diff --git a/clang/lib/CodeGen/CGOpenMPRuntime.cpp b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
index 50ee50b7a167..dd07d0f34339 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntime.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
@@ -2653,8 +2653,10 @@ enum KmpTaskTFields {
   KmpTaskTRoutine,
   /// \brief Partition id for the untied tasks.
   KmpTaskTPartId,
-  /// \brief Function with call of destructors for private variables.
-  KmpTaskTDestructors,
+  /// Function with call of destructors for private variables.
+  Data1,
+  /// Task priority.
+  Data2,
   /// (Taskloops only) Lower bound.
   KmpTaskTLowerBound,
   /// (Taskloops only) Upper bound.
@@ -3178,19 +3180,27 @@ createKmpTaskTRecordDecl(CodeGenModule &CGM, OpenMPDirectiveKind Kind,
   //         void *              shareds;
   //         kmp_routine_entry_t routine;
   //         kmp_int32           part_id;
-  //         kmp_routine_entry_t destructors;
+  //         kmp_cmplrdata_t data1;
+  //         kmp_cmplrdata_t data2;
   // For taskloops additional fields:
   //         kmp_uint64          lb;
   //         kmp_uint64          ub;
   //         kmp_int64           st;
   //         kmp_int32           liter;
   //       };
+  auto *UD = C.buildImplicitRecord("kmp_cmplrdata_t", TTK_Union);
+  UD->startDefinition();
+  addFieldToRecordDecl(C, UD, KmpInt32Ty);
+  addFieldToRecordDecl(C, UD, KmpRoutineEntryPointerQTy);
+  UD->completeDefinition();
+  QualType KmpCmplrdataTy = C.getRecordType(UD);
   auto *RD = C.buildImplicitRecord("kmp_task_t");
   RD->startDefinition();
   addFieldToRecordDecl(C, RD, C.VoidPtrTy);
   addFieldToRecordDecl(C, RD, KmpRoutineEntryPointerQTy);
   addFieldToRecordDecl(C, RD, KmpInt32Ty);
-  addFieldToRecordDecl(C, RD, KmpRoutineEntryPointerQTy);
+  addFieldToRecordDecl(C, RD, KmpCmplrdataTy);
+  addFieldToRecordDecl(C, RD, KmpCmplrdataTy);
   if (isOpenMPTaskLoopDirective(Kind)) {
     QualType KmpUInt64Ty =
         CGM.getContext().getIntTypeForBitwidth(/*DestWidth=*/64, /*Signed=*/0);
@@ -3805,18 +3815,30 @@ CGOpenMPRuntime::emitTaskInit(CodeGenFunction &CGF, SourceLocation Loc,
           /*WithLastIter=*/!Data.LastprivateVars.empty());
     }
   }
+  // Fields of union "kmp_cmplrdata_t" for destructors and priority.
+  enum { Priority = 0, Destructors = 1 };
   // Provide pointer to function with destructors for privates.
-  llvm::Value *DestructorFn =
-      NeedsCleanup ? emitDestructorsFunction(CGM, Loc, KmpInt32Ty,
-                                             KmpTaskTWithPrivatesPtrQTy,
-                                             KmpTaskTWithPrivatesQTy)
-                   : llvm::ConstantPointerNull::get(
-                         cast<llvm::PointerType>(KmpRoutineEntryPtrTy));
-  LValue Destructor = CGF.EmitLValueForField(
-      TDBase, *std::next(KmpTaskTQTyRD->field_begin(), KmpTaskTDestructors));
-  CGF.EmitStoreOfScalar(CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
-                            DestructorFn, KmpRoutineEntryPtrTy),
-                        Destructor);
+  auto FI = std::next(KmpTaskTQTyRD->field_begin(), Data1);
+  auto *KmpCmplrdataUD = (*FI)->getType()->getAsUnionType()->getDecl();
+  if (NeedsCleanup) {
+    llvm::Value *DestructorFn = emitDestructorsFunction(
+        CGM, Loc, KmpInt32Ty, KmpTaskTWithPrivatesPtrQTy,
+        KmpTaskTWithPrivatesQTy);
+    LValue Data1LV = CGF.EmitLValueForField(TDBase, *FI);
+    LValue DestructorsLV = CGF.EmitLValueForField(
+        Data1LV, *std::next(KmpCmplrdataUD->field_begin(), Destructors));
+    CGF.EmitStoreOfScalar(CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+                              DestructorFn, KmpRoutineEntryPtrTy),
+                          DestructorsLV);
+  }
+  // Set priority.
+  if (Data.Priority.getInt()) {
+    LValue Data2LV = CGF.EmitLValueForField(
+        TDBase, *std::next(KmpTaskTQTyRD->field_begin(), Data2));
+    LValue PriorityLV = CGF.EmitLValueForField(
+        Data2LV, *std::next(KmpCmplrdataUD->field_begin(), Priority));
+    CGF.EmitStoreOfScalar(Data.Priority.getPointer(), PriorityLV);
+  }
   Result.NewTask = NewTask;
   Result.TaskEntry = TaskEntry;
   Result.NewTaskNewTaskTTy = NewTaskNewTaskTTy;
diff --git a/clang/lib/CodeGen/CGStmtOpenMP.cpp b/clang/lib/CodeGen/CGStmtOpenMP.cpp
index 2f4a2c7f7b69..89d47e567c20 100644
--- a/clang/lib/CodeGen/CGStmtOpenMP.cpp
+++ b/clang/lib/CodeGen/CGStmtOpenMP.cpp
@@ -2371,6 +2371,10 @@ void CodeGenFunction::EmitOMPTaskBasedDirective(const OMPExecutableDirective &S,
     // TODO: Add codegen for priority clause arg when runtime lib support it.
     auto *Prio = Clause->getPriority();
     Data.Priority.setInt(Prio);
+    Data.Priority.setPointer(EmitScalarConversion(
+        EmitScalarExpr(Prio), Prio->getType(),
+        getContext().getIntTypeForBitwidth(/*DestWidth=*/32, /*Signed=*/1),
+        Prio->getExprLoc()));
   }
   // The first function argument for tasks is a thread id, the second one is a
   // part id (0 for tied tasks, >=0 for untied task).
diff --git a/clang/test/OpenMP/task_codegen.cpp b/clang/test/OpenMP/task_codegen.cpp
index d020e4268a4b..08c9ce33d8e8 100644
--- a/clang/test/OpenMP/task_codegen.cpp
+++ b/clang/test/OpenMP/task_codegen.cpp
@@ -9,7 +9,7 @@
 // CHECK-DAG: [[IDENT_T:%.+]] = type { i32, i32, i32, i32, i8* }
 // CHECK-DAG: [[STRUCT_SHAREDS:%.+]] = type { i8*, [2 x [[STRUCT_S:%.+]]]* }
 // CHECK-DAG: [[STRUCT_SHAREDS1:%.+]] = type { [2 x [[STRUCT_S:%.+]]]* }
-// CHECK-DAG: [[KMP_TASK_T:%.+]] = type { i8*, i32 (i32, i8*)*, i32, i32 (i32, i8*)* }
+// CHECK-DAG: [[KMP_TASK_T:%.+]] = type { i8*, i32 (i32, i8*)*, i32, %union{{.+}}, %union{{.+}} }
 // CHECK-DAG: [[KMP_DEPEND_INFO:%.+]] = type { i64, i64, i8 }
 struct S {
   int a;
@@ -30,13 +30,14 @@ int main() {
 // CHECK: store i8* [[B]], i8** [[B_REF]]
 // CHECK: [[S_REF:%.+]] = getelementptr inbounds [[STRUCT_SHAREDS]], [[STRUCT_SHAREDS]]* [[CAPTURES]], i32 0, i32 1
 // CHECK: store [2 x [[STRUCT_S]]]* [[S]], [2 x [[STRUCT_S]]]** [[S_REF]]
-// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i32 33, i64 32, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_T]]{{.*}}*)* [[TASK_ENTRY1:@.+]] to i32 (i32, i8*)*))
+// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i32 33, i64 40, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_T]]{{.*}}*)* [[TASK_ENTRY1:@.+]] to i32 (i32, i8*)*))
 // CHECK: [[SHAREDS_REF_PTR:%.+]] = getelementptr inbounds [[KMP_TASK_T]], [[KMP_TASK_T]]* [[TASK_PTR:%.+]], i32 0, i32 0
 // CHECK: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS_REF_PTR]]
 // CHECK: [[BITCAST:%.+]] = bitcast [[STRUCT_SHAREDS]]* [[CAPTURES]] to i8*
 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[SHAREDS_REF]], i8* [[BITCAST]], i64 16, i32 8, i1 false)
-// CHECK: [[DESTRUCTORS_REF_PTR:%.+]] = getelementptr inbounds [[KMP_TASK_T]], [[KMP_TASK_T]]* [[TASK_PTR]], i32 0, i32 3
-// CHECK: store i32 (i32, i8*)* null, i32 (i32, i8*)** [[DESTRUCTORS_REF_PTR]]
+// CHECK: [[PRIORITY_REF_PTR:%.+]] = getelementptr inbounds [[KMP_TASK_T]], [[KMP_TASK_T]]* [[TASK_PTR]], i32 0, i32 4
+// CHECK: [[PRIORITY:%.+]] = bitcast %union{{.+}}* [[PRIORITY_REF_PTR]] to i32*
+// CHECK: store i32 {{.+}}, i32* [[PRIORITY]]
 // CHECK: call i32 @__kmpc_omp_task([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i8* [[ORIG_TASK_PTR]])
 #pragma omp task shared(a, b, s) priority(b)
   {
@@ -46,13 +47,11 @@ int main() {
   }
 // CHECK: [[S_REF:%.+]] = getelementptr inbounds [[STRUCT_SHAREDS1]], [[STRUCT_SHAREDS1]]* [[CAPTURES:%.+]], i32 0, i32 0
 // CHECK: store [2 x [[STRUCT_S]]]* [[S]], [2 x [[STRUCT_S]]]** [[S_REF]]
-// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{[^,]+}}, i32 [[GTID]], i32 1, i64 32, i64 8,
+// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{[^,]+}}, i32 [[GTID]], i32 1, i64 40, i64 8,
 // CHECK: [[SHAREDS_REF_PTR:%.+]] = getelementptr inbounds [[KMP_TASK_T]], [[KMP_TASK_T]]* [[TASK_PTR:%.+]], i32 0, i32 0
 // CHECK: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS_REF_PTR]]
 // CHECK: [[BITCAST:%.+]] = bitcast [[STRUCT_SHAREDS1]]* [[CAPTURES]] to i8*
 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[SHAREDS_REF]], i8* [[BITCAST]], i64 8, i32 8, i1 false)
-// CHECK: [[DESTRUCTORS_REF_PTR:%.+]] = getelementptr inbounds [[KMP_TASK_T]], [[KMP_TASK_T]]* [[TASK_PTR]], i32 0, i32 3
-// CHECK: store i32 (i32, i8*)* null, i32 (i32, i8*)** [[DESTRUCTORS_REF_PTR]]
 // CHECK: [[DEP:%.*]] = getelementptr inbounds [4 x [[KMP_DEPEND_INFO]]], [4 x [[KMP_DEPEND_INFO]]]* [[DEPENDENCIES:%.*]], i64 0, i64 0
 // CHECK: [[T0:%.*]] = getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* [[DEP]], i32 0, i32 0
 // CHECK: store i64 ptrtoint (i32* @{{.+}} to i64), i64* [[T0]]
@@ -100,17 +99,13 @@ int main() {
     a = 15;
     s[1].a = 10;
   }
-// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i32 0, i64 32, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_T]]{{.*}}*)* [[TASK_ENTRY2:@.+]] to i32 (i32, i8*)*))
-// CHECK: [[DESTRUCTORS_REF_PTR:%.+]] = getelementptr inbounds [[KMP_TASK_T]]{{.*}}* {{%.+}}, i32 0, i32 3
-// CHECK: store i32 (i32, i8*)* null, i32 (i32, i8*)** [[DESTRUCTORS_REF_PTR]]
+// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i32 0, i64 40, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_T]]{{.*}}*)* [[TASK_ENTRY2:@.+]] to i32 (i32, i8*)*))
 // CHECK: call i32 @__kmpc_omp_task([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i8* [[ORIG_TASK_PTR]])
 #pragma omp task untied
   {
     a = 1;
   }
-// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i32 0, i64 32, i64 1,
-// CHECK: [[DESTRUCTORS_REF_PTR:%.+]] = getelementptr inbounds [[KMP_TASK_T]]{{.*}}* {{%.+}}, i32 0, i32 3
-// CHECK: store i32 (i32, i8*)* null, i32 (i32, i8*)** [[DESTRUCTORS_REF_PTR]]
+// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i32 0, i64 40, i64 1,
 // CHECK: getelementptr inbounds [2 x [[STRUCT_S]]], [2 x [[STRUCT_S]]]* [[S]], i64 0, i64 0
 // CHECK: getelementptr inbounds [2 x [[KMP_DEPEND_INFO]]], [2 x [[KMP_DEPEND_INFO]]]* %{{[^,]+}}, i64 0, i64 0
 // CHECK: getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i32 0, i32 0
@@ -149,9 +144,7 @@ int main() {
   {
     a = 1;
   }
-// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i32 3, i64 32, i64 1,
-// CHECK: [[DESTRUCTORS_REF_PTR:%.+]] = getelementptr inbounds [[KMP_TASK_T]]{{.*}}* {{%.+}}, i32 0, i32 3
-// CHECK: store i32 (i32, i8*)* null, i32 (i32, i8*)** [[DESTRUCTORS_REF_PTR]]
+// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i32 3, i64 40, i64 1,
 // CHECK: getelementptr inbounds [3 x [[KMP_DEPEND_INFO]]], [3 x [[KMP_DEPEND_INFO]]]* %{{[^,]+}}, i64 0, i64 0
 // CHECK: getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i32 0, i32 0
 // CHECK: store i64 ptrtoint (i32* @{{.+}} to i64), i64*
@@ -199,17 +192,13 @@ int main() {
   {
     a = 2;
   }
-// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i32 3, i64 32, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_T]]{{.*}}*)* [[TASK_ENTRY3:@.+]] to i32 (i32, i8*)*))
-// CHECK: [[DESTRUCTORS_REF_PTR:%.+]] = getelementptr inbounds [[KMP_TASK_T]]{{.*}}* {{%.+}}, i32 0, i32 3
-// CHECK: store i32 (i32, i8*)* null, i32 (i32, i8*)** [[DESTRUCTORS_REF_PTR]]
+// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i32 3, i64 40, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_T]]{{.*}}*)* [[TASK_ENTRY3:@.+]] to i32 (i32, i8*)*))
 // CHECK: call i32 @__kmpc_omp_task([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i8* [[ORIG_TASK_PTR]])
 #pragma omp task final(true)
   {
     a = 2;
   }
-// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i32 1, i64 32, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_T]]{{.*}}*)* [[TASK_ENTRY4:@.+]] to i32 (i32, i8*)*))
-// CHECK: [[DESTRUCTORS_REF_PTR:%.+]] = getelementptr inbounds [[KMP_TASK_T]]{{.*}}* {{%.*}}, i32 0, i32 3
-// CHECK: store i32 (i32, i8*)* null, i32 (i32, i8*)** [[DESTRUCTORS_REF_PTR]]
+// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i32 1, i64 40, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_T]]{{.*}}*)* [[TASK_ENTRY4:@.+]] to i32 (i32, i8*)*))
 // CHECK: call i32 @__kmpc_omp_task([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i8* [[ORIG_TASK_PTR]])
   const bool flag = false;
 #pragma omp task final(flag)
@@ -220,9 +209,7 @@ int main() {
 // CHECK: [[CMP:%.+]] = icmp ne i8 [[B_VAL]], 0
 // CHECK: [[FINAL:%.+]] = select i1 [[CMP]], i32 2, i32 0
 // CHECK: [[FLAGS:%.+]] = or i32 [[FINAL]], 1
-// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i32 [[FLAGS]], i64 32, i64 8, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_T]]{{.*}}*)* [[TASK_ENTRY5:@.+]] to i32 (i32, i8*)*))
-// CHECK: [[DESTRUCTORS_REF_PTR:%.+]] = getelementptr inbounds [[KMP_TASK_T]]{{.*}}* {{%.+}}, i32 0, i32 3
-// CHECK: store i32 (i32, i8*)* null, i32 (i32, i8*)** [[DESTRUCTORS_REF_PTR]]
+// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i32 [[FLAGS]], i64 40, i64 8, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_T]]{{.*}}*)* [[TASK_ENTRY5:@.+]] to i32 (i32, i8*)*))
 // CHECK: call i32 @__kmpc_omp_task([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i8* [[ORIG_TASK_PTR]])
   int c __attribute__((aligned(128)));
 #pragma omp task final(b) shared(c)
@@ -230,9 +217,7 @@ int main() {
     a = 4;
     c = 5;
   }
-// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i32 0, i64 32, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_T]]{{.*}}*)* [[TASK_ENTRY6:@.+]] to i32 (i32, i8*)*))
-// CHECK: [[DESTRUCTORS_REF_PTR:%.+]] = getelementptr inbounds [[KMP_TASK_T]]{{.*}}* {{%.+}}, i32 0, i32 3
-// CHECK: store i32 (i32, i8*)* null, i32 (i32, i8*)** [[DESTRUCTORS_REF_PTR]]
+// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i32 0, i64 40, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_T]]{{.*}}*)* [[TASK_ENTRY6:@.+]] to i32 (i32, i8*)*))
 // CHECK: call i32 @__kmpc_omp_task([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i8* [[ORIG_TASK_PTR]])
 #pragma omp task untied
   {
@@ -275,7 +260,6 @@ int main() {
 // CHECK: call i32 @__kmpc_omp_task(%
 
 // CHECK: call i8* @__kmpc_omp_task_alloc(
-// CHECK: store i32 (i32, i8*)* null, i32 (i32, i8*)** %
 // CHECK: call i32 @__kmpc_omp_task(%
 // CHECK: load i32*, i32** %
 // CHECK: store i32 2, i32* %
diff --git a/clang/test/OpenMP/task_firstprivate_codegen.cpp b/clang/test/OpenMP/task_firstprivate_codegen.cpp
index 4419263c3203..0d8e1c4afe4e 100644
--- a/clang/test/OpenMP/task_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/task_firstprivate_codegen.cpp
@@ -24,7 +24,7 @@ struct S {
 
 volatile double g;
 
-// CHECK-DAG: [[KMP_TASK_T_TY:%.+]] = type { i8*, i32 (i32, i8*)*, i32, i32 (i32, i8*)* }
+// CHECK-DAG: [[KMP_TASK_T_TY:%.+]] = type { i8*, i32 (i32, i8*)*, i32, %union{{.+}}, %union{{.+}} }
 // CHECK-DAG: [[S_DOUBLE_TY:%.+]] = type { double }
 // CHECK-DAG: [[PRIVATES_MAIN_TY:%.+]] = type {{.?}}{ [2 x [[S_DOUBLE_TY]]], [[S_DOUBLE_TY]], i32, [2 x i32]
 // CHECK-DAG: [[CAP_MAIN_TY:%.+]] = type {{.*}}{ [2 x i32]*, i32, {{.*}}[2 x [[S_DOUBLE_TY]]]*, [[S_DOUBLE_TY]]*, i{{[0-9]+}}
@@ -58,7 +58,7 @@ int main() {
   // LAMBDA: call{{( x86_thiscallcc)?}} void [[OUTER_LAMBDA:@.+]](
   [&]() {
   // LAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]](
-  // LAMBDA: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 48, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
+  // LAMBDA: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 56, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
 // LAMBDA: [[PRIVATES:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
 // LAMBDA: [[G_PRIVATE_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[PRIVATES]], i{{.+}} 0, i{{.+}} 0
 // LAMBDA: [[G_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 0
@@ -102,7 +102,7 @@ int main() {
   // BLOCKS: call void {{%.+}}(i8
   ^{
   // BLOCKS: define{{.*}} internal{{.*}} void {{.+}}(i8*
-  // BLOCKS: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 48, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
+  // BLOCKS: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 56, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
   // BLOCKS: [[PRIVATES:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
   // BLOCKS: [[G_PRIVATE_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[PRIVATES]], i{{.+}} 0, i{{.+}} 0
   // BLOCKS: [[G_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 0
@@ -191,7 +191,7 @@ int main() {
 //         [[KMP_TASK_T]] task_data;
 //         [[KMP_TASK_MAIN_TY]] privates;
 //       };
-// CHECK: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc([[LOC]], i32 [[GTID]], i32 9, i64 72, i64 40, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
+// CHECK: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc([[LOC]], i32 [[GTID]], i32 9, i64 80, i64 40, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
 // CHECK: [[RES_KMP_TASK:%.+]] = bitcast i8* [[RES]] to [[KMP_TASK_MAIN_TY]]*
 
 // Fill kmp_task_t->shareds by copying from original capture argument.
@@ -242,7 +242,8 @@ int main() {
 
 // Provide pointer to destructor function, which will destroy private variables at the end of the task.
 // CHECK: [[DESTRUCTORS_REF:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{.+}} 0, i{{.+}} 3
-// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_REF]],
+// CHECK: [[DESTRUCTORS_PTR:%.+]] = bitcast %union{{.+}}* [[DESTRUCTORS_REF]] to i32 (i32, i8*)**
+// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_PTR]],
 
 // Start task.
 // CHECK: call i32 @__kmpc_omp_task([[LOC]], i32 [[GTID]], i8* [[RES]])
@@ -387,7 +388,8 @@ int main() {
 
 // Provide pointer to destructor function, which will destroy private variables at the end of the task.
 // CHECK: [[DESTRUCTORS_REF:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{.+}} 0, i{{.+}} 3
-// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_TMAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_REF]],
+// CHECK: [[DESTRUCTORS_PTR:%.+]] = bitcast %union{{.+}}* [[DESTRUCTORS_REF]] to i32 (i32, i8*)**
+// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_TMAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_PTR]],
 
 // Start task.
 // CHECK: call i32 @__kmpc_omp_task([[LOC]], i32 [[GTID]], i8* [[RES]])
diff --git a/clang/test/OpenMP/task_if_codegen.cpp b/clang/test/OpenMP/task_if_codegen.cpp
index 6b97febb9074..4226dceda86e 100644
--- a/clang/test/OpenMP/task_if_codegen.cpp
+++ b/clang/test/OpenMP/task_if_codegen.cpp
@@ -63,11 +63,11 @@ int tmain(T Arg) {
 // CHECK-LABEL: @main
 int main() {
 // CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(
-// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc({{[^,]+}}, i32 [[GTID]], i32 1, i64 32, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[CAP_FN7:[^ ]+]] to i32 (i32, i8*)*))
+// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc({{[^,]+}}, i32 [[GTID]], i32 1, i64 40, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[CAP_FN7:[^ ]+]] to i32 (i32, i8*)*))
 // CHECK: call i32 @__kmpc_omp_task(%{{.+}}* @{{.+}}, i32 [[GTID]], i8* [[ORIG_TASK_PTR]])
 #pragma omp task if (true)
   fn7();
-// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc({{[^,]+}}, i32 [[GTID]], i32 1, i64 32, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[CAP_FN8:[^ ]+]] to i32 (i32, i8*)*))
+// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc({{[^,]+}}, i32 [[GTID]], i32 1, i64 40, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[CAP_FN8:[^ ]+]] to i32 (i32, i8*)*))
 // CHECK: [[TASK_PTR:%.+]] = bitcast i8* [[ORIG_TASK_PTR]] to
 // CHECK: call void @__kmpc_omp_task_begin_if0(%{{.+}}* @{{.+}}, i{{.+}} [[GTID]], i8* [[ORIG_TASK_PTR]])
 // CHECK: call i32 [[CAP_FN8]](i32 [[GTID]], %{{.+}}* [[TASK_PTR]])
@@ -75,7 +75,7 @@ int main() {
 #pragma omp task if (false)
   fn8();
 
-// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc({{[^,]+}}, i32 [[GTID]], i32 1, i64 32, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[CAP_FN9:[^ ]+]] to i32 (i32, i8*)*))
+// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc({{[^,]+}}, i32 [[GTID]], i32 1, i64 40, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[CAP_FN9:[^ ]+]] to i32 (i32, i8*)*))
 // CHECK: [[TASK_PTR:%.+]] = bitcast i8* [[ORIG_TASK_PTR]] to
 // CHECK: br i1 %{{.+}}, label %[[OMP_THEN:.+]], label %[[OMP_ELSE:.+]]
 // CHECK: [[OMP_THEN]]
@@ -89,7 +89,7 @@ int main() {
 // CHECK: [[OMP_END]]
 #pragma omp task if (Arg)
   fn9();
-// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc({{[^,]+}}, i32 [[GTID]], i32 1, i64 32, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[CAP_FN10:[^ ]+]] to i32 (i32, i8*)*))
+// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc({{[^,]+}}, i32 [[GTID]], i32 1, i64 40, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[CAP_FN10:[^ ]+]] to i32 (i32, i8*)*))
 // CHECK: [[TASK_PTR:%.+]] = bitcast i8* [[ORIG_TASK_PTR]] to
 // CHECK: br i1 %{{.+}}, label %[[OMP_THEN:.+]], label %[[OMP_ELSE:.+]]
 // CHECK: [[OMP_THEN]]
@@ -126,7 +126,7 @@ int main() {
 
 // CHECK-LABEL: define {{.+}} @{{.+}}tmain
 // CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(
-// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^,]+}}, i32 [[GTID]], i32 1, i64 32, i64 1, i32 (i32, i8*)* bitcast (i32 (i32,  %{{[^*]+}}*)* [[CAP_FN1:[^ ]+]] to i32 (i32, i8*)*))
+// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^,]+}}, i32 [[GTID]], i32 1, i64 40, i64 1, i32 (i32, i8*)* bitcast (i32 (i32,  %{{[^*]+}}*)* [[CAP_FN1:[^ ]+]] to i32 (i32, i8*)*))
 // CHECK: call i32 @__kmpc_omp_task(%{{.+}}* @{{.+}}, i32 [[GTID]], i8* [[ORIG_TASK_PTR]])
 
 // CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc(
@@ -135,7 +135,7 @@ int main() {
 // CHECK: call i32 [[CAP_FN2:@.+]](i32 [[GTID]], %{{.+}}* [[TASK_PTR]])
 // CHECK: call void @__kmpc_omp_task_complete_if0(%{{.+}}* @{{.+}}, i{{.+}} [[GTID]], i8* [[ORIG_TASK_PTR]])
 
-// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^,]+}}, i32 [[GTID]], i32 1, i64 32, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[CAP_FN3:[^ ]+]] to i32 (i32, i8*)*))
+// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^,]+}}, i32 [[GTID]], i32 1, i64 40, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[CAP_FN3:[^ ]+]] to i32 (i32, i8*)*))
 // CHECK: [[TASK_PTR:%.+]] = bitcast i8* [[ORIG_TASK_PTR]] to
 // CHECK: br i1 %{{.+}}, label %[[OMP_THEN:.+]], label %[[OMP_ELSE:.+]]
 // CHECK: [[OMP_THEN]]
@@ -148,7 +148,7 @@ int main() {
 // CHECK: br label %[[OMP_END]]
 // CHECK: [[OMP_END]]
 
-// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^,]+}}, i32 [[GTID]], i32 1, i64 32, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[CAP_FN4:[^ ]+]] to i32 (i32, i8*)*))
+// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^,]+}}, i32 [[GTID]], i32 1, i64 40, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[CAP_FN4:[^ ]+]] to i32 (i32, i8*)*))
 // CHECK: [[TASK_PTR:%.+]] = bitcast i8* [[ORIG_TASK_PTR]] to
 // CHECK: br i1 %{{.+}}, label %[[OMP_THEN:.+]], label %[[OMP_ELSE:.+]]
 // CHECK: [[OMP_THEN]]
@@ -162,7 +162,7 @@ int main() {
 // CHECK: br label %[[OMP_END]]
 // CHECK: [[OMP_END]]
 
-// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^,]+}}, i32 [[GTID]], i32 1, i64 32, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[CAP_FN5:[^ ]+]] to i32 (i32, i8*)*))
+// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^,]+}}, i32 [[GTID]], i32 1, i64 40, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[CAP_FN5:[^ ]+]] to i32 (i32, i8*)*))
 // CHECK: [[TASK_PTR:%.+]] = bitcast i8* [[ORIG_TASK_PTR]] to
 // CHECK: br i1 %{{.+}}, label %[[OMP_THEN:.+]], label %[[OMP_ELSE:.+]]
 // CHECK: [[OMP_THEN]]
@@ -176,7 +176,7 @@ int main() {
 // CHECK: br label %[[OMP_END]]
 // CHECK: [[OMP_END]]
 
-// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^,]+}}, i32 [[GTID]], i32 1, i64 32, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[CAP_FN6:[^ ]+]] to i32 (i32, i8*)*))
+// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^,]+}}, i32 [[GTID]], i32 1, i64 40, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[CAP_FN6:[^ ]+]] to i32 (i32, i8*)*))
 // CHECK: [[TASK_PTR:%.+]] = bitcast i8* [[ORIG_TASK_PTR]] to
 // CHECK: br i1 %{{.+}}, label %[[OMP_THEN:.+]], label %[[OMP_ELSE:.+]]
 // CHECK: [[OMP_THEN]]
diff --git a/clang/test/OpenMP/task_private_codegen.cpp b/clang/test/OpenMP/task_private_codegen.cpp
index 209f21913ca0..97155a73f1bd 100644
--- a/clang/test/OpenMP/task_private_codegen.cpp
+++ b/clang/test/OpenMP/task_private_codegen.cpp
@@ -24,7 +24,7 @@ struct S {
 
 volatile double g;
 
-// CHECK-DAG: [[KMP_TASK_T_TY:%.+]] = type { i8*, i32 (i32, i8*)*, i32, i32 (i32, i8*)* }
+// CHECK-DAG: [[KMP_TASK_T_TY:%.+]] = type { i8*, i32 (i32, i8*)*, i32, %union{{.+}}, %union{{.+}} }
 // CHECK-DAG: [[S_DOUBLE_TY:%.+]] = type { double }
 // CHECK-DAG: [[CAP_MAIN_TY:%.+]] = type { i8 }
 // CHECK-DAG: [[PRIVATES_MAIN_TY:%.+]] = type {{.?}}{ [2 x [[S_DOUBLE_TY]]], [[S_DOUBLE_TY]], i32, [2 x i32]
@@ -56,7 +56,7 @@ int main() {
   // LAMBDA: call{{( x86_thiscallcc)?}} void [[OUTER_LAMBDA:@.+]](
   [&]() {
   // LAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]](
-  // LAMBDA: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 48, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
+  // LAMBDA: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 56, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
 // LAMBDA: [[PRIVATES:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
 // LAMBDA: call i32 @__kmpc_omp_task(%{{.+}}* @{{.+}}, i32 %{{.+}}, i8* [[RES]])
 // LAMBDA: ret
@@ -92,7 +92,7 @@ int main() {
   // BLOCKS: call void {{%.+}}(i8
   ^{
   // BLOCKS: define{{.*}} internal{{.*}} void {{.+}}(i8*
-  // BLOCKS: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 48, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
+  // BLOCKS: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 56, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
   // BLOCKS: [[PRIVATES:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
   // BLOCKS: call i32 @__kmpc_omp_task(%{{.+}}* @{{.+}}, i32 %{{.+}}, i8* [[RES]])
   // BLOCKS: ret
@@ -158,7 +158,7 @@ int main() {
 //         [[KMP_TASK_T_TY]] task_data;
 //         [[KMP_TASK_MAIN_TY]] privates;
 //       };
-// CHECK: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc([[LOC]], i32 [[GTID]], i32 9, i64 72, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
+// CHECK: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc([[LOC]], i32 [[GTID]], i32 9, i64 80, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
 // CHECK: [[RES_KMP_TASK:%.+]] = bitcast i8* [[RES]] to [[KMP_TASK_MAIN_TY]]*
 
 // CHECK: [[TASK:%.+]] = getelementptr inbounds [[KMP_TASK_MAIN_TY]], [[KMP_TASK_MAIN_TY]]* [[RES_KMP_TASK]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
@@ -182,7 +182,8 @@ int main() {
 
 // Provide pointer to destructor function, which will destroy private variables at the end of the task.
 // CHECK: [[DESTRUCTORS_REF:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{.+}} 0, i{{.+}} 3
-// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_REF]],
+// CHECK: [[DESTRUCTORS_PTR:%.+]] = bitcast %union{{.+}}* [[DESTRUCTORS_REF]] to i32 (i32, i8*)**
+// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_PTR]],
 
 // Start task.
 // CHECK: call i32 @__kmpc_omp_task([[LOC]], i32 [[GTID]], i8* [[RES]])
@@ -295,7 +296,8 @@ int main() {
 
 // Provide pointer to destructor function, which will destroy private variables at the end of the task.
 // CHECK: [[DESTRUCTORS_REF:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{.+}} 0, i{{.+}} 3
-// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_TMAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_REF]],
+// CHECK: [[DESTRUCTORS_PTR:%.+]] = bitcast %union{{.+}}* [[DESTRUCTORS_REF]] to i32 (i32, i8*)**
+// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_TMAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_PTR]],
 
 // Start task.
 // CHECK: call i32 @__kmpc_omp_task([[LOC]], i32 [[GTID]], i8* [[RES]])
diff --git a/clang/test/OpenMP/taskloop_codegen.cpp b/clang/test/OpenMP/taskloop_codegen.cpp
index fff3ccd71d70..e585fcec10d5 100644
--- a/clang/test/OpenMP/taskloop_codegen.cpp
+++ b/clang/test/OpenMP/taskloop_codegen.cpp
@@ -9,32 +9,28 @@
 // CHECK-LABEL: @main
 int main(int argc, char **argv) {
 // CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%ident_t* [[DEFLOC:@.+]])
-// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* [[DEFLOC]], i32 [[GTID]], i32 33, i64 64, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK1:@.+]] to i32 (i32, i8*)*))
+// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* [[DEFLOC]], i32 [[GTID]], i32 33, i64 72, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK1:@.+]] to i32 (i32, i8*)*))
 // CHECK: [[TASK:%.+]] = bitcast i8* [[TASKV]] to [[TDP_TY]]*
 // CHECK: [[TASK_DATA:%.+]] = getelementptr inbounds [[TDP_TY]], [[TDP_TY]]* [[TASK]], i32 0, i32 0
-// CHECK: getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 3
-// CHECK: store i32 (i32, i8*)* null, i32 (i32, i8*)** %{{.+}}
-// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 4
+// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 5
 // CHECK: store i64 0, i64* [[DOWN]],
-// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 5
+// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 6
 // CHECK: store i64 9, i64* [[UP]],
-// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 6
+// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 7
 // CHECK: store i64 1, i64* [[ST]],
 // CHECK: [[ST_VAL:%.+]] = load i64, i64* [[ST]],
 // CHECK: call void @__kmpc_taskloop(%ident_t* [[DEFLOC]], i32 [[GTID]], i8* [[TASKV]], i32 1, i64* [[DOWN]], i64* [[UP]], i64 [[ST_VAL]], i32 0, i32 0, i64 0, i8* null)
 #pragma omp taskloop priority(argc)
   for (int i = 0; i < 10; ++i)
     ;
-// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* [[DEFLOC]], i32 [[GTID]], i32 1, i64 64, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK2:@.+]] to i32 (i32, i8*)*))
+// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* [[DEFLOC]], i32 [[GTID]], i32 1, i64 72, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK2:@.+]] to i32 (i32, i8*)*))
 // CHECK: [[TASK:%.+]] = bitcast i8* [[TASKV]] to [[TDP_TY]]*
 // CHECK: [[TASK_DATA:%.+]] = getelementptr inbounds [[TDP_TY]], [[TDP_TY]]* [[TASK]], i32 0, i32 0
-// CHECK: getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 3
-// CHECK: store i32 (i32, i8*)* null, i32 (i32, i8*)** %{{.+}}
-// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 4
+// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 5
 // CHECK: store i64 0, i64* [[DOWN]],
-// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 5
+// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 6
 // CHECK: store i64 9, i64* [[UP]],
-// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 6
+// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 7
 // CHECK: store i64 1, i64* [[ST]],
 // CHECK: [[ST_VAL:%.+]] = load i64, i64* [[ST]],
 // CHECK: [[GRAINSIZE:%.+]] = zext i32 %{{.+}} to i64
@@ -42,18 +38,16 @@ int main(int argc, char **argv) {
 #pragma omp taskloop nogroup grainsize(argc)
   for (int i = 0; i < 10; ++i)
     ;
-// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* [[DEFLOC]], i32 [[GTID]], i32 1, i64 64, i64 24, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK3:@.+]] to i32 (i32, i8*)*))
+// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* [[DEFLOC]], i32 [[GTID]], i32 1, i64 72, i64 24, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK3:@.+]] to i32 (i32, i8*)*))
 // CHECK: [[TASK:%.+]] = bitcast i8* [[TASKV]] to [[TDP_TY]]*
 // CHECK: [[TASK_DATA:%.+]] = getelementptr inbounds [[TDP_TY]], [[TDP_TY]]* [[TASK]], i32 0, i32 0
-// CHECK: getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 3
-// CHECK: store i32 (i32, i8*)* null, i32 (i32, i8*)** %{{.+}}
 // CHECK: [[IF:%.+]] = icmp ne i32 %{{.+}}, 0
 // CHECK: [[IF_INT:%.+]] = sext i1 [[IF]] to i32
-// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 4
+// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 5
 // CHECK: store i64 0, i64* [[DOWN]],
-// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 5
+// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 6
 // CHECK: store i64 %{{.+}}, i64* [[UP]],
-// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 6
+// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 7
 // CHECK: store i64 1, i64* [[ST]],
 // CHECK: [[ST_VAL:%.+]] = load i64, i64* [[ST]],
 // CHECK: call void @__kmpc_taskloop(%ident_t* [[DEFLOC]], i32 [[GTID]], i8* [[TASKV]], i32 [[IF_INT]], i64* [[DOWN]], i64* [[UP]], i64 [[ST_VAL]], i32 0, i32 2, i64 4, i8* null)
@@ -65,13 +59,13 @@ int main(int argc, char **argv) {
 }
 
 // CHECK: define internal i32 [[TASK1]](
-// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* %{{.+}}, i32 0, i32 4
+// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* %{{.+}}, i32 0, i32 5
 // CHECK: [[DOWN_VAL:%.+]] = load i64, i64* [[DOWN]],
-// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 5
+// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 6
 // CHECK: [[UP_VAL:%.+]] = load i64, i64* [[UP]],
-// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 6
+// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 7
 // CHECK: [[ST_VAL:%.+]] = load i64, i64* [[ST]],
-// CHECK: [[LITER:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 7
+// CHECK: [[LITER:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 8
 // CHECK: [[LITER_VAL:%.+]] = load i32, i32* [[LITER]],
 // CHECK: store i64 [[DOWN_VAL]], i64* [[LB:%[^,]+]],
 // CHECK: store i64 [[UP_VAL]], i64* [[UB:%[^,]+]],
@@ -95,13 +89,13 @@ int main(int argc, char **argv) {
 // CHECK: ret i32 0
 
 // CHECK: define internal i32 [[TASK2]](
-// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* %{{.+}}, i32 0, i32 4
+// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* %{{.+}}, i32 0, i32 5
 // CHECK: [[DOWN_VAL:%.+]] = load i64, i64* [[DOWN]],
-// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 5
+// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 6
 // CHECK: [[UP_VAL:%.+]] = load i64, i64* [[UP]],
-// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 6
+// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 7
 // CHECK: [[ST_VAL:%.+]] = load i64, i64* [[ST]],
-// CHECK: [[LITER:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 7
+// CHECK: [[LITER:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 8
 // CHECK: [[LITER_VAL:%.+]] = load i32, i32* [[LITER]],
 // CHECK: store i64 [[DOWN_VAL]], i64* [[LB:%[^,]+]],
 // CHECK: store i64 [[UP_VAL]], i64* [[UB:%[^,]+]],
@@ -125,13 +119,13 @@ int main(int argc, char **argv) {
 // CHECK: ret i32 0
 
 // CHECK: define internal i32 [[TASK3]](
-// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* %{{.+}}, i32 0, i32 4
+// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* %{{.+}}, i32 0, i32 5
 // CHECK: [[DOWN_VAL:%.+]] = load i64, i64* [[DOWN]],
-// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 5
+// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 6
 // CHECK: [[UP_VAL:%.+]] = load i64, i64* [[UP]],
-// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 6
+// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 7
 // CHECK: [[ST_VAL:%.+]] = load i64, i64* [[ST]],
-// CHECK: [[LITER:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 7
+// CHECK: [[LITER:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 8
 // CHECK: [[LITER_VAL:%.+]] = load i32, i32* [[LITER]],
 // CHECK: store i64 [[DOWN_VAL]], i64* [[LB:%[^,]+]],
 // CHECK: store i64 [[UP_VAL]], i64* [[UB:%[^,]+]],
@@ -147,16 +141,14 @@ struct S {
   int a;
   S(int c) {
 // CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%ident_t* [[DEFLOC:@.+]])
-// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* [[DEFLOC]], i32 [[GTID]], i32 1, i64 64, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK4:@.+]] to i32 (i32, i8*)*))
+// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* [[DEFLOC]], i32 [[GTID]], i32 1, i64 72, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK4:@.+]] to i32 (i32, i8*)*))
 // CHECK: [[TASK:%.+]] = bitcast i8* [[TASKV]] to [[TDP_TY]]*
 // CHECK: [[TASK_DATA:%.+]] = getelementptr inbounds [[TDP_TY]], [[TDP_TY]]* [[TASK]], i32 0, i32 0
-// CHECK: getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 3
-// CHECK: store i32 (i32, i8*)* null, i32 (i32, i8*)** %{{.+}}
-// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 4
+// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 5
 // CHECK: store i64 0, i64* [[DOWN]],
-// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 5
+// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 6
 // CHECK: store i64 %{{.+}}, i64* [[UP]],
-// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 6
+// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 7
 // CHECK: store i64 1, i64* [[ST]],
 // CHECK: [[ST_VAL:%.+]] = load i64, i64* [[ST]],
 // CHECK: [[NUM_TASKS:%.+]] = zext i32 %{{.+}} to i64
@@ -168,13 +160,13 @@ struct S {
 } s(1);
 
 // CHECK: define internal i32 [[TASK4]](
-// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* %{{.+}}, i32 0, i32 4
+// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* %{{.+}}, i32 0, i32 5
 // CHECK: [[DOWN_VAL:%.+]] = load i64, i64* [[DOWN]],
-// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 5
+// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 6
 // CHECK: [[UP_VAL:%.+]] = load i64, i64* [[UP]],
-// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 6
+// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 7
 // CHECK: [[ST_VAL:%.+]] = load i64, i64* [[ST]],
-// CHECK: [[LITER:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 7
+// CHECK: [[LITER:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 8
 // CHECK: [[LITER_VAL:%.+]] = load i32, i32* [[LITER]],
 // CHECK: store i64 [[DOWN_VAL]], i64* [[LB:%[^,]+]],
 // CHECK: store i64 [[UP_VAL]], i64* [[UB:%[^,]+]],
diff --git a/clang/test/OpenMP/taskloop_firstprivate_codegen.cpp b/clang/test/OpenMP/taskloop_firstprivate_codegen.cpp
index 8af56fae3b85..822a5c687dc7 100644
--- a/clang/test/OpenMP/taskloop_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/taskloop_firstprivate_codegen.cpp
@@ -24,7 +24,7 @@ struct S {
 
 volatile double g;
 
-// CHECK-DAG: [[KMP_TASK_T_TY:%.+]] = type { i8*, i32 (i32, i8*)*, i32, i32 (i32, i8*)*, i64, i64, i64, i32 }
+// CHECK-DAG: [[KMP_TASK_T_TY:%.+]] = type { i8*, i32 (i32, i8*)*, i32, %union{{.+}}, %union{{.+}}, i64, i64, i64, i32 }
 // CHECK-DAG: [[S_DOUBLE_TY:%.+]] = type { double }
 // CHECK-DAG: [[PRIVATES_MAIN_TY:%.+]] = type {{.?}}{ [2 x [[S_DOUBLE_TY]]], [[S_DOUBLE_TY]], i32, [2 x i32]
 // CHECK-DAG: [[CAP_MAIN_TY:%.+]] = type {{.*}}{ [2 x i32]*, i32, {{.*}}[2 x [[S_DOUBLE_TY]]]*, [[S_DOUBLE_TY]]*, i{{[0-9]+}}
@@ -58,7 +58,7 @@ int main() {
   // LAMBDA: call{{( x86_thiscallcc)?}} void [[OUTER_LAMBDA:@.+]](
   [&]() {
   // LAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]](
-  // LAMBDA: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 80, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
+  // LAMBDA: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 88, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
 // LAMBDA: [[PRIVATES:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
 // LAMBDA: [[G_PRIVATE_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[PRIVATES]], i{{.+}} 0, i{{.+}} 0
 // LAMBDA: [[G_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 0
@@ -102,7 +102,7 @@ int main() {
   // BLOCKS: call void {{%.+}}(i8
   ^{
   // BLOCKS: define{{.*}} internal{{.*}} void {{.+}}(i8*
-  // BLOCKS: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 80, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
+  // BLOCKS: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 88, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
   // BLOCKS: [[PRIVATES:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
   // BLOCKS: [[G_PRIVATE_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[PRIVATES]], i{{.+}} 0, i{{.+}} 0
   // BLOCKS: [[G_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 0
@@ -191,7 +191,7 @@ int main() {
 //         [[KMP_TASK_T]] task_data;
 //         [[KMP_TASK_MAIN_TY]] privates;
 //       };
-// CHECK: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc([[LOC]], i32 [[GTID]], i32 9, i64 104, i64 40, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
+// CHECK: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc([[LOC]], i32 [[GTID]], i32 9, i64 112, i64 40, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
 // CHECK: [[RES_KMP_TASK:%.+]] = bitcast i8* [[RES]] to [[KMP_TASK_MAIN_TY]]*
 
 // Fill kmp_task_t->shareds by copying from original capture argument.
@@ -242,7 +242,8 @@ int main() {
 
 // Provide pointer to destructor function, which will destroy private variables at the end of the task.
 // CHECK: [[DESTRUCTORS_REF:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{.+}} 0, i{{.+}} 3
-// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_REF]],
+// CHECK: [[DESTRUCTORS_PTR:%.+]] = bitcast %union{{.+}}* [[DESTRUCTORS_REF]] to i32 (i32, i8*)**
+// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_PTR]],
 
 // Start task.
 // CHECK: call void @__kmpc_taskloop([[LOC]], i32 [[GTID]], i8* [[RES]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* bitcast (void ([[KMP_TASK_MAIN_TY]]*, [[KMP_TASK_MAIN_TY]]*, i32)* [[MAIN_DUP:@.+]] to i8*))
@@ -404,7 +405,8 @@ int main() {
 
 // Provide pointer to destructor function, which will destroy private variables at the end of the task.
 // CHECK: [[DESTRUCTORS_REF:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{.+}} 0, i{{.+}} 3
-// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_TMAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_REF]],
+// CHECK: [[DESTRUCTORS_PTR:%.+]] = bitcast %union{{.+}}* [[DESTRUCTORS_REF]] to i32 (i32, i8*)**
+// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_TMAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_PTR]],
 
 // Start task.
 // CHECK: call void @__kmpc_taskloop([[LOC]], i32 [[GTID]], i8* [[RES]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* bitcast (void ([[KMP_TASK_TMAIN_TY]]*, [[KMP_TASK_TMAIN_TY]]*, i32)* [[TMAIN_DUP:@.+]] to i8*))
diff --git a/clang/test/OpenMP/taskloop_lastprivate_codegen.cpp b/clang/test/OpenMP/taskloop_lastprivate_codegen.cpp
index b56fd56ac7ea..8414b6f963f9 100644
--- a/clang/test/OpenMP/taskloop_lastprivate_codegen.cpp
+++ b/clang/test/OpenMP/taskloop_lastprivate_codegen.cpp
@@ -24,7 +24,7 @@ struct S {
 
 volatile double g;
 
-// CHECK-DAG: [[KMP_TASK_T_TY:%.+]] = type { i8*, i32 (i32, i8*)*, i32, i32 (i32, i8*)*, i64, i64, i64, i32 }
+// CHECK-DAG: [[KMP_TASK_T_TY:%.+]] = type { i8*, i32 (i32, i8*)*, i32, %union{{.+}}, %union{{.+}}, i64, i64, i64, i32 }
 // CHECK-DAG: [[S_DOUBLE_TY:%.+]] = type { double }
 // CHECK-DAG: [[PRIVATES_MAIN_TY:%.+]] = type {{.?}}{ [2 x [[S_DOUBLE_TY]]], [[S_DOUBLE_TY]], i32, [2 x i32]
 // CHECK-DAG: [[CAP_MAIN_TY:%.+]] = type { [2 x i32]*, i32*, [2 x [[S_DOUBLE_TY]]]*, [[S_DOUBLE_TY]]*, i{{[0-9]+}}* }
@@ -58,7 +58,7 @@ int main() {
   // LAMBDA: call{{( x86_thiscallcc)?}} void [[OUTER_LAMBDA:@.+]](
   [&]() {
   // LAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]](
-  // LAMBDA: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 80, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
+  // LAMBDA: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 88, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
 // LAMBDA: [[PRIVATES:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
 
 // LAMBDA: call void @__kmpc_taskloop(%{{.+}}* @{{.+}}, i32 %{{.+}}, i8* [[RES]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* bitcast (void ([[KMP_TASK_MAIN_TY:%[^*]+]]*, [[KMP_TASK_MAIN_TY]]*, i32)* [[MAIN_DUP:@.+]] to i8*))
@@ -99,7 +99,7 @@ int main() {
   // BLOCKS: call void {{%.+}}(i8
   ^{
   // BLOCKS: define{{.*}} internal{{.*}} void {{.+}}(i8*
-  // BLOCKS: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 80, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
+  // BLOCKS: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 88, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
   // BLOCKS: [[PRIVATES:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
   // BLOCKS: call void @__kmpc_taskloop(%{{.+}}* @{{.+}}, i32 %{{.+}}, i8* [[RES]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* bitcast (void ([[KMP_TASK_MAIN_TY:%[^*]+]]*, [[KMP_TASK_MAIN_TY]]*, i32)* [[MAIN_DUP:@.+]] to i8*))
   // BLOCKS: ret
@@ -183,7 +183,7 @@ int main() {
 //         [[KMP_TASK_T]] task_data;
 //         [[KMP_TASK_MAIN_TY]] privates;
 //       };
-// CHECK: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc([[LOC]], i32 [[GTID]], i32 9, i64 104, i64 40, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
+// CHECK: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc([[LOC]], i32 [[GTID]], i32 9, i64 112, i64 40, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
 // CHECK: [[RES_KMP_TASK:%.+]] = bitcast i8* [[RES]] to [[KMP_TASK_MAIN_TY]]*
 
 // Fill kmp_task_t->shareds by copying from original capture argument.
@@ -215,7 +215,8 @@ int main() {
 
 // Provide pointer to destructor function, which will destroy private variables at the end of the task.
 // CHECK: [[DESTRUCTORS_REF:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{.+}} 0, i{{.+}} 3
-// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_REF]],
+// CHECK: [[DESTRUCTORS_PTR:%.+]] = bitcast %union{{.+}}* [[DESTRUCTORS_REF]] to i32 (i32, i8*)**
+// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_PTR]],
 
 // Start task.
 // CHECK: call void @__kmpc_taskloop([[LOC]], i32 [[GTID]], i8* [[RES]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* bitcast (void ([[KMP_TASK_MAIN_TY]]*, [[KMP_TASK_MAIN_TY]]*, i32)* [[MAIN_DUP:@.+]] to i8*))
@@ -298,7 +299,7 @@ int main() {
 
 // CHECK: define internal void [[MAIN_DUP]]([[KMP_TASK_MAIN_TY]]*, [[KMP_TASK_MAIN_TY]]*, i32)
 // CHECK: getelementptr inbounds [[KMP_TASK_MAIN_TY]], [[KMP_TASK_MAIN_TY]]* %{{.+}}, i32 0, i32 0
-// CHECK: getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* %{{.+}}, i32 0, i32 7
+// CHECK: getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* %{{.+}}, i32 0, i32 8
 // CHECK: load i32, i32* %
 // CHECK: store i32 %{{.+}}, i32* %
 // CHECK: getelementptr inbounds [[KMP_TASK_MAIN_TY]], [[KMP_TASK_MAIN_TY]]* %{{.+}}, i32 0, i32 1
@@ -388,7 +389,8 @@ int main() {
 
 // Provide pointer to destructor function, which will destroy private variables at the end of the task.
 // CHECK: [[DESTRUCTORS_REF:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{.+}} 0, i{{.+}} 3
-// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_TMAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_REF]],
+// CHECK: [[DESTRUCTORS_PTR:%.+]] = bitcast %union{{.+}}* [[DESTRUCTORS_REF]] to i32 (i32, i8*)**
+// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_TMAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_PTR]],
 
 // Start task.
 // CHECK: call void @__kmpc_taskloop([[LOC]], i32 [[GTID]], i8* [[RES]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* bitcast (void ([[KMP_TASK_TMAIN_TY]]*, [[KMP_TASK_TMAIN_TY]]*, i32)* [[TMAIN_DUP:@.+]] to i8*))
@@ -459,7 +461,7 @@ int main() {
 
 // CHECK: define internal void [[TMAIN_DUP]]([[KMP_TASK_TMAIN_TY]]*, [[KMP_TASK_TMAIN_TY]]*, i32)
 // CHECK: getelementptr inbounds [[KMP_TASK_TMAIN_TY]], [[KMP_TASK_TMAIN_TY]]* %{{.+}}, i32 0, i32 0
-// CHECK: getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* %{{.+}}, i32 0, i32 7
+// CHECK: getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* %{{.+}}, i32 0, i32 8
 // CHECK: load i32, i32* %
 // CHECK: store i32 %{{.+}}, i32* %
 // CHECK: getelementptr inbounds [[KMP_TASK_TMAIN_TY]], [[KMP_TASK_TMAIN_TY]]* %{{.+}}, i32 0, i32 2
diff --git a/clang/test/OpenMP/taskloop_private_codegen.cpp b/clang/test/OpenMP/taskloop_private_codegen.cpp
index 3e53a0bb006c..38b20c579393 100644
--- a/clang/test/OpenMP/taskloop_private_codegen.cpp
+++ b/clang/test/OpenMP/taskloop_private_codegen.cpp
@@ -24,7 +24,7 @@ struct S {
 
 volatile double g;
 
-// CHECK-DAG: [[KMP_TASK_T_TY:%.+]] = type { i8*, i32 (i32, i8*)*, i32, i32 (i32, i8*)*, i64, i64, i64, i32 }
+// CHECK-DAG: [[KMP_TASK_T_TY:%.+]] = type { i8*, i32 (i32, i8*)*, i32, %union{{.+}}, %union{{.+}}, i64, i64, i64, i32 }
 // CHECK-DAG: [[S_DOUBLE_TY:%.+]] = type { double }
 // CHECK-DAG: [[CAP_MAIN_TY:%.+]] = type { i8 }
 // CHECK-DAG: [[PRIVATES_MAIN_TY:%.+]] = type {{.?}}{ [2 x [[S_DOUBLE_TY]]], [[S_DOUBLE_TY]], i32, [2 x i32]
@@ -56,7 +56,7 @@ int main() {
   // LAMBDA: call{{( x86_thiscallcc)?}} void [[OUTER_LAMBDA:@.+]](
   [&]() {
   // LAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]](
-  // LAMBDA: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 80, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
+  // LAMBDA: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 88, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
 // LAMBDA: [[PRIVATES:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
 // LAMBDA: call void @__kmpc_taskloop(%{{.+}}* @{{.+}}, i32 %{{.+}}, i8* [[RES]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* null)
 // LAMBDA: ret
@@ -92,7 +92,7 @@ int main() {
   // BLOCKS: call void {{%.+}}(i8
   ^{
   // BLOCKS: define{{.*}} internal{{.*}} void {{.+}}(i8*
-  // BLOCKS: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 80, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
+  // BLOCKS: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 88, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
   // BLOCKS: [[PRIVATES:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
   // BLOCKS: call void @__kmpc_taskloop(%{{.+}}* @{{.+}}, i32 %{{.+}}, i8* [[RES]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* null)
   // BLOCKS: ret
@@ -158,7 +158,7 @@ int main() {
 //         [[KMP_TASK_T_TY]] task_data;
 //         [[KMP_TASK_MAIN_TY]] privates;
 //       };
-// CHECK: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc([[LOC]], i32 [[GTID]], i32 9, i64 104, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
+// CHECK: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc([[LOC]], i32 [[GTID]], i32 9, i64 112, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
 // CHECK: [[RES_KMP_TASK:%.+]] = bitcast i8* [[RES]] to [[KMP_TASK_MAIN_TY]]*
 
 // CHECK: [[TASK:%.+]] = getelementptr inbounds [[KMP_TASK_MAIN_TY]], [[KMP_TASK_MAIN_TY]]* [[RES_KMP_TASK]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
@@ -182,7 +182,8 @@ int main() {
 
 // Provide pointer to destructor function, which will destroy private variables at the end of the task.
 // CHECK: [[DESTRUCTORS_REF:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{.+}} 0, i{{.+}} 3
-// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_REF]],
+// CHECK: [[DESTRUCTORS_PTR:%.+]] = bitcast %union{{.+}}* [[DESTRUCTORS_REF]] to i32 (i32, i8*)**
+// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_PTR]],
 
 // Start task.
 // CHECK: call void @__kmpc_taskloop([[LOC]], i32 [[GTID]], i8* [[RES]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* bitcast (void ([[KMP_TASK_MAIN_TY]]*, [[KMP_TASK_MAIN_TY]]*, i32)* [[MAIN_DUP:@.+]] to i8*))
@@ -312,7 +313,8 @@ int main() {
 
 // Provide pointer to destructor function, which will destroy private variables at the end of the task.
 // CHECK: [[DESTRUCTORS_REF:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{.+}} 0, i{{.+}} 3
-// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_TMAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_REF]],
+// CHECK: [[DESTRUCTORS_PTR:%.+]] = bitcast %union{{.+}}* [[DESTRUCTORS_REF]] to i32 (i32, i8*)**
+// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_TMAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_PTR]],
 
 // Start task.
 // CHECK: call void @__kmpc_taskloop([[LOC]], i32 [[GTID]], i8* [[RES]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* bitcast (void ([[KMP_TASK_TMAIN_TY]]*, [[KMP_TASK_TMAIN_TY]]*, i32)* [[TMAIN_DUP:@.+]] to i8*))
diff --git a/clang/test/OpenMP/taskloop_simd_codegen.cpp b/clang/test/OpenMP/taskloop_simd_codegen.cpp
index c64b68071ab7..dc60009ff2bd 100644
--- a/clang/test/OpenMP/taskloop_simd_codegen.cpp
+++ b/clang/test/OpenMP/taskloop_simd_codegen.cpp
@@ -9,32 +9,28 @@
 // CHECK-LABEL: @main
 int main(int argc, char **argv) {
 // CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%ident_t* [[DEFLOC:@.+]])
-// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* [[DEFLOC]], i32 [[GTID]], i32 33, i64 64, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK1:@.+]] to i32 (i32, i8*)*))
+// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* [[DEFLOC]], i32 [[GTID]], i32 33, i64 72, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK1:@.+]] to i32 (i32, i8*)*))
 // CHECK: [[TASK:%.+]] = bitcast i8* [[TASKV]] to [[TDP_TY]]*
 // CHECK: [[TASK_DATA:%.+]] = getelementptr inbounds [[TDP_TY]], [[TDP_TY]]* [[TASK]], i32 0, i32 0
-// CHECK: getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 3
-// CHECK: store i32 (i32, i8*)* null, i32 (i32, i8*)** %{{.+}}
-// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 4
+// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 5
 // CHECK: store i64 0, i64* [[DOWN]],
-// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 5
+// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 6
 // CHECK: store i64 9, i64* [[UP]],
-// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 6
+// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 7
 // CHECK: store i64 1, i64* [[ST]],
 // CHECK: [[ST_VAL:%.+]] = load i64, i64* [[ST]],
 // CHECK: call void @__kmpc_taskloop(%ident_t* [[DEFLOC]], i32 [[GTID]], i8* [[TASKV]], i32 1, i64* [[DOWN]], i64* [[UP]], i64 [[ST_VAL]], i32 0, i32 0, i64 0, i8* null)
 #pragma omp taskloop simd priority(argc)
   for (int i = 0; i < 10; ++i)
     ;
-// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* [[DEFLOC]], i32 [[GTID]], i32 1, i64 64, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK2:@.+]] to i32 (i32, i8*)*))
+// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* [[DEFLOC]], i32 [[GTID]], i32 1, i64 72, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK2:@.+]] to i32 (i32, i8*)*))
 // CHECK: [[TASK:%.+]] = bitcast i8* [[TASKV]] to [[TDP_TY]]*
 // CHECK: [[TASK_DATA:%.+]] = getelementptr inbounds [[TDP_TY]], [[TDP_TY]]* [[TASK]], i32 0, i32 0
-// CHECK: getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 3
-// CHECK: store i32 (i32, i8*)* null, i32 (i32, i8*)** %{{.+}}
-// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 4
+// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 5
 // CHECK: store i64 0, i64* [[DOWN]],
-// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 5
+// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 6
 // CHECK: store i64 9, i64* [[UP]],
-// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 6
+// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 7
 // CHECK: store i64 1, i64* [[ST]],
 // CHECK: [[ST_VAL:%.+]] = load i64, i64* [[ST]],
 // CHECK: [[GRAINSIZE:%.+]] = zext i32 %{{.+}} to i64
@@ -42,18 +38,16 @@ int main(int argc, char **argv) {
 #pragma omp taskloop simd nogroup grainsize(argc) simdlen(4)
   for (int i = 0; i < 10; ++i)
     ;
-// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* [[DEFLOC]], i32 [[GTID]], i32 1, i64 64, i64 24, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK3:@.+]] to i32 (i32, i8*)*))
+// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* [[DEFLOC]], i32 [[GTID]], i32 1, i64 72, i64 24, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK3:@.+]] to i32 (i32, i8*)*))
 // CHECK: [[TASK:%.+]] = bitcast i8* [[TASKV]] to [[TDP_TY]]*
 // CHECK: [[TASK_DATA:%.+]] = getelementptr inbounds [[TDP_TY]], [[TDP_TY]]* [[TASK]], i32 0, i32 0
-// CHECK: getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 3
-// CHECK: store i32 (i32, i8*)* null, i32 (i32, i8*)** %{{.+}}
 // CHECK: [[IF:%.+]] = icmp ne i32 %{{.+}}, 0
 // CHECK: [[IF_INT:%.+]] = sext i1 [[IF]] to i32
-// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 4
+// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 5
 // CHECK: store i64 0, i64* [[DOWN]],
-// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 5
+// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 6
 // CHECK: store i64 %{{.+}}, i64* [[UP]],
-// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 6
+// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 7
 // CHECK: store i64 1, i64* [[ST]],
 // CHECK: [[ST_VAL:%.+]] = load i64, i64* [[ST]],
 // CHECK: call void @__kmpc_taskloop(%ident_t* [[DEFLOC]], i32 [[GTID]], i8* [[TASKV]], i32 [[IF_INT]], i64* [[DOWN]], i64* [[UP]], i64 [[ST_VAL]], i32 0, i32 2, i64 4, i8* null)
@@ -65,13 +59,13 @@ int main(int argc, char **argv) {
 }
 
 // CHECK: define internal i32 [[TASK1]](
-// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* %{{.+}}, i32 0, i32 4
+// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* %{{.+}}, i32 0, i32 5
 // CHECK: [[DOWN_VAL:%.+]] = load i64, i64* [[DOWN]],
-// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 5
+// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 6
 // CHECK: [[UP_VAL:%.+]] = load i64, i64* [[UP]],
-// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 6
+// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 7
 // CHECK: [[ST_VAL:%.+]] = load i64, i64* [[ST]],
-// CHECK: [[LITER:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 7
+// CHECK: [[LITER:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 8
 // CHECK: [[LITER_VAL:%.+]] = load i32, i32* [[LITER]],
 // CHECK: store i64 [[DOWN_VAL]], i64* [[LB:%[^,]+]],
 // CHECK: store i64 [[UP_VAL]], i64* [[UB:%[^,]+]],
@@ -95,13 +89,13 @@ int main(int argc, char **argv) {
 // CHECK: ret i32 0
 
 // CHECK: define internal i32 [[TASK2]](
-// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* %{{.+}}, i32 0, i32 4
+// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* %{{.+}}, i32 0, i32 5
 // CHECK: [[DOWN_VAL:%.+]] = load i64, i64* [[DOWN]],
-// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 5
+// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 6
 // CHECK: [[UP_VAL:%.+]] = load i64, i64* [[UP]],
-// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 6
+// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 7
 // CHECK: [[ST_VAL:%.+]] = load i64, i64* [[ST]],
-// CHECK: [[LITER:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 7
+// CHECK: [[LITER:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 8
 // CHECK: [[LITER_VAL:%.+]] = load i32, i32* [[LITER]],
 // CHECK: store i64 [[DOWN_VAL]], i64* [[LB:%[^,]+]],
 // CHECK: store i64 [[UP_VAL]], i64* [[UB:%[^,]+]],
@@ -125,13 +119,13 @@ int main(int argc, char **argv) {
 // CHECK: ret i32 0
 
 // CHECK: define internal i32 [[TASK3]](
-// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* %{{.+}}, i32 0, i32 4
+// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* %{{.+}}, i32 0, i32 5
 // CHECK: [[DOWN_VAL:%.+]] = load i64, i64* [[DOWN]],
-// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 5
+// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 6
 // CHECK: [[UP_VAL:%.+]] = load i64, i64* [[UP]],
-// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 6
+// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 7
 // CHECK: [[ST_VAL:%.+]] = load i64, i64* [[ST]],
-// CHECK: [[LITER:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 7
+// CHECK: [[LITER:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 8
 // CHECK: [[LITER_VAL:%.+]] = load i32, i32* [[LITER]],
 // CHECK: store i64 [[DOWN_VAL]], i64* [[LB:%[^,]+]],
 // CHECK: store i64 [[UP_VAL]], i64* [[UB:%[^,]+]],
@@ -149,16 +143,14 @@ struct S {
   int a;
   S(int c) {
 // CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%ident_t* [[DEFLOC:@.+]])
-// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* [[DEFLOC]], i32 [[GTID]], i32 1, i64 64, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK4:@.+]] to i32 (i32, i8*)*))
+// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* [[DEFLOC]], i32 [[GTID]], i32 1, i64 72, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK4:@.+]] to i32 (i32, i8*)*))
 // CHECK: [[TASK:%.+]] = bitcast i8* [[TASKV]] to [[TDP_TY]]*
 // CHECK: [[TASK_DATA:%.+]] = getelementptr inbounds [[TDP_TY]], [[TDP_TY]]* [[TASK]], i32 0, i32 0
-// CHECK: getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 3
-// CHECK: store i32 (i32, i8*)* null, i32 (i32, i8*)** %{{.+}}
-// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 4
+// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 5
 // CHECK: store i64 0, i64* [[DOWN]],
-// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 5
+// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 6
 // CHECK: store i64 %{{.+}}, i64* [[UP]],
-// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 6
+// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 7
 // CHECK: store i64 1, i64* [[ST]],
 // CHECK: [[ST_VAL:%.+]] = load i64, i64* [[ST]],
 // CHECK: [[NUM_TASKS:%.+]] = zext i32 %{{.+}} to i64
@@ -170,13 +162,13 @@ struct S {
 } s(1);
 
 // CHECK: define internal i32 [[TASK4]](
-// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* %{{.+}}, i32 0, i32 4
+// CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* %{{.+}}, i32 0, i32 5
 // CHECK: [[DOWN_VAL:%.+]] = load i64, i64* [[DOWN]],
-// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 5
+// CHECK: [[UP:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 6
 // CHECK: [[UP_VAL:%.+]] = load i64, i64* [[UP]],
-// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 6
+// CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 7
 // CHECK: [[ST_VAL:%.+]] = load i64, i64* [[ST]],
-// CHECK: [[LITER:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 7
+// CHECK: [[LITER:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* %{{.+}}, i32 0, i32 8
 // CHECK: [[LITER_VAL:%.+]] = load i32, i32* [[LITER]],
 // CHECK: store i64 [[DOWN_VAL]], i64* [[LB:%[^,]+]],
 // CHECK: store i64 [[UP_VAL]], i64* [[UB:%[^,]+]],
diff --git a/clang/test/OpenMP/taskloop_simd_firstprivate_codegen.cpp b/clang/test/OpenMP/taskloop_simd_firstprivate_codegen.cpp
index 5cb2cd81503d..0b87ddd9acb1 100644
--- a/clang/test/OpenMP/taskloop_simd_firstprivate_codegen.cpp
+++ b/clang/test/OpenMP/taskloop_simd_firstprivate_codegen.cpp
@@ -24,7 +24,7 @@ struct S {
 
 volatile double g;
 
-// CHECK-DAG: [[KMP_TASK_T_TY:%.+]] = type { i8*, i32 (i32, i8*)*, i32, i32 (i32, i8*)*, i64, i64, i64, i32 }
+// CHECK-DAG: [[KMP_TASK_T_TY:%.+]] = type { i8*, i32 (i32, i8*)*, i32, %union{{.+}}, %union{{.+}}, i64, i64, i64, i32 }
 // CHECK-DAG: [[S_DOUBLE_TY:%.+]] = type { double }
 // CHECK-DAG: [[PRIVATES_MAIN_TY:%.+]] = type {{.?}}{ [2 x [[S_DOUBLE_TY]]], [[S_DOUBLE_TY]], i32, [2 x i32]
 // CHECK-DAG: [[CAP_MAIN_TY:%.+]] = type {{.*}}{ [2 x i32]*, i32, {{.*}}[2 x [[S_DOUBLE_TY]]]*, [[S_DOUBLE_TY]]*, i{{[0-9]+}}
@@ -58,7 +58,7 @@ int main() {
   // LAMBDA: call{{( x86_thiscallcc)?}} void [[OUTER_LAMBDA:@.+]](
   [&]() {
   // LAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]](
-  // LAMBDA: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 80, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
+  // LAMBDA: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 88, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
 // LAMBDA: [[PRIVATES:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
 // LAMBDA: [[G_PRIVATE_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[PRIVATES]], i{{.+}} 0, i{{.+}} 0
 // LAMBDA: [[G_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 0
@@ -102,7 +102,7 @@ int main() {
   // BLOCKS: call void {{%.+}}(i8
   ^{
   // BLOCKS: define{{.*}} internal{{.*}} void {{.+}}(i8*
-  // BLOCKS: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 80, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
+  // BLOCKS: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 88, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
   // BLOCKS: [[PRIVATES:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
   // BLOCKS: [[G_PRIVATE_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[PRIVATES]], i{{.+}} 0, i{{.+}} 0
   // BLOCKS: [[G_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 0
@@ -191,7 +191,7 @@ int main() {
 //         [[KMP_TASK_T]] task_data;
 //         [[KMP_TASK_MAIN_TY]] privates;
 //       };
-// CHECK: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc([[LOC]], i32 [[GTID]], i32 9, i64 104, i64 40, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
+// CHECK: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc([[LOC]], i32 [[GTID]], i32 9, i64 112, i64 40, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
 // CHECK: [[RES_KMP_TASK:%.+]] = bitcast i8* [[RES]] to [[KMP_TASK_MAIN_TY]]*
 
 // Fill kmp_task_t->shareds by copying from original capture argument.
@@ -242,7 +242,8 @@ int main() {
 
 // Provide pointer to destructor function, which will destroy private variables at the end of the task.
 // CHECK: [[DESTRUCTORS_REF:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{.+}} 0, i{{.+}} 3
-// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_REF]],
+// CHECK: [[DESTRUCTORS_PTR:%.+]] = bitcast %union{{.+}}* [[DESTRUCTORS_REF]] to i32 (i32, i8*)**
+// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_PTR]],
 
 // Start task.
 // CHECK: call void @__kmpc_taskloop([[LOC]], i32 [[GTID]], i8* [[RES]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* bitcast (void ([[KMP_TASK_MAIN_TY]]*, [[KMP_TASK_MAIN_TY]]*, i32)* [[MAIN_DUP:@.+]] to i8*))
@@ -404,7 +405,8 @@ int main() {
 
 // Provide pointer to destructor function, which will destroy private variables at the end of the task.
 // CHECK: [[DESTRUCTORS_REF:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{.+}} 0, i{{.+}} 3
-// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_TMAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_REF]],
+// CHECK: [[DESTRUCTORS_PTR:%.+]] = bitcast %union{{.+}}* [[DESTRUCTORS_REF]] to i32 (i32, i8*)**
+// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_TMAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_PTR]],
 
 // Start task.
 // CHECK: call void @__kmpc_taskloop([[LOC]], i32 [[GTID]], i8* [[RES]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* bitcast (void ([[KMP_TASK_TMAIN_TY]]*, [[KMP_TASK_TMAIN_TY]]*, i32)* [[TMAIN_DUP:@.+]] to i8*))
diff --git a/clang/test/OpenMP/taskloop_simd_lastprivate_codegen.cpp b/clang/test/OpenMP/taskloop_simd_lastprivate_codegen.cpp
index 4191950f3be4..e3562a8f0c62 100644
--- a/clang/test/OpenMP/taskloop_simd_lastprivate_codegen.cpp
+++ b/clang/test/OpenMP/taskloop_simd_lastprivate_codegen.cpp
@@ -24,7 +24,7 @@ struct S {
 
 volatile double g;
 
-// CHECK-DAG: [[KMP_TASK_T_TY:%.+]] = type { i8*, i32 (i32, i8*)*, i32, i32 (i32, i8*)*, i64, i64, i64, i32 }
+// CHECK-DAG: [[KMP_TASK_T_TY:%.+]] = type { i8*, i32 (i32, i8*)*, i32, %union{{.+}}, %union{{.+}}, i64, i64, i64, i32 }
 // CHECK-DAG: [[S_DOUBLE_TY:%.+]] = type { double }
 // CHECK-DAG: [[PRIVATES_MAIN_TY:%.+]] = type {{.?}}{ [2 x [[S_DOUBLE_TY]]], [[S_DOUBLE_TY]], i32, [2 x i32]
 // CHECK-DAG: [[CAP_MAIN_TY:%.+]] = type { [2 x i32]*, i32*, [2 x [[S_DOUBLE_TY]]]*, [[S_DOUBLE_TY]]*, i{{[0-9]+}}* }
@@ -58,7 +58,7 @@ int main() {
   // LAMBDA: call{{( x86_thiscallcc)?}} void [[OUTER_LAMBDA:@.+]](
   [&]() {
   // LAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]](
-  // LAMBDA: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 80, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
+  // LAMBDA: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 88, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
 // LAMBDA: [[PRIVATES:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
 
 // LAMBDA: call void @__kmpc_taskloop(%{{.+}}* @{{.+}}, i32 %{{.+}}, i8* [[RES]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* bitcast (void ([[KMP_TASK_MAIN_TY:%[^*]+]]*, [[KMP_TASK_MAIN_TY]]*, i32)* [[MAIN_DUP:@.+]] to i8*))
@@ -99,7 +99,7 @@ int main() {
   // BLOCKS: call void {{%.+}}(i8
   ^{
   // BLOCKS: define{{.*}} internal{{.*}} void {{.+}}(i8*
-  // BLOCKS: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 80, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
+  // BLOCKS: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 88, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
   // BLOCKS: [[PRIVATES:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
   // BLOCKS: call void @__kmpc_taskloop(%{{.+}}* @{{.+}}, i32 %{{.+}}, i8* [[RES]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* bitcast (void ([[KMP_TASK_MAIN_TY:%[^*]+]]*, [[KMP_TASK_MAIN_TY]]*, i32)* [[MAIN_DUP:@.+]] to i8*))
   // BLOCKS: ret
@@ -183,7 +183,7 @@ int main() {
 //         [[KMP_TASK_T]] task_data;
 //         [[KMP_TASK_MAIN_TY]] privates;
 //       };
-// CHECK: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc([[LOC]], i32 [[GTID]], i32 9, i64 104, i64 40, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
+// CHECK: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc([[LOC]], i32 [[GTID]], i32 9, i64 112, i64 40, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
 // CHECK: [[RES_KMP_TASK:%.+]] = bitcast i8* [[RES]] to [[KMP_TASK_MAIN_TY]]*
 
 // Fill kmp_task_t->shareds by copying from original capture argument.
@@ -215,7 +215,8 @@ int main() {
 
 // Provide pointer to destructor function, which will destroy private variables at the end of the task.
 // CHECK: [[DESTRUCTORS_REF:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{.+}} 0, i{{.+}} 3
-// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_REF]],
+// CHECK: [[DESTRUCTORS_PTR:%.+]] = bitcast %union{{.+}}* [[DESTRUCTORS_REF]] to i32 (i32, i8*)**
+// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_PTR]],
 
 // Start task.
 // CHECK: call void @__kmpc_taskloop([[LOC]], i32 [[GTID]], i8* [[RES]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* bitcast (void ([[KMP_TASK_MAIN_TY]]*, [[KMP_TASK_MAIN_TY]]*, i32)* [[MAIN_DUP:@.+]] to i8*))
@@ -298,7 +299,7 @@ int main() {
 
 // CHECK: define internal void [[MAIN_DUP]]([[KMP_TASK_MAIN_TY]]*, [[KMP_TASK_MAIN_TY]]*, i32)
 // CHECK: getelementptr inbounds [[KMP_TASK_MAIN_TY]], [[KMP_TASK_MAIN_TY]]* %{{.+}}, i32 0, i32 0
-// CHECK: getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* %{{.+}}, i32 0, i32 7
+// CHECK: getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* %{{.+}}, i32 0, i32 8
 // CHECK: load i32, i32* %
 // CHECK: store i32 %{{.+}}, i32* %
 // CHECK: getelementptr inbounds [[KMP_TASK_MAIN_TY]], [[KMP_TASK_MAIN_TY]]* %{{.+}}, i32 0, i32 1
@@ -388,7 +389,8 @@ int main() {
 
 // Provide pointer to destructor function, which will destroy private variables at the end of the task.
 // CHECK: [[DESTRUCTORS_REF:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{.+}} 0, i{{.+}} 3
-// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_TMAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_REF]],
+// CHECK: [[DESTRUCTORS_PTR:%.+]] = bitcast %union{{.+}}* [[DESTRUCTORS_REF]] to i32 (i32, i8*)**
+// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_TMAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_PTR]],
 
 // Start task.
 // CHECK: call void @__kmpc_taskloop([[LOC]], i32 [[GTID]], i8* [[RES]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* bitcast (void ([[KMP_TASK_TMAIN_TY]]*, [[KMP_TASK_TMAIN_TY]]*, i32)* [[TMAIN_DUP:@.+]] to i8*))
@@ -459,7 +461,7 @@ int main() {
 
 // CHECK: define internal void [[TMAIN_DUP]]([[KMP_TASK_TMAIN_TY]]*, [[KMP_TASK_TMAIN_TY]]*, i32)
 // CHECK: getelementptr inbounds [[KMP_TASK_TMAIN_TY]], [[KMP_TASK_TMAIN_TY]]* %{{.+}}, i32 0, i32 0
-// CHECK: getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* %{{.+}}, i32 0, i32 7
+// CHECK: getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* %{{.+}}, i32 0, i32 8
 // CHECK: load i32, i32* %
 // CHECK: store i32 %{{.+}}, i32* %
 // CHECK: getelementptr inbounds [[KMP_TASK_TMAIN_TY]], [[KMP_TASK_TMAIN_TY]]* %{{.+}}, i32 0, i32 2
diff --git a/clang/test/OpenMP/taskloop_simd_private_codegen.cpp b/clang/test/OpenMP/taskloop_simd_private_codegen.cpp
index 358f28520d72..557601e9c93e 100644
--- a/clang/test/OpenMP/taskloop_simd_private_codegen.cpp
+++ b/clang/test/OpenMP/taskloop_simd_private_codegen.cpp
@@ -24,7 +24,7 @@ struct S {
 
 volatile double g;
 
-// CHECK-DAG: [[KMP_TASK_T_TY:%.+]] = type { i8*, i32 (i32, i8*)*, i32, i32 (i32, i8*)*, i64, i64, i64, i32 }
+// CHECK-DAG: [[KMP_TASK_T_TY:%.+]] = type { i8*, i32 (i32, i8*)*, i32, %union{{.+}}, %union{{.+}}, i64, i64, i64, i32 }
 // CHECK-DAG: [[S_DOUBLE_TY:%.+]] = type { double }
 // CHECK-DAG: [[CAP_MAIN_TY:%.+]] = type { i8 }
 // CHECK-DAG: [[PRIVATES_MAIN_TY:%.+]] = type {{.?}}{ [2 x [[S_DOUBLE_TY]]], [[S_DOUBLE_TY]], i32, [2 x i32]
@@ -56,7 +56,7 @@ int main() {
   // LAMBDA: call{{( x86_thiscallcc)?}} void [[OUTER_LAMBDA:@.+]](
   [&]() {
   // LAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]](
-  // LAMBDA: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 80, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
+  // LAMBDA: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 88, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
 // LAMBDA: [[PRIVATES:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
 // LAMBDA: call void @__kmpc_taskloop(%{{.+}}* @{{.+}}, i32 %{{.+}}, i8* [[RES]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* null)
 // LAMBDA: ret
@@ -92,7 +92,7 @@ int main() {
   // BLOCKS: call void {{%.+}}(i8
   ^{
   // BLOCKS: define{{.*}} internal{{.*}} void {{.+}}(i8*
-  // BLOCKS: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 80, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
+  // BLOCKS: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 88, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
   // BLOCKS: [[PRIVATES:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
   // BLOCKS: call void @__kmpc_taskloop(%{{.+}}* @{{.+}}, i32 %{{.+}}, i8* [[RES]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* null)
   // BLOCKS: ret
@@ -158,7 +158,7 @@ int main() {
 //         [[KMP_TASK_T_TY]] task_data;
 //         [[KMP_TASK_MAIN_TY]] privates;
 //       };
-// CHECK: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc([[LOC]], i32 [[GTID]], i32 9, i64 104, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
+// CHECK: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc([[LOC]], i32 [[GTID]], i32 9, i64 112, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
 // CHECK: [[RES_KMP_TASK:%.+]] = bitcast i8* [[RES]] to [[KMP_TASK_MAIN_TY]]*
 
 // CHECK: [[TASK:%.+]] = getelementptr inbounds [[KMP_TASK_MAIN_TY]], [[KMP_TASK_MAIN_TY]]* [[RES_KMP_TASK]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
@@ -182,7 +182,8 @@ int main() {
 
 // Provide pointer to destructor function, which will destroy private variables at the end of the task.
 // CHECK: [[DESTRUCTORS_REF:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{.+}} 0, i{{.+}} 3
-// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_REF]],
+// CHECK: [[DESTRUCTORS_PTR:%.+]] = bitcast %union{{.+}}* [[DESTRUCTORS_REF]] to i32 (i32, i8*)**
+// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_MAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_PTR]],
 
 // Start task.
 // CHECK: call void @__kmpc_taskloop([[LOC]], i32 [[GTID]], i8* [[RES]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* bitcast (void ([[KMP_TASK_MAIN_TY]]*, [[KMP_TASK_MAIN_TY]]*, i32)* [[MAIN_DUP:@.+]] to i8*))
@@ -312,7 +313,8 @@ int main() {
 
 // Provide pointer to destructor function, which will destroy private variables at the end of the task.
 // CHECK: [[DESTRUCTORS_REF:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{.+}} 0, i{{.+}} 3
-// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_TMAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_REF]],
+// CHECK: [[DESTRUCTORS_PTR:%.+]] = bitcast %union{{.+}}* [[DESTRUCTORS_REF]] to i32 (i32, i8*)**
+// CHECK: store i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_TMAIN_TY]]*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** [[DESTRUCTORS_PTR]],
 
 // Start task.
 // CHECK: call void @__kmpc_taskloop([[LOC]], i32 [[GTID]], i8* [[RES]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* bitcast (void ([[KMP_TASK_TMAIN_TY]]*, [[KMP_TASK_TMAIN_TY]]*, i32)* [[TMAIN_DUP:@.+]] to i8*))