!49684 fix: aclFinalize repeat on lite

Merge pull request !49684 from guozhijian/fix_aclFinalize_repeat_lite

docs/api/api_python/mindrecord/mindspore.mindrecord.ImageNetToMR.rst

@@ -15,7 +15,7 @@
               ...
               n15075141 999
 
-        - **image_dir** (str) - ImageNet数据集的目录路径，目录中包含类似n02119789、n02100735、n02110185和n02096294的子目录。
+        - **image_dir** (str) - ImageNet数据集的目录路径，目录中包含类似n01440764、n01443537、n01484850和n15075141的子目录。
         - **destination** (str) - 转换生成的MindRecord文件路径，需提前创建目录并且目录下不能存在同名文件。
         - **partition_number** (int，可选) - 生成MindRecord的文件个数。默认值：1。
 

mindspore/ccsrc/cxx_api/graph/acl/acl_env_guard.cc

@@ -39,33 +39,36 @@ aclError AclInitAdapter::AclInit(const char *config_file) {
 aclError AclInitAdapter::AclFinalize() {
   std::lock_guard<std::mutex> lock(flag_mutex_);
   if (!init_flag_) {
+    MS_LOG(INFO) << "Had been acl finalize.";
     return ACL_ERROR_NONE;
   }
 
+  MS_LOG(INFO) << "Begine acl finalize.";
   init_flag_ = false;
   return aclFinalize();
 }
 
 aclError AclInitAdapter::ForceFinalize() {
   std::lock_guard<std::mutex> lock(flag_mutex_);
+  MS_LOG(INFO) << "Begine force acl finalize.";
   init_flag_ = false;
   return aclFinalize();
 }
 
 AclEnvGuard::AclEnvGuard() : errno_(AclInitAdapter::GetInstance().AclInit(nullptr)) {
   if (errno_ != ACL_ERROR_NONE && errno_ != ACL_ERROR_REPEAT_INITIALIZE) {
-    MS_LOG(ERROR) << "Execute aclInit Failed";
+    MS_LOG(ERROR) << "Execute aclInit Failed.";
     return;
   }
-  MS_LOG(INFO) << "Acl init success";
+  MS_LOG(INFO) << "Acl init success.";
 }
 
 AclEnvGuard::~AclEnvGuard() {
   errno_ = AclInitAdapter::GetInstance().AclFinalize();
   if (errno_ != ACL_ERROR_NONE && errno_ != ACL_ERROR_REPEAT_FINALIZE) {
-    MS_LOG(ERROR) << "Finalize acl failed";
+    MS_LOG(ERROR) << "Finalize acl failed.";
   }
-  MS_LOG(INFO) << "Acl finalize success";
+  MS_LOG(INFO) << "Acl finalize success.";
 }
 
 std::shared_ptr<AclEnvGuard> AclEnvGuard::GetAclEnv() {
@@ -77,11 +80,11 @@ std::shared_ptr<AclEnvGuard> AclEnvGuard::GetAclEnv() {
     acl_env = std::make_shared<AclEnvGuard>();
     aclError ret = acl_env->GetErrno();
     if (ret != ACL_ERROR_NONE && ret != ACL_ERROR_REPEAT_INITIALIZE) {
-      MS_LOG(ERROR) << "Execute aclInit Failed";
+      MS_LOG(ERROR) << "Execute aclInit Failed.";
       return nullptr;
     }
     global_acl_env_ = acl_env;
-    MS_LOG(INFO) << "Acl init success";
+    MS_LOG(INFO) << "Acl init success.";
   }
   return acl_env;
 }

mindspore/ccsrc/minddata/dataset/kernels/image/dvpp/utils/CMakeLists.txt

@@ -17,11 +17,6 @@ if(NOT MSLITE_ENABLE_ACL)
     ${DVPP_UTILS_SRC}
     ${CMAKE_SOURCE_DIR}/mindspore/ccsrc/cxx_api/graph/acl/acl_env_guard.cc
     )
-else()
-    set(DVPP_UTILS_SRC
-    ${DVPP_UTILS_SRC}
-    ${CMAKE_SOURCE_DIR}/../../mindspore/ccsrc/cxx_api/graph/acl/acl_env_guard.cc
-    )
 endif()
 
 add_library(dvpp_utils SHARED ${DVPP_UTILS_SRC})
@@ -30,7 +25,8 @@ enable_target_when_only_build_plugins(dvpp_utils)
 if(MSLITE_ENABLE_ACL)
     find_library(acl_dvpp libacl_dvpp.so ${ASCEND_CANN_RUNTIME_PATH} ${ASCEND_TOOLKIT_RUNTIME_PATH})
     find_library(acl libascendcl.so ${ASCEND_CANN_RUNTIME_PATH} ${ASCEND_TOOLKIT_RUNTIME_PATH})
-    target_link_libraries(dvpp_utils PRIVATE minddata-lite ${acl} ${acl_dvpp} mindspore_core)
+    # find acl_env_guard in ascend_kernel_plugin
+    target_link_libraries(dvpp_utils PRIVATE ascend_kernel_plugin minddata-lite ${acl} ${acl_dvpp} mindspore_core)
 else()
     target_link_libraries(dvpp_utils PRIVATE _c_dataengine ${ACL} ${ACL_DVPP} mindspore_core mindspore_shared_lib)
 endif()

mindspore/ccsrc/minddata/dataset/kernels/image/dvpp/utils/ResourceManager.cc

@@ -66,7 +66,7 @@ void ResourceManager::Release() {
   }
 
   // finalize the acl when the process exit
-  ret = mindspore::AclInitAdapter::GetInstance().AclFinalize();
+  ret = AclInitAdapter::GetInstance().AclFinalize();
   if (ret != APP_ERR_OK) {
     MS_LOG(DEBUG) << "Failed to finalize acl, ret = " << ret << ".";
   }
@@ -96,7 +96,7 @@ APP_ERROR ResourceManager::InitResource(ResourceInfo &resourceInfo) {
     return APP_ERR_OK;
   }
   APP_ERROR ret = APP_ERR_OK;
-  acl_env_ = mindspore::AclEnvGuard::GetAclEnv();
+  acl_env_ = AclEnvGuard::GetAclEnv();
   if (acl_env_ == nullptr) {
     MS_LOG(ERROR) << "Failed to init acl.";
     return APP_ERR_COMM_FAILURE;

mindspore/ccsrc/minddata/dataset/kernels/image/dvpp/utils/ResourceManager.h

@@ -22,7 +22,15 @@
 #include <sys/stat.h>
 #include <unordered_map>
 #include <vector>
+#ifndef BUILD_LITE
 #include "mindspore/ccsrc/cxx_api/graph/acl/acl_env_guard.h"
+using AclEnvGuard = mindspore::AclEnvGuard;
+using AclInitAdapter = mindspore::AclInitAdapter;
+#else
+#include "mindspore/lite/src/extendrt/kernel/ascend/model/acl_env_guard.h"
+using AclEnvGuard = mindspore::kernel::acl::AclEnvGuard;
+using AclInitAdapter = mindspore::kernel::acl::AclInitAdapter;
+#endif
 #include "minddata/dataset/kernels/image/dvpp/utils/CommonDataType.h"
 #include "minddata/dataset/kernels/image/dvpp/utils/ErrorCode.h"
 #include "minddata/dataset/kernels/image/dvpp/utils/resouce_info.h"
@@ -57,7 +65,7 @@ class ResourceManager {
   std::vector<int> deviceIds_;
   std::vector<aclrtContext> contexts_;
   std::unordered_map<int, int> deviceIdMap_;  // Map of device to index
-  std::shared_ptr<mindspore::AclEnvGuard> acl_env_;
+  std::shared_ptr<AclEnvGuard> acl_env_;
 };
 
 #endif

mindspore/lite/src/extendrt/kernel/ascend/model/acl_env_guard.cc

@@ -23,16 +23,83 @@ namespace acl {
 std::shared_ptr<AclEnvGuard> AclEnvGuard::global_acl_env_ = nullptr;
 std::mutex AclEnvGuard::global_acl_env_mutex_;
 
-AclEnvGuard::AclEnvGuard(std::string_view cfg_file) {
-  errno_ = aclInit(cfg_file.data());
-  if (errno_ != ACL_ERROR_NONE && errno_ != ACL_ERROR_REPEAT_INITIALIZE) {
-    MS_LOG(ERROR) << "Execute aclInit Failed";
-    return;
-  }
-  MS_LOG(INFO) << "Acl init success";
+AclInitAdapter &AclInitAdapter::GetInstance() {
+  static AclInitAdapter instance = {};
+  return instance;
 }
 
-AclEnvGuard::~AclEnvGuard() { (void)aclFinalize(); }
+aclError AclInitAdapter::AclInit(const char *config_file) {
+  std::lock_guard<std::mutex> lock(flag_mutex_);
+  if (init_flag_) {
+    return ACL_ERROR_NONE;
+  }
+
+  init_flag_ = true;
+  return aclInit(config_file);
+}
+
+aclError AclInitAdapter::AclFinalize() {
+  std::lock_guard<std::mutex> lock(flag_mutex_);
+  if (!init_flag_) {
+    MS_LOG(INFO) << "Had been acl finalize.";
+    return ACL_ERROR_NONE;
+  }
+
+  MS_LOG(INFO) << "Begin acl finalize.";
+  init_flag_ = false;
+  return aclFinalize();
+}
+
+aclError AclInitAdapter::ForceFinalize() {
+  std::lock_guard<std::mutex> lock(flag_mutex_);
+  MS_LOG(INFO) << "Begin force acl finalize.";
+  init_flag_ = false;
+  return aclFinalize();
+}
+
+AclEnvGuard::AclEnvGuard() : errno_(AclInitAdapter::GetInstance().AclInit(nullptr)) {
+  if (errno_ != ACL_ERROR_NONE && errno_ != ACL_ERROR_REPEAT_INITIALIZE) {
+    MS_LOG(ERROR) << "Execute aclInit Failed.";
+    return;
+  }
+  MS_LOG(INFO) << "Acl init success.";
+}
+
+AclEnvGuard::AclEnvGuard(std::string_view cfg_file) : errno_(AclInitAdapter::GetInstance().AclInit(cfg_file.data())) {
+  if (errno_ != ACL_ERROR_NONE && errno_ != ACL_ERROR_REPEAT_INITIALIZE) {
+    MS_LOG(ERROR) << "Execute aclInit Failed.";
+    return;
+  }
+  MS_LOG(INFO) << "Acl init success.";
+}
+
+AclEnvGuard::~AclEnvGuard() {
+  errno_ = AclInitAdapter::GetInstance().AclFinalize();
+  if (errno_ != ACL_ERROR_NONE && errno_ != ACL_ERROR_REPEAT_FINALIZE) {
+    MS_LOG(ERROR) << "Finalize acl failed.";
+  }
+  MS_LOG(INFO) << "Acl finalize success.";
+}
+
+std::shared_ptr<AclEnvGuard> AclEnvGuard::GetAclEnv() {
+  std::shared_ptr<AclEnvGuard> acl_env;
+
+  std::lock_guard<std::mutex> lock(global_acl_env_mutex_);
+  acl_env = global_acl_env_;
+  if (acl_env != nullptr) {
+    MS_LOG(INFO) << "Acl has been initialized, skip.";
+  } else {
+    acl_env = std::make_shared<AclEnvGuard>();
+    aclError ret = acl_env->GetErrno();
+    if (ret != ACL_ERROR_NONE && ret != ACL_ERROR_REPEAT_INITIALIZE) {
+      MS_LOG(ERROR) << "Execute aclInit Failed.";
+      return nullptr;
+    }
+    global_acl_env_ = acl_env;
+    MS_LOG(INFO) << "Acl init success.";
+  }
+  return acl_env;
+}
 
 std::shared_ptr<AclEnvGuard> AclEnvGuard::GetAclEnv(std::string_view cfg_file) {
   std::shared_ptr<AclEnvGuard> acl_env;
@@ -48,11 +115,11 @@ std::shared_ptr<AclEnvGuard> AclEnvGuard::GetAclEnv(std::string_view cfg_file) {
     acl_env = std::make_shared<AclEnvGuard>(cfg_file);
     aclError ret = acl_env->GetErrno();
     if (ret != ACL_ERROR_NONE && ret != ACL_ERROR_REPEAT_INITIALIZE) {
-      MS_LOG(ERROR) << "Execute aclInit Failed";
+      MS_LOG(ERROR) << "Execute aclInit Failed.";
       return nullptr;
     }
     global_acl_env_ = acl_env;
-    MS_LOG(INFO) << "Acl init success";
+    MS_LOG(INFO) << "Acl init success.";
   }
   return acl_env;
 }

mindspore/lite/src/extendrt/kernel/ascend/model/acl_env_guard.h

@@ -23,11 +23,28 @@
 
 namespace mindspore::kernel {
 namespace acl {
+class AclInitAdapter {
+ public:
+  static AclInitAdapter &GetInstance();
+  aclError AclInit(const char *config_file);
+  aclError AclFinalize();
+  aclError ForceFinalize();
+
+ private:
+  AclInitAdapter() : init_flag_(false) {}
+  ~AclInitAdapter() = default;
+
+  bool init_flag_;
+  std::mutex flag_mutex_;
+};
+
 class AclEnvGuard {
  public:
+  AclEnvGuard();
   explicit AclEnvGuard(std::string_view cfg_file);
   ~AclEnvGuard();
   aclError GetErrno() const { return errno_; }
+  static std::shared_ptr<AclEnvGuard> GetAclEnv();
   static std::shared_ptr<AclEnvGuard> GetAclEnv(std::string_view cfg_file);
 
  private:

mindspore/lite/src/litert/kernel/ascend/src/acl_env_guard.cc

@@ -23,16 +23,83 @@ namespace acl {
 std::shared_ptr<AclEnvGuard> AclEnvGuard::global_acl_env_ = nullptr;
 std::mutex AclEnvGuard::global_acl_env_mutex_;
 
-AclEnvGuard::AclEnvGuard(std::string_view cfg_file) : errno_(ACL_ERROR_NONE) {
-  errno_ = aclInit(cfg_file.data());
-  if (errno_ != ACL_ERROR_NONE && errno_ != ACL_ERROR_REPEAT_INITIALIZE) {
-    MS_LOG(ERROR) << "Execute aclInit Failed";
-    return;
-  }
-  MS_LOG(INFO) << "Acl init success";
+AclInitAdapter &AclInitAdapter::GetInstance() {
+  static AclInitAdapter instance = {};
+  return instance;
 }
 
-AclEnvGuard::~AclEnvGuard() { (void)aclFinalize(); }
+aclError AclInitAdapter::AclInit(const char *config_file) {
+  std::lock_guard<std::mutex> lock(flag_mutex_);
+  if (init_flag_) {
+    return ACL_ERROR_NONE;
+  }
+
+  init_flag_ = true;
+  return aclInit(config_file);
+}
+
+aclError AclInitAdapter::AclFinalize() {
+  std::lock_guard<std::mutex> lock(flag_mutex_);
+  if (!init_flag_) {
+    MS_LOG(INFO) << "Had been acl finalize.";
+    return ACL_ERROR_NONE;
+  }
+
+  MS_LOG(INFO) << "Begine acl finalize.";
+  init_flag_ = false;
+  return aclFinalize();
+}
+
+aclError AclInitAdapter::ForceFinalize() {
+  std::lock_guard<std::mutex> lock(flag_mutex_);
+  MS_LOG(INFO) << "Begine force acl finalize.";
+  init_flag_ = false;
+  return aclFinalize();
+}
+
+AclEnvGuard::AclEnvGuard() : errno_(AclInitAdapter::GetInstance().AclInit(nullptr)) {
+  if (errno_ != ACL_ERROR_NONE && errno_ != ACL_ERROR_REPEAT_INITIALIZE) {
+    MS_LOG(ERROR) << "Execute aclInit Failed.";
+    return;
+  }
+  MS_LOG(INFO) << "Acl init success.";
+}
+
+AclEnvGuard::AclEnvGuard(std::string_view cfg_file) : errno_(AclInitAdapter::GetInstance().AclInit(cfg_file.data())) {
+  if (errno_ != ACL_ERROR_NONE && errno_ != ACL_ERROR_REPEAT_INITIALIZE) {
+    MS_LOG(ERROR) << "Execute aclInit Failed.";
+    return;
+  }
+  MS_LOG(INFO) << "Acl init success.";
+}
+
+AclEnvGuard::~AclEnvGuard() {
+  errno_ = AclInitAdapter::GetInstance().AclFinalize();
+  if (errno_ != ACL_ERROR_NONE && errno_ != ACL_ERROR_REPEAT_FINALIZE) {
+    MS_LOG(ERROR) << "Finalize acl failed.";
+  }
+  MS_LOG(INFO) << "Acl finalize success.";
+}
+
+std::shared_ptr<AclEnvGuard> AclEnvGuard::GetAclEnv() {
+  std::shared_ptr<AclEnvGuard> acl_env;
+
+  std::lock_guard<std::mutex> lock(global_acl_env_mutex_);
+  acl_env = global_acl_env_;
+  if (acl_env != nullptr) {
+    MS_LOG(INFO) << "Acl has been initialized, skip.";
+  } else {
+    acl_env = std::make_shared<AclEnvGuard>();
+    aclError ret = acl_env->GetErrno();
+    if (ret != ACL_ERROR_NONE && ret != ACL_ERROR_REPEAT_INITIALIZE) {
+      MS_LOG(ERROR) << "Execute aclInit Failed.";
+      return nullptr;
+    }
+    global_acl_env_ = acl_env;
+    MS_LOG(INFO) << "Acl init success.";
+  }
+  return acl_env;
+}
 
 std::shared_ptr<AclEnvGuard> AclEnvGuard::GetAclEnv(std::string_view cfg_file) {
   std::shared_ptr<AclEnvGuard> acl_env;
@@ -48,11 +115,11 @@ std::shared_ptr<AclEnvGuard> AclEnvGuard::GetAclEnv(std::string_view cfg_file) {
     acl_env = std::make_shared<AclEnvGuard>(cfg_file);
     aclError ret = acl_env->GetErrno();
     if (ret != ACL_ERROR_NONE && ret != ACL_ERROR_REPEAT_INITIALIZE) {
-      MS_LOG(ERROR) << "Execute aclInit Failed";
+      MS_LOG(ERROR) << "Execute aclInit Failed.";
       return nullptr;
     }
     global_acl_env_ = acl_env;
-    MS_LOG(INFO) << "Acl init success";
+    MS_LOG(INFO) << "Acl init success.";
   }
   return acl_env;
 }

mindspore/lite/src/litert/kernel/ascend/src/acl_env_guard.h

@@ -23,11 +23,28 @@
 
 namespace mindspore::kernel {
 namespace acl {
+class AclInitAdapter {
+ public:
+  static AclInitAdapter &GetInstance();
+  aclError AclInit(const char *config_file);
+  aclError AclFinalize();
+  aclError ForceFinalize();
+
+ private:
+  AclInitAdapter() : init_flag_(false) {}
+  ~AclInitAdapter() = default;
+
+  bool init_flag_;
+  std::mutex flag_mutex_;
+};
+
 class AclEnvGuard {
  public:
+  AclEnvGuard();
   explicit AclEnvGuard(std::string_view cfg_file);
   ~AclEnvGuard();
   aclError GetErrno() const { return errno_; }
+  static std::shared_ptr<AclEnvGuard> GetAclEnv();
   static std::shared_ptr<AclEnvGuard> GetAclEnv(std::string_view cfg_file);
 
  private:

mindspore/python/mindspore/mindrecord/filewriter.py

@@ -326,7 +326,7 @@ class FileWriter:
                 # check the status of worker process
                 for i in range(len(self._paths)):
                     if not self._workers[i].is_alive():
-                        raise RuntimeError("Worker process(pid:{}) has stopped. Please check " \
+                        raise RuntimeError("Worker process(pid:{}) has stopped abnormal. Please check " \
                                            "the above log".format(self._workers[i].pid))
                 continue
             return SUCCESS

mindspore/python/mindspore/mindrecord/tools/imagenet_to_mr.py

@@ -33,7 +33,7 @@ class ImageNetToMR:
     Args:
         map_file (str): The map file that indicates label. This file can be generated by command
             :code:`ls -l [image_dir] | grep -vE "total|\." | awk -F " " '{print $9, NR-1;}' > [file_path]` ,
-            where `image_dir` is image directory contains n02119789, n02100735, n02110185 and n02096294 directory
+            where `image_dir` is image directory contains n01440764, n01443537, n01484850 and n15075141 directory
             and `file_path` is the generated `map_file` . An example of `map_file` is as below:
 
             .. code-block::
@@ -45,7 +45,7 @@ class ImageNetToMR:
               ...
               n15075141 999
 
-        image_dir (str): Image directory contains n02119789, n02100735, n02110185 and n02096294 directory.
+        image_dir (str): Image directory contains n01440764, n01443537, n01484850 and n15075141 directory.
         destination (str): MindRecord file path to transform into, ensure that the directory is created in advance and
             no file with the same name exists in the directory.
         partition_number (int, optional): The partition size. Default: 1.