs < rhs) {                                       \
-      DOMI_LOGE("param[%s] is less than[%s]", #lhs, #rhs); \
-      return ge::PARAM_INVALID;                            \
-    }                                                      \
-  } while (0)
+#define GE_CHECK_GE(lhs, rhs)                                       \
+  do {                                                              \
+    if ((lhs) < (rhs)) {                                            \
+      GELOGE(ge::FAILED, "param[%s] is less than[%s]", #lhs, #rhs); \
+      return ge::PARAM_INVALID;                                     \
+    }                                                               \
+  } while (false)
 
 // Check if the value on the left is less than or equal to the value on the right
-#define GE_CHECK_LE(lhs, rhs)                                 \
-  do {                                                        \
-    if (lhs > rhs) {                                          \
-      DOMI_LOGE("param[%s] is greater than[%s]", #lhs, #rhs); \
-      return ge::PARAM_INVALID;                               \
-    }                                                         \
-  } while (0)
+#define GE_CHECK_LE(lhs, rhs)                                          \
+  do {                                                                 \
+    if ((lhs) > (rhs)) {                                               \
+      GELOGE(ge::FAILED, "param[%s] is greater than[%s]", #lhs, #rhs); \
+      return ge::PARAM_INVALID;                                        \
+    }                                                                  \
+  } while (false)
 
 #define GE_DELETE_NEW_SINGLE(var) \
   do {                            \
-    if (var != nullptr) {         \
-      delete var;                 \
-      var = nullptr;              \
+    if ((var) != nullptr) {       \
+      delete (var);               \
+      (var) = nullptr;            \
     }                             \
-  } while (0)
+  } while (false)
 
 #define GE_DELETE_NEW_ARRAY(var) \
   do {                           \
-    if (var != nullptr) {        \
-      delete[] var;              \
-      var = nullptr;             \
+    if ((var) != nullptr) {      \
+      delete[](var);             \
+      (var) = nullptr;           \
     }                            \
-  } while (0)
+  } while (false)
 
 #define GE_FREE_RT_LOG(addr)                                        \
   do {                                                              \
-    if (addr != nullptr) {                                          \
-      rtError_t error = rtFree(addr);                               \
+    if ((addr) != nullptr) {                                        \
+      const rtError_t error = rtFree(addr);                         \
       if (error != RT_ERROR_NONE) {                                 \
         GELOGE(RT_FAILED, "Call rtFree failed, error: %#x", error); \
       }                                                             \
-      addr = nullptr;                                               \
+      (addr) = nullptr;                                             \
     }                                                               \
-  } while (0)
+  } while (false)
 
+namespace ge {
 /**
  * @ingroup domi_common
  * @brief version of om.proto file
  */
-static constexpr int32_t OM_PROTO_VERSION = 2;
-
-/**
- * Finding an Integer Ceiling Value Without Precision Loss
- */
-#define CEIL(N, n) (((N) + (n)-1) / (n))
+constexpr int32_t OM_PROTO_VERSION = 2;
 
-namespace ge {
-using google::protobuf::Message;
+///
+/// @ingroup domi_common
+/// @brief onverts Vector of a number to a string.
+/// @param [in] v  Vector of a number
+/// @return string
+///
+template <typename T>
+GE_FUNC_VISIBILITY std::string ToString(const std::vector<T> &v) {
+  bool first = true;
+  std::stringstream ss;
+  ss << "[";
+  for (const T &x : v) {
+    if (first) {
+      first = false;
+      ss << x;
+    } else {
+      ss << ", " << x;
+    }
+  }
+  ss << "]";
+  return ss.str();
+}
 
 ///
 /// @ingroup domi_common
-/// @brief Maximum file path length
+/// @brief Converts RepeatedField to String.
+/// @param [in] rpd_field  RepeatedField
+/// @return string
 ///
-const int32_t DOMI_MAX_PATH_LEN = 256;
+template <typename T>
+GE_FUNC_VISIBILITY std::string ToString(const google::protobuf::RepeatedField<T> &rpd_field) {
+  std::stringstream ss;
+  ss << "[";
+  for (const T x : rpd_field) {
+    ss << x;
+    ss << ", ";
+  }
+  // Delete the two extra characters at the end of the line.
+  std::string str = ss.str().substr(0U, ss.str().length() - 2U);
+  str += "]";
+  return str;
+}
+
+///
+///  @ingroup ge_ir_utils
+///  @brief RepeatedPtrField->String
+///  @param [in] const rpd_field  RepeatedPtrField
+///  @return String
+///
+template <typename T>
+GE_FUNC_VISIBILITY std::string ToString(const google::protobuf::RepeatedPtrField<T> &rpd_ptr_field) {
+  std::stringstream ss;
+  ss << "[";
+  for (const T &x : rpd_ptr_field) {
+    ss << x;
+    ss << ", ";
+  }
+  std::string str_ret = ss.str().substr(0U, ss.str().length() - 2U);
+  str_ret += "]";
+  return str_ret;
+}
 
 ///
 /// @ingroup domi_common
@@ -241,7 +295,8 @@ const int32_t DOMI_MAX_PATH_LEN = 256;
 /// @return true success
 /// @return false fail
 ///
-GE_FUNC_VISIBILITY bool ReadProtoFromArray(const void *data, int size, Message *proto);
+GE_FUNC_VISIBILITY bool ReadProtoFromArray(const void *const data, const int32_t size,
+                                           google::protobuf::Message *const proto);
 
 ///
 /// @ingroup domi_proto
@@ -251,17 +306,15 @@ GE_FUNC_VISIBILITY bool ReadProtoFromArray(const void *data, int size, Message *
 /// @return true success
 /// @return false fail
 ///
-GE_FUNC_VISIBILITY bool ReadProtoFromText(const char *file, google::protobuf::Message *message);
-
-GE_FUNC_VISIBILITY bool ReadProtoFromMem(const char *data, int size, google::protobuf::Message *message);
+GE_FUNC_VISIBILITY bool ReadProtoFromText(const char_t *const file, google::protobuf::Message *const message);
 
 ///
 /// @ingroup: domi_common
 /// @brief: get length of file
 /// @param [in] input_file: path of file
-/// @return long： File length. If the file length fails to be obtained, the value -1 is returned.
+/// @return int64_t： File length. If the file length fails to be obtained, the value -1 is returned.
 ///
-GE_FUNC_VISIBILITY extern long GetFileLength(const std::string &input_file);
+GE_FUNC_VISIBILITY extern int64_t GetFileLength(const std::string &input_file);
 
 ///
 /// @ingroup domi_common
@@ -272,9 +325,7 @@ GE_FUNC_VISIBILITY extern long GetFileLength(const std::string &input_file);
 /// @return false fail
 /// @return true success
 ///
-GE_FUNC_VISIBILITY bool ReadBytesFromBinaryFile(const char *file_name, char **buffer, int &length);
-
-GE_FUNC_VISIBILITY bool ReadBytesFromBinaryFile(const char *file_name, std::vector<char> &buffer);
+GE_FUNC_VISIBILITY bool ReadBytesFromBinaryFile(const char_t *const file_name, char_t **const buffer, int32_t &length);
 
 ///
 /// @ingroup domi_common
@@ -283,7 +334,7 @@ GE_FUNC_VISIBILITY bool ReadBytesFromBinaryFile(const char *file_name, std::vect
 /// @return 0 success
 /// @return -1 fail
 ///
-GE_FUNC_VISIBILITY extern int CreateDirectory(const std::string &directory_path);
+GE_FUNC_VISIBILITY extern int32_t CreateDirectory(const std::string &directory_path);
 
 ///
 /// @ingroup domi_common
@@ -292,46 +343,6 @@ GE_FUNC_VISIBILITY extern int CreateDirectory(const std::string &directory_path)
 ///
 GE_FUNC_VISIBILITY std::string CurrentTimeInStr();
 
-///
-/// @ingroup domi_common
-/// @brief onverts Vector of a number to a string.
-/// @param [in] v  Vector of a number
-/// @return string
-///
-template <typename T>
-GE_FUNC_VISIBILITY std::string ToString(std::vector<T> &v) {
-  std::stringstream ss;
-  ss << "[";
-  for (T x : v) {
-    ss << x;
-    ss << ", ";
-  }
-  std::string strRet =
-    ss.str().substr(0, ss.str().length() - 2);  // Delete the two extra characters at the end of the line.
-  strRet += "]";
-  return strRet;
-}
-
-///
-/// @ingroup domi_common
-/// @brief Converts RepeatedField to String.
-/// @param [in] rpd_field  RepeatedField
-/// @return string
-///
-template <typename T>
-GE_FUNC_VISIBILITY std::string ToString(const google::protobuf::RepeatedField<T> &rpd_field) {
-  std::stringstream ss;
-  ss << "[";
-  for (T x : rpd_field) {
-    ss << x;
-    ss << ", ";
-  }
-  std::string strRet =
-    ss.str().substr(0, ss.str().length() - 2);  // Delete the two extra characters at the end of the line.
-  strRet += "]";
-  return strRet;
-}
-
 ///
 /// @ingroup domi_common
 /// @brief Obtains the absolute time (timestamp) of the current system.
@@ -355,7 +366,7 @@ GE_FUNC_VISIBILITY uint32_t GetCurrentSecondTimestap();
 /// @param [in] b
 /// @return false: true: The result is within the normal int64 range.
 ///
-GE_FUNC_VISIBILITY bool CheckInt64MulOverflow(int64_t a, int64_t b);
+GE_FUNC_VISIBILITY bool CheckInt64MulOverflow(const int64_t a, const int64_t b);
 
 ///
 /// @ingroup domi_common
@@ -363,7 +374,7 @@ GE_FUNC_VISIBILITY bool CheckInt64MulOverflow(int64_t a, int64_t b);
 /// @param [in] path of input file
 /// @param [out] Absolute path of a file. If the absolute path cannot be obtained, an empty string is returned
 ///
-GE_FUNC_VISIBILITY std::string RealPath(const char *path);
+GE_FUNC_VISIBILITY std::string RealPath(const char_t *path);
 
 ///
 /// @ingroup domi_common
@@ -387,28 +398,12 @@ GE_FUNC_VISIBILITY bool CheckOutputPathValid(const std::string &file_path, const
 ///
 /// @ingroup domi_common
 /// @brief Check whether the file path meets the whitelist verification requirements.
-/// @param [in] filePath file path
-/// @param [out] result
-///
-GE_FUNC_VISIBILITY bool ValidateStr(const std::string &filePath, const std::string &mode);
-
-///
-/// @ingroup domi_common
-/// @brief Check whether the file is normal file.
-/// @param [in] file_path file path
+/// @param [in] str file path
 /// @param [out] result
 ///
-GE_FUNC_VISIBILITY bool IsValidFile(const char *file_path);
+GE_FUNC_VISIBILITY bool ValidateStr(const std::string &file_path, const std::string &mode);
 
-///
-/// @ingroup domi_common
-/// @brief Check path invalid
-/// @param [in] path, path to be checked
-/// @param [in] length, length of path
-/// @return 0 success
-/// @return -1 fail
-///
-GE_FUNC_VISIBILITY Status CheckPath(const char *path, size_t length);
+GE_FUNC_VISIBILITY Status ConvertToInt32(const std::string &str, int32_t &val);
 }  // namespace ge
 
-#endif  // INC_FRAMEWORK_COMMON_UTIL_H_
+#endif  // AIR_INC_FRAMEWORK_COMMON_UTIL_H_

inc/framework/engine/dnnengine.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -21,16 +21,17 @@
 #include <string>
 #include <vector>
 
-#include "common/ge_inner_error_codes.h"
-#include "common/ge_types.h"
+#include "framework/common/ge_inner_error_codes.h"
+#include "framework/common/ge_types.h"
 #include "graph/types.h"
 
 namespace ge {
-enum PriorityEnum {
+enum class PriorityEnum {
   COST_0 = 0,
-  COST_1,
-  COST_2,
-  COST_3,
+  COST_1 = 1,
+  COST_2 = 2,
+  COST_3 = 3,
+  COST_4 = 4,
   COST_9 = 9,
   COST_10 = 10,
 };
@@ -38,19 +39,37 @@ enum PriorityEnum {
 struct DNNEngineAttribute {
   std::string engine_name;
   std::vector<std::string> mem_type;
-  uint32_t compute_cost;
+  PriorityEnum compute_cost;
   enum RuntimeType runtime_type;  // HOST, DEVICE
   // If engine input format must be specific, set this attribute, else set FORMAT_RESERVED
   Format engine_input_format;
   Format engine_output_format;
+  bool atomic_engine_flag;
 };
 
 class GE_FUNC_VISIBILITY DNNEngine {
  public:
+  DNNEngine() = default;
+  explicit DNNEngine(const DNNEngineAttribute &attrs) {
+    engine_attribute_ = attrs;
+  }
   virtual ~DNNEngine() = default;
-  virtual Status Initialize(const std::map<std::string, std::string> &options) = 0;
-  virtual Status Finalize() = 0;
-  virtual void GetAttributes(DNNEngineAttribute &attr) const = 0;
+  Status Initialize(const std::map<std::string, std::string> &options) const {
+    (void)options;
+    return SUCCESS;
+  }
+  Status Finalize() const {
+    return SUCCESS;
+  }
+  void GetAttributes(DNNEngineAttribute &attr) const {
+    attr = engine_attribute_;
+  }
+  bool IsAtomic() const {
+    return engine_attribute_.atomic_engine_flag;
+  }
+
+ protected:
+  DNNEngineAttribute engine_attribute_;
 };
 }  // namespace ge
 

inc/framework/executor/ge_executor.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -22,42 +22,58 @@
 #include <vector>
 
 #include "common/dynamic_aipp.h"
-#include "common/ge_inner_error_codes.h"
-#include "common/ge_types.h"
-#include "common/types.h"
+#include "framework/common/ge_inner_error_codes.h"
+#include "framework/common/ge_types.h"
+#include "framework/common/types.h"
 #include "graph/tensor.h"
 #include "graph/ge_tensor.h"
-#include "runtime/base.h"
 
 namespace ge {
 class SingleOp;
 class DynamicSingleOp;
+class GeRootModel;
 
 struct RunModelData {
   uint32_t index;  // Data index
   uint32_t modelId;
-  std::vector<DataBuffer> blobs;       // All input/output data buffer
-  uint32_t timestamp;                  // Data creation time
-  uint32_t timeout;                    // Processing timeout
-  uint64_t request_id = 0;             // Request ID
-  uint64_t dynamic_batch_size = 0;     // Dynamic batch size scene, set dynamic size, not supported by default:0
-  uint64_t dynamic_image_height = 0;   // Dynamic image size scene, set image height, not supported by default:0
-  uint64_t dynamic_image_width = 0;    // Dynamic image size scene, set image width, not supported by default:0
-  std::vector<uint64_t> dynamic_dims;  // Dynamic dims scene, set dynamic dims, not supported by default:empty
+  std::vector<DataBuffer> blobs;        // All input/output data buffer
+  uint32_t timestamp;                   // Data creation time
+  uint32_t timeout;                     // Processing timeout
+  uint64_t request_id = 0UL;            // Request ID
+  uint64_t dynamic_batch_size = 0UL;    // Dynamic batch size scene, set dynamic size, not supported by default:0
+  uint64_t dynamic_image_height = 0UL;  // Dynamic image size scene, set image height, not supported by default:0
+  uint64_t dynamic_image_width = 0UL;   // Dynamic image size scene, set image width, not supported by default:0
+  std::vector<uint64_t> dynamic_dims;   // Dynamic dims scene, set dynamic dims, not supported by default:empty
 };
 
 class GE_FUNC_VISIBILITY GeExecutor {
  public:
   GeExecutor();
   ~GeExecutor() = default;
-  ge::Status Initialize();
-  ge::Status Finalize();
 
-  ge::Status UnloadModel(uint32_t modelId);
+  Status Initialize();
+  Status Finalize();
+
+  ///
+  /// @ingroup ge
+  /// @brief Initialize global execute environment.
+  /// @param [in] options: environment variables.
+  /// @return init result
+  ///
+  static Status Initialize(const std::map<std::string, std::string> &options);
+
+  ///
+  /// @ingroup ge
+  /// @brief Finalize global execute environment.
+  /// @return execute result
+  ///
+  static Status FinalizeEx();
+
+  Status UnloadModel(const uint32_t model_id);
 
   // Get input and output descriptor
-  ge::Status GetModelDescInfo(uint32_t model_id, std::vector<ge::TensorDesc> &input_desc,
-                              std::vector<ge::TensorDesc> &output_desc, bool new_model_desc = false);
+  Status GetModelDescInfo(const uint32_t model_id, std::vector<TensorDesc> &input_desc,
+                          std::vector<TensorDesc> &output_desc, const bool new_model_desc = false);
 
   ///
   /// @ingroup ge
@@ -68,7 +84,8 @@ class GE_FUNC_VISIBILITY GeExecutor {
   /// @param [in] batch_size: batch size entered by user in dynamic multi-batch scenario
   /// @return execute result
   ///
-  ge::Status SetDynamicBatchSize(uint32_t model_id, void *dynamic_input_addr, uint64_t length, uint64_t batch_size);
+  Status SetDynamicBatchSize(const uint32_t model_id, void *const dynamic_input_addr, const uint64_t length,
+                             const uint64_t batch_size);
 
   ///
   /// @ingroup ge
@@ -80,8 +97,8 @@ class GE_FUNC_VISIBILITY GeExecutor {
   /// @param [in] image_width: image width entered by user in dynamic multi-resolution scenario
   /// @return execute result
   ///
-  ge::Status SetDynamicImageSize(uint32_t model_id, void *dynamic_input_addr, uint64_t length, uint64_t image_height,
-                                 uint64_t image_width);
+  Status SetDynamicImageSize(const uint32_t model_id, void *const dynamic_input_addr, const uint64_t length,
+                             const uint64_t image_height, const uint64_t image_width);
 
   ///
   /// @ingroup ge
@@ -93,8 +110,8 @@ class GE_FUNC_VISIBILITY GeExecutor {
   /// @param [in] dynamic_dims: array of dynamic dimensions
   /// @return execute result
   ///
-  ge::Status SetDynamicDims(uint32_t model_id, void *dynamic_input_addr, uint64_t length,
-                            const std::vector<uint64_t> &dynamic_dims);
+  Status SetDynamicDims(const uint32_t model_id, void *const dynamic_input_addr, const uint64_t length,
+                        const std::vector<uint64_t> &dynamic_dims);
 
   ///
   /// @ingroup ge
@@ -104,8 +121,8 @@ class GE_FUNC_VISIBILITY GeExecutor {
   /// @param [out] cur_dynamic_dims: current dynamic dims
   /// @return execute result
   ///
-  ge::Status GetCurDynamicDims(uint32_t model_id, const std::vector<uint64_t> &dynamic_dims,
-                               std::vector<uint64_t> &cur_dynamic_dims);
+  Status GetCurDynamicDims(const uint32_t model_id, const std::vector<uint64_t> &dynamic_dims,
+                           std::vector<uint64_t> &cur_dynamic_dims);
 
   ///
   /// @ingroup ge
@@ -115,8 +132,8 @@ class GE_FUNC_VISIBILITY GeExecutor {
   /// @param [out] dynamic_type
   /// @return execute result
   ///
-  ge::Status GetDynamicBatchInfo(uint32_t model_id, std::vector<std::vector<int64_t>> &batch_info,
-                                 int32_t &dynamic_type);
+  Status GetDynamicBatchInfo(const uint32_t model_id, std::vector<std::vector<int64_t>> &batch_info,
+                             int32_t &dynamic_type);
 
   ///
   /// @ingroup ge
@@ -125,7 +142,7 @@ class GE_FUNC_VISIBILITY GeExecutor {
   /// @param [out] batch_info
   /// @return execute result
   ///
-  ge::Status GetCombinedDynamicDims(uint32_t model_id, std::vector<std::vector<int64_t>> &batch_info);
+  Status GetCombinedDynamicDims(const uint32_t model_id, std::vector<std::vector<int64_t>> &batch_info);
 
   ///
   /// @ingroup ge
@@ -134,9 +151,9 @@ class GE_FUNC_VISIBILITY GeExecutor {
   /// @param [out] user_designate_shape_order
   /// @return execute result
   ///
-  ge::Status GetUserDesignateShapeOrder(uint32_t model_id, std::vector<std::string> &user_designate_shape_order);
+  Status GetUserDesignateShapeOrder(const uint32_t model_id, std::vector<std::string> &user_designate_shape_order);
 
-  ge::Status GetCurShape(const uint32_t model_id, std::vector<int64_t> &batch_info, int32_t &dynamic_type);
+  Status GetCurShape(const uint32_t model_id, std::vector<int64_t> &batch_info, int32_t &dynamic_type);
 
   ///
   /// @ingroup ge
@@ -148,22 +165,22 @@ class GE_FUNC_VISIBILITY GeExecutor {
   /// @param [in] aippParms: kAippDynamicPara by user in dynamic aipp
   /// @return execute result
   ///
-  ge::Status SetDynamicAippData(uint32_t model_id, void *dynamic_input_addr, uint64_t length,
-                                const std::vector<kAippDynamicBatchPara> &aippBatchPara,
-                                const kAippDynamicPara &aippParms);
+  Status SetDynamicAippData(const uint32_t model_id, void *const dynamic_input_addr, const uint64_t length,
+                            const std::vector<kAippDynamicBatchPara> &aipp_batch_para,
+                            const kAippDynamicPara &aipp_parms);
 
-  ge::Status GetAIPPInfo(uint32_t model_id, uint32_t index, AippConfigInfo &aipp_info);
+  Status GetAIPPInfo(const uint32_t model_id, const uint32_t index, AippConfigInfo &aipp_info);
 
-  ge::Status GetOpAttr(uint32_t model_id, const std::string &op_name, const std::string &attr_name,
-                       std::string &attr_value);
+  Status GetOpAttr(const uint32_t model_id, const std::string &op_name, const std::string &attr_name,
+                   std::string &attr_value);
 
-  ge::Status GetModelAttr(uint32_t model_id, std::vector<std::string> &dynamic_output_shape_info);
+  Status GetModelAttr(const uint32_t model_id, std::vector<std::string> &dynamic_output_shape_info);
 
-  ge::Status GetAippType(uint32_t model_id, uint32_t index, InputAippType &type, size_t &aipp_index);
+  Status GetAippType(const uint32_t model_id, const uint32_t index, InputAippType &type, size_t &aipp_index);
 
-  ge::Status CommandHandle(const ge::Command &command);
+  Status CommandHandle(const Command &command) const;
 
-  ge::Status SetDump(const DumpConfig &dump_config);
+  Status SetDump(const DumpConfig &dump_config);
 
   ///
   /// @ingroup ge
@@ -173,7 +190,7 @@ class GE_FUNC_VISIBILITY GeExecutor {
   /// @return SUCCESS
   /// @return FAILED
   ///
-  ge::Status GetMaxUsedMemory(uint32_t model_id, uint32_t &max_size);
+  Status GetMaxUsedMemory(const uint32_t model_id, uint32_t &max_size);
 
   ///
   /// @ingroup ge
@@ -182,7 +199,7 @@ class GE_FUNC_VISIBILITY GeExecutor {
   /// @param [out] ModelData &model_data: Offline model memory data
   /// @return SUCCESS handle successfully / others handle failed
   ///
-  ge::Status LoadDataFromFile(const std::string &path, ge::ModelData &model_data);
+  Status LoadDataFromFile(const std::string &path, ModelData &model_data);
 
   ///
   /// @ingroup ge
@@ -195,8 +212,8 @@ class GE_FUNC_VISIBILITY GeExecutor {
   /// @param [out] uint32_t &model_id: Corresponding identification after model loading
   /// @return SUCCESS handle successfully / others handle failed
   ///
-  ge::Status LoadModelFromData(uint32_t &model_id, const ge::ModelData &model_data, void *dev_ptr, size_t mem_size,
-                               void *weight_ptr, size_t weight_size);
+  Status LoadModelFromData(uint32_t &model_id, const ModelData &model_data, void *const dev_ptr, const size_t mem_size,
+                           void *const weight_ptr, const size_t weight_size);
 
   ///
   /// @ingroup ge
@@ -207,9 +224,20 @@ class GE_FUNC_VISIBILITY GeExecutor {
   /// @param [in] output_queue_ids: input queue ids create from user.
   /// @return: 0 for success / others for fail
   ///
-  ge::Status LoadModelWithQ(uint32_t &model_id, const ge::ModelData &model_data,
-                            const std::vector<uint32_t> &input_queue_ids,
-                            const std::vector<uint32_t> &output_queue_ids);
+  Status LoadModelWithQ(uint32_t &model_id, const ModelData &model_data, const std::vector<uint32_t> &input_queue_ids,
+                        const std::vector<uint32_t> &output_queue_ids);
+
+  ///
+  /// @ingroup ge
+  /// @brief Load task list from ModelData with queue.
+  /// @param [out] model_id: model id allocate from manager.
+  /// @param [in] root_model: Instance of GeRootModel.
+  /// @param [in] input_queue_ids: input queue ids create from user.
+  /// @param [in] output_queue_ids: input queue ids create from user.
+  /// @return: 0 for success / others for fail
+  ///
+  Status LoadModelWithQ(uint32_t &model_id, const std::shared_ptr<GeRootModel> &root_model,
+                        const std::vector<uint32_t> &input_queue_ids, const std::vector<uint32_t> &output_queue_ids);
 
   ///
   /// @ingroup ge
@@ -221,8 +249,17 @@ class GE_FUNC_VISIBILITY GeExecutor {
   /// @param [out] domi::OutputData *output_data: Model output data
   /// @return SUCCESS handle successfully / others handle failed
   ///
-  ge::Status ExecModel(uint32_t model_id, void *stream, const ge::RunModelData &input_data,
-                       ge::RunModelData &output_data, bool async_mode = false);
+  Status ExecModel(const uint32_t model_id, void *const stream, const RunModelData &input_data,
+                   RunModelData &output_data, const bool async_mode = false);
+
+  ///
+  /// @ingroup ge
+  /// @brief Load task list from root_model without input queue or output queue.
+  /// @param [out] model_id: model id allocate from manager.
+  /// @param [in] root_model: Instance of GeRootModel.
+  /// @return: 0 for success / others for fail
+  ///
+  Status LoadModelWithoutQ(uint32_t &model_id, const std::shared_ptr<GeRootModel> &root_model) const;
 
   ///
   /// @ingroup ge
@@ -236,9 +273,9 @@ class GE_FUNC_VISIBILITY GeExecutor {
   /// @param [out] std::vector<GeTensorDesc> &output_desc: description of model output data
   /// @return SUCCESS handle successfully / others handle failed
   ///
-  ge::Status ExecModel(uint32_t model_id, void *stream, const ge::RunModelData &run_input_data,
-                       const std::vector<GeTensorDesc> &input_desc, ge::RunModelData &run_output_data,
-                       std::vector<GeTensorDesc> &output_desc, bool async_mode = false);
+  Status ExecModel(const uint32_t model_id, void *const stream, const RunModelData &run_input_data,
+                   const std::vector<GeTensorDesc> &input_desc, RunModelData &run_output_data,
+                   std::vector<GeTensorDesc> &output_desc, const bool async_mode = false);
 
   ///
   /// @ingroup ge
@@ -248,7 +285,7 @@ class GE_FUNC_VISIBILITY GeExecutor {
   /// @param [out] size_t &weight_size Weight memory space size
   /// @return SUCCESS handle successfully / others handle failed
   ///
-  ge::Status GetMemAndWeightSize(const std::string &path, size_t &mem_size, size_t &weight_size);
+  Status GetMemAndWeightSize(const std::string &path, size_t &mem_size, size_t &weight_size);
 
   ///
   /// @ingroup ge
@@ -259,39 +296,45 @@ class GE_FUNC_VISIBILITY GeExecutor {
   /// @param [out] size_t &weight_size Weight memory space size
   /// @return SUCCESS handle successfully / others handle failed
   ///
-  ge::Status GetMemAndWeightSize(const void *model_data, size_t model_size, size_t &mem_size, size_t &weight_size);
+  Status GetMemAndWeightSize(const void *const model_data, const size_t model_size, size_t &mem_size,
+                             size_t &weight_size);
+
+  static Status LoadSingleOp(const std::string &model_name, const ModelData &model_data, void *const stream,
+                             SingleOp **const single_op);
+
+  static Status LoadSingleOpV2(const std::string &model_name, const ModelData &model_data, void *const stream,
+                               SingleOp **const single_op, const uint64_t model_id);
 
-  static ge::Status LoadSingleOp(const std::string &modelName, const ge::ModelData &modelData, void *stream,
-                                 SingleOp **single_op);
+  static Status ExecuteAsync(SingleOp *const executor, const std::vector<DataBuffer> &inputs,
+                             std::vector<DataBuffer> &outputs);
 
-  static ge::Status LoadSingleOpV2(const std::string &modelName, const ge::ModelData &modelData, void *stream,
-                                   SingleOp **single_op, const uint64_t model_id);
+  static Status LoadDynamicSingleOp(const std::string &model_name, const ModelData &model_data, void *const stream,
+                                    DynamicSingleOp **const single_op);
 
-  static ge::Status ExecuteAsync(SingleOp *executor, const std::vector<DataBuffer> &inputs,
-                                 std::vector<DataBuffer> &outputs);
+  static Status LoadDynamicSingleOpV2(const std::string &model_name, const ModelData &model_data, void *const stream,
+                                      DynamicSingleOp **const single_op, const uint64_t model_id);
 
-  static ge::Status LoadDynamicSingleOp(const std::string &model_name, const ge::ModelData &modelData, void *stream,
-                                        DynamicSingleOp **single_op);
+  static Status UnloadSingleOp(const uint64_t op_id);
 
-  static ge::Status LoadDynamicSingleOpV2(const std::string &model_name, const ge::ModelData &modelData, void *stream,
-                                          DynamicSingleOp **single_op, const uint64_t model_id);
+  static Status UnloadDynamicSingleOp(const uint64_t op_id);
 
-  static ge::Status ExecuteAsync(DynamicSingleOp *executor, const std::vector<GeTensorDesc> &input_desc,
-                                 const std::vector<DataBuffer> &inputs, std::vector<GeTensorDesc> &output_desc,
-                                 std::vector<DataBuffer> &outputs);
+  static Status ExecuteAsync(DynamicSingleOp *const executor, const std::vector<GeTensorDesc> &input_desc,
+                             const std::vector<DataBuffer> &inputs, std::vector<GeTensorDesc> &output_desc,
+                             std::vector<DataBuffer> &outputs);
 
-  static ge::Status ReleaseSingleOpResource(void *stream);
+  static Status ReleaseSingleOpResource(void *const stream);
 
-  static ge::Status GetDeviceIdByModelId(uint32_t model_id, uint32_t &device_id);
+  static Status GetDeviceIdByModelId(const uint32_t model_id, uint32_t &device_id);
 
-  ge::Status GetBatchInfoSize(uint32_t model_id, size_t &shape_count);
-  ge::Status GetOrigInputInfo(uint32_t model_id, uint32_t index, OriginInputInfo &orig_input_info);
-  ge::Status GetAllAippInputOutputDims(uint32_t model_id, uint32_t index, std::vector<InputOutputDims> &input_dims,
-                                       std::vector<InputOutputDims> &output_dims);
-  ge::Status GetOpDescInfo(uint32_t device_id, uint32_t stream_id, uint32_t task_id, OpDescInfo &op_desc_info);
+  Status GetBatchInfoSize(const uint32_t model_id, size_t &shape_count);
+  Status GetOrigInputInfo(const uint32_t model_id, const uint32_t index, OriginInputInfo &orig_input_info);
+  Status GetAllAippInputOutputDims(const uint32_t model_id, const uint32_t index,
+                                   std::vector<InputOutputDims> &input_dims, std::vector<InputOutputDims> &output_dims);
+  Status GetOpDescInfo(const uint32_t device_id, const uint32_t stream_id, const uint32_t task_id,
+                       OpDescInfo &op_desc_info);
 
  private:
-  static bool isInit_;
+  static std::atomic_bool is_inited_;
 };
 }  // namespace ge
 

inc/framework/ge_runtime/davinci_model.h

@@ -1,113 +0,0 @@
-/**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#ifndef INC_FRAMEWORK_GE_RUNTIME_DAVINCI_MODEL_H_
-#define INC_FRAMEWORK_GE_RUNTIME_DAVINCI_MODEL_H_
-
-#include <memory>
-#include <vector>
-
-#include "ge_runtime/op_info.h"
-#include "ge_runtime/task_info.h"
-
-namespace ge {
-namespace model_runner {
-class DavinciModel {
- public:
-  DavinciModel(const std::vector<std::shared_ptr<TaskInfo>> &task_info_list,
-               const std::vector<std::shared_ptr<OpInfo>> &data_info_list,
-               const std::vector<std::shared_ptr<OpInfo>> &output_info_list,
-               const std::vector<std::shared_ptr<OpInfo>> &constant_info_list,
-               const std::vector<model_runner::OpInfoPtr> &variable_info_list,
-               const std::vector<uint32_t> &wait_active_stream_list,
-               const std::vector<uint32_t> &force_copy_stream_list, uint64_t mem_size = 0, uint64_t weight_size = 0,
-               uint64_t var_size = 0, uintptr_t logic_mem_base = 0, uintptr_t logic_weight_base = 0,
-               uintptr_t logic_var_base = 0, uint32_t stream_num = 0, uint32_t batch_num = 0, uint32_t event_num = 0,
-               int32_t priority = 0)
-      : task_info_list_(task_info_list),
-        data_info_list_(data_info_list),
-        output_info_list_(output_info_list),
-        constant_info_list_(constant_info_list),
-        variable_info_list_(variable_info_list),
-        wait_active_stream_list_(wait_active_stream_list),
-        force_copy_stream_list_(force_copy_stream_list),
-        mem_size_(mem_size),
-        weight_size_(weight_size),
-        var_size_(var_size),
-        logic_mem_base_(logic_mem_base),
-        logic_weight_base_(logic_weight_base),
-        logic_var_base_(logic_var_base),
-        stream_num_(stream_num),
-        batch_num_(batch_num),
-        event_num_(event_num),
-        priority_(priority) {}
-  ~DavinciModel() {}
-
-  uint64_t GetMemSize() const { return mem_size_; }
-  uint64_t GetWeightSize() const { return weight_size_; }
-  uint64_t GetVarSize() const { return var_size_; }
-
-  uintptr_t GetLogicMemBase() const { return logic_mem_base_; }
-  uintptr_t GetLogicWeightBase() const { return logic_weight_base_; }
-  uintptr_t GetLogicVarBase() const { return logic_var_base_; }
-
-  uint32_t GetStreamNum() const { return stream_num_; }
-  uint32_t GetBatchNum() const { return batch_num_; }
-  uint32_t GetEventNum() const { return event_num_; }
-
-  const std::vector<uint32_t> &GetWaitActiveStreams() const { return wait_active_stream_list_; }
-  const std::vector<uint32_t> &GetForceCopyStreams() const { return force_copy_stream_list_; }
-
-  int32_t GetPriority() const { return priority_; }
-
-  const std::vector<std::shared_ptr<TaskInfo>> &GetTaskInfoList() const { return task_info_list_; }
-  const std::vector<std::shared_ptr<OpInfo>> &GetDataInfoList() const { return data_info_list_; }
-  const std::vector<std::shared_ptr<OpInfo>> &GetOutputInfoList() const { return output_info_list_; }
-  const std::vector<std::shared_ptr<OpInfo>> &GetConstantInfoList() const { return output_info_list_; }
-  const std::vector<model_runner::OpInfoPtr> &GetVariableInfoList() const { return variable_info_list_; }
-
- private:
-  std::vector<std::shared_ptr<TaskInfo>> task_info_list_;
-  std::vector<std::shared_ptr<OpInfo>> data_info_list_;
-  std::vector<std::shared_ptr<OpInfo>> output_info_list_;
-  std::vector<std::shared_ptr<OpInfo>> constant_info_list_;
-  std::vector<model_runner::OpInfoPtr> variable_info_list_;
-
-  std::vector<uint32_t> wait_active_stream_list_;
-  std::vector<uint32_t> force_copy_stream_list_;
-
-  uint64_t mem_size_;
-  uint64_t weight_size_;
-  uint64_t var_size_;
-
-  uintptr_t logic_mem_base_;
-  uintptr_t logic_weight_base_;
-  uintptr_t logic_var_base_;
-
-  uint32_t stream_num_;
-  uint32_t batch_num_;
-  uint32_t event_num_;
-
-  int32_t priority_;
-
-  // Disable to copy constructor and assignment operator
-  DavinciModel &operator=(const DavinciModel &) = delete;
-  DavinciModel(const DavinciModel &) = delete;
-};
-}  // namespace model_runner
-}  // namespace ge
-
-#endif  // INC_FRAMEWORK_GE_RUNTIME_DAVINCI_MODEL_H_

inc/framework/ge_runtime/model_runner.h

@@ -1,68 +0,0 @@
-/**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#ifndef INC_FRAMEWORK_GE_RUNTIME_MODEL_RUNNER_H_
-#define INC_FRAMEWORK_GE_RUNTIME_MODEL_RUNNER_H_
-
-#include <memory>
-#include <unordered_map>
-#include <vector>
-
-#include "common/ge_inner_error_codes.h"
-#include "common/ge_types.h"
-#include "ge_runtime/davinci_model.h"
-
-namespace ge {
-namespace model_runner {
-class RuntimeModel;
-using RuntimeInfo = std::tuple<uint32_t, uint32_t, void *>;
-class ModelRunner {
- public:
-  static ModelRunner &Instance();
-
-  bool LoadDavinciModel(uint32_t device_id, uint64_t session_id, uint32_t model_id,
-                        std::shared_ptr<DavinciModel> davinci_model, std::shared_ptr<ModelListener> listener);
-
-  bool DistributeTask(uint32_t model_id);
-
-  bool LoadModelComplete(uint32_t model_id);
-
-  const std::vector<uint32_t> &GetTaskIdList(uint32_t model_id) const;
-
-  const std::vector<uint32_t> &GetStreamIdList(uint32_t model_id) const;
-
-  const std::map<std::string, std::shared_ptr<RuntimeInfo>> &GetRuntimeInfoMap(uint32_t model_id) const;
-
-  void *GetModelHandle(uint32_t model_id) const;
-
-  bool UnloadModel(uint32_t model_id);
-
-  bool RunModel(uint32_t model_id, const InputData &input_data, OutputData *output_data);
-
-  bool GetInputOutputDescInfo(uint32_t model_id, bool zero_copy, std::vector<InputOutputDescInfo> *input_desc,
-                              std::vector<InputOutputDescInfo> *output_desc, std::vector<uint32_t> *input_format,
-                              std::vector<uint32_t> *output_format);
-
- private:
-  ModelRunner() = default;
-  ~ModelRunner() = default;
-
-  std::unordered_map<uint32_t, std::shared_ptr<RuntimeModel>> runtime_models_;
-};
-}  // namespace model_runner
-}  // namespace ge
-
-#endif  // INC_FRAMEWORK_GE_RUNTIME_MODEL_RUNNER_H_

inc/framework/ge_runtime/op_info.h

@@ -1,72 +0,0 @@
-/**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#ifndef INC_FRAMEWORK_GE_RUNTIME_OP_INFO_H_
-#define INC_FRAMEWORK_GE_RUNTIME_OP_INFO_H_
-
-#include <memory>
-#include <string>
-#include <vector>
-
-namespace ge {
-namespace model_runner {
-struct TensorInfo {
-  int64_t GetShapeSize() const {
-    int64_t res = 1;
-    if (dims.empty()) {
-      return 0;
-    }
-    for (auto dim : dims) {
-      res *= dim;
-    }
-    return res;
-  }
-
-  int64_t GetDim(uint32_t index) {
-    if (index >= dims.size()) {
-      return 0;
-    }
-    return dims[index];
-  }
-
-  std::vector<int64_t> dims;
-  uint32_t datatype;
-  uint32_t format;
-  uint32_t real_dim_cnt;
-  uint32_t size;
-  bool is_output;
-};
-
-struct OpInfo {
-  uint32_t index;
-  std::string name;
-  std::string type;
-  bool var_is_broadcast;
-  std::vector<uintptr_t> input_addrs;
-  std::vector<uintptr_t> output_addrs;
-  std::vector<TensorInfo> input_tensors;
-  std::vector<TensorInfo> output_tensors;
-  std::vector<TensorInfo> weight_tensors;
-  std::vector<std::string> src_name;
-  std::vector<int64_t> src_index;
-  std::string weight_data;
-};
-
-using TensorInfoPtr = std::shared_ptr<TensorInfo>;
-using OpInfoPtr = std::shared_ptr<OpInfo>;
-}  // namespace model_runner
-}  // namespace ge
-#endif  // INC_FRAMEWORK_GE_RUNTIME_OP_INFO_H_

inc/framework/ge_runtime/task_info.h

@@ -1,405 +0,0 @@
-/**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#ifndef INC_FRAMEWORK_GE_RUNTIME_TASK_INFO_H_
-#define INC_FRAMEWORK_GE_RUNTIME_TASK_INFO_H_
-
-#include <stdint.h>
-#include <memory>
-#include <string>
-#include <utility>
-#include <vector>
-
-#include "cce/taskdown_api.h"
-
-namespace ge {
-namespace model_runner {
-enum TaskInfoType {
-  CCE = 0,
-  TBE,
-  AICPU,
-  LABEL_SET,
-  LABEL_SWITCH,
-  LABEL_GOTO,
-  EVENT_RECORD,
-  EVENT_WAIT,
-  FUSION_START,
-  FUSION_END,
-  HCCL,
-  PROFILER_TRACE,
-  MEMCPY_ASYNC,
-  STREAM_SWITCH,
-  STREAM_ACTIVE,
-  // Insert new task type here
-  REVSERVED = 23
-};
-
-class TaskInfo {
- public:
-  virtual ~TaskInfo() {}
-  uint32_t stream_id() const { return stream_id_; }
-  TaskInfoType type() const { return type_; }
-  std::string op_name() const { return op_name_; }
-  bool dump_flag() const { return dump_flag_; }
-
- protected:
-  TaskInfo(const std::string &op_name, uint32_t stream_id, TaskInfoType type, bool dump_flag)
-      : op_name_(op_name), stream_id_(stream_id), type_(type), dump_flag_(dump_flag) {}
-
- private:
-  std::string op_name_;
-  uint32_t stream_id_;
-  TaskInfoType type_;
-  bool dump_flag_;
-};
-
-class CceTaskInfo : public TaskInfo {
- public:
-  CceTaskInfo(const std::string &op_name, uint32_t stream_id, const cce::ccOpContext &ctx, const std::string &stub_func,
-              uint32_t block_dim, const std::vector<uint8_t> &args, uint32_t args_size,
-              const std::vector<uint8_t> &sm_desc, const std::vector<uint8_t> &flow_table,
-              const std::vector<uint8_t> &args_offset, bool is_flowtable)
-      : TaskInfo(op_name, stream_id, TaskInfoType::CCE, false),
-        ctx_(ctx),
-        stub_func_(stub_func),
-        block_dim_(block_dim),
-        args_(args),
-        args_size_(args_size),
-        sm_desc_(sm_desc),
-        flow_table_(flow_table),
-        args_offset_(args_offset),
-        is_flowtable_(is_flowtable) {}
-  ~CceTaskInfo() override {}
-
-  cce::ccOpContext cc_context() const { return ctx_; }
-  std::string stub_func() const { return stub_func_; }
-  uint32_t block_dim() const { return block_dim_; }
-  const std::vector<uint8_t> &args() const { return args_; }
-  uint32_t args_size() const { return args_size_; }
-  const std::vector<uint8_t> &sm_desc() const { return sm_desc_; }
-  const std::vector<uint8_t> &flow_table() const { return flow_table_; }
-  const std::vector<uint8_t> &args_offset() const { return args_offset_; }
-  bool is_flowtable() const { return is_flowtable_; }
-
- private:
-  cce::ccOpContext ctx_;
-  std::string stub_func_;
-  uint32_t block_dim_;
-  std::vector<uint8_t> args_;
-  uint32_t args_size_;
-  std::vector<uint8_t> sm_desc_;
-  std::vector<uint8_t> flow_table_;
-  std::vector<uint8_t> args_offset_;
-  bool is_flowtable_;
-};
-
-class TbeTaskInfo : public TaskInfo {
- public:
-  TbeTaskInfo(const std::string &op_name, uint32_t stream_id, const std::string &stub_func, uint32_t block_dim,
-              const std::vector<uint8_t> &args, uint32_t args_size, const std::vector<uint8_t> &sm_desc, void *binary,
-              uint32_t binary_size, const std::vector<uint8_t> &meta_data, const std::vector<void *> &input_data_addrs,
-              const std::vector<void *> &output_data_addrs, const std::vector<void *> &workspace_addrs, bool dump_flag)
-      : TaskInfo(op_name, stream_id, TaskInfoType::TBE, dump_flag),
-        stub_func_(stub_func),
-        block_dim_(block_dim),
-        args_(args),
-        args_size_(args_size),
-        sm_desc_(sm_desc),
-        binary_(binary),
-        binary_size_(binary_size),
-        meta_data_(meta_data),
-        input_data_addrs_(input_data_addrs),
-        output_data_addrs_(output_data_addrs),
-        workspace_addrs_(workspace_addrs) {}
-  ~TbeTaskInfo() override {}
-
-  const std::string &stub_func() const { return stub_func_; }
-  uint32_t block_dim() const { return block_dim_; }
-  const std::vector<uint8_t> &args() const { return args_; }
-  uint32_t args_size() const { return args_size_; }
-  const std::vector<uint8_t> &sm_desc() const { return sm_desc_; }
-  void *binary() const { return binary_; }
-  uint32_t binary_size() const { return binary_size_; }
-  const std::vector<uint8_t> &meta_data() const { return meta_data_; }
-  const std::vector<void *> &input_data_addrs() const { return input_data_addrs_; }
-  const std::vector<void *> &output_data_addrs() const { return output_data_addrs_; }
-  const std::vector<void *> &workspace_addrs() const { return workspace_addrs_; }
-
-  void SetBinary(void *binary, uint32_t binary_size) {
-    binary_ = binary;
-    binary_size_ = binary_size;
-  }
-
- private:
-  std::string stub_func_;
-  uint32_t block_dim_;
-  std::vector<uint8_t> args_;
-  uint32_t args_size_;
-  std::vector<uint8_t> sm_desc_;
-  void *binary_;
-  uint32_t binary_size_;
-  std::vector<uint8_t> meta_data_;
-  std::vector<void *> input_data_addrs_;
-  std::vector<void *> output_data_addrs_;
-  std::vector<void *> workspace_addrs_;
-};
-
-class AicpuTaskInfo : public TaskInfo {
- public:
-  AicpuTaskInfo(const std::string &op_name, uint32_t stream_id, const string &so_name, const std::string &kernel_name,
-                const std::string &node_def, const std::string &ext_info, const std::vector<void *> &input_data_addrs,
-                const std::vector<void *> &output_data_addrs, bool dump_flag)
-      : TaskInfo(op_name, stream_id, TaskInfoType::AICPU, dump_flag),
-        so_name_(so_name),
-        kernel_name_(kernel_name),
-        node_def_(node_def),
-        ext_info_(ext_info),
-        input_data_addrs_(input_data_addrs),
-        output_data_addrs_(output_data_addrs) {}
-  ~AicpuTaskInfo() override {}
-
-  const std::string &so_name() const { return so_name_; }
-  const std::string &kernel_name() const { return kernel_name_; }
-  const std::string &node_def() const { return node_def_; }
-  const std::vector<void *> &input_data_addrs() const { return input_data_addrs_; }
-  const std::vector<void *> &output_data_addrs() const { return output_data_addrs_; }
-  const std::string &ext_info() const { return ext_info_; }
-
- private:
-  std::string so_name_;
-  std::string kernel_name_;
-  std::string node_def_;
-  std::string ext_info_;
-  std::vector<void *> input_data_addrs_;
-  std::vector<void *> output_data_addrs_;
-};
-
-class LabelSetTaskInfo : public TaskInfo {
- public:
-  LabelSetTaskInfo(const std::string &op_name, uint32_t stream_id, uint32_t label_id)
-      : TaskInfo(op_name, stream_id, TaskInfoType::LABEL_SET, false), label_id_(label_id) {}
-  ~LabelSetTaskInfo() override {}
-  uint32_t label_id() const { return label_id_; }
-
- private:
-  uint32_t label_id_;
-};
-
-class LabelGotoTaskInfo : public TaskInfo {
- public:
-  LabelGotoTaskInfo(const std::string &op_name, uint32_t stream_id, uint32_t label_id)
-      : TaskInfo(op_name, stream_id, TaskInfoType::LABEL_GOTO, false), label_id_(label_id) {}
-  ~LabelGotoTaskInfo() override {}
-  uint32_t label_id() const { return label_id_; }
-
- private:
-  uint32_t label_id_;
-};
-
-class LabelSwitchTaskInfo : public TaskInfo {
- public:
-  LabelSwitchTaskInfo(const std::string &op_name, uint32_t stream_id, uint32_t label_size,
-                      const std::vector<uint32_t> &label_list, void *cond)
-      : TaskInfo(op_name, stream_id, TaskInfoType::LABEL_SWITCH, false),
-        label_size_(label_size),
-        label_list_(label_list),
-        cond_(cond) {}
-  ~LabelSwitchTaskInfo() override {}
-  uint32_t label_size() const { return label_size_; }
-  const std::vector<uint32_t> &label_list() const { return label_list_; }
-  void *cond() const { return cond_; }
-
- private:
-  uint32_t label_size_;
-  std::vector<uint32_t> label_list_;
-  void *cond_;
-};
-
-class EventTaskInfo : public TaskInfo {
- public:
-  uint32_t event_id() const { return event_id_; }
-
- protected:
-  EventTaskInfo(const std::string &op_name, uint32_t stream_id, TaskInfoType type, uint32_t event_id)
-      : TaskInfo(op_name, stream_id, type, false), event_id_(event_id) {}
-  ~EventTaskInfo() override {}
-
-  uint32_t event_id_;
-};
-
-class EventRecordTaskInfo : public EventTaskInfo {
- public:
-  EventRecordTaskInfo(const std::string &op_name, uint32_t stream_id, uint32_t event_id)
-      : EventTaskInfo(op_name, stream_id, TaskInfoType::EVENT_RECORD, event_id) {}
-  ~EventRecordTaskInfo() override {}
-};
-
-class EventWaitTaskInfo : public EventTaskInfo {
- public:
-  EventWaitTaskInfo(const std::string &op_name, uint32_t stream_id, uint32_t event_id)
-      : EventTaskInfo(op_name, stream_id, TaskInfoType::EVENT_WAIT, event_id) {}
-  ~EventWaitTaskInfo() override {}
-};
-
-class FusionStartTaskInfo : public TaskInfo {
- public:
-  explicit FusionStartTaskInfo(const std::string &op_name, uint32_t stream_id)
-      : TaskInfo(op_name, stream_id, TaskInfoType::FUSION_START, false) {}
-  ~FusionStartTaskInfo() override {}
-};
-
-class FusionEndTaskInfo : public TaskInfo {
- public:
-  explicit FusionEndTaskInfo(const std::string &op_name, uint32_t stream_id)
-      : TaskInfo(op_name, stream_id, TaskInfoType::FUSION_END, false) {}
-  ~FusionEndTaskInfo() override {}
-};
-
-class HcclTaskInfo : public TaskInfo {
- public:
-  HcclTaskInfo(const std::string &op_name, uint32_t stream_id, const std::string hccl_type, void *input_data_addr,
-               void *output_data_addr, int64_t workspace_size, int64_t hccl_stream_num,
-               const std::vector<uint8_t> &private_def, void *ops_kernel_store, int32_t count, int64_t root_id,
-               int64_t op_type, int64_t data_type, const std::string &group, bool dump_flag)
-      : TaskInfo(op_name, stream_id, TaskInfoType::HCCL, dump_flag),
-        hccl_type_(hccl_type),
-        input_data_addr_(input_data_addr),
-        output_data_addr_(output_data_addr),
-        workspace_size_(workspace_size),
-        hccl_stream_num_(hccl_stream_num),
-        private_def_(private_def),
-        ops_kernel_store_(ops_kernel_store),
-        count_(count),
-        root_id_(root_id),
-        op_type_(op_type),
-        data_type_(data_type),
-        group_(group) {}
-  ~HcclTaskInfo() override {}
-
-  const std::string &hccl_type() const { return hccl_type_; }
-  void *input_data_addr() const { return input_data_addr_; }
-  void *output_data_addr() const { return output_data_addr_; }
-  int64_t workspace_size() const { return workspace_size_; }
-  int64_t hccl_stream_num() const { return hccl_stream_num_; }
-  const std::vector<uint8_t> &private_def() const { return private_def_; }
-  void *ops_kernel_store() const { return ops_kernel_store_; }
-  int32_t count() const { return count_; }
-  int64_t root_id() const { return root_id_; }
-  int64_t op_type() const { return op_type_; }
-  int64_t data_type() const { return data_type_; }
-  const std::string &group() const { return group_; }
-
- private:
-  std::string hccl_type_;
-  void *input_data_addr_;
-  void *output_data_addr_;
-  int64_t workspace_size_;
-  int64_t hccl_stream_num_;
-  std::vector<uint8_t> private_def_;
-  void *ops_kernel_store_;
-  int32_t count_;
-  int64_t root_id_;
-  int64_t op_type_;
-  int64_t data_type_;
-  std::string group_;
-};
-
-class ProfilerTraceTaskInfo : public TaskInfo {
- public:
-  ProfilerTraceTaskInfo(const std::string &op_name, uint32_t stream_id, uint64_t log_id, bool notify, uint32_t flat)
-      : TaskInfo(op_name, stream_id, TaskInfoType::PROFILER_TRACE, false),
-        log_id_(log_id),
-        notify_(notify),
-        flat_(flat) {}
-  ~ProfilerTraceTaskInfo() override {}
-
-  uint64_t log_id() const { return log_id_; }
-  bool notify() const { return notify_; }
-  uint32_t flat() const { return flat_; }
-
- private:
-  uint64_t log_id_;
-  bool notify_;
-  uint32_t flat_;
-};
-
-class MemcpyAsyncTaskInfo : public TaskInfo {
- public:
-  MemcpyAsyncTaskInfo(const std::string &op_name, uint32_t stream_id, void *dst, uint64_t dst_max, void *src,
-                      uint64_t count, uint32_t kind, bool dump_flag)
-      : TaskInfo(op_name, stream_id, TaskInfoType::MEMCPY_ASYNC, dump_flag),
-        dst_(dst),
-        dst_max_(dst_max),
-        src_(src),
-        count_(count),
-        kind_(kind) {}
-  ~MemcpyAsyncTaskInfo() override {}
-
-  void *dst() const { return dst_; }
-  uint64_t dst_max() const { return dst_max_; }
-  void *src() const { return src_; }
-  uint64_t count() const { return count_; }
-  uint32_t kind() const { return kind_; }
-
- private:
-  void *dst_;
-  uint64_t dst_max_;
-  void *src_;
-  uint64_t count_;
-  int32_t kind_;
-};
-
-class StreamSwitchTaskInfo : public TaskInfo {
- public:
-  StreamSwitchTaskInfo(const std::string &op_name, uint32_t stream_id, int64_t true_stream_id, void *input_addr,
-                       void *value_addr, int64_t cond, int64_t data_type)
-      : TaskInfo(op_name, stream_id, TaskInfoType::STREAM_SWITCH, false),
-        true_stream_id_(true_stream_id),
-        input_addr_(input_addr),
-        value_addr_(value_addr),
-        cond_(cond),
-        data_type_(data_type) {}
-  ~StreamSwitchTaskInfo() override {}
-
-  int64_t true_stream_id() const { return true_stream_id_; }
-  void *input_addr() const { return input_addr_; }
-  void *value_addr() const { return value_addr_; }
-  int64_t cond() const { return cond_; }
-  int64_t data_type() const { return data_type_; }
-
- private:
-  int64_t true_stream_id_;
-  void *input_addr_;
-  void *value_addr_;
-  int64_t cond_;
-  int64_t data_type_;
-};
-
-class StreamActiveTaskInfo : public TaskInfo {
- public:
-  StreamActiveTaskInfo(const std::string &op_name, uint32_t stream_id, uint32_t active_stream_id)
-      : TaskInfo(op_name, stream_id, TaskInfoType::STREAM_ACTIVE, false), active_stream_id_(active_stream_id) {}
-  ~StreamActiveTaskInfo() override {}
-
-  uint32_t active_stream_id() const { return active_stream_id_; }
-
- private:
-  uint32_t active_stream_id_;
-};
-}  // namespace model_runner
-}  // namespace ge
-
-#endif  // INC_FRAMEWORK_GE_RUNTIME_TASK_INFO_H_

inc/framework/generator/ge_generator.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -21,26 +21,31 @@
 #include <memory>
 #include <string>
 #include <vector>
-#include "ge/ge_ir_build.h"
-#include "common/ge_inner_error_codes.h"
-#include "common/ge_types.h"
+#include "external/ge/ge_ir_build.h"
+#include "framework/common/ge_inner_error_codes.h"
+#include "framework/common/ge_types.h"
 #include "graph/ge_tensor.h"
 #include "graph/graph.h"
 #include "graph/op_desc.h"
 #include "graph/detail/attributes_holder.h"
-#include "omg/omg_inner_types.h"
+#include "framework/omg/omg_inner_types.h"
 
 namespace ge {
+const std::string kAttrSupportDynamicShape = "support_dynamicshape";
+
 class GeRootModel;
 class GE_FUNC_VISIBILITY GeGenerator {
  public:
+  using InOutTensorRef = std::pair<const std::vector<ge::GeTensor> &, const std::vector<ge::GeTensor> &>;
   static GeGenerator &GetInstance() {
     static GeGenerator Instance;
     return Instance;
   }
   GeGenerator() = default;
 
-  ~GeGenerator() { (void)Finalize(); }
+  ~GeGenerator() {
+    (void)Finalize();
+  }
 
   GeGenerator(const GeGenerator &) = delete;
 
@@ -54,7 +59,7 @@ class GE_FUNC_VISIBILITY GeGenerator {
   Status GenerateOfflineModel(const Graph &graph, const std::string &file_name_prefix,
                               const std::vector<GeTensor> &inputs = std::vector<GeTensor>());
 
-  Status GenerateOnlineModel(const Graph &graph, const vector<GeTensor> &inputs, ge::ModelBufferData &model);
+  Status GenerateOnlineModel(const Graph &graph, const std::vector<GeTensor> &inputs, ge::ModelBufferData &model);
 
   Status GenerateInfershapeGraph(const Graph &graph);
 
@@ -81,10 +86,16 @@ class GE_FUNC_VISIBILITY GeGenerator {
   /// @param [in] compile_flag: op build flag, accurate build is 0, fuzz build is 1
   /// @param [out] model_buff: model buff of op.
   /// @return SUCCESS or FAILED
-  Status BuildSingleOpModel(OpDescPtr &op_desc, const vector<GeTensor> &inputs, const vector<GeTensor> &outputs,
-                            OpEngineType engine_type, ModelBufferData &model_buff);
-  Status BuildSingleOpModel(OpDescPtr &op_desc, const vector<GeTensor> &inputs, const vector<GeTensor> &outputs,
-                            OpEngineType engine_type, int32_t compile_flag, ModelBufferData &model_buff);
+  Status BuildSingleOpModel(OpDescPtr &op_desc, const std::vector<GeTensor> &inputs,
+                            const std::vector<GeTensor> &outputs, OpEngineType engine_type,
+                            ModelBufferData &model_buff);
+  Status BuildSingleOpModel(OpDescPtr &op_desc, const std::vector<GeTensor> &inputs,
+                            const std::vector<GeTensor> &outputs, OpEngineType engine_type, int32_t compile_flag,
+                            ModelBufferData &model_buff);
+  Status BuildSingleOpModel(OpDescPtr &op_desc, const std::vector<GeTensor> &inputs,
+                            const std::vector<GeTensor> &outputs, OpEngineType engine_type, int32_t compile_flag,
+                            ModelBufferData &model_buff, GraphStage graph_stage, ComputeGraphPtr &compute_graph);
+
   ///
   /// @ingroup ge
   /// @brief: Build single Op into model buff.
@@ -94,22 +105,33 @@ class GE_FUNC_VISIBILITY GeGenerator {
   /// @param [in] graph_name: graph name.
   /// @param [out] graph: graph of single op.
   /// @return SUCCESS or FAILED
-  Status BuildSingleOpGraph(OpDescPtr &op_desc, const vector<GeTensor> &inputs, const vector<GeTensor> &outputs,
-                            std::string graph_name, Graph &graph);
+  Status BuildSingleOpGraph(const OpDescPtr &op_desc, const InOutTensorRef &inputs_outputs, std::string graph_name,
+                            Graph &graph, std::vector<std::pair<std::string, std::string>> &inputs_name_type) const;
+  Status BuildOriginalGraphInfo(OpDescPtr &op_desc, const std::vector<GeTensor> &inputs,
+                                const std::vector<GeTensor> &outputs, const std::string &model_file_name,
+                                bool is_offline, int32_t compile_flag, GraphStage graph_stage, Graph &graph,
+                                ComputeGraphPtr &compute_graph, bool &fuzz_compile_flag,
+                                std::vector<std::pair<std::string, std::string>> &inputs_name_type);
 
  private:
-  Status GenerateModel(const Graph &graph, const string &file_name_prefix, const vector<GeTensor> &inputs,
+  Status GenerateModel(const Graph &graph, const std::string &file_name_prefix, const std::vector<GeTensor> &inputs,
                        ge::ModelBufferData &model, bool is_offline = true);
-  Status BuildSingleOp(OpDescPtr &op_desc, const vector<GeTensor> &inputs, const vector<GeTensor> &outputs,
-                       const string &model_file_name, OpEngineType engine_type, ModelBufferData &model_buff,
-                       bool is_offline = true, int32_t compile_flag = 0);
+  Status BuildSingleOp(OpDescPtr &op_desc, const std::vector<GeTensor> &inputs, const std::vector<GeTensor> &outputs,
+                       const std::string &model_file_name, OpEngineType engine_type, ModelBufferData &model_buff,
+                       ComputeGraphPtr &comp_graph, bool is_offline = true, int32_t compile_flag = 0,
+                       GraphStage graph_stage = GraphStage::GRAPH_STAGE_RESERVED);
   bool CheckNoAicore(const ComputeGraphPtr &graph);
-  void RemoveConst(const vector<GeTensor> &inputs, vector<GeTensor> &outputs);
-  Status CheckForSingleOp(OpDescPtr &op_desc, const vector<GeTensor> &inputs, const vector<GeTensor> &outputs);
-  Status InferFormatForSingleOp(OpDescPtr &op_desc, Graph &graph);
+  void RemoveConst(const std::vector<GeTensor> &inputs, std::vector<GeTensor> &outputs) const;
+  Status CheckForSingleOp(const OpDescPtr &op_desc, const std::vector<GeTensor> &inputs,
+                          const std::vector<GeTensor> &outputs) const;
+  Status InferFormatForSingleOp(const OpDescPtr &op_desc, const Graph &graph) const;
 
   using GeRootModelPtr = std::shared_ptr<ge::GeRootModel>;
   Status SetModelNameForDump(const GeRootModelPtr &ge_root_model);
+  Status CreateGeneralizedBuildAttrs(const GeRootModelPtr &ge_root_model, const std::vector<GeTensor> &inputs,
+                                     const std::vector<GeTensor> &outputs,
+                                     const std::vector<std::pair<std::string, std::string>> &inputs_name_type,
+                                     std::vector<ge::NamedAttrs> &generalized_build_attrs) const;
 
   class Impl;
 

inc/framework/generator/generator_api.h

@@ -1,187 +0,0 @@
-/**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#ifndef INC_FRAMEWORK_GENERATOR_GENERATOR_API_H_
-#define INC_FRAMEWORK_GENERATOR_GENERATOR_API_H_
-
-#if defined(_MSC_VER)
-#ifdef FUNC_VISIBILITY
-#define GE_FUNC_VISIBILITY _declspec(dllexport)
-#else
-#define GE_FUNC_VISIBILITY
-#endif
-#else
-#ifdef FUNC_VISIBILITY
-#define GE_FUNC_VISIBILITY __attribute__((visibility("default")))
-#else
-#define GE_FUNC_VISIBILITY
-#endif
-#endif
-
-#include <stdint.h>
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-typedef uint32_t Status_t;
-
-typedef void *OpAttr_t;
-typedef void *OpTensor_t;
-
-///
-/// @ingroup ge
-/// @brief Generate offline model for the op.
-/// @param [in] op_type: type name of the op.
-/// @param [in] in_tensor: input description array (created by OpTensorCreate).
-/// @param [in] in_num: number of in_tensor.
-/// @param [in] out_tensor: output description array (created by OpTensorCreate).
-/// @param [in] out_num: number of out_tensor.
-/// @param [in] attr: the attributes of the op (created by OpAttrCreate).
-/// @param [in] om_file: file name for the om to save.
-/// @return 0 for success / others for fail
-///
-GE_FUNC_VISIBILITY extern Status_t OpTaskGernerator(const char *op_type, const OpTensor_t *in_tensor, int in_num,
-                                                    const OpTensor_t *out_tensor, int out_num, const OpAttr_t attr,
-                                                    const char *om_file);
-
-///
-/// @ingroup ge
-/// @brief Create Tensor Description.
-/// @param [in] format: tensor format of the data.
-/// @param [in] datatype: tensor type of the data.
-/// @param [in] shape: tensor shape array.
-/// @param [in] num: number of shape.
-/// @return OpTensor_t for success / nullptr for failure
-///
-GE_FUNC_VISIBILITY extern OpTensor_t OpTensorCreate(int format, int datatype, const int64_t *shape, int num);
-
-///
-/// @ingroup ge
-/// @brief Destroy Tensor Description.
-/// @param [in] OpTensor_t tensor: created by OpTensorCreate.
-/// @param [out] none
-/// @return 0 for success / others for failure.
-///
-GE_FUNC_VISIBILITY extern Status_t OpTensorDestroy(OpTensor_t tensor);
-
-///
-/// @ingroup ge
-/// @brief Create an attribute holder.
-/// @param [in] none
-/// @param [out] none
-/// @return OpAttr_t for success / nullptr for failure.
-///
-GE_FUNC_VISIBILITY extern OpAttr_t OpAttrCreate();
-
-///
-/// @ingroup ge
-/// @brief Destroy Attribute holder.
-/// @param [in] OpAttr_t attr: created by OpAttrCreate.
-/// @param [out] none
-/// @return 0 for success / others for failure.
-///
-GE_FUNC_VISIBILITY extern Status_t OpAttrDestroy(OpAttr_t attr);
-
-///
-/// @ingroup ge
-/// @brief Set a boolean attribute to the attribute holder.
-/// @param [in] attr: attribute holder (created by OpAttrCreate).
-/// @param [in] name: attribute name (can`t be nullptr, end with '\0').
-/// @param [in] value: attributed value.
-/// @return 0 for success / others for failure.
-///
-GE_FUNC_VISIBILITY extern Status_t SetAttrBool(OpAttr_t attr, const char *name, bool value);
-
-///
-/// @ingroup ge
-/// @brief Set an integer attribute to the attribute holder.
-/// @param [in] attr: attribute holder (created by OpAttrCreate).
-/// @param [in] name: attribute name (can`t be nullptr, end with '\0').
-/// @param [in] value: attribute value.
-/// @return 0 for success / others for failure.
-///
-GE_FUNC_VISIBILITY extern Status_t SetAttrInt(OpAttr_t attr, const char *name, int64_t value);
-
-///
-/// @ingroup ge
-/// @brief Set a float attribute to the attribute holder.
-/// @param [in] attr: attribute holder (created by OpAttrCreate).
-/// @param [in] name: attribute name (can`t be nullptr, end with '\0').
-/// @param [in] value: attribute value.
-/// @return 0 for success / others for failure.
-///
-GE_FUNC_VISIBILITY extern Status_t SetAttrFloat(OpAttr_t attr, const char *name, float value);
-
-///
-/// @ingroup ge
-/// @brief Set a string attribute to the attribute holder.
-/// @param [in] attr: attribute holder (created by OpAttrCreate).
-/// @param [in] name: attribute name (can`t be nullptr, end with '\0').
-/// @param [in] value: attribute value (can`t be nullptr, end with '\0').
-/// @return 0 for success / others for failure.
-///
-GE_FUNC_VISIBILITY extern Status_t SetAttrString(OpAttr_t attr, const char *name, const char *value);
-
-///
-/// @ingroup ge
-/// @brief Set a boolean array attribute to the attribute holder.
-/// @param [in] attr: attribute holder (created by OpAttrCreate).
-/// @param [in] name: attribute name (can`t be nullptr, end with '\0').
-/// @param [in] value: attribute value array.
-/// @param [in] num: number of value array.
-/// @return 0 for success / others for failure.
-///
-GE_FUNC_VISIBILITY extern Status_t SetAttrBoolList(OpAttr_t attr, const char *name, const bool *value, int num);
-
-///
-/// @ingroup ge
-/// @brief Set an integer array attribute to the attribute holder.
-/// @param [in] attr: attribute holder (created by OpAttrCreate).
-/// @param [in] name: attribute name (can`t be nullptr, end with '\0').
-/// @param [in] value: attribute value array.
-/// @param [in] num: number of value array.
-/// @return 0 for success / others for failure.
-///
-GE_FUNC_VISIBILITY extern Status_t SetAttrIntList(OpAttr_t attr, const char *name, const int64_t *value, int num);
-
-///
-/// @ingroup ge
-/// @brief Set a float array attribute to the attribute holder.
-/// @param [in] attr: attribute holder (created by OpAttrCreate).
-/// @param [in] name: attribute name (can`t be nullptr, end with '\0').
-/// @param [in] value: attribute value array.
-/// @param [in] num: number of value array.
-/// @return 0 for success / others for failure.
-///
-GE_FUNC_VISIBILITY extern Status_t SetAttrFloatList(OpAttr_t attr, const char *name, const float *value, int num);
-
-///
-/// @ingroup ge
-/// @brief Set a string array attribute to the attribute holder.
-/// @param [in] attr: attribute holder (created by OpAttrCreate).
-/// @param [in] name: attribute name (can`t be nullptr, end with '\0').
-/// @param [in] value: attribute value array (each value can`t be nullptr, end with '\0').
-/// @param [in] num: number of value array.
-/// @return 0 for success / others for failure.
-///
-GE_FUNC_VISIBILITY extern Status_t SetAttrStringList(OpAttr_t attr, const char *name, const char **value, int num);
-
-#ifdef __cplusplus
-}
-#endif
-
-#endif  // INC_FRAMEWORK_GENERATOR_GENERATOR_API_H_

inc/framework/memory/memory_api.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -17,11 +17,7 @@
 #ifndef INC_FRAMEWORK_MEMORY_MEMORY_API_H_
 #define INC_FRAMEWORK_MEMORY_MEMORY_API_H_
 
-#include <string>
-#include <vector>
-
-#include "ge/ge_api_error_codes.h"
-#include "graph//types.h"
+#include "external/ge/ge_api_error_codes.h"
 #include "runtime/mem.h"
 
 namespace ge {

inc/framework/memory/memory_assigner.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -19,7 +19,7 @@
 
 #include <utility>
 
-#include "common/ge_inner_error_codes.h"
+#include "framework/common/ge_inner_error_codes.h"
 #include "graph/node.h"
 
 namespace ge {
@@ -33,7 +33,7 @@ class GE_FUNC_VISIBILITY MemoryAssigner {
 
   MemoryAssigner &operator=(const MemoryAssigner &) = delete;
 
-  Status AssignMemory(bool is_loop_graph, map<uint64_t, size_t> &mem_offset, size_t &zero_copy_mem_size);
+  Status AssignMemory(std::map<uint64_t, size_t> &mem_offset, size_t &zero_copy_mem_size);
 
  private:
   ge::ComputeGraphPtr compute_graph_;

inc/framework/omg/ge_init.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -18,7 +18,7 @@
 #define INC_FRAMEWORK_OMG_GE_INIT_H_
 #include <map>
 #include <string>
-#include "common/ge_inner_error_codes.h"
+#include "framework/common/ge_inner_error_codes.h"
 
 namespace ge {
 class GE_FUNC_VISIBILITY GEInit {

inc/framework/omg/model_tool.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.

inc/framework/omg/omg.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -17,10 +17,12 @@
 #ifndef INC_FRAMEWORK_OMG_OMG_H_
 #define INC_FRAMEWORK_OMG_OMG_H_
 
-#include <google/protobuf/message.h>
 #include <string>
 #include <unordered_map>
 #include <vector>
+
+#include <google/protobuf/message.h>
+#include "external/ge/ge_api_types.h"
 #include "framework/omg/omg_inner_types.h"
 #include "framework/omg/parser/parser_inner_ctx.h"
 #include "proto/ge_ir.pb.h"
@@ -31,20 +33,14 @@
 #include "graph/model.h"
 #include "runtime/kernel.h"
 
-using domi::Status;
-using std::pair;
-using std::string;
-using std::unordered_map;
-using std::vector;
-
 namespace ge {
 /**
  * @ingroup domi_omg
  * @brief init omg context
  * @return void
  */
-GE_FUNC_VISIBILITY Status InitDomiOmgContext(const string &input_shape, const string &input_format,
-                                             const string &net_format, bool is_dynamic_input);
+GE_FUNC_VISIBILITY domi::Status InitDomiOmgContext(const std::string &input_shape, const std::string &input_format,
+                                                   const std::string &net_format, bool is_dynamic_input);
 
 /**
  * @ingroup domi_omg
@@ -61,10 +57,10 @@ GE_FUNC_VISIBILITY Status InitDomiOmgContext(const string &input_shape, const st
  * @param [in] atc_params multiply atc params
  * @return Status result code
  */
-GE_FUNC_VISIBILITY Status ParseGraph(ge::Graph &graph, const std::map<string, string> &atc_params,
-                                     const char *model_file, const char *weights_file, domi::FrameworkType type,
-                                     const char *op_conf = nullptr, const char *target = nullptr,
-                                     RunMode run_mode = GEN_OM_MODEL, bool is_dynamic_input = false);
+GE_FUNC_VISIBILITY domi::Status ParseGraph(ge::Graph &graph, const std::map<std::string, std::string> &atc_params,
+                                           const char *model_file, const char *weights_file, domi::FrameworkType type,
+                                           const char *op_conf = nullptr, const char *target = nullptr,
+                                           RunMode run_mode = RunMode::GEN_OM_MODEL, bool is_dynamic_input = false);
 
 /**
  * @ingroup domi_omg
@@ -74,9 +70,9 @@ GE_FUNC_VISIBILITY Status ParseGraph(ge::Graph &graph, const std::map<string, st
  * @param [key] encrypted key
  * @return Status result code
  */
-GE_FUNC_VISIBILITY Status ConvertOm(const char *model_file, const char *json_file, bool is_covert_to_json);
+GE_FUNC_VISIBILITY domi::Status ConvertOm(const char *model_file, const char *json_file, bool is_covert_to_json);
 
-GE_FUNC_VISIBILITY Status ConvertPbtxtToJson(const char *model_file, const char *json_file);
+GE_FUNC_VISIBILITY domi::Status ConvertPbtxtToJson(const char *model_file, const char *json_file);
 /**
  * @ingroup domi_omg
  * @brief convert the model file in protobuf format into a JSON file.
@@ -86,23 +82,24 @@ GE_FUNC_VISIBILITY Status ConvertPbtxtToJson(const char *model_file, const char
  * @param [key] encrypted key
  * @return Status result code
  */
-GE_FUNC_VISIBILITY Status ConvertFwkModelToJson(domi::FrameworkType framework, const char *model_file,
-                                                const char *json_file);
+GE_FUNC_VISIBILITY domi::Status ConvertFwkModelToJson(const domi::FrameworkType framework, const char *model_file,
+                                                      const char *json_file);
 
-GE_FUNC_VISIBILITY void GetGroupName(ge::proto::ModelDef &model);
+GE_FUNC_VISIBILITY void GetGroupName(ge::proto::ModelDef &model_def);
 
-GE_FUNC_VISIBILITY void FindParserSo(const string &path, vector<string> &fileList, string &caffe_parser_path);
+GE_FUNC_VISIBILITY void FindParserSo(const std::string &path, std::vector<std::string> &file_list,
+                                     std::string &caffe_parser_path);
 
-GE_FUNC_VISIBILITY Status DumpInfershapeJson(const ge::Graph &graph, const char *json_file);
+GE_FUNC_VISIBILITY domi::Status DumpInfershapeJson(const ge::Graph &graph, const char *json_file);
 
-GE_FUNC_VISIBILITY Status SetOutputNodeInfo(ge::Graph &graph, const std::string &output_type,
-                                            const std::string &output_format);
+GE_FUNC_VISIBILITY domi::Status SetOutputNodeInfo(ge::Graph &graph, const std::string &output_type,
+                                                  const std::string &output);
 
-GE_FUNC_VISIBILITY Status GetOutputLeaf(ge::NodePtr node,
-                                        std::vector<std::pair<ge::NodePtr, int32_t>> &output_nodes_info);
+GE_FUNC_VISIBILITY domi::Status GetOutputLeaf(ge::NodePtr node,
+                                              std::vector<std::pair<ge::NodePtr, int32_t>> &output_nodes_info);
 
-GE_FUNC_VISIBILITY void GetOutputNodesNameAndIndex(std::vector<std::pair<ge::NodePtr, int32_t>> &output_nodes_info,
-                                                   std::vector<std::string> &output_nodes_name);
+GE_FUNC_VISIBILITY void CreateOutputNodesInfo(std::vector<std::pair<ge::NodePtr, int32_t>> &output_nodes_info,
+                                              std::vector<std::string> &output_nodes_name);
 
 GE_FUNC_VISIBILITY void UpdateOmgCtxWithParserCtx();
 

inc/framework/omg/omg_inner_types.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -31,12 +31,7 @@
 using domi::DOMI_TENSOR_ND;
 using domi::DOMI_TENSOR_RESERVED;
 using domi::domiTensorFormat_t;
-using domi::FRAMEWORK_RESERVED;
-using domi::FrameworkType;
-using std::map;
-using std::string;
 using std::unordered_map;
-using std::vector;
 
 namespace ge {
 /**
@@ -51,34 +46,13 @@ enum RunMode {
   DISPLAY_OM_INFO = 6  // display model info
 };
 
-///
-/// @ingroup domi_omg
-/// @brief high-precision mode
-///
-enum HighPrecisionMode {
-  // the FP16 high-precision function is disabled in common mode
-  HIGH_PRECISION_DEFAULT = 0,
-
-  // high-precision mode, enabling FP16 high-precision mode (Convolution/FullConnect/AvgPooling are involved)
-  HIGH_PRECISION_FP16 = 1
-};
-
-///
-/// @ingroup domi_omg
-/// @brief description buffer data
-///
-struct OMGBufferData {
-  void *data;
-  uint32_t length;
-};
-
 struct OmgContext {
-  OmgContext() { format = DOMI_TENSOR_ND; }
-  domiTensorFormat_t format;
+  OmgContext() : format(domi::DOMI_TENSOR_ND) {}
+  domi::domiTensorFormat_t format;
 
   // format of the input specified by the command line
-  std::unordered_map<std::string, domiTensorFormat_t> input_nodes_format_map;
-  std::vector<domiTensorFormat_t> output_formats;
+  std::unordered_map<std::string, domi::domiTensorFormat_t> input_nodes_format_map;
+  std::vector<domi::domiTensorFormat_t> output_formats;
 
   // user-designate input dims
   std::vector<std::pair<std::string, std::vector<int64_t>>> user_input_dims;
@@ -96,18 +70,18 @@ struct OmgContext {
   // default out nodes (this is used for determing the orders)
   std::vector<std::pair<std::string, int32_t>> default_out_nodes;
   // save the output node of the network, value = topName,
-  // topName indicates the output name of the operator.
-  std::vector<std::string> user_out_nodes_top_vec;
+  // tensorName indicates the output name of the operator.
+  std::vector<std::string> user_out_tensors;
   // net out nodes (where user_out_nodes or leaf nodes)
   std::vector<std::string> net_out_nodes;
-  // net out nodes top names(only caffe has top)
-  std::vector<std::string> out_top_names;
-  // net data nodes top names(only caffe has top)
-  std::vector<std::string> data_top_names;
+  // net out nodes tensor names(caffe or onnx)
+  std::vector<std::string> out_tensor_names;
+  // net data nodes tensor names(caffe or onnx)
+  std::vector<std::string> data_tensor_names;
   // preferential format used by the entire network
-  domiTensorFormat_t net_format = DOMI_TENSOR_RESERVED;
+  domi::domiTensorFormat_t net_format = domi::DOMI_TENSOR_RESERVED;
   domi::FrameworkType type = domi::FRAMEWORK_RESERVED;
-  RunMode run_mode = ONLY_PRE_CHECK;
+  RunMode run_mode = RunMode::ONLY_PRE_CHECK;
   bool train_flag = false;
 
   std::string output_type;
@@ -118,14 +92,13 @@ struct OmgContext {
   std::string dynamic_image_size;
   std::string dynamic_dims;
   std::string dynamic_node_type;
-  std::vector<std::vector<int64_t>> user_real_input_dims;
-  std::vector<int64_t> cur_dynamic_dims;
   bool need_multi_batch = false;
   std::vector<NodePtr> data_nodes;
   std::vector<NodePtr> getnext_nosink_nodes;
   bool fuzz_compile_flag = false;
   std::string atc_cmdline;
   bool user_attr_index_valid = false;
+  bool is_online_model = false;
 };
 }  // namespace ge
 

inc/framework/omg/omg_types.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.

inc/framework/omg/parser/model_parser.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -25,8 +25,6 @@
 #include "graph/ge_tensor.h"
 #include "graph/graph.h"
 #include "graph/op_desc.h"
-#include "graph/operator.h"
-#include "graph/range_vistor.h"
 #include "graph/utils/attr_utils.h"
 #include "graph/utils/graph_utils.h"
 #include "graph/utils/op_desc_utils.h"
@@ -36,7 +34,7 @@ using Status = domi::Status;
 
 namespace domi {
 using GetGraphCallback = std::function<std::unique_ptr<google::protobuf::Message>(
-  const google::protobuf::Message *root_proto, const std::string &graph)>;
+    const google::protobuf::Message *root_proto, const std::string &graph)>;
 
 using GetGraphCallbackV2 = std::function<std::string(const std::string &subgraph_name)>;
 
@@ -109,7 +107,11 @@ class GE_FUNC_VISIBILITY ModelParser {
    * @return SUCCESS
    * @return Others failed
    */
-  virtual Status ToJson(const char *model_file, const char *json_file) { return domi::SUCCESS; }
+  virtual Status ToJson(const char *model_file, const char *json_file) {
+    (void)model_file;
+    (void)json_file;
+    return SUCCESS;
+  }
 
   /*
    * @ingroup domi_omg
@@ -129,7 +131,11 @@ class GE_FUNC_VISIBILITY ModelParser {
    * @return SUCCESS
    * @return Others failed
    */
-  virtual Status ParseProto(const std::string &serialized_proto, ge::ComputeGraphPtr &graph) { return UNSUPPORTED; }
+  virtual Status ParseProto(const std::string &serialized_proto, ge::ComputeGraphPtr &graph) {
+    (void)serialized_proto;
+    (void)graph;
+    return UNSUPPORTED;
+  }
 
   /**
    * @ingroup domi_omg
@@ -142,8 +148,22 @@ class GE_FUNC_VISIBILITY ModelParser {
    */
   virtual Status ParseProtoWithSubgraph(const std::string &serialized_proto, GetGraphCallbackV2 callback,
                                         ge::ComputeGraphPtr &graph) {
+    (void)serialized_proto;
+    (void)callback;
+    (void)graph;
     return UNSUPPORTED;
   }
+
+  virtual bool HasError() {
+    return false;
+  }
+
+  virtual Status Save(const std::string &file) {
+    (void)file;
+    return SUCCESS;
+  }
+
+  virtual void Clear(){};
 };
 }  // namespace domi
 

inc/framework/omg/parser/op_parser.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -19,10 +19,8 @@
 
 #include <google/protobuf/text_format.h>
 #include "framework/omg/parser/parser_types.h"
-#include "omg/omg_inner_types.h"
+#include "framework/omg/omg_inner_types.h"
 #include "proto/om.pb.h"
-#include "graph/ge_tensor.h"
-#include "graph/op_desc.h"
 #include "graph/utils/op_desc_utils.h"
 
 using google::protobuf::Message;
@@ -50,7 +48,7 @@ class GE_FUNC_VISIBILITY OpParser {
    * @return SUCCESS
    * @return FAILED
    */
-  virtual Status ParseParams(const Message *op_src, ge::OpDescPtr &op_desc) = 0;
+  virtual domi::Status ParseParams(const google::protobuf::Message *op_src, ge::OpDescPtr &op_desc) = 0;
 
   /**
    * @ingroup domi_omg
@@ -60,7 +58,7 @@ class GE_FUNC_VISIBILITY OpParser {
    * @return SUCCESS
    * @return FAILED
    */
-  virtual Status ParseParams(const Message *op_src, ge::Operator &op_dest) = 0;
+  virtual domi::Status ParseParams(const google::protobuf::Message *op_src, ge::Operator &op_dest) = 0;
 
   /**
    * @ingroup domi_omg
@@ -70,7 +68,7 @@ class GE_FUNC_VISIBILITY OpParser {
    * @return SUCCESS
    * @return FAILED
    */
-  virtual Status ParseWeights(const Message *op_src, ge::NodePtr &node) = 0;
+  virtual domi::Status ParseWeights(const google::protobuf::Message *op_src, ge::NodePtr &node) = 0;
 
   /**
    * @ingroup domi_omg
@@ -80,7 +78,7 @@ class GE_FUNC_VISIBILITY OpParser {
    * @return SUCCESS
    * @return FAILED
    */
-  virtual Status GetFormat(const Message *op_src, domi::domiTensorFormat_t &format) {
+  virtual domi::Status GetFormat(const google::protobuf::Message *op_src, domi::domiTensorFormat_t &format) {
     (void)op_src;
     // Indicates that the op does not provide a value for format
     format = domi::DOMI_TENSOR_RESERVED;

inc/framework/omg/parser/parser_api.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -17,15 +17,14 @@
 #ifndef INC_FRAMEWORK_OMG_PARSER_PARSER_API_H_
 #define INC_FRAMEWORK_OMG_PARSER_PARSER_API_H_
 
-#include <iostream>
 #include <map>
 #include <string>
-#include "ge/ge_api_error_codes.h"
+#include "external/ge/ge_api_error_codes.h"
 
 namespace ge {
 // Initialize parser
-GE_FUNC_VISIBILITY Status ParserInitialize(const std::map<std::string, std::string>& options);
+GE_FUNC_VISIBILITY Status ParserInitialize(const std::map<std::string, std::string> &options);
 // Finalize parser, release all resources
 GE_FUNC_VISIBILITY Status ParserFinalize();
 }  // namespace ge
-#endif // INC_FRAMEWORK_OMG_PARSER_PARSER_API_H_
+#endif  // INC_FRAMEWORK_OMG_PARSER_PARSER_API_H_

inc/framework/omg/parser/parser_factory.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -23,14 +23,13 @@
 #include <string>
 #include "framework/omg/omg_inner_types.h"
 #include "framework/omg/parser/parser_types.h"
-
-using Status = domi::Status;
+#include "external/register/register.h"
 
 namespace domi {
 class WeightsParser;
 class ModelParser;
 
-typedef std::shared_ptr<ModelParser> (*MODEL_PARSER_CREATOR_FUN)(void);
+using MODEL_PARSER_CREATOR_FUN = std::shared_ptr<ModelParser> (*)(void);
 
 // Create modelparser for different frameworks
 class GE_FUNC_VISIBILITY ModelParserFactory {
@@ -63,7 +62,7 @@ class GE_FUNC_VISIBILITY ModelParserFactory {
 
 class GE_FUNC_VISIBILITY ModelParserRegisterar {
  public:
-  ModelParserRegisterar(const domi::FrameworkType type, MODEL_PARSER_CREATOR_FUN fun) {
+  ModelParserRegisterar(const domi::FrameworkType type, MODEL_PARSER_CREATOR_FUN const fun) noexcept {
     ModelParserFactory::Instance()->RegisterCreator(type, fun);
   }
   ~ModelParserRegisterar() {}
@@ -82,7 +81,7 @@ class GE_FUNC_VISIBILITY ModelParserRegisterar {
   }                                                              \
   ModelParserRegisterar g_##type##_Model_Parser_Creator(type, Creator_##type##_Model_Parser)
 
-typedef std::shared_ptr<WeightsParser> (*WEIGHTS_PARSER_CREATOR_FUN)(void);
+using WEIGHTS_PARSER_CREATOR_FUN = std::shared_ptr<WeightsParser> (*)(void);
 
 // Create weightsparser for different frameworks
 class GE_FUNC_VISIBILITY WeightsParserFactory {
@@ -115,7 +114,7 @@ class GE_FUNC_VISIBILITY WeightsParserFactory {
 
 class GE_FUNC_VISIBILITY WeightsParserRegisterar {
  public:
-  WeightsParserRegisterar(const domi::FrameworkType type, WEIGHTS_PARSER_CREATOR_FUN fun) {
+  WeightsParserRegisterar(const domi::FrameworkType type, WEIGHTS_PARSER_CREATOR_FUN const fun) noexcept {
     WeightsParserFactory::Instance()->RegisterCreator(type, fun);
   }
   ~WeightsParserRegisterar() {}
@@ -133,6 +132,12 @@ class GE_FUNC_VISIBILITY WeightsParserRegisterar {
     return std::shared_ptr<WeightsParser>(ptr);                      \
   }                                                                  \
   WeightsParserRegisterar g_##type##_Weights_Parser_Creator(type, Creator_##type##_Weights_Parser)
-};  // namespace domi
+
+class GE_FUNC_VISIBILITY OpRegTbeParserFactory {
+ public:
+  static OpRegTbeParserFactory *Instance();
+  void Finalize(const domi::OpRegistrationData &reg_data);
+};
+}  // namespace domi
 
 #endif  // INC_FRAMEWORK_OMG_PARSER_PARSER_FACTORY_H_

inc/framework/omg/parser/parser_inner_ctx.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -21,7 +21,6 @@
 #include <memory>
 #include <string>
 #include <unordered_map>
-#include <utility>
 #include <vector>
 #include "external/register/register_fmk_types.h"
 #include "external/register/register_types.h"
@@ -30,8 +29,8 @@
 namespace ge {
 struct ParserContext {
   // format of the input specified by the command line
-  std::unordered_map<std::string, domiTensorFormat_t> input_nodes_format_map;
-  std::vector<domiTensorFormat_t> output_formats;
+  std::unordered_map<std::string, domi::domiTensorFormat_t> input_nodes_format_map;
+  std::vector<domi::domiTensorFormat_t> output_formats;
   // user-designate input dims
   std::vector<std::pair<std::string, std::vector<int64_t>>> user_input_dims;
   std::map<std::string, std::vector<int64_t>> input_dims;
@@ -46,20 +45,20 @@ struct ParserContext {
   // operator
   std::map<std::string, std::vector<int32_t>> out_nodes_map;
   // save the output node of the network, value = topName,
-  // topName indicates the output name of the operator.
-  std::vector<std::string> user_out_nodes_top_vec;
+  // tensorName indicates the output name of the operator.
+  std::vector<std::string> user_out_tensors;
   // net out nodes (where user_out_nodes or leaf nodes)
   std::vector<std::string> net_out_nodes;
-  // net data nodes top names(only caffe has top)
-  std::vector<std::string> data_top_names;
-  // net out nodes top names(only caffe has top)
-  std::vector<std::string> out_top_names;
+  // net out nodes tensor names(caffe or onnx)
+  std::vector<std::string> out_tensor_names;
+  // net data nodes tensor names(caffe or onnx)
+  std::vector<std::string> data_tensor_names;
   // Whether to use dynamic batch size or dynamic image size
   bool is_dynamic_input = false;
   bool train_flag = false;
   domi::domiTensorFormat_t format = domi::DOMI_TENSOR_ND;
   domi::FrameworkType type = domi::FRAMEWORK_RESERVED;
-  RunMode run_mode = GEN_OM_MODEL;
+  RunMode run_mode = RunMode::GEN_OM_MODEL;
   // save caffe custom proto path, used by caffe parse
   std::string custom_proto_path;
   // save caffe proto path, used by caffe parse

inc/framework/omg/parser/parser_types.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -17,7 +17,7 @@
 #ifndef PARSER_COMMON_TYPES_H_
 #define PARSER_COMMON_TYPES_H_
 
-#include <stdint.h>
+#include <cstdint>
 #include <string>
 
 #include "register/register_types.h"
@@ -133,6 +133,7 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const char *_IF;
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const char *STATELESSIF;
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const char *IF;
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const char *CASE;
+FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const char *STATELESSCASE;
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const char *_WHILE;
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const char *WHILE;
 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const char *STATELESSWHILE;

inc/framework/omg/parser/weights_parser.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -17,13 +17,12 @@
 #ifndef INC_FRAMEWORK_OMG_PARSER_WEIGHTS_PARSER_H_
 #define INC_FRAMEWORK_OMG_PARSER_WEIGHTS_PARSER_H_
 
+#include "external/register/register_error_codes.h"
 #include "graph/graph.h"
 #include "graph/attr_value.h"
 #include "graph/compute_graph.h"
 #include "graph/ge_tensor.h"
 #include "graph/op_desc.h"
-#include "graph/operator.h"
-#include "graph/range_vistor.h"
 #include "graph/utils/attr_utils.h"
 #include "graph/utils/op_desc_utils.h"
 #include "graph/utils/tensor_utils.h"
@@ -68,6 +67,17 @@ class GE_FUNC_VISIBILITY WeightsParser {
    * @author
    */
   virtual Status ParseFromMemory(const char *input, uint32_t lengt, ge::ComputeGraphPtr &graph) = 0;
+
+  virtual bool HasError() {
+    return false;
+  }
+
+  virtual Status Save(const std::string &file) {
+    (void)file;
+    return SUCCESS;
+  }
+
+  virtual void Clear() {}
 };
 }  // namespace domi
 

inc/framework/omg/version.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -18,12 +18,9 @@
 #define INC_FRAMEWORK_OMG_VERSION_H_
 
 #include <memory>
-#include <set>
-#include <string>
-#include <vector>
 
-#include "common/debug/log.h"
-#include "common/string_util.h"
+#include "framework/common/debug/log.h"
+#include "framework/common/string_util.h"
 #include "framework/common/debug/ge_log.h"
 
 namespace ge {
@@ -33,8 +30,8 @@ class GE_FUNC_VISIBILITY PlatformVersionManager {
   ~PlatformVersionManager() = delete;
   static Status GetPlatformVersion(std::string &ver) {
     ver = "1.11.z";
-    std::vector<std::string> version_splits = StringUtils::Split(ver, '.');
-    GE_IF_BOOL_EXEC(version_splits.size() < 3, GELOGW("Read platform version error!"); return FAILED;);
+    const std::vector<std::string> version_splits = StringUtils::Split(ver, '.');
+    GE_IF_BOOL_EXEC(version_splits.size() < 3U, GELOGW("Read platform version error!"); return FAILED;);
 
     GELOGI("Read current platform version: %s.", ver.c_str());
     return SUCCESS;

inc/framework/pne/pne_model.h

@@ -0,0 +1,128 @@
+/**
+ * Copyright (c) Huawei Technologies Co., Ltd. 2022. All rights reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef INC_FRAMEWORK_PNE_MODEL_H_
+#define INC_FRAMEWORK_PNE_MODEL_H_
+
+#include <map>
+#include <string>
+#include <vector>
+
+#include "graph/compute_graph.h"
+#include "framework/common/debug/log.h"
+#include "framework/common/ge_inner_error_codes.h"
+#include "framework/common/ge_types.h"
+#include "framework/engine/dnnengine.h"
+#include "external/ge/ge_ir_build.h"
+
+namespace ge {
+const std::string PNE_ID_NPU = "NPU";
+const std::string PNE_ID_CPU = "HOST_CPU";
+
+struct ModelRelation;
+class PneModel {
+ public:
+  PneModel() = default;
+  explicit PneModel(const ComputeGraphPtr &root_graph) : root_graph_(root_graph){};
+  virtual ~PneModel() = default;
+  PneModel(const PneModel &other) = delete;
+  PneModel &operator=(const PneModel &other) = delete;
+
+ public:
+  inline Status AddSubModel(const shared_ptr<PneModel> &submodel, std::string type = "") {
+    if (submodel == nullptr) {
+      GELOGE(INTERNAL_ERROR, "submodel is nullptr, type = %s", type.c_str());
+      return INTERNAL_ERROR;
+    }
+    submodel->SetModelType(type);
+    if (!submodels_.emplace(submodel->GetModelName(), submodel).second) {
+      GELOGE(INTERNAL_ERROR, "submodel already exist, name = %s, type = %s", submodel->GetModelName().c_str(),
+             type.c_str());
+      return INTERNAL_ERROR;
+    }
+    return SUCCESS;
+  }
+
+  inline const std::shared_ptr<PneModel> GetSubmodel(const std::string &name) const {
+    const auto &it = submodels_.find(name);
+    if (it == submodels_.end()) {
+      return nullptr;
+    }
+    return it->second;
+  }
+
+  inline const std::map<std::string, std::shared_ptr<PneModel>> &GetSubmodels() const {
+    return submodels_;
+  }
+
+  inline void SetModelType(const std::string &type) {
+    model_type_ = type;
+  }
+
+  inline const std::string &GetModelType() const {
+    return model_type_;
+  }
+
+  inline void SetModelName(const std::string &model_name) {
+    model_name_ = model_name;
+  }
+
+  inline const std::string &GetModelName() const {
+    return model_name_;
+  }
+
+  inline void SetRootGraph(const ComputeGraphPtr graph) {
+    root_graph_ = graph;
+  }
+
+  inline const ComputeGraphPtr &GetRootGraph() const {
+    return root_graph_;
+  }
+
+  inline void SetModelRelation(std::shared_ptr<ModelRelation> model_relation) {
+    model_relation_ = std::move(model_relation);
+  }
+
+  inline const std::shared_ptr<ModelRelation> GetModelRelation() const {
+    return model_relation_;
+  }
+
+ public:
+  virtual Status SerializeModel(ModelBufferData &model_buff) = 0;
+
+  virtual Status UnSerializeModel(const ModelBufferData &model_buff) = 0;
+
+  virtual void SetModelId(const uint32_t model_id) {
+    model_id_ = model_id;
+  }
+
+  virtual uint32_t GetModelId() const {
+    return model_id_;
+  }
+
+ private:
+  std::map<std::string, std::shared_ptr<PneModel>> submodels_;
+  std::shared_ptr<ModelRelation> model_relation_;
+  ComputeGraphPtr root_graph_ = nullptr;
+  std::string model_name_;
+  std::string model_type_;
+  uint32_t model_id_ = INVALID_MODEL_ID;
+};
+
+using PneModelPtr = std::shared_ptr<PneModel>;
+}  // namespace ge
+
+#endif  // INC_FRAMEWORK_PNE_MODEL_H_

inc/framework/pne/process_node_engine.h

@@ -0,0 +1,67 @@
+/**
+ * Copyright (c) Huawei Technologies Co., Ltd. 2022. All rights reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef INC_FRAMEWORK_PROCESS_NODE_ENGINE_H_
+#define INC_FRAMEWORK_PROCESS_NODE_ENGINE_H_
+
+#include <map>
+#include <string>
+#include <vector>
+
+#include "framework/common/ge_inner_error_codes.h"
+#include "framework/common/ge_types.h"
+#include "graph/manager/graph_manager_utils.h"
+#include "framework/pne/pne_model.h"
+
+namespace ge {
+class ProcessNodeEngineImpl {
+ public:
+  virtual Status OptimizeGraph(const std::vector<GeTensor> &inputs, ComputeGraphPtr &compute_graph) = 0;
+
+  virtual Status BuildGraph(ComputeGraphPtr &compute_graph, PneModelPtr &model) = 0;
+};
+
+using ProcessNodeEngineImplPtr = std::shared_ptr<ProcessNodeEngineImpl>;
+
+class ProcessNodeEngine {
+ public:
+  ProcessNodeEngine() = default;
+  virtual ~ProcessNodeEngine() = default;
+  ProcessNodeEngine(const ProcessNodeEngine &other) = delete;
+  ProcessNodeEngine &operator=(const ProcessNodeEngine &other) = delete;
+
+ public:
+  virtual Status Initialize(const std::map<std::string, std::string> &options) = 0;
+
+  virtual Status Finalize() = 0;
+
+  virtual Status OptimizeGraph(const std::vector<GeTensor> &inputs, ComputeGraphPtr &compute_graph) = 0;
+
+  virtual Status BuildGraph(ComputeGraphPtr &compute_graph, PneModelPtr &model) = 0;
+
+  virtual const std::string &GetEngineName(const ge::NodePtr &node_ptr = nullptr) const = 0;
+
+  virtual void SetImpl(ProcessNodeEngineImplPtr impl) = 0;
+
+ protected:
+  std::string engine_id_;
+  ProcessNodeEngineImplPtr impl_ = nullptr;
+};
+
+using ProcessNodeEnginePtr = std::shared_ptr<ProcessNodeEngine>;
+}  // namespace ge
+
+#endif  // INC_FRAMEWORK_PROCESS_NODE_ENGINE_H_

inc/framework/runtime/gert_api.h

@@ -0,0 +1,27 @@
+/**
+ * Copyright (c) Huawei Technologies Co., Ltd. 2022. All rights reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef AIR_CXX_INC_FRAMEWORK_RUNTIME_GERT_API_H_
+#define AIR_CXX_INC_FRAMEWORK_RUNTIME_GERT_API_H_
+#include "model_v2_executor.h"
+#include "common/ge_types.h"
+
+namespace gert {
+std::unique_ptr<ModelV2Executor> LoadExecutorFromFile(const char *file_path, ge::graphStatus &error_code);
+std::unique_ptr<ModelV2Executor> LoadExecutorFromModelData(const ge::ModelData &model_data,
+                                                           ge::graphStatus &error_code);
+}  // namespace gert
+#endif  // AIR_CXX_INC_FRAMEWORK_RUNTIME_GERT_API_H_

inc/framework/runtime/model_desc.h

@@ -0,0 +1,94 @@
+/**
+ * Copyright (c) Huawei Technologies Co., Ltd. 2022. All rights reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef AIR_CXX_INC_FRAMEWORK_RUNTIME_MODEL_DESC_H_
+#define AIR_CXX_INC_FRAMEWORK_RUNTIME_MODEL_DESC_H_
+#include "common/ge_types.h"
+#include "exe_graph/runtime/shape.h"
+#include "exe_graph/runtime/continuous_vector.h"
+#include "exe_graph/runtime/storage_format.h"
+#include "exe_graph/runtime/storage_shape.h"
+
+namespace gert {
+class ShapeRange {
+ public:
+  const Shape &GetMin() const;
+  const Shape &GetMax() const;
+  Shape &MutableMin();
+  Shape &MutableMax();
+
+ private:
+  Shape min_;
+  Shape max_;
+};
+
+class ModelIoDesc {
+ public:
+  const char *GetName() const;
+  int32_t GetDataType() const;
+  ge::Format GetStorageFormat() const;
+  ge::Format GetOriginFormat() const;
+  int64_t GetSize() const;
+  const Shape &GetStorageShape() const;
+  const Shape &GetOriginShape() const;
+  const ShapeRange &GetOriginShapeRange() const;
+  const ShapeRange &GetStorageShapeRange() const;
+
+  void SetName(const char *name);
+  void SetDataType(int32_t data_type);
+  void SetStorageFormat(ge::Format format);
+  void SetOriginFormat(ge::Format format);
+  Shape &MutableStorageShape();
+  Shape &MutableOriginShape();
+  ShapeRange &MutableOriginShapeRange();
+  ShapeRange &MutableStorageShapeRange();
+
+ private:
+  const char *name_;
+  int32_t data_type_;
+  StorageFormat format_;
+  StorageShape shape_;
+  ShapeRange storage_shape_range_;
+  ShapeRange origin_shape_range_;
+};
+
+class ModelDesc {
+ public:
+  static size_t CalcSize(size_t input_num, size_t output_num);
+  const ModelIoDesc *GetInputDesc(size_t index) const;
+  const ModelIoDesc *GetAllInputsDesc(size_t &input_num) const;
+
+  const ModelIoDesc *GetOutputDesc(size_t index) const;
+  const ModelIoDesc *GetAllOutputsDesc(size_t &output_num) const;
+
+  ModelIoDesc *MutableInputDesc(size_t index);
+  ModelIoDesc *MutableOutputDesc(size_t index);
+  ModelIoDesc *AllMutableIoDesc(size_t &input_num, size_t &output_num);
+  void SetInputNum(size_t input_num);
+  void SetOutputNum(size_t output_num);
+
+  ge::graphStatus GetDynamicBatchInfo(std::vector<std::vector<int64_t>> &batch_info, int32_t &dynamic_type) const;
+  ge::graphStatus GetUserDesignateShapeOrder(std::vector<std::string> &user_designate_shape_order) const;
+  ge::graphStatus GetModelAttrs(std::vector<std::string> &attrs) const;
+
+ private:
+  size_t input_num_;
+  size_t output_num_;
+  ContinuousVector model_io_descs_;
+};
+}  // namespace gert
+
+#endif  // AIR_CXX_INC_FRAMEWORK_RUNTIME_MODEL_DESC_H_

inc/framework/runtime/model_v2_executor.h

@@ -0,0 +1,142 @@
+/**
+ * Copyright (c) Huawei Technologies Co., Ltd. 2022. All rights reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef AIR_CXX_RUNTIME_V2_CORE_MODEL_V_2_EXECUTOR_H_
+#define AIR_CXX_RUNTIME_V2_CORE_MODEL_V_2_EXECUTOR_H_
+#include <memory>
+#include "graph/compute_graph.h"
+#include "graph/ge_error_codes.h"
+#include "model_desc.h"
+#include "runtime/stream.h"
+#include "exe_graph/runtime/tensor.h"
+
+namespace gert {
+enum SubExeGraphType { kInitExeGraph, kMainExeGraph, kDeInitExeGraph, kSubExeGraphTypeEnd };
+static constexpr char *kSubExeGraphTypeStrs[kSubExeGraphTypeEnd] = {(char *)"Init", (char *)"Main", (char *)"DeInit"};
+inline const char *GetSubExeGraphTypeStr(SubExeGraphType type) {
+  return kSubExeGraphTypeStrs[type];
+}
+
+class ResourceGuard {
+ public:
+  void *ResetExecutionData(std::unique_ptr<uint8_t[]> execution_data);
+  void ResetAnyValue(std::unique_ptr<uint8_t[]> any_values, size_t count);
+  void PushNode(void *node);
+  void PushWatcher(void *watcher);
+  void *ResetNodesArray(std::unique_ptr<uint8_t[]> nodes_array);
+  void *ResetStartNodesArray(std::unique_ptr<uint8_t[]> start_nodes_array);
+  void *ResetNodesIndgreeArray(std::unique_ptr<uint8_t[]> nodes_indgree_array);
+  void *ResetNodesWaitIndgreeArray(std::unique_ptr<uint8_t[]> nodes_indgree_array);
+  void *ResetInputsArray(std::unique_ptr<uint8_t[]> inputs_array);
+  void *ResetOutputsArray(std::unique_ptr<uint8_t[]> outputs_array);
+  void *ResetWatchersArray(std::unique_ptr<uint8_t[]> watchers_array);
+  void *ResetReadyQueue(void *ready_queue);
+  void *ResetBuffer(std::unique_ptr<uint8_t[]> buffer);
+  void *ResetComputeNodeInfo(std::unique_ptr<uint8_t[]> compute_node_info);
+  void *ResetKernelExtendInfo(std::unique_ptr<uint8_t[]> kernel_extend_info);
+  void *ResetModelDesc(std::unique_ptr<uint8_t[]> model_desc);
+
+  ~ResourceGuard();
+
+ private:
+  std::unique_ptr<uint8_t[]> execution_data_holder_;
+  size_t any_values_num_;
+  std::unique_ptr<uint8_t[]> any_values_guard_;
+
+  std::vector<std::unique_ptr<void, decltype(&free)>> nodes_guarder_;
+  std::vector<std::unique_ptr<void, decltype(&free)>> watchers_guarder_;
+  std::unique_ptr<uint8_t[]> continuous_buffer_guarder_;
+  std::unique_ptr<uint8_t[]> buffer_guarder_;
+  std::unique_ptr<uint8_t[]> compute_node_info_guarder_;
+  std::unique_ptr<uint8_t[]> kernel_extend_info_guarder_;
+  std::unique_ptr<uint8_t[]> model_desc_guarder_;
+
+  std::unique_ptr<uint8_t[]> nodes_array_guarder_;
+  std::unique_ptr<uint8_t[]> start_nodes_array_guarder_;
+  std::unique_ptr<uint8_t[]> nodes_indgree_array_guarder_;
+  std::unique_ptr<uint8_t[]> nodes_wait_indgree_array_guarder_;
+  std::unique_ptr<uint8_t[]> inputs_array_guarder_;
+  std::unique_ptr<uint8_t[]> outputs_array_guarder_;
+  std::unique_ptr<uint8_t[]> watchers_array_guarder_;
+  std::unique_ptr<void, decltype(&free)> ready_queue_guarder_{nullptr, nullptr};
+};
+
+struct ModelExecuteArg {
+  rtStream_t stream;
+};
+static_assert(std::is_standard_layout<ModelExecuteArg>::value, "The class ModelExecuteArg must be a POD");
+
+class ExeGraphExecutor {
+ public:
+  // todo unload时释放anyvalue资源
+  ge::graphStatus Load() {
+    return ge::GRAPH_SUCCESS;
+  }
+  ge::graphStatus UnLoad() {
+    return ge::GRAPH_SUCCESS;
+  }
+
+  /**
+   * 设置图执行的输入/输出，需要注意的是，使用者需要自己保证inputs/outputs刷新完全！！！
+   */
+  ge::graphStatus SpecifyInputs(void **inputs, size_t start, size_t num);
+  ge::graphStatus SpecifyOutputs(void **outputs, size_t num);
+  ge::graphStatus Execute();
+
+  const void *GetExecutionData() const {
+    return execution_data_;
+  }
+
+  ResourceGuard &GetResourceGuard();
+  void *SetExecutionData(std::unique_ptr<uint8_t[]> execution_data);
+
+ private:
+  friend class ModelV2ExecutorTestHelper;
+
+  void *execution_data_;
+  ResourceGuard resource_guard_;
+};
+class ModelV2Executor {
+ public:
+  static std::unique_ptr<ModelV2Executor> Create(const ge::ComputeGraphPtr &root_graph);
+
+  ge::graphStatus Load();
+  ge::graphStatus Execute(const ModelExecuteArg &arg, Tensor **inputs, size_t input_num, Tensor **outputs,
+                          size_t output_num);
+  ge::graphStatus ExecuteSync(Tensor **inputs, size_t input_num, Tensor **outputs, size_t output_num);
+  ge::graphStatus UnLoad();
+
+  const ModelDesc &GetModelDesc() const;
+  void SetModelDesc(ModelDesc *model_desc);
+  ModelV2Executor(const ModelV2Executor &) = delete;
+  ModelV2Executor(ModelV2Executor &&) = delete;
+  ModelV2Executor &operator=(const ModelV2Executor &) = delete;
+  ModelV2Executor &operator=(ModelV2Executor &&) = delete;
+
+ private:
+  friend class ModelV2ExecutorBuilder;
+  friend class ModelV2ExecutorTestHelper;
+  ModelV2Executor() = default;
+
+ private:
+  std::array<ExeGraphExecutor, kSubExeGraphTypeEnd> graphs_;
+  ResourceGuard resource_guard_;
+  ModelDesc *model_desc_ = nullptr;
+  rtStream_t default_stream_ = nullptr;
+};
+}  // namespace gert
+
+#endif  // AIR_CXX_RUNTIME_V2_CORE_MODEL_V_2_EXECUTOR_H_

metadef

@@ -1 +1 @@
-Subproject commit a725349b65aef2940555af2ddb7b9461fbe0d5fd
+Subproject commit e4d1efc47349f13af1bcdb53ba408118779fc27e

scripts/format_source_code.sh

@@ -0,0 +1,107 @@
+#!/bin/bash
+# Copyright 2019-2020 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+
+set -e
+
+CLANG_FORMAT=$(which clang-format) || (echo "Please install 'clang-format' tool first"; exit 1)
+
+version=$("${CLANG_FORMAT}" --version | sed -n "s/.*\ \([0-9]*\)\.[0-9]*\.[0-9]*.*/\1/p")
+if [[ "${version}" -lt "8" ]]; then
+  echo "clang-format's version must be at least 8.0.0"
+  exit 1
+fi
+
+CURRENT_PATH=$(pwd)
+SCRIPTS_PATH=$(dirname "$0")
+
+echo "CURRENT_PATH=${CURRENT_PATH}"
+echo "SCRIPTS_PATH=${SCRIPTS_PATH}"
+
+# print usage message
+function usage()
+{
+  echo "Format the specified source files to conform the code style."
+  echo "Usage:"
+  echo "bash $0 [-a] [-c] [-l] [-h]"
+  echo "e.g. $0 -c"
+  echo ""
+  echo "Options:"
+  echo "    -a format of all files"
+  echo "    -c format of the files changed compared to last commit, default case"
+  echo "    -l format of the files changed in last commit"
+  echo "    -h Print usage"
+}
+
+# check and set options
+function checkopts()
+{
+  # init variable
+  mode="changed"    # default format changed files
+
+  # Process the options
+  while getopts 'aclh' opt
+  do
+    case "${opt}" in
+      a)
+        mode="all"
+        ;;
+      c)
+        mode="changed"
+        ;;
+      l)
+        mode="lastcommit"
+        ;;
+      h)
+        usage
+        exit 0
+        ;;
+      *)
+        echo "Unknown option ${opt}!"
+        usage
+        exit 1
+    esac
+  done
+}
+
+# init variable
+# check options
+checkopts "$@"
+
+# switch to project root path, which contains clang-format config file '.clang-format'
+cd "${SCRIPTS_PATH}/.." || exit 1
+
+FMT_FILE_LIST='__format_files_list__'
+
+if [[ "X${mode}" == "Xall" ]]; then
+  find src -type f -name "*" | grep "\.h$\|\.cc$" > "${FMT_FILE_LIST}" || true
+  find inc -type f -name "*" | grep "\.h$\|\.cc$" >> "${FMT_FILE_LIST}" || true
+elif [[ "X${mode}" == "Xchanged" ]]; then
+  # --diff-filter=ACMRTUXB will ignore deleted files in commit
+  git diff --diff-filter=ACMRTUXB --name-only | grep "^inc\|^src" | grep "\.h$\|\.cc$" >> "${FMT_FILE_LIST}" || true
+else  # "X${mode}" == "Xlastcommit"
+  git diff --diff-filter=ACMRTUXB --name-only HEAD~ HEAD | grep "^inc\|^src" | grep "\.h$\|\.cc$" > "${FMT_FILE_LIST}" || true
+fi
+
+while read line; do
+  if [ -f "${line}" ]; then
+    ${CLANG_FORMAT} -i "${line}"
+  fi
+done < "${FMT_FILE_LIST}"
+
+rm "${FMT_FILE_LIST}"
+cd "${CURRENT_PATH}" || exit 1
+
+echo "Specified cpp source files have been format successfully."

third_party/fwkacllib/inc/aicpu/aicpu_schedule/aicpu_op_type_list.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright (c) Huawei Technologies Co., Ltd. 2019-2020. All rights reserved.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -17,6 +17,7 @@
 #ifndef AICPU_OP_TYPE_LIST_H_
 #define AICPU_OP_TYPE_LIST_H_
 
+extern "C" {
 enum OpKernelType {
     TF_KERNEL,
     CPU_KERNEL
@@ -29,19 +30,13 @@ enum ReturnCode {
 };
 
 #pragma pack(push, 1)
-//One byte alignment
+// One byte alignment
 struct SysOpInfo {
     uint64_t opLen;
     uint64_t opType;
     OpKernelType kernelsType;
 };
 
-struct OpParamInfo {
-    uint64_t num;
-    uint64_t dtypeList;
-    uint64_t formatList;
-};
-
 struct SysOpCheckInfo {
     uint64_t opListNum;
     uint64_t offSetLen;
@@ -57,4 +52,6 @@ struct SysOpCheckResp {
     uint64_t opParamInfoList;
 };
 #pragma pack(pop)
+}
+
 #endif  // AICPU_OP_TYPE_LIST_H_

third_party/fwkacllib/inc/aicpu/common/aicpu_task_struct.h

@@ -14,13 +14,15 @@
  * limitations under the License.
  */
 
-#ifndef AICPU_TASK_STRUCT_H_
-#define AICPU_TASK_STRUCT_H_
+#ifndef AICPU_TASK_STRUCT_H
+#define AICPU_TASK_STRUCT_H
 
 #include <cstdint>
 
 namespace aicpu {
 
+using char_t = char;
+
 #pragma pack(push, 1)
 struct AicpuParamHead
 {
@@ -29,9 +31,76 @@ struct AicpuParamHead
     uint32_t        extInfoLength;             // extInfo struct Length
     uint64_t        extInfoAddr;               // extInfo address
 };
+
+enum class AicpuConfigMsgType {
+    AICPU_CONFIG_MSG_TYPE_BUF_FREE      = 0,  /* free buf */
+    AICPU_CONFIG_MSG_TYPE_BUF_RESET     = 1,  /* reset buf */
+    AICPU_CONFIG_MSG_TYPE_BUF_SET_ADDR  = 2,  /* set buf addr to aicpu */
+};
+
+enum class AicpuErrMsgType {
+    ERR_MSG_TYPE_NULL   = 0,
+    ERR_MSG_TYPE_AICORE = 1,
+    ERR_MSG_TYPE_AICPU  = 2,
+};
+
+enum class AicpuExtInfoMsgType {
+    EXT_MODEL_ID_MSG_TYPE = 0,
+};
+
+struct AicpuConfigMsg {
+    uint8_t msgType;
+    uint8_t reserved1;
+    uint16_t bufLen;
+    uint32_t offset;
+    uint64_t bufAddr;
+    uint32_t tsId;
+    uint32_t reserved2;
+};
+
+struct AicpuModelIdInfo {
+    uint32_t modelId;
+    uint32_t extendModelId;
+    uint32_t extendInfo[13];
+};
+
+// 64 bytes
+struct AicpuExtendInfo {
+    uint8_t msgType;
+    uint8_t version;
+    uint8_t reserved[2];
+    union {
+        AicpuModelIdInfo modelIdMap;
+    };
+};
+
+struct AicoreErrMsgInfo {
+    uint8_t errType;
+    uint8_t version;
+    uint8_t reserved1[2];    /* reserved1, 4 byte alignment */
+    uint32_t errorCode;
+    uint32_t modelId;
+    uint32_t taskId;
+    uint32_t streamId;
+    uint64_t transactionId;
+    uint8_t reserved2[228];  /* the total byte is 256, reserved2 len = 256 - other lens */
+};
+
+struct AicpuErrMsgInfo {
+    uint8_t errType;
+    uint8_t version;
+    uint8_t reserved1[2];    /* reserved1, 4 byte alignment */
+    uint32_t errorCode;
+    uint32_t modelId;
+    uint32_t streamId;
+    uint64_t transactionId;
+    char_t opName[64];      /* op name str */
+    char_t errDesc[128];    /* err msg desc info */
+    uint8_t reserved2[40];  /* the total byte is 256, reserved2 len = 256 - other lens */
+};
 #pragma pack(pop)
 
 }  // namespace aicpu
 
-#endif  // AICPU_TASK_STRUCT_H_
+#endif  // AICPU_TASK_STRUCT_H
 

third_party/fwkacllib/inc/cce/fwk_adpt_struct.h

@@ -21,7 +21,7 @@
 
 namespace aicpu {
 namespace FWKAdapter {
-
+using char_t = char;
 // API RETURN CODE
 enum FWKAdptAPIRetCode {
   FWK_ADPT_SUCCESS = 0,                  // success
@@ -63,6 +63,8 @@ enum FWKTaskExtInfoType {
   FWK_ADPT_EXT_BITMAP,
   FWK_ADPT_EXT_TOPIC_TYPE,
   FWK_ADPT_EXT_ASYNCWAIT,
+  FWK_ADPT_EXT_UNKNOWN_SHAPE_INPUT_INDEX,
+  FWK_ADPT_EXT_UNKNOWN_SHAPE_OUTPUT_INDEX,
   FWK_ADPT_EXT_INVALID
 };
 
@@ -113,7 +115,7 @@ struct StrFWKKernel {
 typedef StrFWKKernel FWKOperateParam;
 
 // Extent info ShapeAndType
-const uint32_t kMaxShapeDims = 8;
+const uint32_t kMaxShapeDims = 8U;
 #pragma pack(push, 1)
 struct ShapeAndType {
   int32_t type;
@@ -122,13 +124,13 @@ struct ShapeAndType {
 #pragma pack(pop)
 
 // Extend info structure for extInfoAddr
-const uint32_t kExtInfoHeadSize = 8;
+const uint32_t kExtInfoHeadSize = 8U;
 
 #pragma pack(push, 1)
 struct ExtInfo {
   int32_t  infoType;    // extend type
   uint32_t infoLen;     // length for infoMsg
-  char     infoMsg[0];  // extend value
+  char_t  infoMsg[0];  // extend value
 };
 #pragma pack(pop)
 
@@ -143,9 +145,9 @@ struct ResultSummary {
 
 #pragma pack(push, 1)
 struct AsyncWait {
-  uint8_t waitType; // wait type, FWK_ADPT_WAIT_TYPE_EVENT: event wait
-  uint32_t waitId; // wait id, GE refresh
-  uint32_t timeOut; // reserved
+  uint8_t waitType;  // wait type, FWk_ADPT_WAIT_TPYE_EVENT: event wait
+  uint32_t waitId;  // wait id, GE refresh
+  uint32_t timeOut;  // reserved
   uint64_t reserved;
 };
 #pragma pack(pop)

third_party/fwkacllib/inc/cce/taskdown_common.hpp

@@ -27,15 +27,16 @@ namespace cce {
 #define CC_FUSION_OP_MAX 32
 
 typedef enum tagccKernelType {
-  CCE_AI_CORE = 0, /* cce aicore */
-  CCE_AI_CPU = 1,  /* cce aicpu */
-  TE = 2,          /* te operator*/
-  CUSTOMIZED = 3,  /* customized operator */
-  TE_AI_CORE = 4,  /* te aicore operator*/
-  TE_AI_CPU = 5,   /* te aicpu operator */
-  AI_CPU = 6,      /* aicpu */
-  CUST_AI_CPU = 7, /* custom aicpu*/
-  INVALID = 8,     /* unknown kernel type */
+  CCE_AI_CORE = 0,   /* cce aicore */
+  CCE_AI_CPU = 1,    /* cce aicpu */
+  TE = 2,            /* te operator*/
+  CUSTOMIZED = 3,    /* customized operator */
+  TE_AI_CORE = 4,    /* te aicore operator*/
+  TE_AI_CPU = 5,     /* te aicpu operator */
+  AI_CPU = 6,        /* aicpu */
+  CUST_AI_CPU = 7,   /* custom aicpu*/
+  HOST_CPU = 8,      /* host cpu */
+  INVALID = 10000    /* unknown kernel type */
 } ccKernelType;
 
 typedef struct tagOpContext {

third_party/fwkacllib/inc/common/type_def.h

@@ -0,0 +1,52 @@
+/**
+* Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved.
+*
+* This program is distributed in the hope that it will be useful,
+* but WITHOUT ANY WARRANTY; without even the implied warranty of
+* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+*
+* Description:interface.
+* Create: 2021-12-21
+*/
+#ifndef AICPU_TYPE_DEF_H
+#define AICPU_TYPE_DEF_H
+
+#include <cstdint>
+#include <cstddef>
+#ifndef char_t
+typedef char char_t;
+#endif
+
+#ifndef float32_t
+typedef float float32_t;
+#endif
+
+#ifndef float64_t
+typedef double float64_t;
+#endif
+
+inline uint64_t PtrToValue(const void *ptr)
+{
+    return static_cast<const uint64_t>(reinterpret_cast<const uintptr_t>(ptr));
+}
+
+inline void *ValueToPtr(const uint64_t value)
+{
+    return reinterpret_cast<void *>(static_cast<const uintptr_t>(value));
+}
+
+template<typename TI, typename TO>
+inline TO *PtrToPtr(TI *ptr)
+{
+    return reinterpret_cast<TO *>(ptr);
+}
+
+template<typename T>
+inline T *PtrAdd(T * const ptr, const size_t maxIdx, const size_t idx)
+{
+    if ((ptr != nullptr) && (idx < maxIdx)) {
+        return reinterpret_cast<T *>(ptr + idx);
+    }
+    return nullptr;
+}
+#endif  // AICPU_TYPE_DEF_H

third_party/fwkacllib/inc/external/runtime/rt_error_codes.h

@@ -39,6 +39,7 @@ static const int32_t ACL_ERROR_RT_INVALID_MEMORY_TYPE        = 107016; // invali
 static const int32_t ACL_ERROR_RT_INVALID_HANDLE             = 107017; // invalid handle
 static const int32_t ACL_ERROR_RT_INVALID_MALLOC_TYPE        = 107018; // invalid malloc type
 static const int32_t ACL_ERROR_RT_WAIT_TIMEOUT               = 107019; // wait timeout
+static const int32_t ACL_ERROR_RT_TASK_TIMEOUT               = 107020; // task timeout
 
 static const int32_t ACL_ERROR_RT_FEATURE_NOT_SUPPORT        = 207000; // feature not support
 static const int32_t ACL_ERROR_RT_MEMORY_ALLOCATION          = 207001; // memory allocation error
@@ -52,6 +53,12 @@ static const int32_t ACL_ERROR_RT_NO_STREAM_RESOURCE         = 207008; // no str
 static const int32_t ACL_ERROR_RT_NO_NOTIFY_RESOURCE         = 207009; // no notify resource
 static const int32_t ACL_ERROR_RT_NO_MODEL_RESOURCE          = 207010; // no model resource
 static const int32_t ACL_ERROR_RT_NO_CDQ_RESOURCE            = 207011; // no cdq resource
+static const int32_t ACL_ERROR_RT_OVER_LIMIT                 = 207012; // over limit
+static const int32_t ACL_ERROR_RT_QUEUE_EMPTY                = 207013; // queue is empty
+static const int32_t ACL_ERROR_RT_QUEUE_FULL                 = 207014; // queue is full
+static const int32_t ACL_ERROR_RT_REPEATED_INIT              = 207015; // repeated init
+static const int32_t ACL_ERROR_RT_AIVEC_OVER_FLOW            = 207016; // aivec over flow
+static const int32_t ACL_ERROR_RT_OVER_FLOW                  = 207017; // common over flow
 
 static const int32_t ACL_ERROR_RT_INTERNAL_ERROR             = 507000; // runtime internal error
 static const int32_t ACL_ERROR_RT_TS_ERROR                   = 507001; // ts internel error
@@ -91,6 +98,14 @@ static const int32_t ACL_ERROR_RT_VECTOR_CORE_TIMEOUT        = 507034; // vector
 static const int32_t ACL_ERROR_RT_VECTOR_CORE_EXCEPTION      = 507035; // vector core exception
 static const int32_t ACL_ERROR_RT_VECTOR_CORE_TRAP_EXCEPTION = 507036; // vector core trap exception
 static const int32_t ACL_ERROR_RT_CDQ_BATCH_ABNORMAL         = 507037; // cdq alloc batch abnormal
+static const int32_t ACL_ERROR_RT_DIE_MODE_CHANGE_ERROR      = 507038; // can not change die mode
+static const int32_t ACL_ERROR_RT_DIE_SET_ERROR              = 507039; // single die mode can not set die
+static const int32_t ACL_ERROR_RT_INVALID_DIEID              = 507040; // invalid die id
+static const int32_t ACL_ERROR_RT_DIE_MODE_NOT_SET           = 507041; // die mode not set
+static const int32_t ACL_ERROR_RT_AICORE_TRAP_READ_OVERFLOW       = 507042; // aic trap read overflow
+static const int32_t ACL_ERROR_RT_AICORE_TRAP_WRITE_OVERFLOW      = 507043; // aic trap write overflow
+static const int32_t ACL_ERROR_RT_VECTOR_CORE_TRAP_READ_OVERFLOW  = 507044; // aiv trap read overflow
+static const int32_t ACL_ERROR_RT_VECTOR_CORE_TRAP_WRITE_OVERFLOW = 507045; // aiv trap write overflow
 
 static const int32_t ACL_ERROR_RT_DRV_INTERNAL_ERROR         = 507899; // drv internal error
 static const int32_t ACL_ERROR_RT_AICPU_INTERNAL_ERROR       = 507900; // aicpu internal error
@@ -99,5 +114,4 @@ static const int32_t ACL_ERROR_RT_SOCKET_CLOSE               = 507901; // hdc di
 #ifdef __cplusplus
 }
 #endif
-
 #endif // __INC_EXTERNEL_RT_ERROR_CODES_H__

third_party/fwkacllib/inc/hccl/base.h

@@ -86,6 +86,47 @@ enum OriginalGraphShapeType {
     SHAPE_RESERVED  /**< reserved */
 };
 
+enum HcclEventType {
+    HCCL_EVENT_SEND_COMPLETION = 0,
+    HCCL_EVENT_RECV_REQUEST,
+    HCCL_EVENT_RECV_COMPLETION,
+    HCCL_EVENT_CONGESTION_RELIEF,
+    HCCL_EVENT_RESERVED /**< reserved */
+};
+
+const u32 TAG_MAX_LEN = 127; // 最大的tag 长度
+using TagAttr = struct TagAttrDef {
+    char name[TAG_MAX_LEN + 1]; // tag标识
+    // tag标识的接收数据，调用者是否会主动调用接收接口，0 = 否, 1 = 会(预留，暂不支持)。
+    // 对于activeRecv = 0，当接收侧收到数据或者发送请求时，主动通知调用者。
+    uint32_t activeRecv;
+    uint32_t sendCredit; // 配置该tag允许inflight的send个数
+    uint32_t eventId;
+};
+
+using HcclEventMsg = struct HcclEventMsgDef {
+    HcclComm comm;
+    u32 peerRank;
+    u32 tag;
+    // 0:HCCL_SEND_COMPLETION; 1:HCCL_RECV_COMPLETION; 2:HCCL_RECV_REQUEST; 3:HCCL_CONGESTION_RELIEF
+    u32 hcclEventType;
+    union {
+        struct {
+            u32 reserver;
+        } sendCompletionItem;
+        struct {
+            u32 reserver;
+        } recvRequestItem;
+        struct {
+            u32 reserver;
+        } recvCompletionItem;
+        struct CongestionReliefItem {
+            u32 reserver;
+        } congestionReliefItem;
+    } desc;
+};
+
+
 /**
 * @brief stream handle.
 */
@@ -124,29 +165,54 @@ struct HcomRemoteAccessAddrInfo {
 };
 
 struct HcomAllToAllVParams {
-  void *sendbuf;
-  void *sendcounts;
-  void *sdispls;
-  HcclDataType sendtype;
-  void *recvbuf;
-  void *recvcounts;
-  void *rdispls;
-  HcclDataType recvtype;
-  const char *group;
+    void *sendbuf;  // device mem
+    void *sendcounts;  // device mem;  Type: uint_64
+    void *sdispls;  // device mem;  Type: uint_64
+    HcclDataType sendtype;
+    void *recvbuf;  // device mem
+    void *recvcounts;  // device mem;  Type: uint_64 
+    void *rdispls;  // device mem;  Type: uint_64
+    HcclDataType recvtype;
+    const char *group;  // not used now
 };
 
 struct HcomGatherAllToAllVParams {
- void *addrInfo;
- void *addrInfoCountPerRank;
- void *recvbuf;
- void *recvcounts;
- void *rdispls;
- void *gatheredbuf;
- s32 addrLength;
- HcclDataType recvtype;
- const char *group;
+    void *addrInfo;  // device mem;  contains host VA[uint_64]:  [addr, length, addr, length, addr, length, ...]
+    void *addrInfoCountPerRank;  // device mem;  length: ranksize;  contains addrInfoCounts for every rank
+    void *recvbuf;  // device mem
+    void *recvcounts;  // device mem;  Type: uint_64
+    void *rdispls;  // device mem;  Type: uint_64
+    void *gatheredbuf;  // device mem
+    s32 addrLength;
+    HcclDataType recvtype;
+    const char *group;  // not used now
 };
 
+typedef enum workMode {
+HCCL_MODE_NORMAL = 0, // 不支持任何Probe any，仅支持精确的probe
+HCCL_MODE_ANY = 1     // 仅支持ANY_SOURCE + ANY_TAG的probe
+} WorkMode;
+
+typedef struct tagCommAttr {
+    WorkMode mode;  // 通信域内的probe工作模式
+    uint32_t deviceId = 0;
+} CommAttr;
+
+typedef void* HcclMessage;
+typedef void* HcclRequest;
+
+typedef struct {
+    int srcRank;    // 接收/探测到的msg/信封的发送端rank_id，MPI标准定义，调用者可以访问
+    int tag;        // 接收/探测到的msg/信封的tag，MPI标准定义，调用者可以访问
+    int error;      // 接收/探测的错误码0：no error，others：传输过程出错，MPI标准定义，调用者可以访问
+    int cancelled;  // 指定实现，不建议调用者访问
+    int count;      // 接收/探测到的payload大小，指定实现，不建议调用者访问
+} HcclStatus;
+
+#define HCCL_REQUEST_NULL   NULL
+
+#define HCCL_TAG_ANY (1 << 30)
+
 #ifdef __cplusplus
 }
 #endif // __cplusplus

third_party/fwkacllib/inc/hccl/hccl_types.h

@@ -1,101 +0,0 @@
-/**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-/**
- * @file hccl_types.h
- * @brief HCCL data type definition 
- * 
- */
- 
-#ifndef HCCL_TYPES_H_
-#define HCCL_TYPES_H_
-
-#include <stdint.h>
-
-#ifdef __cplusplus
-extern "C" {
-#endif // __cplusplus
-
-/**
- * @brief HCCL functions return value definition
- */
-typedef enum {
-    HCCL_SUCCESS = 0,               /**< success */
-    HCCL_E_PARA = 1,                /**< parameter error */
-    HCCL_E_PTR = 2,                 /**< empty pointer */
-    HCCL_E_MEMORY = 3,              /**< memory error */
-    HCCL_E_INTERNAL = 4,            /**< internal error */
-    HCCL_E_NOT_SUPPORT = 5,         /**< not support feature */
-    HCCL_E_NOT_FOUND = 6,           /**< not found specific resource */
-    HCCL_E_UNAVAIL = 7,             /**< resource unavailable */
-    HCCL_E_SYSCALL = 8,             /**< call system interface error */
-    HCCL_E_TIMEOUT = 9,             /**< timeout */
-    HCCL_E_OPEN_FILE_FAILURE = 10,  /**< open file fail */
-    HCCL_E_TCP_CONNECT = 11,        /**< tcp connect fail */
-    HCCL_E_ROCE_CONNECT = 12,       /**< roce connect fail */
-    HCCL_E_TCP_TRANSFER = 13,       /**< tcp transfer fail */
-    HCCL_E_ROCE_TRANSFER = 14,      /**< roce transfer fail */
-    HCCL_E_RUNTIME = 15,            /**< call runtime api fail */
-    HCCL_E_DRV = 16,                /**< call driver api fail */
-    HCCL_E_PROFILING = 17,          /**< call profiling api fail */
-    HCCL_E_CCE = 18,                /**< call cce api fail */
-    HCCL_E_NETWORK = 19,            /**< call network api fail */
-    HCCL_E_RESERVED                 /**< reserved */
-} HcclResult;
-
-/**
- * @brief handle to HCCL communicator
- */
-typedef void *HcclComm;
-
-/**
- * @brief HCCL Reduction opperation
- */
-typedef enum {
-    HCCL_REDUCE_SUM = 0,    /**< sum */
-    HCCL_REDUCE_PROD = 1,   /**< prod */
-    HCCL_REDUCE_MAX = 2,    /**< max */
-    HCCL_REDUCE_MIN = 3,    /**< min */
-    HCCL_REDUCE_RESERVED    /**< reserved */
-} HcclReduceOp;
-
-/**
- * @brief HCCL data type
- */
-typedef enum {
-    HCCL_DATA_TYPE_INT8 = 0,    /**< int8 */
-    HCCL_DATA_TYPE_INT16 = 1,   /**< int16 */
-    HCCL_DATA_TYPE_INT32 = 2,   /**< int32 */
-    HCCL_DATA_TYPE_FP16 = 3,    /**< fp16 */
-    HCCL_DATA_TYPE_FP32 = 4,    /**< fp32 */
-    HCCL_DATA_TYPE_INT64 = 5,    /**< int64 */
-    HCCL_DATA_TYPE_UINT64 = 6,    /**< uint64 */
-    HCCL_DATA_TYPE_RESERVED     /**< reserved */
-} HcclDataType;
-
-const uint32_t HCCL_ROOT_INFO_BYTES =  4108; // 4108: root info length
-
-/**
- * @brief HCCL root info
- */
-typedef struct HcclRootInfoDef {
-    char internal[HCCL_ROOT_INFO_BYTES];
-} HcclRootInfo;
-
-#ifdef __cplusplus
-}
-#endif // __cplusplus
-#endif // HCCL_TYPES_H_

third_party/fwkacllib/inc/hccl/hcom.h

@@ -126,58 +126,6 @@ extern HcclResult HcomSetGradFusionByIndex(const char *group, u32 segmentNum, co
  * @return HcclResult
  */
 extern HcclResult HcomSetGradFusionBySize(const char *group, u32 segmentNum, const float *sizeList);
-
-/**
- * @brief Initialize hcom executor.
- *
- * @param void
- * @return HcclResult
- */
-HcclResult HcomExecInitialize();
-
-/**
- * @brief Finalize hcom executor.
- *
- * @param void
- * @return HcclResult
- */
-HcclResult HcomExecFinalize();
-
-/**
- * @brief Put collective communication operation into hcom executor.
- *
- * @param opInfo information about collective communication operation.
- * @param callback callback after collective communication operation.
- * @return HcclResult
- */
-HcclResult HcomExecEnqueueOperation(HcomOperation opInfo, std::function<void(HcclResult status)> callback);
-
-/**
- * @brief Put remote access operation into hcom executor.
- *
- * @param remoteAccessType operation type (read or write).
- * @param addrInfos address information about collective communication operation.
- * @param callback callback after collective communication operation.
- * @return HcclResult
- */
-HcclResult HcomExecEnqueueRemoteAccess(const std::string& remoteAccessType,
-                                       const std::vector<HcomRemoteAccessAddrInfo>& addrInfos,
-                                       std::function<void(HcclResult status)> callback);
-
-HcclResult HcomExecEnqueueAllToAllV(HcomAllToAllVParams params, std::function<void(HcclResult status)> callback);
-
-HcclResult HcomExecEnqueueGatherAllToAllV(HcomGatherAllToAllVParams params,
-                                          std::function<void(HcclResult status)> callback);
-
-/**
- * @brief Register memories and init resources for remote access.
- *
- * @param addrList memory addresses for remote access.
- * @param count number of remote memory addresses.
- * @return HcclResult
- */
-extern HcclResult HcomRegRemoteAccessMem(const MemRegisterAddr* addrList, u32 count);
-
 #ifdef __cplusplus
 }
 #endif // __cplusplus