Index: llvm/include/llvm/Frontend/OpenMP/OMPKinds.def
===================================================================
--- llvm/include/llvm/Frontend/OpenMP/OMPKinds.def
+++ llvm/include/llvm/Frontend/OpenMP/OMPKinds.def
@@ -427,7 +427,7 @@
           VoidPtrPtr, VoidPtrPtr, Int64Ptr, Int64Ptr, VoidPtrPtr, VoidPtrPtr)
 __OMP_RTL(__tgt_target_data_begin_mapper_issue, false, Void, IdentPtr, Int64, Int32,
           VoidPtrPtr, VoidPtrPtr, Int64Ptr, Int64Ptr, VoidPtrPtr, VoidPtrPtr, AsyncInfoPtr)
-__OMP_RTL(__tgt_target_data_begin_mapper_wait, false, Void, Int64, AsyncInfoPtr)
+__OMP_RTL(__tgt_target_data_begin_mapper_wait, false, Void, IdentPtr, Int64, AsyncInfoPtr)
 __OMP_RTL(__tgt_target_data_end_mapper, false, Void, IdentPtr, Int64, Int32, VoidPtrPtr,
           VoidPtrPtr, Int64Ptr, Int64Ptr, VoidPtrPtr, VoidPtrPtr)
 __OMP_RTL(__tgt_target_data_end_nowait_mapper, false, Void, IdentPtr, Int64, Int32,
Index: openmp/libomptarget/include/omptarget.h
===================================================================
--- openmp/libomptarget/include/omptarget.h
+++ openmp/libomptarget/include/omptarget.h
@@ -187,10 +187,17 @@
 
   __tgt_async_info AsyncInfo;
   DeviceTy &Device;
+  // If the flag is set, synchronize is called in the delete function.
+  bool ShouldSync;
 
 public:
-  AsyncInfoTy(DeviceTy &Device) : Device(Device) {}
-  ~AsyncInfoTy() { synchronize(); }
+  AsyncInfoTy(DeviceTy &Device, bool ShouldSync = true)
+      : Device(Device), ShouldSync(ShouldSync) {}
+
+  ~AsyncInfoTy() {
+    if (ShouldSync)
+      synchronize();
+  }
 
   /// Implicit conversion to the __tgt_async_info which is used in the
   /// plugin interface.
@@ -278,6 +285,17 @@
                                     void **Args, int64_t *ArgSizes,
                                     int64_t *ArgTypes, map_var_info_t *ArgNames,
                                     void **ArgMappers);
+// Issues the memory transfer asynchronously and sets a handle.
+void __tgt_target_data_begin_mapper_issue(
+    ident_t *Loc, int64_t DeviceId, int32_t ArgNum, void **ArgsBase,
+    void **Args, int64_t *ArgSizes, int64_t *ArgTypes, map_var_info_t *ArgNames,
+    void **ArgMappers, __tgt_async_info *Handle);
+
+// Waits in the set handle from "__tgt_target_data_begin_mapper_issue"
+// for the memory transfer to finish.
+void __tgt_target_data_begin_mapper_wait(ident_t *Loc, int64_t DeviceId,
+                                         __tgt_async_info *Handle);
+
 void __tgt_target_data_begin_nowait_mapper(
     ident_t *Loc, int64_t DeviceId, int32_t ArgNum, void **ArgsBase,
     void **Args, int64_t *ArgSizes, int64_t *ArgTypes, map_var_info_t *ArgNames,
Index: openmp/libomptarget/src/exports
===================================================================
--- openmp/libomptarget/src/exports
+++ openmp/libomptarget/src/exports
@@ -15,6 +15,8 @@
     __tgt_target_nowait;
     __tgt_target_teams_nowait;
     __tgt_target_data_begin_mapper;
+    __tgt_target_data_begin_mapper_issue;
+    __tgt_target_data_begin_mapper_wait;
     __tgt_target_data_end_mapper;
     __tgt_target_data_update_mapper;
     __tgt_target_mapper;
Index: openmp/libomptarget/src/interface.cpp
===================================================================
--- openmp/libomptarget/src/interface.cpp
+++ openmp/libomptarget/src/interface.cpp
@@ -70,6 +70,15 @@
                                            int64_t *ArgTypes,
                                            map_var_info_t *ArgNames,
                                            void **ArgMappers) {
+  __tgt_target_data_begin_mapper_issue(Loc, DeviceId, ArgNum, ArgsBase, Args,
+                                       ArgSizes, ArgTypes, ArgNames, ArgMappers,
+                                       nullptr);
+}
+
+EXTERN void __tgt_target_data_begin_mapper_issue(
+    ident_t *Loc, int64_t DeviceId, int32_t ArgNum, void **ArgsBase,
+    void **Args, int64_t *ArgSizes, int64_t *ArgTypes, map_var_info_t *ArgNames,
+    void **ArgMappers, __tgt_async_info *Handle) {
   TIMESCOPE_WITH_IDENT(Loc);
   DP("Entering data begin region for device %" PRId64 " with %d mappings\n",
      DeviceId, ArgNum);
@@ -92,14 +101,39 @@
   }
 #endif
 
-  AsyncInfoTy AsyncInfo(Device);
+  AsyncInfoTy AsyncInfo(Device, !(Handle));
   int Rc = targetDataBegin(Loc, Device, ArgNum, ArgsBase, Args, ArgSizes,
                            ArgTypes, ArgNames, ArgMappers, AsyncInfo);
-  if (Rc == OFFLOAD_SUCCESS)
-    Rc = AsyncInfo.synchronize();
+
+  if (Rc == OFFLOAD_SUCCESS) {
+    if (Handle)
+      Handle->Queue = ((__tgt_async_info *)AsyncInfo)->Queue;
+    else
+      Rc = AsyncInfo.synchronize();
+  }
+
   handleTargetOutcome(Rc == OFFLOAD_SUCCESS, Loc);
 }
 
+EXTERN void __tgt_target_data_begin_mapper_wait(ident_t *Loc, int64_t DeviceId,
+                                                __tgt_async_info *Handle) {
+
+  TIMESCOPE_WITH_IDENT(Loc);
+  assert((Handle && Handle->Queue) && "Incomplete data mapping");
+
+  DP("Entering wait function in data begin region for device %" PRId64 "\n",
+     DeviceId);
+
+  if (checkDevice(DeviceId, Loc)) {
+    DP("Not offloading to device %" PRId64 "\n", DeviceId);
+    return;
+  }
+
+  DeviceTy &Device = *PM->Devices[DeviceId];
+  if (Device.RTL->synchronize)
+    Device.RTL->synchronize(DeviceId, Handle);
+}
+
 EXTERN void __tgt_target_data_begin_nowait_mapper(
     ident_t *Loc, int64_t DeviceId, int32_t ArgNum, void **ArgsBase,
     void **Args, int64_t *ArgSizes, int64_t *ArgTypes, map_var_info_t *ArgNames,
Index: openmp/libomptarget/src/omptarget.cpp
===================================================================
--- openmp/libomptarget/src/omptarget.cpp
+++ openmp/libomptarget/src/omptarget.cpp
@@ -273,19 +273,8 @@
   return PM->TargetOffloadPolicy == tgt_disabled;
 }
 
-// If offload is enabled, ensure that device DeviceID has been initialized,
-// global ctors have been executed, and global data has been mapped.
-//
-// The return bool indicates if the offload is to the host device
-// There are three possible results:
-// - Return false if the taregt device is ready for offload
-// - Return true without reporting a runtime error if offload is
-//   disabled, perhaps because the initial device was specified.
-// - Report a runtime error and return true.
-//
-// If DeviceID == OFFLOAD_DEVICE_DEFAULT, set DeviceID to the default device.
-// This step might be skipped if offload is disabled.
-bool checkDeviceAndCtors(int64_t &DeviceID, ident_t *Loc) {
+// If offload is enabled, ensure that device DeviceID has been initialized.
+bool checkDevice(int64_t &DeviceID, ident_t *Loc) {
   if (isOffloadDisabled()) {
     DP("Offload is disabled\n");
     return true;
@@ -316,6 +305,26 @@
     return true;
   }
 
+  return false;
+}
+
+// If offload is enabled, ensure that device DeviceID has been initialized,
+// global ctors have been executed, and global data has been mapped.
+//
+// The return bool indicates if the offload is to the host device
+// There are three possible results:
+// - Return false if the taregt device is ready for offload
+// - Return true without reporting a runtime error if offload is
+//   disabled, perhaps because the initial device was specified.
+// - Report a runtime error and return true.
+//
+// If DeviceID == OFFLOAD_DEVICE_DEFAULT, set DeviceID to the default device.
+// This step might be skipped if offload is disabled.
+bool checkDeviceAndCtors(int64_t &DeviceID, ident_t *Loc) {
+
+  if(checkDevice(DeviceID, Loc))
+    return true;
+
   // Get device info.
   DeviceTy &Device = *PM->Devices[DeviceID];
 
Index: openmp/libomptarget/src/private.h
===================================================================
--- openmp/libomptarget/src/private.h
+++ openmp/libomptarget/src/private.h
@@ -46,6 +46,7 @@
                   AsyncInfoTy &AsyncInfo);
 
 extern void handleTargetOutcome(bool Success, ident_t *Loc);
+extern bool checkDevice(int64_t &DeviceID, ident_t *Loc);
 extern bool checkDeviceAndCtors(int64_t &DeviceID, ident_t *Loc);
 extern void *targetAllocExplicit(size_t Size, int DeviceNum, int Kind,
                                  const char *Name);