diff --git a/clang/lib/CodeGen/CGStmtOpenMP.cpp b/clang/lib/CodeGen/CGStmtOpenMP.cpp
--- a/clang/lib/CodeGen/CGStmtOpenMP.cpp
+++ b/clang/lib/CodeGen/CGStmtOpenMP.cpp
@@ -4852,6 +4852,8 @@
       // a pointer to this memory.
       for (auto &Pair : UntiedLocalVars) {
         QualType VDType = Pair.first->getType().getNonReferenceType();
+        if (Pair.first->getType()->isLValueReferenceType())
+          VDType = CGF.getContext().getPointerType(VDType);
         if (isAllocatableDecl(Pair.first)) {
           llvm::Value *Ptr = CGF.Builder.CreateLoad(Pair.second.first);
           Address Replacement(
diff --git a/clang/test/OpenMP/taskloop_untied_codegen.cpp b/clang/test/OpenMP/taskloop_untied_codegen.cpp
new file mode 100644
--- /dev/null
+++ b/clang/test/OpenMP/taskloop_untied_codegen.cpp
@@ -0,0 +1,26 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 2
+// RUN: %clang_cc1 -verify -triple aarch64-unknown-linux-gnu -fopenmp -x c++ -std=c++11 -emit-llvm %s -o - | FileCheck %s
+// expected-no-diagnostics
+
+// CHECK-LABEL: define dso_local void @_Z15taskloop_untiedv
+// CHECK-SAME: () #[[ATTR0:[0-9]+]] {
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[WORK:%.*]] = alloca [100 x float], align 4
+// CHECK-NEXT:    [[AGG_CAPTURED:%.*]] = alloca [[STRUCT_ANON:%.*]], align 1
+// CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
+// CHECK-NEXT:    [[TMP1:%.*]] = call ptr @__kmpc_omp_task_alloc(ptr @[[GLOB1]], i32 [[TMP0]], i32 0, i64 472, i64 1, ptr @.omp_task_entry.)
+// CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], ptr [[TMP1]], i32 0, i32 0
+// CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES]], ptr [[TMP1]], i32 0, i32 1
+// CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T:%.*]], ptr [[TMP3]], i32 0, i32 3
+// CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[TMP4]], ptr align 4 [[WORK]], i64 400, i1 false)
+// CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T:%.*]], ptr [[TMP2]], i32 0, i32 2
+// CHECK-NEXT:    store i32 0, ptr [[TMP5]], align 8
+// CHECK-NEXT:    [[TMP6:%.*]] = call i32 @__kmpc_omp_task(ptr @[[GLOB1]], i32 [[TMP0]], ptr [[TMP1]])
+// CHECK-NEXT:    ret void
+//
+void taskloop_untied() {
+  float work[100];
+#pragma omp task untied
+  for (auto cb : work)
+    cb = 1.0;
+}