Apply copilot suggestions

Author: zhaomaosu

unified-runtime/source/loader/layers/sanitizer/asan/asan_ddi.cpp

@@ -1436,6 +1436,11 @@ ur_result_t UR_APICALL urEnqueueCooperativeKernelLaunchExp(
     /// are not NULL, phEvent must not refer to an element of the
     /// phEventWaitList array.
     ur_event_handle_t *phEvent) {
+  // This mutex is to prevent concurrent kernel launches across different queues
+  // as the DeviceTSAN local shadow memory does not support concurrent
+  // kernel launches now.
+  std::scoped_lock<ur_shared_mutex> Guard(
+      getAsanInterceptor()->KernelLaunchMutex);
 
   UR_LOG_L(getContext()->logger, DEBUG,
            "==== urEnqueueCooperativeKernelLaunchExp");

unified-runtime/source/loader/layers/sanitizer/msan/msan_ddi.cpp

@@ -421,6 +421,12 @@ ur_result_t urEnqueueKernelLaunch(
     /// [out][optional] return an event object that identifies this
     /// particular kernel execution instance.
     ur_event_handle_t *phEvent) {
+  // This mutex is to prevent concurrent kernel launches across different queues
+  // as the DeviceMSAN local/private shadow memory does not support concurrent
+  // kernel launches now.
+  std::scoped_lock<ur_shared_mutex> Guard(
+      getMsanInterceptor()->KernelLaunchMutex);
+
   UR_LOG_L(getContext()->logger, DEBUG, "==== urEnqueueKernelLaunch");
 
   USMLaunchInfo LaunchInfo(GetContext(hQueue), GetDevice(hQueue),
@@ -1346,6 +1352,11 @@ ur_result_t UR_APICALL urEnqueueCooperativeKernelLaunchExp(
     /// are not NULL, phEvent must not refer to an element of the
     /// phEventWaitList array.
     ur_event_handle_t *phEvent) {
+  // This mutex is to prevent concurrent kernel launches across different queues
+  // as the DeviceTSAN local shadow memory does not support concurrent
+  // kernel launches now.
+  std::scoped_lock<ur_shared_mutex> Guard(
+      getMsanInterceptor()->KernelLaunchMutex);
 
   UR_LOG_L(getContext()->logger, DEBUG,
            "==== urEnqueueCooperativeKernelLaunchExp");

unified-runtime/source/loader/layers/sanitizer/msan/msan_interceptor.hpp

@@ -337,6 +337,8 @@ class MsanInterceptor {
 
   bool isNormalExit() { return m_NormalExit; }
 
+  ur_shared_mutex KernelLaunchMutex;
+
 private:
   /// Initialize Global Variables & Kernel Name at first Launch
   ur_result_t prepareLaunch(std::shared_ptr<msan::DeviceInfo> &DeviceInfo,

unified-runtime/source/loader/layers/sanitizer/tsan/tsan_ddi.cpp

@@ -1163,6 +1163,12 @@ ur_result_t urEnqueueKernelLaunch(
     /// [out][optional] return an event object that identifies this
     /// particular kernel execution instance.
     ur_event_handle_t *phEvent) {
+  // This mutex is to prevent concurrent kernel launches across different queues
+  // as the DeviceTSAN local shadow memory does not support concurrent
+  // kernel launches now.
+  std::scoped_lock<ur_shared_mutex> Guard(
+      getTsanInterceptor()->KernelLaunchMutex);
+
   UR_LOG_L(getContext()->logger, DEBUG, "==== urEnqueueKernelLaunch");
 
   LaunchInfo LaunchInfo(GetContext(hQueue), GetDevice(hQueue), pGlobalWorkSize,
@@ -1212,6 +1218,12 @@ ur_result_t UR_APICALL urEnqueueCooperativeKernelLaunchExp(
     /// are not NULL, phEvent must not refer to an element of the
     /// phEventWaitList array.
     ur_event_handle_t *phEvent) {
+  // This mutex is to prevent concurrent kernel launches across different queues
+  // as the DeviceTSAN local shadow memory does not support concurrent
+  // kernel launches now.
+  std::scoped_lock<ur_shared_mutex> Guard(
+      getTsanInterceptor()->KernelLaunchMutex);
+
   UR_LOG_L(getContext()->logger, DEBUG,
            "==== urEnqueueCooperativeKernelLaunchExp");
 

unified-runtime/source/loader/layers/sanitizer/tsan/tsan_interceptor.hpp

@@ -226,6 +226,8 @@ class TsanInterceptor {
     return m_KernelMap[Kernel];
   }
 
+  ur_shared_mutex KernelLaunchMutex;
+
 private:
   ur_result_t updateShadowMemory(std::shared_ptr<ContextInfo> &CI,
                                  std::shared_ptr<DeviceInfo> &DI,

unified-runtime/source/loader/layers/sanitizer/tsan/tsan_shadow.cpp

@@ -196,13 +196,13 @@ ur_result_t ShadowMemoryGPU::AllocLocalShadow(ur_queue_handle_t Queue,
                                               uptr &End) {
   const size_t LocalMemorySize = GetDeviceLocalMemorySize(Device);
   const size_t RequiredShadowSize = NumWG * LocalMemorySize;
-  static size_t LastAllocedSize = 0;
-  if (RequiredShadowSize > LastAllocedSize) {
+  static size_t LastAllocatedSize = 0;
+  if (RequiredShadowSize > LastAllocatedSize) {
     if (LocalShadowOffset) {
       UR_CALL(getContext()->urDdiTable.USM.pfnFree(Context,
                                                    (void *)LocalShadowOffset));
       LocalShadowOffset = 0;
-      LastAllocedSize = 0;
+      LastAllocatedSize = 0;
     }
 
     UR_CALL(getContext()->urDdiTable.USM.pfnDeviceAlloc(
@@ -216,12 +216,12 @@ ur_result_t ShadowMemoryGPU::AllocLocalShadow(ur_queue_handle_t Queue,
       UR_CALL(getContext()->urDdiTable.USM.pfnFree(Context,
                                                    (void *)LocalShadowOffset));
       LocalShadowOffset = 0;
-      LastAllocedSize = 0;
+      LastAllocatedSize = 0;
 
       return URes;
     }
 
-    LastAllocedSize = RequiredShadowSize;
+    LastAllocatedSize = RequiredShadowSize;
   }
 
   Begin = LocalShadowOffset;