Index: clang/lib/CodeGen/CGOpenMPRuntime.h
===================================================================
--- clang/lib/CodeGen/CGOpenMPRuntime.h
+++ clang/lib/CodeGen/CGOpenMPRuntime.h
@@ -121,6 +121,7 @@
   bool Nogroup = false;
   bool IsReductionWithTaskMod = false;
   bool IsWorksharingReduction = false;
+  bool HasNowaitClause = false;
 };
 
 /// Class intended to support codegen of all kind of the reduction clauses.
Index: clang/lib/CodeGen/CGOpenMPRuntime.cpp
===================================================================
--- clang/lib/CodeGen/CGOpenMPRuntime.cpp
+++ clang/lib/CodeGen/CGOpenMPRuntime.cpp
@@ -4739,7 +4739,7 @@
       Region->emitUntiedSwitch(CGF);
   };
 
-  llvm::Value *DepWaitTaskArgs[6];
+  llvm::Value *DepWaitTaskArgs[7];
   if (!Data.Dependences.empty()) {
     DepWaitTaskArgs[0] = UpLoc;
     DepWaitTaskArgs[1] = ThreadID;
@@ -4747,6 +4747,8 @@
     DepWaitTaskArgs[3] = DependenciesArray.getPointer();
     DepWaitTaskArgs[4] = CGF.Builder.getInt32(0);
     DepWaitTaskArgs[5] = llvm::ConstantPointerNull::get(CGF.VoidPtrTy);
+    DepWaitTaskArgs[6] =
+        llvm::ConstantInt::get(CGF.Int32Ty, Data.HasNowaitClause);
   }
   auto &M = CGM.getModule();
   auto &&ElseCodeGen = [this, &M, &TaskArgs, ThreadID, NewTaskNewTaskTTy,
@@ -4758,9 +4760,9 @@
     // ndeps_noalias, kmp_depend_info_t *noalias_dep_list); if dependence info
     // is specified.
     if (!Data.Dependences.empty())
-      CGF.EmitRuntimeCall(
-          OMPBuilder.getOrCreateRuntimeFunction(M, OMPRTL___kmpc_omp_wait_deps),
-          DepWaitTaskArgs);
+      CGF.EmitRuntimeCall(OMPBuilder.getOrCreateRuntimeFunction(
+                              M, OMPRTL___kmpc_omp_taskwait_deps_51),
+                          DepWaitTaskArgs);
     // Call proxy_task_entry(gtid, new_task);
     auto &&CodeGen = [TaskEntry, ThreadID, NewTaskNewTaskTTy,
                       Loc](CodeGenFunction &CGF, PrePostActionTy &Action) {
@@ -5799,7 +5801,7 @@
 
   if (CGF.CGM.getLangOpts().OpenMPIRBuilder && Data.Dependences.empty()) {
     // TODO: Need to support taskwait with dependences in the OpenMPIRBuilder.
-    OMPBuilder.createTaskwait(CGF.Builder);
+    OMPBuilder.createTaskwait(CGF.Builder, Data.HasNowaitClause);
   } else {
     llvm::Value *ThreadID = getThreadID(CGF, Loc);
     llvm::Value *UpLoc = emitUpdateLocation(CGF, Loc);
@@ -5808,34 +5810,38 @@
     llvm::Value *NumOfElements;
     std::tie(NumOfElements, DependenciesArray) =
         emitDependClause(CGF, Data.Dependences, Loc);
-    llvm::Value *DepWaitTaskArgs[6];
     if (!Data.Dependences.empty()) {
+      llvm::Value *DepWaitTaskArgs[7];
       DepWaitTaskArgs[0] = UpLoc;
       DepWaitTaskArgs[1] = ThreadID;
       DepWaitTaskArgs[2] = NumOfElements;
       DepWaitTaskArgs[3] = DependenciesArray.getPointer();
       DepWaitTaskArgs[4] = CGF.Builder.getInt32(0);
       DepWaitTaskArgs[5] = llvm::ConstantPointerNull::get(CGF.VoidPtrTy);
+      DepWaitTaskArgs[6] =
+          llvm::ConstantInt::get(CGF.Int32Ty, Data.HasNowaitClause);
 
       CodeGenFunction::RunCleanupsScope LocalScope(CGF);
 
-      // Build void __kmpc_omp_wait_deps(ident_t *, kmp_int32 gtid,
+      // Build void __kmpc_omp_taskwait_deps_51(ident_t *, kmp_int32 gtid,
       // kmp_int32 ndeps, kmp_depend_info_t *dep_list, kmp_int32
-      // ndeps_noalias, kmp_depend_info_t *noalias_dep_list); if dependence info
-      // is specified.
-      CGF.EmitRuntimeCall(
-          OMPBuilder.getOrCreateRuntimeFunction(M, OMPRTL___kmpc_omp_wait_deps),
-          DepWaitTaskArgs);
+      // ndeps_noalias, kmp_depend_info_t *noalias_dep_list,
+      // kmp_int32 has_no_wait); if dependence info is specified.
+      CGF.EmitRuntimeCall(OMPBuilder.getOrCreateRuntimeFunction(
+                              M, OMPRTL___kmpc_omp_taskwait_deps_51),
+                          DepWaitTaskArgs);
 
     } else {
 
       // Build call kmp_int32 __kmpc_omp_taskwait(ident_t *loc, kmp_int32
       // global_tid);
-      llvm::Value *Args[] = {UpLoc, ThreadID};
+      llvm::Value *HasNowaitClauseVal =
+          llvm::ConstantInt::get(CGF.IntTy, Data.HasNowaitClause);
+      llvm::Value *Args[3] = {UpLoc, ThreadID, HasNowaitClauseVal};
       // Ignore return result until untied tasks are supported.
-      CGF.EmitRuntimeCall(
-          OMPBuilder.getOrCreateRuntimeFunction(M, OMPRTL___kmpc_omp_taskwait),
-          Args);
+      CGF.EmitRuntimeCall(OMPBuilder.getOrCreateRuntimeFunction(
+                              M, OMPRTL___kmpc_omp_taskwait_51),
+                          Args);
     }
   }
 
Index: clang/lib/CodeGen/CGStmtOpenMP.cpp
===================================================================
--- clang/lib/CodeGen/CGStmtOpenMP.cpp
+++ clang/lib/CodeGen/CGStmtOpenMP.cpp
@@ -5257,6 +5257,7 @@
   OMPTaskDataTy Data;
   // Build list of dependences
   buildDependences(S, Data);
+  Data.HasNowaitClause = S.hasClausesOfKind<OMPNowaitClause>();
   CGM.getOpenMPRuntime().emitTaskwaitCall(*this, S.getBeginLoc(), Data);
 }
 
Index: clang/test/OpenMP/target_depend_codegen.cpp
===================================================================
--- clang/test/OpenMP/target_depend_codegen.cpp
+++ clang/test/OpenMP/target_depend_codegen.cpp
@@ -89,7 +89,7 @@
 // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP_START]], i[[SZ]] 2
 // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP_START]], i[[SZ]] 3
 // CHECK:       [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8*
-// CHECK:       call void @__kmpc_omp_wait_deps(%struct.ident_t* @1, i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null)
+// CHECK:       call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* @1, i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null, i32 0)
 // CHECK:       call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @1, i32 [[GTID]], i8* [[TASK]])
 // CHECK:       call i32 [[TASK_ENTRY0]](i32 [[GTID]], [[TASK_TY0]]* [[BC_TASK]])
 // CHECK:       call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @1, i32 [[GTID]], i8* [[TASK]])
@@ -209,7 +209,7 @@
 // CHECK:       [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* @1, i32 [[GTID]], i32 1, i[[SZ]] {{48|24}}, i[[SZ]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY2:@.+]] to i32 (i32, i8*)*))
 // CHECK:       [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY2:%.+]]*
 // CHECK:       [[DEP:%.+]] = bitcast %struct.kmp_depend_info* %{{.+}} to i8*
-// CHECK:       call void @__kmpc_omp_wait_deps(%struct.ident_t* @1, i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null)
+// CHECK:       call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* @1, i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null, i32 0)
 // CHECK:       call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @1, i32 [[GTID]], i8* [[TASK]])
 // CHECK:       call i32 [[TASK_ENTRY2]](i32 [[GTID]], [[TASK_TY2]]* [[BC_TASK]])
 // CHECK:       call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @1, i32 [[GTID]], i8* [[TASK]])
Index: clang/test/OpenMP/target_enter_data_depend_codegen.cpp
===================================================================
--- clang/test/OpenMP/target_enter_data_depend_codegen.cpp
+++ clang/test/OpenMP/target_enter_data_depend_codegen.cpp
@@ -184,7 +184,7 @@
   // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 2
   // CK1: store i8 3, i8* [[DEP_ATTRS]]
   // CK1: [[BC:%.+]] = bitcast %struct.kmp_depend_info* [[MAIN_DEP]] to i8*
-  // CK1: call void @__kmpc_omp_wait_deps(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 3, i8* [[BC]], i32 0, i8* null)
+  // CK1: call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 3, i8* [[BC]], i32 0, i8* null, i32 0)
   // CK1: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]])
   // CK1: = call i32 [[TASK_ENTRY2]](i32 %{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]])
   // CK1: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]])
@@ -273,7 +273,7 @@
   // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 2
   // CK1: store i8 3, i8* [[DEP_ATTRS]]
   // CK1: [[BC:%.+]] = bitcast %struct.kmp_depend_info* [[MAIN_DEP]] to i8*
-  // CK1: call void @__kmpc_omp_wait_deps(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 4, i8* [[BC]], i32 0, i8* null)
+  // CK1: call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 4, i8* [[BC]], i32 0, i8* null, i32 0)
   // CK1: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]])
   // CK1: = call i32 [[TASK_ENTRY3]](i32 %{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]])
   // CK1: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]])
@@ -371,7 +371,7 @@
   // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 2
   // CK1: store i8 1, i8* [[DEP_ATTRS]]
   // CK1: [[BC:%.+]] = bitcast %struct.kmp_depend_info* [[MAIN_DEP]] to i8*
-  // CK1: call void @__kmpc_omp_wait_deps(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 5, i8* [[BC]], i32 0, i8* null)
+  // CK1: call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 5, i8* [[BC]], i32 0, i8* null, i32 0)
   // CK1: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]])
   // CK1: = call i32 [[TASK_ENTRY4]](i32 %{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]])
   // CK1: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]])
Index: clang/test/OpenMP/target_exit_data_depend_codegen.cpp
===================================================================
--- clang/test/OpenMP/target_exit_data_depend_codegen.cpp
+++ clang/test/OpenMP/target_exit_data_depend_codegen.cpp
@@ -184,7 +184,7 @@
   // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 2
   // CK1: store i8 3, i8* [[DEP_ATTRS]]
   // CK1: [[BC:%.+]] = bitcast %struct.kmp_depend_info* [[MAIN_DEP]] to i8*
-  // CK1: call void @__kmpc_omp_wait_deps(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 3, i8* [[BC]], i32 0, i8* null)
+  // CK1: call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 3, i8* [[BC]], i32 0, i8* null, i32 0)
   // CK1: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]])
   // CK1: = call i32 [[TASK_ENTRY2]](i32 %{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]])
   // CK1: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]])
@@ -273,7 +273,7 @@
   // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 2
   // CK1: store i8 3, i8* [[DEP_ATTRS]]
   // CK1: [[BC:%.+]] = bitcast %struct.kmp_depend_info* [[MAIN_DEP]] to i8*
-  // CK1: call void @__kmpc_omp_wait_deps(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 4, i8* [[BC]], i32 0, i8* null)
+  // CK1: call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 4, i8* [[BC]], i32 0, i8* null, i32 0)
   // CK1: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]])
   // CK1: = call i32 [[TASK_ENTRY3]](i32 %{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]])
   // CK1: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]])
@@ -371,7 +371,7 @@
   // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 2
   // CK1: store i8 1, i8* [[DEP_ATTRS]]
   // CK1: [[BC:%.+]] = bitcast %struct.kmp_depend_info* [[MAIN_DEP]] to i8*
-  // CK1: call void @__kmpc_omp_wait_deps(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 5, i8* [[BC]], i32 0, i8* null)
+  // CK1: call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 5, i8* [[BC]], i32 0, i8* null, i32 0)
   // CK1: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]])
   // CK1: = call i32 [[TASK_ENTRY4]](i32 %{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]])
   // CK1: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]])
Index: clang/test/OpenMP/target_parallel_depend_codegen.cpp
===================================================================
--- clang/test/OpenMP/target_parallel_depend_codegen.cpp
+++ clang/test/OpenMP/target_parallel_depend_codegen.cpp
@@ -87,7 +87,7 @@
 // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 2
 // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 3
 // CHECK:       [[DEP:%.+]] = bitcast %struct.kmp_depend_info* %{{.+}} to i8*
-// CHECK:       call void @__kmpc_omp_wait_deps(%struct.ident_t* @1, i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null)
+// CHECK:       call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* @1, i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null, i32 0)
 // CHECK:       call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @1, i32 [[GTID]], i8* [[TASK]])
 // CHECK:       call i32 [[TASK_ENTRY0]](i32 [[GTID]], [[TASK_TY0]]* [[BC_TASK]])
 // CHECK:       call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @1, i32 [[GTID]], i8* [[TASK]])
@@ -164,7 +164,7 @@
 // CHECK:       [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY2:%.+]]*
 // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 0
 // CHECK:       [[DEP:%.+]] = bitcast %struct.kmp_depend_info* %{{.+}} to i8*
-// CHECK:       call void @__kmpc_omp_wait_deps(%struct.ident_t* @1, i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null)
+// CHECK:       call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* @1, i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null, i32 0)
 // CHECK:       call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @1, i32 [[GTID]], i8* [[TASK]])
 // CHECK:       call i32 [[TASK_ENTRY2]](i32 [[GTID]], [[TASK_TY2]]* [[BC_TASK]])
 // CHECK:       call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @1, i32 [[GTID]], i8* [[TASK]])
Index: clang/test/OpenMP/target_parallel_for_depend_codegen.cpp
===================================================================
--- clang/test/OpenMP/target_parallel_for_depend_codegen.cpp
+++ clang/test/OpenMP/target_parallel_for_depend_codegen.cpp
@@ -87,7 +87,7 @@
   // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 2
   // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 3
   // CHECK:       [[DEP:%.+]] = bitcast %struct.kmp_depend_info* %{{.+}} to i8*
-  // CHECK:       call void @__kmpc_omp_wait_deps(%struct.ident_t* [[IN]], i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null)
+  // CHECK:       call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* [[IN]], i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null, i32 0)
   // CHECK:       call void @__kmpc_omp_task_begin_if0(%struct.ident_t* [[IN]], i32 [[GTID]], i8* [[TASK]])
   // CHECK:       call i32 [[TASK_ENTRY0]](i32 [[GTID]], [[TASK_TY0]]* [[BC_TASK]])
   // CHECK:       call void @__kmpc_omp_task_complete_if0(%struct.ident_t* [[IN]], i32 [[GTID]], i8* [[TASK]])
@@ -160,7 +160,7 @@
   // CHECK:       [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY2:%.+]]*
   // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 0
   // CHECK:       [[DEP:%.+]] = bitcast %struct.kmp_depend_info* %{{.+}} to i8*
-  // CHECK:       call void @__kmpc_omp_wait_deps(%struct.ident_t* [[IN]], i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null)
+  // CHECK:       call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* [[IN]], i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null, i32 0)
   // CHECK:       call void @__kmpc_omp_task_begin_if0(%struct.ident_t* [[IN]], i32 [[GTID]], i8* [[TASK]])
   // CHECK:       call i32 [[TASK_ENTRY2]](i32 [[GTID]], [[TASK_TY2]]* [[BC_TASK]])
   // CHECK:       call void @__kmpc_omp_task_complete_if0(%struct.ident_t* [[IN]], i32 [[GTID]], i8* [[TASK]])
Index: clang/test/OpenMP/target_parallel_for_simd_depend_codegen.cpp
===================================================================
--- clang/test/OpenMP/target_parallel_for_simd_depend_codegen.cpp
+++ clang/test/OpenMP/target_parallel_for_simd_depend_codegen.cpp
@@ -87,7 +87,7 @@
   // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 2
   // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 3
   // CHECK:       [[DEP:%.+]] = bitcast %struct.kmp_depend_info* %{{.+}} to i8*
-  // CHECK:       call void @__kmpc_omp_wait_deps(%struct.ident_t* [[IN]], i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null)
+  // CHECK:       call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* [[IN]], i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null, i32 0)
   // CHECK:       call void @__kmpc_omp_task_begin_if0(%struct.ident_t* [[IN]], i32 [[GTID]], i8* [[TASK]])
   // CHECK:       call i32 [[TASK_ENTRY0]](i32 [[GTID]], [[TASK_TY0]]* [[BC_TASK]])
   // CHECK:       call void @__kmpc_omp_task_complete_if0(%struct.ident_t* [[IN]], i32 [[GTID]], i8* [[TASK]])
@@ -160,7 +160,7 @@
   // CHECK:       [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY2:%.+]]*
   // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 0
   // CHECK:       [[DEP:%.+]] = bitcast %struct.kmp_depend_info* %{{.+}} to i8*
-  // CHECK:       call void @__kmpc_omp_wait_deps(%struct.ident_t* [[IN]], i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null)
+  // CHECK:       call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* [[IN]], i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null, i32 0)
   // CHECK:       call void @__kmpc_omp_task_begin_if0(%struct.ident_t* [[IN]], i32 [[GTID]], i8* [[TASK]])
   // CHECK:       call i32 [[TASK_ENTRY2]](i32 [[GTID]], [[TASK_TY2]]* [[BC_TASK]])
   // CHECK:       call void @__kmpc_omp_task_complete_if0(%struct.ident_t* [[IN]], i32 [[GTID]], i8* [[TASK]])
Index: clang/test/OpenMP/target_simd_depend_codegen.cpp
===================================================================
--- clang/test/OpenMP/target_simd_depend_codegen.cpp
+++ clang/test/OpenMP/target_simd_depend_codegen.cpp
@@ -87,7 +87,7 @@
 // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 2
 // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 3
 // CHECK:       [[DEP:%.+]] = bitcast %struct.kmp_depend_info* %{{.+}} to i8*
-// CHECK:       call void @__kmpc_omp_wait_deps(%struct.ident_t* @1, i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null)
+// CHECK:       call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* @1, i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null, i32 0)
 // CHECK:       call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @1, i32 [[GTID]], i8* [[TASK]])
 // CHECK:       call i32 [[TASK_ENTRY0]](i32 [[GTID]], [[TASK_TY0]]* [[BC_TASK]])
 // CHECK:       call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @1, i32 [[GTID]], i8* [[TASK]])
@@ -163,7 +163,7 @@
 // CHECK:       [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY2:%.+]]*
 // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 0
 // CHECK:       [[DEP:%.+]] = bitcast %struct.kmp_depend_info* %{{.+}} to i8*
-// CHECK:       call void @__kmpc_omp_wait_deps(%struct.ident_t* @1, i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null)
+// CHECK:       call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* @1, i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null, i32 0)
 // CHECK:       call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @1, i32 [[GTID]], i8* [[TASK]])
 // CHECK:       call i32 [[TASK_ENTRY2]](i32 [[GTID]], [[TASK_TY2]]* [[BC_TASK]])
 // CHECK:       call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @1, i32 [[GTID]], i8* [[TASK]])
Index: clang/test/OpenMP/target_teams_depend_codegen.cpp
===================================================================
--- clang/test/OpenMP/target_teams_depend_codegen.cpp
+++ clang/test/OpenMP/target_teams_depend_codegen.cpp
@@ -87,7 +87,7 @@
 // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 2
 // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 3
 // CHECK:       [[DEP:%.+]] = bitcast %struct.kmp_depend_info* %{{.+}} to i8*
-// CHECK:       call void @__kmpc_omp_wait_deps(%struct.ident_t* @1, i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null)
+// CHECK:       call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* @1, i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null, i32 0)
 // CHECK:       call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @1, i32 [[GTID]], i8* [[TASK]])
 // CHECK:       call i32 [[TASK_ENTRY0]](i32 [[GTID]], [[TASK_TY0]]* [[BC_TASK]])
 // CHECK:       call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @1, i32 [[GTID]], i8* [[TASK]])
@@ -163,7 +163,7 @@
 // CHECK:       [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY2:%.+]]*
 // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 0
 // CHECK:       [[DEP:%.+]] = bitcast %struct.kmp_depend_info* %{{.+}} to i8*
-// CHECK:       call void @__kmpc_omp_wait_deps(%struct.ident_t* @1, i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null)
+// CHECK:       call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* @1, i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null, i32 0)
 // CHECK:       call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @1, i32 [[GTID]], i8* [[TASK]])
 // CHECK:       call i32 [[TASK_ENTRY2]](i32 [[GTID]], [[TASK_TY2]]* [[BC_TASK]])
 // CHECK:       call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @1, i32 [[GTID]], i8* [[TASK]])
Index: clang/test/OpenMP/target_teams_distribute_depend_codegen.cpp
===================================================================
--- clang/test/OpenMP/target_teams_distribute_depend_codegen.cpp
+++ clang/test/OpenMP/target_teams_distribute_depend_codegen.cpp
@@ -87,7 +87,7 @@
   // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 2
   // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 3
   // CHECK:       [[DEP:%.+]] = bitcast %struct.kmp_depend_info* %{{.+}} to i8*
-  // CHECK:       call void @__kmpc_omp_wait_deps(%struct.ident_t* [[ID]], i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null)
+  // CHECK:       call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* [[ID]], i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null, i32 0)
   // CHECK:       call void @__kmpc_omp_task_begin_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]])
   // CHECK:       call i32 [[TASK_ENTRY0]](i32 [[GTID]], [[TASK_TY0]]* [[BC_TASK]])
   // CHECK:       call void @__kmpc_omp_task_complete_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]])
@@ -160,7 +160,7 @@
   // CHECK:       [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY2:%.+]]*
   // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 0
   // CHECK:       [[DEP:%.+]] = bitcast %struct.kmp_depend_info* %{{.+}} to i8*
-  // CHECK:       call void @__kmpc_omp_wait_deps(%struct.ident_t* [[ID]], i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null)
+  // CHECK:       call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* [[ID]], i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null, i32 0)
   // CHECK:       call void @__kmpc_omp_task_begin_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]])
   // CHECK:       call i32 [[TASK_ENTRY2]](i32 [[GTID]], [[TASK_TY2]]* [[BC_TASK]])
   // CHECK:       call void @__kmpc_omp_task_complete_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]])
Index: clang/test/OpenMP/target_teams_distribute_parallel_for_depend_codegen.cpp
===================================================================
--- clang/test/OpenMP/target_teams_distribute_parallel_for_depend_codegen.cpp
+++ clang/test/OpenMP/target_teams_distribute_parallel_for_depend_codegen.cpp
@@ -87,7 +87,7 @@
   // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 2
   // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 3
   // CHECK:       [[DEP:%.+]] = bitcast %struct.kmp_depend_info* %{{.+}} to i8*
-  // CHECK:       call void @__kmpc_omp_wait_deps(%struct.ident_t* [[ID]], i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null)
+  // CHECK:       call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* [[ID]], i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null, i32 0)
   // CHECK:       call void @__kmpc_omp_task_begin_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]])
   // CHECK:       call i32 [[TASK_ENTRY0]](i32 [[GTID]], [[TASK_TY0]]* [[BC_TASK]])
   // CHECK:       call void @__kmpc_omp_task_complete_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]])
@@ -160,7 +160,7 @@
   // CHECK:       [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY2:%.+]]*
   // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 0
   // CHECK:       [[DEP:%.+]] = bitcast %struct.kmp_depend_info* %{{.+}} to i8*
-  // CHECK:       call void @__kmpc_omp_wait_deps(%struct.ident_t* [[ID]], i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null)
+  // CHECK:       call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* [[ID]], i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null, i32 0)
   // CHECK:       call void @__kmpc_omp_task_begin_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]])
   // CHECK:       call i32 [[TASK_ENTRY2]](i32 [[GTID]], [[TASK_TY2]]* [[BC_TASK]])
   // CHECK:       call void @__kmpc_omp_task_complete_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]])
Index: clang/test/OpenMP/target_teams_distribute_parallel_for_simd_depend_codegen.cpp
===================================================================
--- clang/test/OpenMP/target_teams_distribute_parallel_for_simd_depend_codegen.cpp
+++ clang/test/OpenMP/target_teams_distribute_parallel_for_simd_depend_codegen.cpp
@@ -87,7 +87,7 @@
   // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 2
   // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 3
   // CHECK:       [[DEP:%.+]] = bitcast %struct.kmp_depend_info* %{{.+}} to i8*
-  // CHECK:       call void @__kmpc_omp_wait_deps(%struct.ident_t* [[ID]], i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null)
+  // CHECK:       call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* [[ID]], i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null, i32 0)
   // CHECK:       call void @__kmpc_omp_task_begin_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]])
   // CHECK:       call i32 [[TASK_ENTRY0]](i32 [[GTID]], [[TASK_TY0]]* [[BC_TASK]])
   // CHECK:       call void @__kmpc_omp_task_complete_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]])
@@ -160,7 +160,7 @@
   // CHECK:       [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY2:%.+]]*
   // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 0
   // CHECK:       [[DEP:%.+]] = bitcast %struct.kmp_depend_info* %{{.+}} to i8*
-  // CHECK:       call void @__kmpc_omp_wait_deps(%struct.ident_t* [[ID]], i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null)
+  // CHECK:       call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* [[ID]], i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null, i32 0)
   // CHECK:       call void @__kmpc_omp_task_begin_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]])
   // CHECK:       call i32 [[TASK_ENTRY2]](i32 [[GTID]], [[TASK_TY2]]* [[BC_TASK]])
   // CHECK:       call void @__kmpc_omp_task_complete_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]])
Index: clang/test/OpenMP/target_teams_distribute_simd_depend_codegen.cpp
===================================================================
--- clang/test/OpenMP/target_teams_distribute_simd_depend_codegen.cpp
+++ clang/test/OpenMP/target_teams_distribute_simd_depend_codegen.cpp
@@ -101,7 +101,7 @@
   // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 2
   // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 3
   // CHECK:       [[DEP:%.+]] = bitcast %struct.kmp_depend_info* %{{.+}} to i8*
-  // CHECK:       call void @__kmpc_omp_wait_deps(%struct.ident_t* [[ID]], i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null)
+  // CHECK:       call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* [[ID]], i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null, i32 0)
   // CHECK:       call void @__kmpc_omp_task_begin_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]])
   // CHECK:       call i32 [[TASK_ENTRY0]](i32 [[GTID]], [[TASK_TY0]]* [[BC_TASK]])
   // CHECK:       call void @__kmpc_omp_task_complete_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]])
@@ -199,7 +199,7 @@
   // CHECK:       [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY2:%.+]]*
   // CHECK:       getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* %{{.+}}, i[[SZ]] 0
   // CHECK:       [[DEP:%.+]] = bitcast %struct.kmp_depend_info* %{{.+}} to i8*
-  // CHECK:       call void @__kmpc_omp_wait_deps(%struct.ident_t* [[ID]], i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null)
+  // CHECK:       call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* [[ID]], i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null, i32 0)
   // CHECK:       call void @__kmpc_omp_task_begin_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]])
   // CHECK:       call i32 [[TASK_ENTRY2]](i32 [[GTID]], [[TASK_TY2]]* [[BC_TASK]])
   // CHECK:       call void @__kmpc_omp_task_complete_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]])
Index: clang/test/OpenMP/target_update_depend_codegen.cpp
===================================================================
--- clang/test/OpenMP/target_update_depend_codegen.cpp
+++ clang/test/OpenMP/target_update_depend_codegen.cpp
@@ -184,7 +184,7 @@
   // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 2
   // CK1: store i8 3, i8* [[DEP_ATTRS]]
   // CK1: [[BC:%.+]] = bitcast %struct.kmp_depend_info* [[MAIN_DEP]] to i8*
-  // CK1: call void @__kmpc_omp_wait_deps(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 3, i8* [[BC]], i32 0, i8* null)
+  // CK1: call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 3, i8* [[BC]], i32 0, i8* null, i32 0)
   // CK1: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]])
   // CK1: = call i32 [[TASK_ENTRY2]](i32 %{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]])
   // CK1: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]])
@@ -273,7 +273,7 @@
   // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 2
   // CK1: store i8 3, i8* [[DEP_ATTRS]]
   // CK1: [[BC:%.+]] = bitcast %struct.kmp_depend_info* [[MAIN_DEP]] to i8*
-  // CK1: call void @__kmpc_omp_wait_deps(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 4, i8* [[BC]], i32 0, i8* null)
+  // CK1: call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 4, i8* [[BC]], i32 0, i8* null, i32 0)
   // CK1: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]])
   // CK1: = call i32 [[TASK_ENTRY3]](i32 %{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]])
   // CK1: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]])
@@ -371,7 +371,7 @@
   // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 2
   // CK1: store i8 1, i8* [[DEP_ATTRS]]
   // CK1: [[BC:%.+]] = bitcast %struct.kmp_depend_info* [[MAIN_DEP]] to i8*
-  // CK1: call void @__kmpc_omp_wait_deps(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 5, i8* [[BC]], i32 0, i8* null)
+  // CK1: call void @__kmpc_omp_taskwait_deps_51(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 5, i8* [[BC]], i32 0, i8* null, i32 0)
   // CK1: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]])
   // CK1: = call i32 [[TASK_ENTRY4]](i32 %{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]])
   // CK1: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]])
Index: clang/test/OpenMP/task_codegen.cpp
===================================================================
--- clang/test/OpenMP/task_codegen.cpp
+++ clang/test/OpenMP/task_codegen.cpp
@@ -1093,7 +1093,7 @@
 // CHECK1-NEXT:    call void @_ZN1SD1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[REF_TMP_I]]) #[[ATTR4]]
 // CHECK1-NEXT:    store i32 10, ptr [[TMP15]], align 4
 // CHECK1-NEXT:    [[TMP43:%.*]] = load i32, ptr [[DOTGLOBAL_TID__ADDR_I]], align 4, !noalias !112
-// CHECK1-NEXT:    [[TMP44:%.*]] = call i32 @__kmpc_omp_taskwait(ptr @[[GLOB1]], i32 [[TMP43]])
+// CHECK1-NEXT:    [[TMP44:%.*]] = call i32 @__kmpc_omp_taskwait_51(ptr @[[GLOB1]], i32 [[TMP43]], i32 0)
 // CHECK1-NEXT:    [[TMP45:%.*]] = load ptr, ptr [[DOTPART_ID__ADDR_I]], align 8, !noalias !112
 // CHECK1-NEXT:    store i32 5, ptr [[TMP45]], align 4
 // CHECK1-NEXT:    [[TMP46:%.*]] = load i32, ptr [[DOTGLOBAL_TID__ADDR_I]], align 4, !noalias !112
@@ -2213,7 +2213,7 @@
 // CHECK2-NEXT:    call void @_ZN1SD1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[REF_TMP_I]]) #[[ATTR4]]
 // CHECK2-NEXT:    store i32 10, ptr [[TMP15]], align 4
 // CHECK2-NEXT:    [[TMP43:%.*]] = load i32, ptr [[DOTGLOBAL_TID__ADDR_I]], align 4, !noalias !112
-// CHECK2-NEXT:    [[TMP44:%.*]] = call i32 @__kmpc_omp_taskwait(ptr @[[GLOB1]], i32 [[TMP43]])
+// CHECK2-NEXT:    [[TMP44:%.*]] = call i32 @__kmpc_omp_taskwait_51(ptr @[[GLOB1]], i32 [[TMP43]], i32 0)
 // CHECK2-NEXT:    [[TMP45:%.*]] = load ptr, ptr [[DOTPART_ID__ADDR_I]], align 8, !noalias !112
 // CHECK2-NEXT:    store i32 5, ptr [[TMP45]], align 4
 // CHECK2-NEXT:    [[TMP46:%.*]] = load i32, ptr [[DOTGLOBAL_TID__ADDR_I]], align 4, !noalias !112
@@ -3415,7 +3415,7 @@
 // CHECK2-51-NEXT:    call void @_ZN1SD1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[REF_TMP_I]]) #[[ATTR4]]
 // CHECK2-51-NEXT:    store i32 10, ptr [[TMP15]], align 4
 // CHECK2-51-NEXT:    [[TMP43:%.*]] = load i32, ptr [[DOTGLOBAL_TID__ADDR_I]], align 4, !noalias !122
-// CHECK2-51-NEXT:    [[TMP44:%.*]] = call i32 @__kmpc_omp_taskwait(ptr @[[GLOB1]], i32 [[TMP43]])
+// CHECK2-51-NEXT:    [[TMP44:%.*]] = call i32 @__kmpc_omp_taskwait_51(ptr @[[GLOB1]], i32 [[TMP43]], i32 0)
 // CHECK2-51-NEXT:    [[TMP45:%.*]] = load ptr, ptr [[DOTPART_ID__ADDR_I]], align 8, !noalias !122
 // CHECK2-51-NEXT:    store i32 5, ptr [[TMP45]], align 4
 // CHECK2-51-NEXT:    [[TMP46:%.*]] = load i32, ptr [[DOTGLOBAL_TID__ADDR_I]], align 4, !noalias !122
@@ -4953,7 +4953,7 @@
 // CHECK3-NEXT:    call void @_ZN1SD1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[REF_TMP_I]]) #[[ATTR4]]
 // CHECK3-NEXT:    store i32 10, ptr [[S2_I]], align 4, !noalias !112
 // CHECK3-NEXT:    [[OMP_GLOBAL_THREAD_NUM13_I:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
-// CHECK3-NEXT:    [[TMP31:%.*]] = call i32 @__kmpc_omp_taskwait(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM13_I]])
+// CHECK3-NEXT:    [[TMP31:%.*]] = call i32 @__kmpc_omp_taskwait_51(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM13_I]], i32 0)
 // CHECK3-NEXT:    [[TMP32:%.*]] = load ptr, ptr [[DOTPART_ID__ADDR_I]], align 8, !noalias !112
 // CHECK3-NEXT:    store i32 4, ptr [[TMP32]], align 4
 // CHECK3-NEXT:    [[OMP_GLOBAL_THREAD_NUM14_I:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB21]])
@@ -5969,7 +5969,7 @@
 // CHECK4-NEXT:    call void @_ZN1SD1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[REF_TMP_I]]) #[[ATTR4]]
 // CHECK4-NEXT:    store i32 10, ptr [[S2_I]], align 4, !noalias !112
 // CHECK4-NEXT:    [[OMP_GLOBAL_THREAD_NUM13_I:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
-// CHECK4-NEXT:    [[TMP31:%.*]] = call i32 @__kmpc_omp_taskwait(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM13_I]])
+// CHECK4-NEXT:    [[TMP31:%.*]] = call i32 @__kmpc_omp_taskwait_51(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM13_I]], i32 0)
 // CHECK4-NEXT:    [[TMP32:%.*]] = load ptr, ptr [[DOTPART_ID__ADDR_I]], align 8, !noalias !112
 // CHECK4-NEXT:    store i32 4, ptr [[TMP32]], align 4
 // CHECK4-NEXT:    [[OMP_GLOBAL_THREAD_NUM14_I:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB21]])
@@ -7069,7 +7069,7 @@
 // CHECK3-51-NEXT:    call void @_ZN1SD1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[REF_TMP_I]]) #[[ATTR4]]
 // CHECK3-51-NEXT:    store i32 10, ptr [[S2_I]], align 4, !noalias !122
 // CHECK3-51-NEXT:    [[OMP_GLOBAL_THREAD_NUM13_I:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
-// CHECK3-51-NEXT:    [[TMP31:%.*]] = call i32 @__kmpc_omp_taskwait(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM13_I]])
+// CHECK3-51-NEXT:    [[TMP31:%.*]] = call i32 @__kmpc_omp_taskwait_51(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM13_I]], i32 0)
 // CHECK3-51-NEXT:    [[TMP32:%.*]] = load ptr, ptr [[DOTPART_ID__ADDR_I]], align 8, !noalias !122
 // CHECK3-51-NEXT:    store i32 4, ptr [[TMP32]], align 4
 // CHECK3-51-NEXT:    [[OMP_GLOBAL_THREAD_NUM14_I:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB23]])
@@ -8503,7 +8503,7 @@
 // CHECK4-51-NEXT:    call void @_ZN1SD1Ev(ptr noundef nonnull align 4 dereferenceable(4) [[REF_TMP_I]]) #[[ATTR4]]
 // CHECK4-51-NEXT:    store i32 10, ptr [[S2_I]], align 4, !noalias !112
 // CHECK4-51-NEXT:    [[OMP_GLOBAL_THREAD_NUM13_I:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
-// CHECK4-51-NEXT:    [[TMP31:%.*]] = call i32 @__kmpc_omp_taskwait(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM13_I]])
+// CHECK4-51-NEXT:    [[TMP31:%.*]] = call i32 @__kmpc_omp_taskwait_51(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM13_I]], i32 0)
 // CHECK4-51-NEXT:    [[TMP32:%.*]] = load ptr, ptr [[DOTPART_ID__ADDR_I]], align 8, !noalias !112
 // CHECK4-51-NEXT:    store i32 4, ptr [[TMP32]], align 4
 // CHECK4-51-NEXT:    [[OMP_GLOBAL_THREAD_NUM14_I:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB21]])
Index: clang/test/OpenMP/task_if_codegen.cpp
===================================================================
--- clang/test/OpenMP/task_if_codegen.cpp
+++ clang/test/OpenMP/task_if_codegen.cpp
@@ -198,7 +198,7 @@
 // CHECK1-NEXT:    [[TMP25:%.*]] = call i32 @__kmpc_omp_task_with_deps(ptr @[[GLOB1]], i32 [[TMP0]], ptr [[TMP15]], i32 1, ptr [[TMP18]], i32 0, ptr null)
 // CHECK1-NEXT:    br label [[OMP_IF_END7:%.*]]
 // CHECK1:       omp_if.else6:
-// CHECK1-NEXT:    call void @__kmpc_omp_wait_deps(ptr @[[GLOB1]], i32 [[TMP0]], i32 1, ptr [[TMP18]], i32 0, ptr null)
+// CHECK1-NEXT:    call void @__kmpc_omp_taskwait_deps_51(ptr @[[GLOB1]], i32 [[TMP0]], i32 1, ptr [[TMP18]], i32 0, ptr null, i32 0)
 // CHECK1-NEXT:    call void @__kmpc_omp_task_begin_if0(ptr @[[GLOB1]], i32 [[TMP0]], ptr [[TMP15]])
 // CHECK1-NEXT:    [[TMP26:%.*]] = call i32 @.omp_task_entry..9(i32 [[TMP0]], ptr [[TMP15]]) #[[ATTR3]]
 // CHECK1-NEXT:    call void @__kmpc_omp_task_complete_if0(ptr @[[GLOB1]], i32 [[TMP0]], ptr [[TMP15]])
@@ -404,7 +404,7 @@
 // CHECK1-NEXT:    [[TMP26:%.*]] = call i32 @__kmpc_omp_task_with_deps(ptr @[[GLOB1]], i32 [[TMP0]], ptr [[TMP15]], i32 1, ptr [[TMP18]], i32 0, ptr null)
 // CHECK1-NEXT:    br label [[OMP_IF_END7:%.*]]
 // CHECK1:       omp_if.else6:
-// CHECK1-NEXT:    call void @__kmpc_omp_wait_deps(ptr @[[GLOB1]], i32 [[TMP0]], i32 1, ptr [[TMP18]], i32 0, ptr null)
+// CHECK1-NEXT:    call void @__kmpc_omp_taskwait_deps_51(ptr @[[GLOB1]], i32 [[TMP0]], i32 1, ptr [[TMP18]], i32 0, ptr null, i32 0)
 // CHECK1-NEXT:    call void @__kmpc_omp_task_begin_if0(ptr @[[GLOB1]], i32 [[TMP0]], ptr [[TMP15]])
 // CHECK1-NEXT:    [[TMP27:%.*]] = call i32 @.omp_task_entry..17(i32 [[TMP0]], ptr [[TMP15]]) #[[ATTR3]]
 // CHECK1-NEXT:    call void @__kmpc_omp_task_complete_if0(ptr @[[GLOB1]], i32 [[TMP0]], ptr [[TMP15]])
@@ -429,7 +429,7 @@
 // CHECK1-NEXT:    [[TMP39:%.*]] = call i32 @__kmpc_omp_task_with_deps(ptr @[[GLOB1]], i32 [[TMP0]], ptr [[TMP28]], i32 1, ptr [[TMP31]], i32 0, ptr null)
 // CHECK1-NEXT:    br label [[OMP_IF_END14:%.*]]
 // CHECK1:       omp_if.else13:
-// CHECK1-NEXT:    call void @__kmpc_omp_wait_deps(ptr @[[GLOB1]], i32 [[TMP0]], i32 1, ptr [[TMP31]], i32 0, ptr null)
+// CHECK1-NEXT:    call void @__kmpc_omp_taskwait_deps_51(ptr @[[GLOB1]], i32 [[TMP0]], i32 1, ptr [[TMP31]], i32 0, ptr null, i32 0)
 // CHECK1-NEXT:    call void @__kmpc_omp_task_begin_if0(ptr @[[GLOB1]], i32 [[TMP0]], ptr [[TMP28]])
 // CHECK1-NEXT:    [[TMP40:%.*]] = call i32 @.omp_task_entry..19(i32 [[TMP0]], ptr [[TMP28]]) #[[ATTR3]]
 // CHECK1-NEXT:    call void @__kmpc_omp_task_complete_if0(ptr @[[GLOB1]], i32 [[TMP0]], ptr [[TMP28]])
@@ -454,7 +454,7 @@
 // CHECK1-NEXT:    [[TMP52:%.*]] = call i32 @__kmpc_omp_task_with_deps(ptr @[[GLOB1]], i32 [[TMP0]], ptr [[TMP41]], i32 1, ptr [[TMP44]], i32 0, ptr null)
 // CHECK1-NEXT:    br label [[OMP_IF_END21:%.*]]
 // CHECK1:       omp_if.else20:
-// CHECK1-NEXT:    call void @__kmpc_omp_wait_deps(ptr @[[GLOB1]], i32 [[TMP0]], i32 1, ptr [[TMP44]], i32 0, ptr null)
+// CHECK1-NEXT:    call void @__kmpc_omp_taskwait_deps_51(ptr @[[GLOB1]], i32 [[TMP0]], i32 1, ptr [[TMP44]], i32 0, ptr null, i32 0)
 // CHECK1-NEXT:    call void @__kmpc_omp_task_begin_if0(ptr @[[GLOB1]], i32 [[TMP0]], ptr [[TMP41]])
 // CHECK1-NEXT:    [[TMP53:%.*]] = call i32 @.omp_task_entry..21(i32 [[TMP0]], ptr [[TMP41]]) #[[ATTR3]]
 // CHECK1-NEXT:    call void @__kmpc_omp_task_complete_if0(ptr @[[GLOB1]], i32 [[TMP0]], ptr [[TMP41]])
Index: clang/test/OpenMP/taskwait_ast_print.cpp
===================================================================
--- clang/test/OpenMP/taskwait_ast_print.cpp
+++ clang/test/OpenMP/taskwait_ast_print.cpp
@@ -1,10 +1,10 @@
-// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=50 -ast-print %s | FileCheck %s
-// RUN: %clang_cc1 -fopenmp -fopenmp-version=50 -x c++ -std=c++11 -emit-pch -o %t %s
-// RUN: %clang_cc1 -fopenmp -fopenmp-version=50 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=51 -ast-print %s | FileCheck %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=51 -x c++ -std=c++11 -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=51 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s
 
-// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=50 -ast-print %s | FileCheck %s
-// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=50 -x c++ -std=c++11 -emit-pch -o %t %s
-// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=50 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s
+// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=51 -ast-print %s | FileCheck %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=51 -x c++ -std=c++11 -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=51 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s
 // expected-no-diagnostics
 
 #ifndef HEADER
@@ -19,6 +19,23 @@
 #pragma omp taskwait depend(in:a, argc)
   return a + argc;
 }
+
+template <class T>
+T nmain(T argc) {
+  static T a;
+#pragma omp taskwait
+#pragma omp taskwait nowait
+  return a + argc;
+}
+
+template <class T>
+T ndmain(T argc) {
+  static T a;
+#pragma omp taskwait
+#pragma omp taskwait depend(in:a, argc) nowait
+  return a + argc;
+}
+
 // CHECK:      static T a;
 // CHECK-NEXT: #pragma omp taskwait{{$}}
 // CHECK-NEXT: #pragma omp taskwait depend(in : a,argc){{$}}
@@ -29,6 +46,26 @@
 // CHECK-NEXT: #pragma omp taskwait
 // CHECK-NEXT: #pragma omp taskwait depend(in : a,argc){{$}}
 
+// CHECK:      static T a;
+// CHECK-NEXT: #pragma omp taskwait{{$}}
+// CHECK-NEXT: #pragma omp taskwait nowait{{$}}
+// CHECK:      static int a;
+// CHECK-NEXT: #pragma omp taskwait
+// CHECK-NEXT: #pragma omp taskwait nowait{{$}}
+// CHECK:      static char a;
+// CHECK-NEXT: #pragma omp taskwait
+// CHECK-NEXT: #pragma omp taskwait nowait{{$}}
+
+// CHECK:      static T a;
+// CHECK-NEXT: #pragma omp taskwait{{$}}
+// CHECK-NEXT: #pragma omp taskwait depend(in : a,argc) nowait{{$}}
+// CHECK:      static int a;
+// CHECK-NEXT: #pragma omp taskwait
+// CHECK-NEXT: #pragma omp taskwait depend(in : a,argc) nowait{{$}}
+// CHECK:      static char a;
+// CHECK-NEXT: #pragma omp taskwait
+// CHECK-NEXT: #pragma omp taskwait depend(in : a,argc) nowait{{$}}
+
 int main(int argc, char **argv) {
   static int a;
 // CHECK: static int a;
@@ -36,7 +73,7 @@
 #pragma omp taskwait depend(out:a, argc)
   // CHECK-NEXT: #pragma omp taskwait
   // CHECK-NEXT: #pragma omp taskwait depend(out : a,argc)
-  return tmain(argc) + tmain(argv[0][0]) + a;
+  return tmain(argc) + tmain(argv[0][0]) + a + nmain(argc) + nmain(argv[0][0]) + ndmain(argc) + ndmain(argv[0][0]);
 }
 
 #endif
Index: clang/test/OpenMP/taskwait_codegen.cpp
===================================================================
--- clang/test/OpenMP/taskwait_codegen.cpp
+++ clang/test/OpenMP/taskwait_codegen.cpp
@@ -29,11 +29,11 @@
 
 // CHECK-LABEL: @main
 // CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(ptr @{{.+}})
-// CHECK: call i32 @__kmpc_omp_taskwait(ptr @{{.+}}, i32 [[GTID]])
+// CHECK: call i32 @__kmpc_omp_taskwait_51(ptr @{{.+}}, i32 [[GTID]], i32 0)
 
 // CHECK-LABEL: tmain
 // CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(ptr @{{.+}})
-// CHECK: call i32 @__kmpc_omp_taskwait(ptr @{{.+}}, i32 [[GTID]])
+// CHECK: call i32 @__kmpc_omp_taskwait_51(ptr @{{.+}}, i32 [[GTID]], i32 0)
 
 
 #endif
Index: clang/test/OpenMP/taskwait_depend_codegen.cpp
===================================================================
--- clang/test/OpenMP/taskwait_depend_codegen.cpp
+++ clang/test/OpenMP/taskwait_depend_codegen.cpp
@@ -32,7 +32,7 @@
 
 // CHECK-LABEL: tmain
 // CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(ptr @{{.+}})
-// CHECK: call void @__kmpc_omp_wait_deps(ptr @{{.+}}, i32 [[GTID]], i32 1, ptr %{{.}}, i32 0, ptr null)
+// CHECK: call void @__kmpc_omp_taskwait_deps_51(ptr @{{.+}}, i32 [[GTID]], i32 1, ptr %{{.}}, i32 0, ptr null, i32 0)
 
 
 #endif
Index: clang/test/OpenMP/taskwait_depend_nowait_codegen.cpp
===================================================================
--- /dev/null
+++ clang/test/OpenMP/taskwait_depend_nowait_codegen.cpp
@@ -0,0 +1,50 @@
+// RUN: %clang_cc1 -no-opaque-pointers -verify -triple x86_64-apple-darwin10 -fopenmp -fopenmp-version=51 -x c++ -emit-llvm %s -o - | FileCheck %s
+// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=51 -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s
+// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=51 -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s
+
+// RUN: %clang_cc1 -no-opaque-pointers -verify -triple x86_64-apple-darwin10 -fopenmp-simd -fopenmp-version=51 -x c++ -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=51 -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s
+// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=51 -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
+// expected-no-diagnostics
+#ifndef HEADER
+#define HEADER
+
+void foo() {}
+
+template <class T>
+T tmain(T &argc) {
+  static T a;
+  #pragma omp taskwait depend(in:argc)
+  return a + argc;
+}
+
+template <class T>
+T nmain(T &argc) {
+  static T a;
+  #pragma omp taskwait depend(in:argc) nowait
+  return a + argc;
+}
+
+int main(int argc, char **argv) {
+  int n = 0;
+  #pragma omp task shared(n,argc) depend(out:n)
+     n = argc;
+  return tmain(n) + nmain(n);
+}
+
+// CHECK-LABEL: @main
+// CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%{{.+}}* @{{.+}})
+// CHECK: [[ALLOC:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{.+}}* @{{.+}}, i32 [[GTID]], i32 1, i64 40, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* @{{.+}} to i32 (i32, i8*)*))
+// CHECK: %{{.+}} = call i32 @__kmpc_omp_task_with_deps(%{{.+}}* @{{.+}}, i32 [[GTID]], i8* [[ALLOC]], i32 1, i8* %{{[0-9]*}}, i32 0, i8* null)
+
+// CHECK-LABEL: define {{.*}} @{{.*}}tmain{{.*}}
+// CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%{{.+}}* @{{.+}})
+// CHECK: call void @__kmpc_omp_taskwait_deps_51(%{{.+}}* @{{.+}}, i32 [[GTID]], i32 1, i8* %{{.}}, i32 0, i8* null, i32 0)
+
+// CHECK-LABEL: define {{.*}} @{{.*}}nmain{{.*}}
+// CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%{{.+}}* @{{.+}})
+// CHECK: call void @__kmpc_omp_taskwait_deps_51(%{{.+}}* @{{.+}}, i32 [[GTID]], i32 1, i8* %{{.}}, i32 0, i8* null, i32 1)
+
+
+#endif
Index: clang/test/OpenMP/taskwait_nowait_codegen.cpp
===================================================================
--- /dev/null
+++ clang/test/OpenMP/taskwait_nowait_codegen.cpp
@@ -0,0 +1,51 @@
+// RUN: %clang_cc1 -no-opaque-pointers -verify -triple x86_64-apple-darwin10 -fopenmp -fopenmp-version=51 -x c++ -emit-llvm %s -o - | FileCheck %s
+// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=51 -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s
+// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=51 -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s
+//
+// RUN: %clang_cc1 -no-opaque-pointers -verify -triple x86_64-apple-darwin10 -fopenmp -fopenmp-version=51 -fopenmp-enable-irbuilder -x c++ -emit-llvm %s -o - | FileCheck %s
+// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=51 -fopenmp-enable-irbuilder -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s
+// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=51 -fopenmp-enable-irbuilder -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s
+
+// RUN: %clang_cc1 -no-opaque-pointers -verify -triple x86_64-apple-darwin10 -fopenmp-simd -fopenmp-version=51 -x c++ -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=51 -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s
+// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=51 -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
+// expected-no-diagnostics
+#ifndef HEADER
+#define HEADER
+
+void foo() {}
+
+template <class T>
+T tmain(T argc) {
+  static T a;
+#pragma omp taskwait
+  return a + argc;
+}
+
+template <class T>
+T no_wait(T argc) {
+  static T a;
+#pragma omp taskwait nowait
+  return a + argc;
+}
+
+int main(int argc, char **argv) {
+#pragma omp taskwait
+  return tmain(argc)+no_wait(argc);
+}
+
+// CHECK-LABEL: @main
+// CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%{{.+}}* @{{.+}})
+// CHECK: call i32 @__kmpc_omp_taskwait_51(%{{.+}}* @{{.+}}, i32 [[GTID]], i32 0)
+
+// CHECK-LABEL: define {{.*}} @{{.*}}tmain{{.*}}
+// CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%{{.+}}* @{{.+}})
+// CHECK: call i32 @__kmpc_omp_taskwait_51(%{{.+}}* @{{.+}}, i32 [[GTID]], i32 0)
+//
+// CHECK-LABEL: define {{.*}} @{{.*}}no_wait{{.*}}
+// CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%{{.+}}* @{{.+}})
+// CHECK: call i32 @__kmpc_omp_taskwait_51(%{{.+}}* @{{.+}}, i32 [[GTID]], i32 1)
+
+
+#endif
Index: llvm/include/llvm/Frontend/OpenMP/OMP.td
===================================================================
--- llvm/include/llvm/Frontend/OpenMP/OMP.td
+++ llvm/include/llvm/Frontend/OpenMP/OMP.td
@@ -563,7 +563,8 @@
 }
 def OMP_TaskWait : Directive<"taskwait"> {
   let allowedClauses = [
-    VersionedClause<OMPC_Depend, 50>
+    VersionedClause<OMPC_Depend, 50>,
+    VersionedClause<OMPC_NoWait, 51>
   ];
 }
 def OMP_TaskGroup : Directive<"taskgroup"> {
Index: llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
===================================================================
--- llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
+++ llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
@@ -729,7 +729,8 @@
   /// Generator for '#omp taskwait'
   ///
   /// \param Loc The location where the taskwait directive was encountered.
-  void createTaskwait(const LocationDescription &Loc);
+  void createTaskwait(const LocationDescription &Loc,
+                      bool HasNowaitClause = false);
 
   /// Generator for '#omp taskyield'
   ///
@@ -1018,7 +1019,8 @@
   /// Generate a taskwait runtime call.
   ///
   /// \param Loc The location at which the request originated and is fulfilled.
-  void emitTaskwaitImpl(const LocationDescription &Loc);
+  void emitTaskwaitImpl(const LocationDescription &Loc,
+                        bool HasNowaitClause = false);
 
   /// Generate a taskyield runtime call.
   ///
Index: llvm/include/llvm/Frontend/OpenMP/OMPKinds.def
===================================================================
--- llvm/include/llvm/Frontend/OpenMP/OMPKinds.def
+++ llvm/include/llvm/Frontend/OpenMP/OMPKinds.def
@@ -203,6 +203,7 @@
 __OMP_RTL(__kmpc_get_hardware_thread_id_in_block, false, Int32, )
 __OMP_RTL(__kmpc_fork_call, true, Void, IdentPtr, Int32, ParallelTaskPtr)
 __OMP_RTL(__kmpc_omp_taskwait, false, Int32, IdentPtr, Int32)
+__OMP_RTL(__kmpc_omp_taskwait_51, false, Int32, IdentPtr, Int32, Int32)
 __OMP_RTL(__kmpc_omp_taskyield, false, Int32, IdentPtr, Int32, /* Int */ Int32)
 __OMP_RTL(__kmpc_push_num_threads, false, Void, IdentPtr, Int32,
           /* Int */ Int32)
@@ -370,6 +371,8 @@
 
 __OMP_RTL(__kmpc_omp_wait_deps, false, Void, IdentPtr, Int32, Int32,
           /* kmp_depend_info_t */ VoidPtr, Int32, VoidPtr)
+__OMP_RTL(__kmpc_omp_taskwait_deps_51, false, Void, IdentPtr, Int32, Int32,
+          /* kmp_depend_info_t */ VoidPtr, Int32, VoidPtr, Int32)
 __OMP_RTL(__kmpc_cancellationpoint, false, Int32, IdentPtr, Int32, Int32)
 
 __OMP_RTL(__kmpc_fork_teams, true, Void, IdentPtr, Int32, ParallelTaskPtr)
@@ -633,6 +636,8 @@
                 ParamAttrs(ReadOnlyPtrAttrs, AttributeSet(), ReadOnlyPtrAttrs))
 __OMP_RTL_ATTRS(__kmpc_omp_taskwait, BarrierAttrs, AttributeSet(),
                 ParamAttrs(ReadOnlyPtrAttrs))
+__OMP_RTL_ATTRS(__kmpc_omp_taskwait_51, BarrierAttrs, AttributeSet(),
+                ParamAttrs(ReadOnlyPtrAttrs))
 __OMP_RTL_ATTRS(__kmpc_omp_taskyield, InaccessibleArgOnlyAttrs, AttributeSet(),
                 ParamAttrs(ReadOnlyPtrAttrs))
 __OMP_RTL_ATTRS(__kmpc_push_num_threads, InaccessibleArgOnlyAttrs,
@@ -895,6 +900,9 @@
 __OMP_RTL_ATTRS(__kmpc_omp_wait_deps, BarrierAttrs, AttributeSet(),
                 ParamAttrs(ReadOnlyPtrAttrs, AttributeSet(), AttributeSet(),
                            ReadOnlyPtrAttrs))
+__OMP_RTL_ATTRS(__kmpc_omp_taskwait_deps_51, BarrierAttrs, AttributeSet(),
+                ParamAttrs(ReadOnlyPtrAttrs, AttributeSet(), AttributeSet(),
+                           ReadOnlyPtrAttrs))
 __OMP_RTL_ATTRS(__kmpc_cancellationpoint, DefaultAttrs, AttributeSet(),
                 ParamAttrs(ReadOnlyPtrAttrs))
 
Index: llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
===================================================================
--- llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
+++ llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
@@ -1250,23 +1250,26 @@
   emitFlush(Loc);
 }
 
-void OpenMPIRBuilder::emitTaskwaitImpl(const LocationDescription &Loc) {
+void OpenMPIRBuilder::emitTaskwaitImpl(const LocationDescription &Loc,
+                                       bool HasNowaitClause) {
   // Build call kmp_int32 __kmpc_omp_taskwait(ident_t *loc, kmp_int32
   // global_tid);
   uint32_t SrcLocStrSize;
   Constant *SrcLocStr = getOrCreateSrcLocStr(Loc, SrcLocStrSize);
   Value *Ident = getOrCreateIdent(SrcLocStr, SrcLocStrSize);
-  Value *Args[] = {Ident, getOrCreateThreadID(Ident)};
+  Value *NowaitClauseValue = ConstantInt::get(Int32, HasNowaitClause);
+  Value *Args[] = {Ident, getOrCreateThreadID(Ident), NowaitClauseValue};
 
   // Ignore return result until untied tasks are supported.
-  Builder.CreateCall(getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_omp_taskwait),
-                     Args);
+  Builder.CreateCall(
+      getOrCreateRuntimeFunctionPtr(OMPRTL___kmpc_omp_taskwait_51), Args);
 }
 
-void OpenMPIRBuilder::createTaskwait(const LocationDescription &Loc) {
+void OpenMPIRBuilder::createTaskwait(const LocationDescription &Loc,
+                                     bool HasNowaitClause) {
   if (!updateToLocation(Loc))
     return;
-  emitTaskwaitImpl(Loc);
+  emitTaskwaitImpl(Loc, HasNowaitClause);
 }
 
 void OpenMPIRBuilder::emitTaskyieldImpl(const LocationDescription &Loc) {
@@ -2389,7 +2392,7 @@
   case OMPScheduleType::BaseRuntimeSimd:
     assert(!ChunkSize &&
            "schedule type does not support user-defined chunk sizes");
-    [[fallthrough]];
+    LLVM_FALLTHROUGH;
   case OMPScheduleType::BaseDynamicChunked:
   case OMPScheduleType::BaseGuidedChunked:
   case OMPScheduleType::BaseGuidedIterativeChunked:
Index: mlir/test/Target/LLVMIR/openmp-llvm.mlir
===================================================================
--- mlir/test/Target/LLVMIR/openmp-llvm.mlir
+++ mlir/test/Target/LLVMIR/openmp-llvm.mlir
@@ -7,7 +7,7 @@
   omp.barrier
 
   // CHECK: [[OMP_THREAD1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @{{[0-9]+}})
-  // CHECK-NEXT:  [[RET_VAL:%.*]] = call i32 @__kmpc_omp_taskwait(ptr @{{[0-9]+}}, i32 [[OMP_THREAD1]])
+  // CHECK-NEXT:  [[RET_VAL:%.*]] = call i32 @__kmpc_omp_taskwait_51(ptr @{{[0-9]+}}, i32 [[OMP_THREAD1]], i32 0)
   omp.taskwait
 
   // CHECK: [[OMP_THREAD2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @{{[0-9]+}})
Index: openmp/runtime/src/dllexports
===================================================================
--- openmp/runtime/src/dllexports
+++ openmp/runtime/src/dllexports
@@ -282,6 +282,7 @@
     __kmpc_omp_task_alloc                   191
     __kmpc_omp_task                         192
     __kmpc_omp_taskwait                     193
+    __kmpc_omp_taskwait_51                  293
     __kmpc_omp_task_begin_if0               196
     __kmpc_omp_task_complete_if0            197
     __kmpc_omp_task_parts                   198
@@ -345,6 +346,7 @@
         __kmpc_fork_teams                   241
         __kmpc_omp_task_with_deps           242
         __kmpc_omp_wait_deps                243
+        __kmpc_omp_taskwait_deps_51         292
         __kmpc_cancel                       244
         __kmpc_cancellationpoint            245
         __kmpc_cancel_barrier               246
Index: openmp/runtime/src/kmp.h
===================================================================
--- openmp/runtime/src/kmp.h
+++ openmp/runtime/src/kmp.h
@@ -3973,6 +3973,10 @@
 KMP_EXPORT kmp_int32 __kmpc_omp_task_parts(ident_t *loc_ref, kmp_int32 gtid,
                                            kmp_task_t *new_task);
 KMP_EXPORT kmp_int32 __kmpc_omp_taskwait(ident_t *loc_ref, kmp_int32 gtid);
+/* __kmpc_omp_taskwait_51 : Function for OpenMP 5.1 nowait clause.
+ *                          Placeholder for taskwait with nowait clause.*/
+KMP_EXPORT kmp_int32 __kmpc_omp_taskwait_51(ident_t *loc_ref, kmp_int32 gtid,
+                                            kmp_int32 has_no_wait);
 
 KMP_EXPORT kmp_int32 __kmpc_omp_taskyield(ident_t *loc_ref, kmp_int32 gtid,
                                           int end_part);
@@ -3997,6 +4001,15 @@
                                      kmp_depend_info_t *dep_list,
                                      kmp_int32 ndeps_noalias,
                                      kmp_depend_info_t *noalias_dep_list);
+/* __kmpc_omp_taskwait_deps_51 : Function for OpenMP 5.1 nowait clause.
+ *                               Placeholder for taskwait with nowait clause.*/
+KMP_EXPORT void __kmpc_omp_taskwait_deps_51(ident_t *loc_ref, kmp_int32 gtid,
+                                            kmp_int32 ndeps,
+                                            kmp_depend_info_t *dep_list,
+                                            kmp_int32 ndeps_noalias,
+                                            kmp_depend_info_t *noalias_dep_list,
+                                            kmp_int32 has_no_wait);
+
 extern kmp_int32 __kmp_omp_task(kmp_int32 gtid, kmp_task_t *new_task,
                                 bool serialize_immediate);
 
Index: openmp/runtime/src/kmp_taskdeps.cpp
===================================================================
--- openmp/runtime/src/kmp_taskdeps.cpp
+++ openmp/runtime/src/kmp_taskdeps.cpp
@@ -744,10 +744,24 @@
 void __kmpc_omp_wait_deps(ident_t *loc_ref, kmp_int32 gtid, kmp_int32 ndeps,
                           kmp_depend_info_t *dep_list, kmp_int32 ndeps_noalias,
                           kmp_depend_info_t *noalias_dep_list) {
-  KA_TRACE(10, ("__kmpc_omp_wait_deps(enter): T#%d loc=%p\n", gtid, loc_ref));
+  __kmpc_omp_taskwait_deps_51(loc_ref, gtid, ndeps, dep_list, ndeps_noalias,
+                              noalias_dep_list, false);
+}
 
+/* __kmpc_omp_taskwait_deps_51 : Function for OpenMP 5.1 nowait clause.
+                                 Placeholder for taskwait with nowait clause.
+                                 Earlier code of __kmpc_omp_wait_deps() is now
+                                 in this function.
+*/
+void __kmpc_omp_taskwait_deps_51(ident_t *loc_ref, kmp_int32 gtid,
+                                 kmp_int32 ndeps, kmp_depend_info_t *dep_list,
+                                 kmp_int32 ndeps_noalias,
+                                 kmp_depend_info_t *noalias_dep_list,
+                                 kmp_int32 has_no_wait) {
+  KA_TRACE(10, ("__kmpc_omp_taskwait_deps(enter): T#%d loc=%p nowait#%d\n",
+                gtid, loc_ref, has_no_wait));
   if (ndeps == 0 && ndeps_noalias == 0) {
-    KA_TRACE(10, ("__kmpc_omp_wait_deps(exit): T#%d has no dependences to "
+    KA_TRACE(10, ("__kmpc_omp_taskwait_deps(exit): T#%d has no dependences to "
                   "wait upon : loc=%p\n",
                   gtid, loc_ref));
     return;
@@ -839,7 +853,7 @@
   ignore = ignore || current_task->td_dephash == NULL;
 
   if (ignore) {
-    KA_TRACE(10, ("__kmpc_omp_wait_deps(exit): T#%d has no blocking "
+    KA_TRACE(10, ("__kmpc_omp_taskwait_deps(exit): T#%d has no blocking "
                   "dependences : loc=%p\n",
                   gtid, loc_ref));
 #if OMPT_SUPPORT
@@ -854,7 +868,7 @@
   if (!__kmp_check_deps(gtid, &node, NULL, &current_task->td_dephash,
                         DEP_BARRIER, ndeps, dep_list, ndeps_noalias,
                         noalias_dep_list)) {
-    KA_TRACE(10, ("__kmpc_omp_wait_deps(exit): T#%d has no blocking "
+    KA_TRACE(10, ("__kmpc_omp_taskwait_deps(exit): T#%d has no blocking "
                   "dependences : loc=%p\n",
                   gtid, loc_ref));
 #if OMPT_SUPPORT
@@ -875,6 +889,7 @@
 #if OMPT_SUPPORT
   __ompt_taskwait_dep_finish(current_task, taskwait_task_data);
 #endif /* OMPT_SUPPORT */
-  KA_TRACE(10, ("__kmpc_omp_wait_deps(exit): T#%d finished waiting : loc=%p\n",
+  KA_TRACE(10, ("__kmpc_omp_taskwait_deps(exit): T#%d finished waiting : loc=%p\
+                \n",
                 gtid, loc_ref));
 }
Index: openmp/runtime/src/kmp_tasking.cpp
===================================================================
--- openmp/runtime/src/kmp_tasking.cpp
+++ openmp/runtime/src/kmp_tasking.cpp
@@ -2049,7 +2049,8 @@
 template <bool ompt>
 static kmp_int32 __kmpc_omp_taskwait_template(ident_t *loc_ref, kmp_int32 gtid,
                                               void *frame_address,
-                                              void *return_address) {
+                                              void *return_address,
+                                              kmp_int32 has_no_wait) {
   kmp_taskdata_t *taskdata = nullptr;
   kmp_info_t *thread;
   int thread_finished = FALSE;
@@ -2162,23 +2163,34 @@
 OMPT_NOINLINE
 static kmp_int32 __kmpc_omp_taskwait_ompt(ident_t *loc_ref, kmp_int32 gtid,
                                           void *frame_address,
-                                          void *return_address) {
+                                          void *return_address,
+                                          kmp_int32 has_no_wait) {
   return __kmpc_omp_taskwait_template<true>(loc_ref, gtid, frame_address,
-                                            return_address);
+                                            return_address, has_no_wait);
 }
 #endif // OMPT_SUPPORT && OMPT_OPTIONAL
 
 // __kmpc_omp_taskwait: Wait until all tasks generated by the current task are
 // complete
 kmp_int32 __kmpc_omp_taskwait(ident_t *loc_ref, kmp_int32 gtid) {
+  return __kmpc_omp_taskwait_51(loc_ref, gtid, false);
+}
+
+/* __kmpc_omp_taskwait_51 : Function for OpenMP 5.1 nowait clause.
+ *                          Placeholder for taskwait with nowait clause.
+ *                          The code is a copy of __kmpc_omp_taskwait.*/
+kmp_int32 __kmpc_omp_taskwait_51(ident_t *loc_ref, kmp_int32 gtid,
+                                 kmp_int32 has_no_wait) {
 #if OMPT_SUPPORT && OMPT_OPTIONAL
   if (UNLIKELY(ompt_enabled.enabled)) {
     OMPT_STORE_RETURN_ADDRESS(gtid);
     return __kmpc_omp_taskwait_ompt(loc_ref, gtid, OMPT_GET_FRAME_ADDRESS(0),
-                                    OMPT_LOAD_RETURN_ADDRESS(gtid));
+                                    OMPT_LOAD_RETURN_ADDRESS(gtid),
+                                    has_no_wait);
   }
 #endif
-  return __kmpc_omp_taskwait_template<false>(loc_ref, gtid, NULL, NULL);
+  return __kmpc_omp_taskwait_template<false>(loc_ref, gtid, NULL, NULL,
+                                             has_no_wait);
 }
 
 // __kmpc_omp_taskyield: switch to a different task