From 3e1d72fdae87dd9a526acf945098f33ba0bddeed Mon Sep 17 00:00:00 2001
From: sunxinlei <sunxinlei1@huawei.com>
Date: Sun, 29 Sep 2024 15:18:19 +0800
Subject: [PATCH] fix double free core

---
 torch_npu/csrc/core/npu/NPUException.h | 4 ++--
 torch_npu/csrc/core/npu/NPUQueue.cpp   | 8 ++++----
 2 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/torch_npu/csrc/core/npu/NPUException.h b/torch_npu/csrc/core/npu/NPUException.h
index 0f03ee1865..aab033c0c4 100644
--- a/torch_npu/csrc/core/npu/NPUException.h
+++ b/torch_npu/csrc/core/npu/NPUException.h
@@ -89,8 +89,8 @@ std::string formatErrorCode(SubModule submodule, ErrCode errorCode);
 #define GRAPH_ERROR(error) formatErrorCode(SubModule::GRAPH, error)
 #define PROF_ERROR(error) formatErrorCode(SubModule::PROF, error)
 
-#define DEVICE_TASK_ABORT "107022"
-#define DEVICE_MEM_ERROR "507053"
+#define DEVICE_TASK_ABORT "reason=[device task abort]"
+#define DEVICE_MEM_ERROR "reason=[device mem error]"
 
 inline const char* getErrorFunction(const char* msg)
 {
diff --git a/torch_npu/csrc/core/npu/NPUQueue.cpp b/torch_npu/csrc/core/npu/NPUQueue.cpp
index edbd5a8655..43c1499d53 100644
--- a/torch_npu/csrc/core/npu/NPUQueue.cpp
+++ b/torch_npu/csrc/core/npu/NPUQueue.cpp
@@ -475,7 +475,7 @@ void Repository::Dequeue() {
   uint64_t u = 1;
 
   SetReadWorking(true);
-  while (ret == false && GetStatus() != RepoStatus::CAN_EXIT) {
+  while (ret == false && GetStatus() != RepoStatus::CAN_EXIT && GetStatus() != RepoStatus::STOP_EXIT) {
     ret = ReadQueue();
     if (ret == false) {
       if (GetStatus() == RepoStatus::NEED_EXIT) {
@@ -483,9 +483,9 @@ void Repository::Dequeue() {
         break;
       }
 
-      if (GetStatus() == RepoStatus::ERROR_EXIT) {
-        break;
-      }
+        if (GetStatus() == RepoStatus::ERROR_EXIT) {
+          break;
+        }
 
       SetReadWorking(false);
       __sync_synchronize();
-- 
Gitee