[clang] Convert remaining OpenMP tests to opaque pointers

Reviewed By: nikic

Differential Revision: https://reviews.llvm.org/D150733

14 files changed, 3136 insertions, 3666 deletions

diff --git a/clang/test/OpenMP/declare_reduction_codegen.cpp b/clang/test/OpenMP/declare_reduction_codegen.cpp
index c07bfdf99ee2..acaf501bb431 100644
--- a/clang/test/OpenMP/declare_reduction_codegen.cpp
+++ b/clang/test/OpenMP/declare_reduction_codegen.cpp
@@ -1,14 +1,14 @@
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -x c++ -emit-llvm %s -triple x86_64-linux -fexceptions -fcxx-exceptions -o - -femit-all-decls -disable-llvm-passes | FileCheck %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -std=c++11 -triple x86_64-linux -fexceptions -fcxx-exceptions -emit-pch -o %t %s -femit-all-decls -disable-llvm-passes
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -triple x86_64-linux -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -femit-all-decls -disable-llvm-passes | FileCheck --check-prefixes=CHECK-LOAD,OMP50-LOAD %s
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -emit-llvm %s -triple x86_64-linux -fexceptions -fcxx-exceptions -o - -femit-all-decls -disable-llvm-passes | FileCheck %s
+// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-linux -fexceptions -fcxx-exceptions -emit-pch -o %t %s -femit-all-decls -disable-llvm-passes
+// RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-linux -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -femit-all-decls -disable-llvm-passes | FileCheck --check-prefixes=CHECK-LOAD,OMP50-LOAD %s
 
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=45 -x c++ -emit-llvm %s -triple x86_64-linux -fexceptions -fcxx-exceptions -o - -femit-all-decls -disable-llvm-passes | FileCheck %s --check-prefixes=CHECK,OMP45
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple x86_64-linux -fexceptions -fcxx-exceptions -emit-pch -o %t %s -femit-all-decls -disable-llvm-passes
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=45 -x c++ -triple x86_64-linux -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -femit-all-decls -disable-llvm-passes | FileCheck --check-prefixes=CHECK-LOAD,OMP45-LOAD %s
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -emit-llvm %s -triple x86_64-linux -fexceptions -fcxx-exceptions -o - -femit-all-decls -disable-llvm-passes | FileCheck %s --check-prefixes=CHECK,OMP45
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple x86_64-linux -fexceptions -fcxx-exceptions -emit-pch -o %t %s -femit-all-decls -disable-llvm-passes
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple x86_64-linux -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -femit-all-decls -disable-llvm-passes | FileCheck --check-prefixes=CHECK-LOAD,OMP45-LOAD %s
 
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -x c++ -emit-llvm %s -triple x86_64-linux -fexceptions -fcxx-exceptions -o - -femit-all-decls -disable-llvm-passes | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -std=c++11 -triple x86_64-linux -fexceptions -fcxx-exceptions -emit-pch -o %t %s -femit-all-decls -disable-llvm-passes
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple x86_64-linux -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -femit-all-decls -disable-llvm-passes | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -emit-llvm %s -triple x86_64-linux -fexceptions -fcxx-exceptions -o - -femit-all-decls -disable-llvm-passes | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-linux -fexceptions -fcxx-exceptions -emit-pch -o %t %s -femit-all-decls -disable-llvm-passes
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-linux -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -femit-all-decls -disable-llvm-passes | FileCheck --check-prefix SIMD-ONLY0 %s
 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
 // expected-no-diagnostics
 
@@ -30,7 +30,7 @@ void add(short &out, short &in) {}
 // OMP45: foo_reduction_array
 void foo_reduction_array() {
   short y[1];
-  // OMP45: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(
+  // OMP45: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(
 #pragma omp parallel for reduction(my_add : y)
   for (int i = 0; i < 1; i++) {
   }
@@ -39,32 +39,32 @@ void foo_reduction_array() {
 // OMP45: define internal void @
 
 #pragma omp declare reduction(+ : int, char : omp_out *= omp_in)
-// CHECK: define internal {{.*}}void @{{[^(]+}}(i32* noalias noundef %0, i32* noalias noundef %1)
-// CHECK: [[MUL:%.+]] = mul nsw i32
-// CHECK-NEXT: store i32 [[MUL]], i32*
+// CHECK: define internal {{.*}}void @{{[^(]+}}(ptr noalias noundef %0, ptr noalias noundef %1)
+// CHECK: [[MUL:%.+]] = mul nsw i32 %{{.+}}, %{{.+}}
+// CHECK-NEXT: store i32 [[MUL]], ptr
 // CHECK-NEXT: ret void
 // CHECK-NEXT: }
-// CHECK-LOAD: define internal {{.*}}void @{{[^(]+}}(i32* noalias noundef %0, i32* noalias noundef %1)
-// CHECK-LOAD: [[MUL:%.+]] = mul nsw i32
-// CHECK-LOAD-NEXT: store i32 [[MUL]], i32*
+// CHECK-LOAD: define internal {{.*}}void @{{[^(]+}}(ptr noalias noundef %0, ptr noalias noundef %1)
+// CHECK-LOAD: [[MUL:%.+]] = mul nsw i32 %{{.+}}, %{{.+}}
+// CHECK-LOAD-NEXT: store i32 [[MUL]], ptr
 // CHECK-LOAD-NEXT: ret void
 // CHECK-LOAD-NEXT: }
 
-// CHECK: define internal {{.*}}void @{{[^(]+}}(i8* noalias noundef %0, i8* noalias noundef %1)
+// CHECK: define internal {{.*}}void @{{[^(]+}}(ptr noalias noundef %0, ptr noalias noundef %1)
 // CHECK: sext i8
 // CHECK: sext i8
-// CHECK: [[MUL:%.+]] = mul nsw i32
+// CHECK: [[MUL:%.+]] = mul nsw i32 %{{.+}}, %{{.+}}
 // CHECK-NEXT: [[TRUNC:%.+]] = trunc i32 [[MUL]] to i8
-// CHECK-NEXT: store i8 [[TRUNC]], i8*
+// CHECK-NEXT: store i8 [[TRUNC]], ptr
 // CHECK-NEXT: ret void
 // CHECK-NEXT: }
 
-// CHECK-LOAD: define internal {{.*}}void @{{[^(]+}}(i8* noalias noundef %0, i8* noalias noundef %1)
+// CHECK-LOAD: define internal {{.*}}void @{{[^(]+}}(ptr noalias noundef %0, ptr noalias noundef %1)
 // CHECK-LOAD: sext i8
 // CHECK-LOAD: sext i8
-// CHECK-LOAD: [[MUL:%.+]] = mul nsw i32
+// CHECK-LOAD: [[MUL:%.+]] = mul nsw i32 %{{.+}}, %{{.+}}
 // CHECK-LOAD-NEXT: [[TRUNC:%.+]] = trunc i32 [[MUL]] to i8
-// CHECK-LOAD-NEXT: store i8 [[TRUNC]], i8*
+// CHECK-LOAD-NEXT: store i8 [[TRUNC]], ptr
 // CHECK-LOAD-NEXT: ret void
 // CHECK-LOAD-NEXT: }
 
@@ -79,26 +79,26 @@ struct SSS {
 
 SSS<int> d;
 
-// CHECK: define internal {{.*}}void @{{[^(]+}}(i32* noalias noundef %0, i32* noalias noundef %1)
-// CHECK: [[XOR:%.+]] = xor i32
-// CHECK-NEXT: store i32 [[XOR]], i32*
+// CHECK: define internal {{.*}}void @{{[^(]+}}(ptr noalias noundef %0, ptr noalias noundef %1)
+// CHECK: [[XOR:%.+]] = xor i32 %{{.+}}, %{{.+}}
+// CHECK-NEXT: store i32 [[XOR]], ptr
 // CHECK-NEXT: ret void
 // CHECK-NEXT: }
 
-// CHECK: define internal {{.*}}void @{{[^(]+}}(i32* noalias noundef %0, i32* noalias noundef %1)
-// CHECK: [[ADD:%.+]] = add nsw i32 24,
-// CHECK-NEXT: store i32 [[ADD]], i32*
+// CHECK: define internal {{.*}}void @{{[^(]+}}(ptr noalias noundef %0, ptr noalias noundef %1)
+// CHECK: [[ADD:%.+]] = add nsw i32 24, %{{.+}}
+// CHECK-NEXT: store i32 [[ADD]], ptr
 // CHECK-NEXT: ret void
 // CHECK-NEXT: }
 
-// CHECK: define internal {{.*}}void @{{[^(]+}}(i32* noalias noundef %0, i32* noalias noundef %1)
-// CHECK: call void @_ZN3SSSIiE6ssssssERi(i32* noundef nonnull align {{[0-9]+}} dereferenceable{{.*}})
+// CHECK: define internal {{.*}}void @{{[^(]+}}(ptr noalias noundef %0, ptr noalias noundef %1)
+// CHECK: call void @_ZN3SSSIiE6ssssssERi(ptr noundef nonnull align {{[0-9]+}} dereferenceable{{.*}})
 // CHECK-NEXT: ret void
 // CHECK-NEXT: }
 
-// CHECK: define internal {{.*}}void @{{[^(]+}}(i32* noalias noundef %0, i32* noalias noundef %1)
-// CHECK: [[ADD:%.+]] = add nsw i32 18,
-// CHECK-NEXT: store i32 [[ADD]], i32*
+// CHECK: define internal {{.*}}void @{{[^(]+}}(ptr noalias noundef %0, ptr noalias noundef %1)
+// CHECK: [[ADD:%.+]] = add nsw i32 18, %{{.+}}
+// CHECK-NEXT: store i32 [[ADD]], ptr
 // CHECK-NEXT: ret void
 // CHECK-NEXT: }
 
@@ -106,26 +106,26 @@ template <typename T>
 void init(T &lhs, T &rhs) {}
 
 #pragma omp declare reduction(fun : SSS < int > : omp_out = omp_in) initializer(init(omp_priv, omp_orig))
-// CHECK: define internal {{.*}}void @{{[^(]+}}([[SSS_INT]]* noalias noundef %0, [[SSS_INT]]* noalias noundef %1)
+// CHECK: define internal {{.*}}void @{{[^(]+}}(ptr noalias noundef %0, ptr noalias noundef %1)
 // CHECK: call void @llvm.memcpy
 // CHECK-NEXT: ret void
 // CHECK-NEXT: }
-// CHECK: define internal {{.*}}void @{{[^(]+}}([[SSS_INT]]* noalias noundef %0, [[SSS_INT]]* noalias noundef %1)
+// CHECK: define internal {{.*}}void @{{[^(]+}}(ptr noalias noundef %0, ptr noalias noundef %1)
 // CHECK: call {{.*}}void @_Z4initI3SSSIiEEvRT_S3_(
 // CHECK-NEXT: ret void
 // CHECK-NEXT: }
 
-// CHECK-LOAD: define internal {{.*}}void @{{[^(]+}}([[SSS_INT]]* noalias noundef %0, [[SSS_INT]]* noalias noundef %1)
+// CHECK-LOAD: define internal {{.*}}void @{{[^(]+}}(ptr noalias noundef %0, ptr noalias noundef %1)
 // CHECK-LOAD: call void @llvm.memcpy
 // CHECK-LOAD-NEXT: ret void
 // CHECK-LOAD-NEXT: }
-// CHECK-LOAD: define internal {{.*}}void @{{[^(]+}}([[SSS_INT]]* noalias noundef %0, [[SSS_INT]]* noalias noundef %1)
+// CHECK-LOAD: define internal {{.*}}void @{{[^(]+}}(ptr noalias noundef %0, ptr noalias noundef %1)
 // CHECK-LOAD: call {{.*}}void @_Z4initI3SSSIiEEvRT_S3_(
 // CHECK-LOAD-NEXT: ret void
 // CHECK-LOAD-NEXT: }
 
-// CHECK: define {{.*}}void @_Z4initI3SSSIiEEvRT_S3_(%struct.SSS* {{.+}}, %struct.SSS* {{.+}})
-// CHECK-LOAD: define {{.*}}void @_Z4initI3SSSIiEEvRT_S3_(%struct.SSS* {{.+}}, %struct.SSS* {{.+}})
+// CHECK: define {{.*}}void @_Z4initI3SSSIiEEvRT_S3_(ptr {{.+}}, ptr {{.+}})
+// CHECK-LOAD: define {{.*}}void @_Z4initI3SSSIiEEvRT_S3_(ptr {{.+}}, ptr {{.+}})
 
 template <typename T>
 T foo(T a) {
@@ -167,9 +167,9 @@ int main() {
 #pragma omp parallel reduction(fun : sss)
   {
   }
-  // CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(
-  // CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(
-  // CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call({{[^@]*}} @{{[^@]*}}[[REGION:@[^ ]+]]
+  // CHECK: call {{.*}}void (ptr, i32, ptr, ...) @__kmpc_fork_call(
+  // CHECK: call {{.*}}void (ptr, i32, ptr, ...) @__kmpc_fork_call(
+  // CHECK: call {{.*}}void (ptr, i32, ptr, ...) @__kmpc_fork_call({{[^@]*}} @{{[^@]*}}[[REGION:@[^,]+]]
   // CHECK-LABEL: work
   work<A>();
   // CHECK-LABEL: foo
@@ -178,66 +178,66 @@ int main() {
 
 // CHECK: define internal {{.*}}void [[REGION]](
 // CHECK: [[SSS_PRIV:%.+]] = alloca %struct.SSS,
-// CHECK: invoke {{.*}} @_ZN3SSSIiEC1Ev(%struct.SSS* {{[^,]*}} [[SSS_PRIV]])
+// CHECK: invoke {{.*}} @_ZN3SSSIiEC1Ev(ptr {{[^,]*}} [[SSS_PRIV]])
 // CHECK-NOT: {{call |invoke }}
 // CHECK: call {{.*}}i32 @__kmpc_reduce_nowait(
 
 // CHECK-LABEL: i32 @{{.+}}foo{{[^(].+}}(i32
 // CHECK-LOAD-LABEL: i32 @{{.+}}foo{{[^(].+}}(i32
 
-// OMP45-LOAD: define internal {{.*}}void @{{[^(]+}}(i32* noalias noundef %0, i32* noalias noundef %1)
-// OMP45-LOAD: [[XOR:%.+]] = xor i32
-// OMP45-LOAD-NEXT: store i32 [[XOR]], i32*
+// OMP45-LOAD: define internal {{.*}}void @{{[^(]+}}(ptr noalias noundef %0, ptr noalias noundef %1)
+// OMP45-LOAD: [[XOR:%.+]] = xor i32 %{{.+}}, %{{.+}}
+// OMP45-LOAD-NEXT: store i32 [[XOR]], ptr
 // OMP45-LOAD-NEXT: ret void
 // OMP45-LOAD-NEXT: }
 
-// OMP45-LOAD: define internal {{.*}}void @{{[^(]+}}(i32* noalias noundef %0, i32* noalias noundef %1)
-// OMP45-LOAD: [[ADD:%.+]] = add nsw i32 24,
-// OMP45-LOAD-NEXT: store i32 [[ADD]], i32*
+// OMP45-LOAD: define internal {{.*}}void @{{[^(]+}}(ptr noalias noundef %0, ptr noalias noundef %1)
+// OMP45-LOAD: [[ADD:%.+]] = add nsw i32 24, %{{.+}}
+// OMP45-LOAD-NEXT: store i32 [[ADD]], ptr
 // OMP45-LOAD-NEXT: ret void
 // OMP45-LOAD-NEXT: }
 
-// CHECK: define internal {{.*}}void @{{[^(]+}}(i32* noalias noundef %0, i32* noalias noundef %1)
-// CHECK: [[ADD:%.+]] = add nsw i32
-// CHECK-NEXT: store i32 [[ADD]], i32*
+// CHECK: define internal {{.*}}void @{{[^(]+}}(ptr noalias noundef %0, ptr noalias noundef %1)
+// CHECK: [[ADD:%.+]] = add nsw i32 %{{.+}}, %{{.+}}
+// CHECK-NEXT: store i32 [[ADD]], ptr
 // CHECK-NEXT: ret void
 // CHECK-NEXT: }
-// CHECK-LOAD: define internal {{.*}}void @{{[^(]+}}(i32* noalias noundef %0, i32* noalias noundef %1)
-// CHECK-LOAD: [[ADD:%.+]] = add nsw i32
-// CHECK-LOAD-NEXT: store i32 [[ADD]], i32*
+// CHECK-LOAD: define internal {{.*}}void @{{[^(]+}}(ptr noalias noundef %0, ptr noalias noundef %1)
+// CHECK-LOAD: [[ADD:%.+]] = add nsw i32 %{{.+}}, %{{.+}}
+// CHECK-LOAD-NEXT: store i32 [[ADD]], ptr
 // CHECK-LOAD-NEXT: ret void
 // CHECK-LOAD-NEXT: }
 
-// CHECK: define internal {{.*}}void @{{[^(]+}}(i32* noalias noundef %0, i32* noalias noundef %1)
-// CHECK: [[MUL:%.+]] = mul nsw i32 15,
-// CHECK-NEXT: store i32 [[MUL]], i32*
+// CHECK: define internal {{.*}}void @{{[^(]+}}(ptr noalias noundef %0, ptr noalias noundef %1)
+// CHECK: [[MUL:%.+]] = mul nsw i32 15, %{{.+}}
+// CHECK-NEXT: store i32 [[MUL]], ptr
 // CHECK-NEXT: ret void
 // CHECK-NEXT: }
-// CHECK-LOAD: define internal {{.*}}void @{{[^(]+}}(i32* noalias noundef %0, i32* noalias noundef %1)
-// CHECK-LOAD: [[MUL:%.+]] = mul nsw i32 15,
-// CHECK-LOAD-NEXT: store i32 [[MUL]], i32*
+// CHECK-LOAD: define internal {{.*}}void @{{[^(]+}}(ptr noalias noundef %0, ptr noalias noundef %1)
+// CHECK-LOAD: [[MUL:%.+]] = mul nsw i32 15, %{{.+}}
+// CHECK-LOAD-NEXT: store i32 [[MUL]], ptr
 // CHECK-LOAD-NEXT: ret void
 // CHECK-LOAD-NEXT: }
 
-// CHECK: define internal {{.*}}void @{{[^(]+}}(i32* noalias noundef %0, i32* noalias noundef %1)
-// CHECK: [[DIV:%.+]] = sdiv i32
-// CHECK-NEXT: store i32 [[DIV]], i32*
+// CHECK: define internal {{.*}}void @{{[^(]+}}(ptr noalias noundef %0, ptr noalias noundef %1)
+// CHECK: [[DIV:%.+]] = sdiv i32 %{{.+}}, %{{.+}}
+// CHECK-NEXT: store i32 [[DIV]], ptr
 // CHECK-NEXT: ret void
 // CHECK-NEXT: }
-// CHECK-LOAD: define internal {{.*}}void @{{[^(]+}}(i32* noalias noundef %0, i32* noalias noundef %1)
-// CHECK-LOAD: [[DIV:%.+]] = sdiv i32
-// CHECK-LOAD-NEXT: store i32 [[DIV]], i32*
+// CHECK-LOAD: define internal {{.*}}void @{{[^(]+}}(ptr noalias noundef %0, ptr noalias noundef %1)
+// CHECK-LOAD: [[DIV:%.+]] = sdiv i32 %{{.+}}, %{{.+}}
+// CHECK-LOAD-NEXT: store i32 [[DIV]], ptr
 // CHECK-LOAD-NEXT: ret void
 // CHECK-LOAD-NEXT: }
 
-// CHECK: define internal {{.*}}void @{{[^(]+}}(i32* noalias noundef %0, i32* noalias noundef %1)
-// CHECK: [[SUB:%.+]] = sub nsw i32 11,
-// CHECK-NEXT: store i32 [[SUB]], i32*
+// CHECK: define internal {{.*}}void @{{[^(]+}}(ptr noalias noundef %0, ptr noalias noundef %1)
+// CHECK: [[SUB:%.+]] = sub nsw i32 11, %{{.+}}
+// CHECK-NEXT: store i32 [[SUB]], ptr
 // CHECK-NEXT: ret void
 // CHECK-NEXT: }
-// CHECK-LOAD: define internal {{.*}}void @{{[^(]+}}(i32* noalias noundef %0, i32* noalias noundef %1)
-// CHECK-LOAD: [[SUB:%.+]] = sub nsw i32 11,
-// CHECK-LOAD-NEXT: store i32 [[SUB]], i32*
+// CHECK-LOAD: define internal {{.*}}void @{{[^(]+}}(ptr noalias noundef %0, ptr noalias noundef %1)
+// CHECK-LOAD: [[SUB:%.+]] = sub nsw i32 11, %{{.+}}
+// CHECK-LOAD-NEXT: store i32 [[SUB]], ptr
 // CHECK-LOAD-NEXT: ret void
 // CHECK-LOAD-NEXT: }
 
diff --git a/clang/test/OpenMP/parallel_for_simd_codegen.cpp b/clang/test/OpenMP/parallel_for_simd_codegen.cpp
index 9c2c6414751e..893b4bd556f6 100644
--- a/clang/test/OpenMP/parallel_for_simd_codegen.cpp
+++ b/clang/test/OpenMP/parallel_for_simd_codegen.cpp
@@ -1,22 +1,22 @@
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=45 -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s --check-prefix=OMP45 --check-prefix=CHECK
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=45 -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -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=45 -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck %s --check-prefix=TERM_DEBUG
-
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -DOMP5 -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s --check-prefix=OMP50 --check-prefix=CHECK
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -DOMP5 -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -DOMP5 -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s
-// RUN: %clang_cc1 -no-opaque-pointers -verify -triple x86_64-apple-darwin10 -fopenmp -DOMP5 -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck %s --check-prefix=TERM_DEBUG
-
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=45 -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -verify -triple x86_64-apple-darwin10 -fopenmp-simd -fopenmp-version=45 -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -DOMP5 -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -DOMP5 -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -DOMP5 -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -verify -triple x86_64-apple-darwin10 -fopenmp-simd -DOMP5 -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s --check-prefix=OMP45 --check-prefix=CHECK
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s
+// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -fopenmp-version=45 -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck %s --check-prefix=TERM_DEBUG
+
+// RUN: %clang_cc1 -verify -fopenmp -DOMP5 -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s --check-prefix=OMP50 --check-prefix=CHECK
+// RUN: %clang_cc1 -fopenmp -DOMP5 -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -DOMP5 -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s
+// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -DOMP5 -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck %s --check-prefix=TERM_DEBUG
+
+// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -fopenmp-version=45 -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+
+// RUN: %clang_cc1 -verify -fopenmp-simd -DOMP5 -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -fopenmp-simd -DOMP5 -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -DOMP5 -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -DOMP5 -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
 // expected-no-diagnostics
 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
 #ifndef HEADER
@@ -25,249 +25,249 @@
 long long get_val() { extern void mayThrow(); mayThrow(); return 0; }
 double *g_ptr;
 
-// CHECK-LABEL: define {{.*void}} @{{.*}}simple{{.*}}(float* noundef {{.+}}, float* noundef {{.+}}, float* noundef {{.+}}, float* noundef {{.+}})
+// CHECK-LABEL: define {{.*void}} @{{.*}}simple{{.*}}(ptr noundef {{.+}}, ptr noundef {{.+}}, ptr noundef {{.+}}, ptr noundef {{.+}})
 void simple(float *a, float *b, float *c, float *d) {
-// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(
+// CHECK: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(
 // CHECK: [[K0:%.+]] = call {{.*}}i64 @{{.*}}get_val
-// CHECK-NEXT: store i64 [[K0]], i64* [[K_VAR:%[^,]+]]
-// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(
-// CHECK: store i32 12, i32* [[LIN_VAR:%[^,]+]]
-// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(
-// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(
-// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(
-// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(
-// CHECK: store i32 -1, i32* [[A:%.+]],
-// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(
-// CHECK: store i32 -1, i32* [[R:%[^,]+]],
-// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(
+// CHECK-NEXT: store i64 [[K0]], ptr [[K_VAR:%[^,]+]]
+// CHECK: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(
+// CHECK: store i32 12, ptr [[LIN_VAR:%[^,]+]]
+// CHECK: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(
+// CHECK: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(
+// CHECK: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(
+// CHECK: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(
+// CHECK: store i32 -1, ptr [[A:%.+]],
+// CHECK: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(
+// CHECK: store i32 -1, ptr [[R:%[^,]+]],
+// CHECK: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(
   #pragma omp parallel for simd
-// CHECK: call void @__kmpc_for_static_init_4(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i32* [[LB:%[^,]+]], i32* [[UB:%[^,]+]], i32* [[STRIDE:%[^,]+]], i32 1, i32 1)
-// CHECK: [[UB_VAL:%.+]] = load i32, i32* [[UB]],
+// CHECK: call void @__kmpc_for_static_init_4(ptr {{[^,]+}}, i32 %{{[^,]+}}, i32 34, ptr %{{[^,]+}}, ptr [[LB:%[^,]+]], ptr [[UB:%[^,]+]], ptr [[STRIDE:%[^,]+]], i32 1, i32 1)
+// CHECK: [[UB_VAL:%.+]] = load i32, ptr [[UB]],
 // CHECK: [[CMP:%.+]] = icmp sgt i32 [[UB_VAL]], 5
 // CHECK: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]]
 // CHECK: [[TRUE]]:
 // CHECK: br label %[[SWITCH:[^,]+]]
 // CHECK: [[FALSE]]:
-// CHECK: [[UB_VAL:%.+]] = load i32, i32* [[UB]],
+// CHECK: [[UB_VAL:%.+]] = load i32, ptr [[UB]],
 // CHECK: br label %[[SWITCH]]
 // CHECK: [[SWITCH]]:
 // CHECK: [[UP:%.+]] = phi i32 [ 5, %[[TRUE]] ], [ [[UB_VAL]], %[[FALSE]] ]
-// CHECK: store i32 [[UP]], i32* [[UB]],
-// CHECK: [[LB_VAL:%.+]] = load i32, i32* [[LB]],
-// CHECK: store i32 [[LB_VAL]], i32* [[OMP_IV:%[^,]+]],
+// CHECK: store i32 [[UP]], ptr [[UB]],
+// CHECK: [[LB_VAL:%.+]] = load i32, ptr [[LB]],
+// CHECK: store i32 [[LB_VAL]], ptr [[OMP_IV:%[^,]+]],
 
-// CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]
-// CHECK: [[UB_VAL:%.+]] = load i32, i32* [[UB]]
+// CHECK: [[IV:%.+]] = load i32, ptr [[OMP_IV]]
+// CHECK: [[UB_VAL:%.+]] = load i32, ptr [[UB]]
 // CHECK-NEXT: [[CMP:%.+]] = icmp sle i32 [[IV]], [[UB_VAL]]
 // CHECK-NEXT: br i1 [[CMP]], label %[[SIMPLE_LOOP1_BODY:.+]], label %[[SIMPLE_LOOP1_END:[^,]+]]
   for (int i = 3; i < 32; i += 5) {
 // CHECK: [[SIMPLE_LOOP1_BODY]]:
 // Start of body: calculate i from IV:
-// CHECK: [[IV1_1:%.+]] = load i32, i32* [[OMP_IV]]
+// CHECK: [[IV1_1:%.+]] = load i32, ptr [[OMP_IV]]
 // CHECK: [[CALC_I_1:%.+]] = mul nsw i32 [[IV1_1]], 5
 // CHECK-NEXT: [[CALC_I_2:%.+]] = add nsw i32 3, [[CALC_I_1]]
-// CHECK-NEXT: store i32 [[CALC_I_2]], i32* [[LC_I:.+]]
+// CHECK-NEXT: store i32 [[CALC_I_2]], ptr [[LC_I:.+]]
 // ... loop body ...
 // End of body: store into a[i]:
-// CHECK: store float [[RESULT:%.+]], float*
+// CHECK: store float [[RESULT:%.+]], ptr
     a[i] = b[i] * c[i] * d[i];
-// CHECK: [[IV1_2:%.+]] = load i32, i32* [[OMP_IV]]
+// CHECK: [[IV1_2:%.+]] = load i32, ptr [[OMP_IV]]
 // CHECK-NEXT: [[ADD1_2:%.+]] = add nsw i32 [[IV1_2]], 1
-// CHECK-NEXT: store i32 [[ADD1_2]], i32* [[OMP_IV]]
+// CHECK-NEXT: store i32 [[ADD1_2]], ptr [[OMP_IV]]
 // br label %{{.+}}, !llvm.loop !{{.+}}
   }
 // CHECK: [[SIMPLE_LOOP1_END]]:
-// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}})
+// CHECK: call void @__kmpc_for_static_fini(ptr {{.+}}, i32 %{{.+}})
 
   long long k = get_val();
 
   #pragma omp parallel for simd linear(k : 3) schedule(dynamic)
-// CHECK: [[K0LOAD:%.+]] = load i64, i64* [[K_VAR:%[^,]+]]
-// CHECK-NEXT: store i64 [[K0LOAD]], i64* [[LIN0:%[^,]+]]
+// CHECK: [[K0LOAD:%.+]] = load i64, ptr [[K_VAR:%[^,]+]]
+// CHECK-NEXT: store i64 [[K0LOAD]], ptr [[LIN0:%[^,]+]]
 
-// CHECK: call void @__kmpc_dispatch_init_4(%struct.ident_t* {{.+}}, i32 %{{.+}}, i32 {{35|1073741859}}, i32 0, i32 8, i32 1, i32 1)
-// CHECK: [[NEXT:%.+]] = call i32 @__kmpc_dispatch_next_4(%struct.ident_t* {{.+}}, i32 %{{.+}}, i32* %{{.+}}, i32* [[LB:%.+]], i32* [[UB:%.+]], i32* %{{.+}})
+// CHECK: call void @__kmpc_dispatch_init_4(ptr {{.+}}, i32 %{{.+}}, i32 {{35|1073741859}}, i32 0, i32 8, i32 1, i32 1)
+// CHECK: [[NEXT:%.+]] = call i32 @__kmpc_dispatch_next_4(ptr {{.+}}, i32 %{{.+}}, ptr %{{.+}}, ptr [[LB:%.+]], ptr [[UB:%.+]], ptr %{{.+}})
 // CHECK: [[COND:%.+]] = icmp ne i32 [[NEXT]], 0
 // CHECK: br i1 [[COND]], label %[[CONT:.+]], label %[[END:.+]]
 // CHECK: [[CONT]]:
-// CHECK: [[LB_VAL:%.+]] = load i32, i32* [[LB]],
-// CHECK: store i32 [[LB_VAL]], i32* [[OMP_IV2:%[^,]+]],
+// CHECK: [[LB_VAL:%.+]] = load i32, ptr [[LB]],
+// CHECK: store i32 [[LB_VAL]], ptr [[OMP_IV2:%[^,]+]],
 
-// CHECK: [[IV2:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.access.group
-// CHECK: [[UB_VAL:%.+]] = load i32, i32* [[UB]]{{.*}}!llvm.access.group
+// CHECK: [[IV2:%.+]] = load i32, ptr [[OMP_IV2]]{{.*}}!llvm.access.group
+// CHECK: [[UB_VAL:%.+]] = load i32, ptr [[UB]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[CMP2:%.+]] = icmp sle i32 [[IV2]], [[UB_VAL]]
 // CHECK-NEXT: br i1 [[CMP2]], label %[[SIMPLE_LOOP2_BODY:.+]], label %[[SIMPLE_LOOP2_END:[^,]+]]
   for (int i = 10; i > 1; i--) {
 // CHECK: [[SIMPLE_LOOP2_BODY]]:
 // Start of body: calculate i from IV:
-// CHECK: [[IV2_0:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.access.group
+// CHECK: [[IV2_0:%.+]] = load i32, ptr [[OMP_IV2]]{{.*}}!llvm.access.group
 // FIXME: It is interesting, why the following "mul 1" was not constant folded?
 // CHECK-NEXT: [[IV2_1:%.+]] = mul nsw i32 [[IV2_0]], 1
 // CHECK-NEXT: [[LC_I_1:%.+]] = sub nsw i32 10, [[IV2_1]]
-// CHECK-NEXT: store i32 [[LC_I_1]], i32* {{.+}}, !llvm.access.group
+// CHECK-NEXT: store i32 [[LC_I_1]], ptr {{.+}}, !llvm.access.group
 //
-// CHECK-NEXT: [[LIN0_1:%.+]] = load i64, i64* [[LIN0]]{{.*}}!llvm.access.group
-// CHECK-NEXT: [[IV2_2:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.access.group
+// CHECK-NEXT: [[LIN0_1:%.+]] = load i64, ptr [[LIN0]]{{.*}}!llvm.access.group
+// CHECK-NEXT: [[IV2_2:%.+]] = load i32, ptr [[OMP_IV2]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[LIN_MUL1:%.+]] = mul nsw i32 [[IV2_2]], 3
 // CHECK-NEXT: [[LIN_EXT1:%.+]] = sext i32 [[LIN_MUL1]] to i64
 // CHECK-NEXT: [[LIN_ADD1:%.+]] = add nsw i64 [[LIN0_1]], [[LIN_EXT1]]
 // Update of the privatized version of linear variable!
-// CHECK-NEXT: store i64 [[LIN_ADD1]], i64* [[K_PRIVATIZED:%[^,]+]]
+// CHECK-NEXT: store i64 [[LIN_ADD1]], ptr [[K_PRIVATIZED:%[^,]+]]
     a[k]++;
     k = k + 3;
-// CHECK: [[IV2_2:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.access.group
+// CHECK: [[IV2_2:%.+]] = load i32, ptr [[OMP_IV2]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[ADD2_2:%.+]] = add nsw i32 [[IV2_2]], 1
-// CHECK-NEXT: store i32 [[ADD2_2]], i32* [[OMP_IV2]]{{.*}}!llvm.access.group
+// CHECK-NEXT: store i32 [[ADD2_2]], ptr [[OMP_IV2]]{{.*}}!llvm.access.group
 // br label {{.+}}, !llvm.loop ![[SIMPLE_LOOP2_ID]]
   }
 // CHECK: [[SIMPLE_LOOP2_END]]:
 //
 // Update linear vars after loop, as the loop was operating on a private version.
-// CHECK: [[LIN0_2:%.+]] = load i64, i64* [[K_PRIVATIZED]]
-// CHECK-NEXT: store i64 [[LIN0_2]], i64* %{{.+}}
+// CHECK: [[LIN0_2:%.+]] = load i64, ptr [[K_PRIVATIZED]]
+// CHECK-NEXT: store i64 [[LIN0_2]], ptr %{{.+}}
 
   int lin = 12;
   #pragma omp parallel for simd linear(lin : get_val()), linear(g_ptr)
 
 // CHECK: alloca i32,
 // Init linear private var.
-// CHECK: [[LIN_VAR:%.+]] = load i32*, i32** %
-// CHECK: [[LIN_LOAD:%.+]] = load i32, i32* [[LIN_VAR]]
-// CHECK-NEXT: store i32 [[LIN_LOAD]], i32* [[LIN_START:%[^,]+]]
+// CHECK: [[LIN_VAR:%.+]] = load ptr, ptr %lin
+// CHECK: [[LIN_LOAD:%.+]] = load i32, ptr [[LIN_VAR]]
+// CHECK-NEXT: store i32 [[LIN_LOAD]], ptr [[LIN_START:%[^,]+]]
 // Remember linear step.
 // CHECK: [[CALL_VAL:%.+]] = invoke
-// CHECK: store i64 [[CALL_VAL]], i64* [[LIN_STEP:%[^,]+]]
+// CHECK: store i64 [[CALL_VAL]], ptr [[LIN_STEP:%[^,]+]]
 
-// CHECK: [[GLIN_LOAD:%.+]] = load double*, double** [[GLIN_VAR:%.+]],
-// CHECK-NEXT: store double* [[GLIN_LOAD]], double** [[GLIN_START:%[^,]+]]
+// CHECK: [[GLIN_LOAD:%.+]] = load ptr, ptr [[GLIN_VAR:%.+]],
+// CHECK-NEXT: store ptr [[GLIN_LOAD]], ptr [[GLIN_START:%[^,]+]]
 
-// CHECK: call void @__kmpc_for_static_init_8u(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 1)
-// CHECK: [[UB_VAL:%.+]] = load i64, i64* [[UB]],
+// CHECK: call void @__kmpc_for_static_init_8u(ptr {{[^,]+}}, i32 %{{[^,]+}}, i32 34, ptr %{{[^,]+}}, ptr [[LB:%[^,]+]], ptr [[UB:%[^,]+]], ptr [[STRIDE:%[^,]+]], i64 1, i64 1)
+// CHECK: [[UB_VAL:%.+]] = load i64, ptr [[UB]],
 // CHECK: [[CMP:%.+]] = icmp ugt i64 [[UB_VAL]], 3
 // CHECK: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]]
 // CHECK: [[TRUE]]:
 // CHECK: br label %[[SWITCH:[^,]+]]
 // CHECK: [[FALSE]]:
-// CHECK: [[UB_VAL:%.+]] = load i64, i64* [[UB]],
+// CHECK: [[UB_VAL:%.+]] = load i64, ptr [[UB]],
 // CHECK: br label %[[SWITCH]]
 // CHECK: [[SWITCH]]:
 // CHECK: [[UP:%.+]] = phi i64 [ 3, %[[TRUE]] ], [ [[UB_VAL]], %[[FALSE]] ]
-// CHECK: store i64 [[UP]], i64* [[UB]],
-// CHECK: [[LB_VAL:%.+]] = load i64, i64* [[LB]],
-// CHECK: store i64 [[LB_VAL]], i64* [[OMP_IV3:%[^,]+]],
+// CHECK: store i64 [[UP]], ptr [[UB]],
+// CHECK: [[LB_VAL:%.+]] = load i64, ptr [[LB]],
+// CHECK: store i64 [[LB_VAL]], ptr [[OMP_IV3:%[^,]+]],
 
-// CHECK: [[IV3:%.+]] = load i64, i64* [[OMP_IV3]]
-// CHECK: [[UB_VAL:%.+]] = load i64, i64* [[UB]]
+// CHECK: [[IV3:%.+]] = load i64, ptr [[OMP_IV3]]
+// CHECK: [[UB_VAL:%.+]] = load i64, ptr [[UB]]
 // CHECK-NEXT: [[CMP3:%.+]] = icmp ule i64 [[IV3]], [[UB_VAL]]
 // CHECK-NEXT: br i1 [[CMP3]], label %[[SIMPLE_LOOP3_BODY:.+]], label %[[SIMPLE_LOOP3_END:[^,]+]]
   for (unsigned long long it = 2000; it >= 600; it-=400) {
 // CHECK: [[SIMPLE_LOOP3_BODY]]:
 // Start of body: calculate it from IV:
-// CHECK: [[IV3_0:%.+]] = load i64, i64* [[OMP_IV3]]
+// CHECK: [[IV3_0:%.+]] = load i64, ptr [[OMP_IV3]]
 // CHECK-NEXT: [[LC_IT_1:%.+]] = mul i64 [[IV3_0]], 400
 // CHECK-NEXT: [[LC_IT_2:%.+]] = sub i64 2000, [[LC_IT_1]]
-// CHECK-NEXT: store i64 [[LC_IT_2]], i64* {{.+}}
+// CHECK-NEXT: store i64 [[LC_IT_2]], ptr {{.+}}
 //
 // Linear start and step are used to calculate current value of the linear variable.
-// CHECK: [[LINSTART:.+]] = load i32, i32* [[LIN_START]]
-// CHECK: [[LINSTEP:.+]] = load i64, i64* [[LIN_STEP]]
-// CHECK-NOT: store i32 {{.+}}, i32* [[LIN_VAR]]
-// CHECK: store i32 {{.+}}, i32* [[LIN_PRIV:%[^,]+]],
-// CHECK: [[GLINSTART:.+]] = load double*, double** [[GLIN_START]]
-// CHECK-NEXT: [[IV3_1:%.+]] = load i64, i64* [[OMP_IV3]]
+// CHECK: [[LINSTART:.+]] = load i32, ptr [[LIN_START]]
+// CHECK: [[LINSTEP:.+]] = load i64, ptr [[LIN_STEP]]
+// CHECK-NOT: store i32 {{.+}}, ptr [[LIN_VAR]]
+// CHECK: store i32 {{.+}}, ptr [[LIN_PRIV:%[^,]+]],
+// CHECK: [[GLINSTART:.+]] = load ptr, ptr [[GLIN_START]]
+// CHECK-NEXT: [[IV3_1:%.+]] = load i64, ptr [[OMP_IV3]]
 // CHECK-NEXT: [[MUL:%.+]] = mul i64 [[IV3_1]], 1
 // CHECK: [[GEP:%.+]] = getelementptr{{.*}}[[GLINSTART]]
-// CHECK-NEXT: store double* [[GEP]], double** [[G_PTR_CUR:%[^,]+]]
+// CHECK-NEXT: store ptr [[GEP]], ptr [[G_PTR_CUR:%[^,]+]]
     *g_ptr++ = 0.0;
-// CHECK: [[GEP_VAL:%.+]] = load double{{.*}}[[G_PTR_CUR]]
+// CHECK: [[GEP_VAL:%.+]] = load ptr, ptr [[G_PTR_CUR]]
 // CHECK: store double{{.*}}[[GEP_VAL]]
     a[it + lin]++;
 // CHECK: [[FLT_INC:%.+]] = fadd float
 // CHECK-NEXT: store float [[FLT_INC]],
-// CHECK: [[IV3_2:%.+]] = load i64, i64* [[OMP_IV3]]
+// CHECK: [[IV3_2:%.+]] = load i64, ptr [[OMP_IV3]]
 // CHECK-NEXT: [[ADD3_2:%.+]] = add i64 [[IV3_2]], 1
-// CHECK-NEXT: store i64 [[ADD3_2]], i64* [[OMP_IV3]]
+// CHECK-NEXT: store i64 [[ADD3_2]], ptr [[OMP_IV3]]
   }
 // CHECK: [[SIMPLE_LOOP3_END]]:
-// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}})
+// CHECK: call void @__kmpc_for_static_fini(ptr {{.+}}, i32 %{{.+}})
 //
 // Linear start and step are used to calculate final value of the linear variables.
-// CHECK: [[LIN:%.+]] = load i32, i32* [[LIN_PRIV]]
-// CHECK: store i32 [[LIN]], i32* [[LIN_VAR]],
-// CHECK: [[GLIN:%.+]] = load double*, double** [[G_PTR_CUR]]
-// CHECK: store double* [[GLIN]], double** [[GLIN_VAR]],
+// CHECK: [[LIN:%.+]] = load i32, ptr [[LIN_PRIV]]
+// CHECK: store i32 [[LIN]], ptr [[LIN_VAR]],
+// CHECK: [[GLIN:%.+]] = load ptr, ptr [[G_PTR_CUR]]
+// CHECK: store ptr [[GLIN]], ptr [[GLIN_VAR]],
 
   #pragma omp parallel for simd
-// CHECK: call void @__kmpc_for_static_init_4(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i32* [[LB:%[^,]+]], i32* [[UB:%[^,]+]], i32* [[STRIDE:%[^,]+]], i32 1, i32 1)
-// CHECK: [[UB_VAL:%.+]] = load i32, i32* [[UB]],
+// CHECK: call void @__kmpc_for_static_init_4(ptr {{[^,]+}}, i32 %{{[^,]+}}, i32 34, ptr %{{[^,]+}}, ptr [[LB:%[^,]+]], ptr [[UB:%[^,]+]], ptr [[STRIDE:%[^,]+]], i32 1, i32 1)
+// CHECK: [[UB_VAL:%.+]] = load i32, ptr [[UB]],
 // CHECK: [[CMP:%.+]] = icmp sgt i32 [[UB_VAL]], 3
 // CHECK: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]]
 // CHECK: [[TRUE]]:
 // CHECK: br label %[[SWITCH:[^,]+]]
 // CHECK: [[FALSE]]:
-// CHECK: [[UB_VAL:%.+]] = load i32, i32* [[UB]],
+// CHECK: [[UB_VAL:%.+]] = load i32, ptr [[UB]],
 // CHECK: br label %[[SWITCH]]
 // CHECK: [[SWITCH]]:
 // CHECK: [[UP:%.+]] = phi i32 [ 3, %[[TRUE]] ], [ [[UB_VAL]], %[[FALSE]] ]
-// CHECK: store i32 [[UP]], i32* [[UB]],
-// CHECK: [[LB_VAL:%.+]] = load i32, i32* [[LB]],
-// CHECK: store i32 [[LB_VAL]], i32* [[OMP_IV4:%[^,]+]],
+// CHECK: store i32 [[UP]], ptr [[UB]],
+// CHECK: [[LB_VAL:%.+]] = load i32, ptr [[LB]],
+// CHECK: store i32 [[LB_VAL]], ptr [[OMP_IV4:%[^,]+]],
 
-// CHECK: [[IV4:%.+]] = load i32, i32* [[OMP_IV4]]
-// CHECK: [[UB_VAL:%.+]] = load i32, i32* [[UB]]
+// CHECK: [[IV4:%.+]] = load i32, ptr [[OMP_IV4]]
+// CHECK: [[UB_VAL:%.+]] = load i32, ptr [[UB]]
 // CHECK-NEXT: [[CMP4:%.+]] = icmp sle i32 [[IV4]], [[UB_VAL]]
 // CHECK-NEXT: br i1 [[CMP4]], label %[[SIMPLE_LOOP4_BODY:.+]], label %[[SIMPLE_LOOP4_END:[^,]+]]
   for (short it = 6; it <= 20; it-=-4) {
 // CHECK: [[SIMPLE_LOOP4_BODY]]:
 // Start of body: calculate it from IV:
-// CHECK: [[IV4_0:%.+]] = load i32, i32* [[OMP_IV4]]
+// CHECK: [[IV4_0:%.+]] = load i32, ptr [[OMP_IV4]]
 // CHECK-NEXT: [[LC_IT_1:%.+]] = mul nsw i32 [[IV4_0]], 4
 // CHECK-NEXT: [[LC_IT_2:%.+]] = add nsw i32 6, [[LC_IT_1]]
 // CHECK-NEXT: [[LC_IT_3:%.+]] = trunc i32 [[LC_IT_2]] to i16
-// CHECK-NEXT: store i16 [[LC_IT_3]], i16*
+// CHECK-NEXT: store i16 [[LC_IT_3]], ptr
 
-// CHECK: [[IV4_2:%.+]] = load i32, i32* [[OMP_IV4]]
+// CHECK: [[IV4_2:%.+]] = load i32, ptr [[OMP_IV4]]
 // CHECK-NEXT: [[ADD4_2:%.+]] = add nsw i32 [[IV4_2]], 1
-// CHECK-NEXT: store i32 [[ADD4_2]], i32* [[OMP_IV4]]
+// CHECK-NEXT: store i32 [[ADD4_2]], ptr [[OMP_IV4]]
   }
 // CHECK: [[SIMPLE_LOOP4_END]]:
-// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}})
+// CHECK: call void @__kmpc_for_static_fini(ptr {{.+}}, i32 %{{.+}})
 
   #pragma omp parallel for simd
-// CHECK: call void @__kmpc_for_static_init_4(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i32* [[LB:%[^,]+]], i32* [[UB:%[^,]+]], i32* [[STRIDE:%[^,]+]], i32 1, i32 1)
-// CHECK: [[UB_VAL:%.+]] = load i32, i32* [[UB]],
+// CHECK: call void @__kmpc_for_static_init_4(ptr {{[^,]+}}, i32 %{{[^,]+}}, i32 34, ptr %{{[^,]+}}, ptr [[LB:%[^,]+]], ptr [[UB:%[^,]+]], ptr [[STRIDE:%[^,]+]], i32 1, i32 1)
+// CHECK: [[UB_VAL:%.+]] = load i32, ptr [[UB]],
 // CHECK: [[CMP:%.+]] = icmp sgt i32 [[UB_VAL]], 25
 // CHECK: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]]
 // CHECK: [[TRUE]]:
 // CHECK: br label %[[SWITCH:[^,]+]]
 // CHECK: [[FALSE]]:
-// CHECK: [[UB_VAL:%.+]] = load i32, i32* [[UB]],
+// CHECK: [[UB_VAL:%.+]] = load i32, ptr [[UB]],
 // CHECK: br label %[[SWITCH]]
 // CHECK: [[SWITCH]]:
 // CHECK: [[UP:%.+]] = phi i32 [ 25, %[[TRUE]] ], [ [[UB_VAL]], %[[FALSE]] ]
-// CHECK: store i32 [[UP]], i32* [[UB]],
-// CHECK: [[LB_VAL:%.+]] = load i32, i32* [[LB]],
-// CHECK: store i32 [[LB_VAL]], i32* [[OMP_IV5:%[^,]+]],
+// CHECK: store i32 [[UP]], ptr [[UB]],
+// CHECK: [[LB_VAL:%.+]] = load i32, ptr [[LB]],
+// CHECK: store i32 [[LB_VAL]], ptr [[OMP_IV5:%[^,]+]],
 
-// CHECK: [[IV5:%.+]] = load i32, i32* [[OMP_IV5]]
-// CHECK: [[UB_VAL:%.+]] = load i32, i32* [[UB]]
+// CHECK: [[IV5:%.+]] = load i32, ptr [[OMP_IV5]]
+// CHECK: [[UB_VAL:%.+]] = load i32, ptr [[UB]]
 // CHECK-NEXT: [[CMP5:%.+]] = icmp sle i32 [[IV5]], [[UB_VAL]]
 // CHECK-NEXT: br i1 [[CMP5]], label %[[SIMPLE_LOOP5_BODY:.+]], label %[[SIMPLE_LOOP5_END:[^,]+]]
   for (unsigned char it = 'z'; it >= 'a'; it+=-1) {
 // CHECK: [[SIMPLE_LOOP5_BODY]]:
 // Start of body: calculate it from IV:
-// CHECK: [[IV5_0:%.+]] = load i32, i32* [[OMP_IV5]]
+// CHECK: [[IV5_0:%.+]] = load i32, ptr [[OMP_IV5]]
 // CHECK-NEXT: [[IV5_1:%.+]] = mul nsw i32 [[IV5_0]], 1
 // CHECK-NEXT: [[LC_IT_1:%.+]] = sub nsw i32 122, [[IV5_1]]
 // CHECK-NEXT: [[LC_IT_2:%.+]] = trunc i32 [[LC_IT_1]] to i8
-// CHECK-NEXT: store i8 [[LC_IT_2]], i8* {{.+}},
+// CHECK-NEXT: store i8 [[LC_IT_2]], ptr {{.+}},
 
-// CHECK: [[IV5_2:%.+]] = load i32, i32* [[OMP_IV5]]
+// CHECK: [[IV5_2:%.+]] = load i32, ptr [[OMP_IV5]]
 // CHECK-NEXT: [[ADD5_2:%.+]] = add nsw i32 [[IV5_2]], 1
-// CHECK-NEXT: store i32 [[ADD5_2]], i32* [[OMP_IV5]]
+// CHECK-NEXT: store i32 [[ADD5_2]], ptr [[OMP_IV5]]
   }
 // CHECK: [[SIMPLE_LOOP5_END]]:
-// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}})
+// CHECK: call void @__kmpc_for_static_fini(ptr {{.+}}, i32 %{{.+}})
 
 // CHECK-NOT: mul i32 %{{.+}}, 10
   #pragma omp parallel for simd
@@ -278,97 +278,97 @@ void simple(float *a, float *b, float *c, float *d) {
   {
   A = -1;
   #pragma omp parallel for simd lastprivate(A)
-// CHECK: call void @__kmpc_for_static_init_8(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 1)
-// CHECK: [[UB_VAL:%.+]] = load i64, i64* [[UB]],
+// CHECK: call void @__kmpc_for_static_init_8(ptr {{[^,]+}}, i32 %{{[^,]+}}, i32 34, ptr %{{[^,]+}}, ptr [[LB:%[^,]+]], ptr [[UB:%[^,]+]], ptr [[STRIDE:%[^,]+]], i64 1, i64 1)
+// CHECK: [[UB_VAL:%.+]] = load i64, ptr [[UB]],
 // CHECK: [[CMP:%.+]] = icmp sgt i64 [[UB_VAL]], 6
 // CHECK: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]]
 // CHECK: [[TRUE]]:
 // CHECK: br label %[[SWITCH:[^,]+]]
 // CHECK: [[FALSE]]:
-// CHECK: [[UB_VAL:%.+]] = load i64, i64* [[UB]],
+// CHECK: [[UB_VAL:%.+]] = load i64, ptr [[UB]],
 // CHECK: br label %[[SWITCH]]
 // CHECK: [[SWITCH]]:
 // CHECK: [[UP:%.+]] = phi i64 [ 6, %[[TRUE]] ], [ [[UB_VAL]], %[[FALSE]] ]
-// CHECK: store i64 [[UP]], i64* [[UB]],
-// CHECK: [[LB_VAL:%.+]] = load i64, i64* [[LB]],
-// CHECK: store i64 [[LB_VAL]], i64* [[OMP_IV7:%[^,]+]],
+// CHECK: store i64 [[UP]], ptr [[UB]],
+// CHECK: [[LB_VAL:%.+]] = load i64, ptr [[LB]],
+// CHECK: store i64 [[LB_VAL]], ptr [[OMP_IV7:%[^,]+]],
 
 // CHECK: br label %[[SIMD_LOOP7_COND:[^,]+]]
 // CHECK: [[SIMD_LOOP7_COND]]:
-// CHECK-NEXT: [[IV7:%.+]] = load i64, i64* [[OMP_IV7]]
-// CHECK-NEXT: [[UB_VAL:%.+]] = load i64, i64* [[UB]]
+// CHECK-NEXT: [[IV7:%.+]] = load i64, ptr [[OMP_IV7]]
+// CHECK-NEXT: [[UB_VAL:%.+]] = load i64, ptr [[UB]]
 // CHECK-NEXT: [[CMP7:%.+]] = icmp sle i64 [[IV7]], [[UB_VAL]]
 // CHECK-NEXT: br i1 [[CMP7]], label %[[SIMPLE_LOOP7_BODY:.+]], label %[[SIMPLE_LOOP7_END:[^,]+]]
   for (long long i = -10; i < 10; i += 3) {
 // CHECK: [[SIMPLE_LOOP7_BODY]]:
 // Start of body: calculate i from IV:
-// CHECK: [[IV7_0:%.+]] = load i64, i64* [[OMP_IV7]]
+// CHECK: [[IV7_0:%.+]] = load i64, ptr [[OMP_IV7]]
 // CHECK-NEXT: [[LC_IT_1:%.+]] = mul nsw i64 [[IV7_0]], 3
 // CHECK-NEXT: [[LC_IT_2:%.+]] = add nsw i64 -10, [[LC_IT_1]]
-// CHECK-NEXT: store i64 [[LC_IT_2]], i64* [[LC:%[^,]+]],
-// CHECK-NEXT: [[LC_VAL:%.+]] = load i64, i64* [[LC]]
+// CHECK-NEXT: store i64 [[LC_IT_2]], ptr [[LC:%[^,]+]],
+// CHECK-NEXT: [[LC_VAL:%.+]] = load i64, ptr [[LC]]
 // CHECK-NEXT: [[CONV:%.+]] = trunc i64 [[LC_VAL]] to i32
-// CHECK-NEXT: store i32 [[CONV]], i32* [[A_PRIV:%[^,]+]],
+// CHECK-NEXT: store i32 [[CONV]], ptr [[A_PRIV:%[^,]+]],
     A = i;
-// CHECK: [[IV7_2:%.+]] = load i64, i64* [[OMP_IV7]]
+// CHECK: [[IV7_2:%.+]] = load i64, ptr [[OMP_IV7]]
 // CHECK-NEXT: [[ADD7_2:%.+]] = add nsw i64 [[IV7_2]], 1
-// CHECK-NEXT: store i64 [[ADD7_2]], i64* [[OMP_IV7]]
+// CHECK-NEXT: store i64 [[ADD7_2]], ptr [[OMP_IV7]]
   }
 // CHECK: [[SIMPLE_LOOP7_END]]:
-// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}})
-// CHECK: load i32, i32*
+// CHECK: call void @__kmpc_for_static_fini(ptr {{.+}}, i32 %{{.+}})
+// CHECK: load i32, ptr
 // CHECK: icmp ne i32 %{{.+}}, 0
 // CHECK: br i1 %{{.+}}, label
-// CHECK: [[A_PRIV_VAL:%.+]] = load i32, i32* [[A_PRIV]],
-// CHECK-NEXT: store i32 [[A_PRIV_VAL]], i32* %{{.+}},
+// CHECK: [[A_PRIV_VAL:%.+]] = load i32, ptr [[A_PRIV]],
+// CHECK-NEXT: store i32 [[A_PRIV_VAL]], ptr %{{.+}},
 // CHECK-NEXT: br label
   }
   int R;
   {
   R = -1;
-// CHECK: store i32 1, i32* [[R_PRIV:%[^,]+]],
+// CHECK: store i32 1, ptr [[R_PRIV:%[^,]+]],
   #pragma omp parallel for simd reduction(*:R)
-// CHECK: call void @__kmpc_for_static_init_8(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 1)
-// CHECK: [[UB_VAL:%.+]] = load i64, i64* [[UB]],
+// CHECK: call void @__kmpc_for_static_init_8(ptr {{[^,]+}}, i32 %{{[^,]+}}, i32 34, ptr %{{[^,]+}}, ptr [[LB:%[^,]+]], ptr [[UB:%[^,]+]], ptr [[STRIDE:%[^,]+]], i64 1, i64 1)
+// CHECK: [[UB_VAL:%.+]] = load i64, ptr [[UB]],
 // CHECK: [[CMP:%.+]] = icmp sgt i64 [[UB_VAL]], 6
 // CHECK: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]]
 // CHECK: [[TRUE]]:
 // CHECK: br label %[[SWITCH:[^,]+]]
 // CHECK: [[FALSE]]:
-// CHECK: [[UB_VAL:%.+]] = load i64, i64* [[UB]],
+// CHECK: [[UB_VAL:%.+]] = load i64, ptr [[UB]],
 // CHECK: br label %[[SWITCH]]
 // CHECK: [[SWITCH]]:
 // CHECK: [[UP:%.+]] = phi i64 [ 6, %[[TRUE]] ], [ [[UB_VAL]], %[[FALSE]] ]
-// CHECK: store i64 [[UP]], i64* [[UB]],
-// CHECK: [[LB_VAL:%.+]] = load i64, i64* [[LB]],
-// CHECK: store i64 [[LB_VAL]], i64* [[OMP_IV8:%[^,]+]],
+// CHECK: store i64 [[UP]], ptr [[UB]],
+// CHECK: [[LB_VAL:%.+]] = load i64, ptr [[LB]],
+// CHECK: store i64 [[LB_VAL]], ptr [[OMP_IV8:%[^,]+]],
 
 // CHECK: br label %[[SIMD_LOOP8_COND:[^,]+]]
 // CHECK: [[SIMD_LOOP8_COND]]:
-// CHECK-NEXT: [[IV8:%.+]] = load i64, i64* [[OMP_IV8]]
-// CHECK-NEXT: [[UB_VAL:%.+]] = load i64, i64* [[UB]]
+// CHECK-NEXT: [[IV8:%.+]] = load i64, ptr [[OMP_IV8]]
+// CHECK-NEXT: [[UB_VAL:%.+]] = load i64, ptr [[UB]]
 // CHECK-NEXT: [[CMP8:%.+]] = icmp sle i64 [[IV8]], [[UB_VAL]]
 // CHECK-NEXT: br i1 [[CMP8]], label %[[SIMPLE_LOOP8_BODY:.+]], label %[[SIMPLE_LOOP8_END:[^,]+]]
   for (long long i = -10; i < 10; i += 3) {
 // CHECK: [[SIMPLE_LOOP8_BODY]]:
 // Start of body: calculate i from IV:
-// CHECK: [[IV8_0:%.+]] = load i64, i64* [[OMP_IV8]]
+// CHECK: [[IV8_0:%.+]] = load i64, ptr [[OMP_IV8]]
 // CHECK-NEXT: [[LC_IT_1:%.+]] = mul nsw i64 [[IV8_0]], 3
 // CHECK-NEXT: [[LC_IT_2:%.+]] = add nsw i64 -10, [[LC_IT_1]]
-// CHECK-NEXT: store i64 [[LC_IT_2]], i64* [[LC:%[^,]+]],
-// CHECK-NEXT: [[LC_VAL:%.+]] = load i64, i64* [[LC]]
-// CHECK: store i32 %{{.+}}, i32* [[R_PRIV]],
+// CHECK-NEXT: store i64 [[LC_IT_2]], ptr [[LC:%[^,]+]],
+// CHECK-NEXT: [[LC_VAL:%.+]] = load i64, ptr [[LC]]
+// CHECK: store i32 %{{.+}}, ptr [[R_PRIV]],
     R *= i;
-// CHECK: [[IV8_2:%.+]] = load i64, i64* [[OMP_IV8]]
+// CHECK: [[IV8_2:%.+]] = load i64, ptr [[OMP_IV8]]
 // CHECK-NEXT: [[ADD8_2:%.+]] = add nsw i64 [[IV8_2]], 1
-// CHECK-NEXT: store i64 [[ADD8_2]], i64* [[OMP_IV8]]
+// CHECK-NEXT: store i64 [[ADD8_2]], ptr [[OMP_IV8]]
   }
 // CHECK: [[SIMPLE_LOOP8_END]]:
-// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}})
+// CHECK: call void @__kmpc_for_static_fini(ptr {{.+}}, i32 %{{.+}})
 // CHECK: call i32 @__kmpc_reduce_nowait(
-// CHECK: [[R_PRIV_VAL:%.+]] = load i32, i32* [[R_PRIV]],
+// CHECK: [[R_PRIV_VAL:%.+]] = load i32, ptr [[R_PRIV]],
 // CHECK: [[RED:%.+]] = mul nsw i32 %{{.+}}, [[R_PRIV_VAL]]
-// CHECK-NEXT: store i32 [[RED]], i32* %{{.+}},
+// CHECK-NEXT: store i32 [[RED]], ptr %{{.+}},
 // CHECK-NEXT: call void @__kmpc_end_reduce_nowait(
   }
 }
@@ -387,57 +387,57 @@ int templ1(T a, T *z) {
 }
 
 // Instatiation templ1<float,2>
-// CHECK-LABEL: define {{.*i32}} @{{.*}}templ1{{.*}}(float noundef {{.+}}, float* noundef {{.+}})
-// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(
+// CHECK-LABEL: define {{.*i32}} @{{.*}}templ1{{.*}}(float noundef {{.+}}, ptr noundef {{.+}})
+// CHECK: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(
 void inst_templ1() {
   float a;
   float z[100];
   templ1<float,2> (a, z);
 }
 
-// OMP50: call void @__kmpc_for_static_init_8(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 1)
-// OMP50: [[UB_VAL:%.+]] = load i64, i64* [[UB]],
+// OMP50: call void @__kmpc_for_static_init_8(ptr {{[^,]+}}, i32 %{{[^,]+}}, i32 34, ptr %{{[^,]+}}, ptr [[LB:%[^,]+]], ptr [[UB:%[^,]+]], ptr [[STRIDE:%[^,]+]], i64 1, i64 1)
+// OMP50: [[UB_VAL:%.+]] = load i64, ptr [[UB]],
 // OMP50: [[CMP:%.+]] = icmp sgt i64 [[UB_VAL]], 15
 // OMP50: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]]
 // OMP50: [[TRUE]]:
 // OMP50: br label %[[SWITCH:[^,]+]]
 // OMP50: [[FALSE]]:
-// OMP50: [[UB_VAL:%.+]] = load i64, i64* [[UB]],
+// OMP50: [[UB_VAL:%.+]] = load i64, ptr [[UB]],
 // OMP50: br label %[[SWITCH]]
 // OMP50: [[SWITCH]]:
 // OMP50: [[UP:%.+]] = phi i64 [ 15, %[[TRUE]] ], [ [[UB_VAL]], %[[FALSE]] ]
-// OMP50: store i64 [[UP]], i64* [[UB]],
-// OMP50: [[LB_VAL:%.+]] = load i64, i64* [[LB]],
-// OMP50: store i64 [[LB_VAL]], i64* [[T1_OMP_IV:%[^,]+]],
+// OMP50: store i64 [[UP]], ptr [[UB]],
+// OMP50: [[LB_VAL:%.+]] = load i64, ptr [[LB]],
+// OMP50: store i64 [[LB_VAL]], ptr [[T1_OMP_IV:%[^,]+]],
 
 // ...
-// OMP50: [[IV:%.+]] = load i64, i64* [[T1_OMP_IV]]
-// OMP50-NEXT: [[UB_VAL:%.+]] = load i64, i64* [[UB]]
+// OMP50: [[IV:%.+]] = load i64, ptr [[T1_OMP_IV]]
+// OMP50-NEXT: [[UB_VAL:%.+]] = load i64, ptr [[UB]]
 // OMP50-NEXT: [[CMP1:%.+]] = icmp sle i64 [[IV]], [[UB_VAL]]
 // OMP50-NEXT: br i1 [[CMP1]], label %[[T1_BODY:.+]], label %[[T1_END:[^,]+]]
 // OMP50: [[T1_BODY]]:
 // Loop counters i and j updates:
-// OMP50: [[IV1:%.+]] = load i64, i64* [[T1_OMP_IV]]
+// OMP50: [[IV1:%.+]] = load i64, ptr [[T1_OMP_IV]]
 // OMP50-NEXT: [[I_1:%.+]] = sdiv i64 [[IV1]], 4
 // OMP50-NEXT: [[I_1_MUL1:%.+]] = mul nsw i64 [[I_1]], 1
 // OMP50-NEXT: [[I_1_ADD0:%.+]] = add nsw i64 0, [[I_1_MUL1]]
 // OMP50-NEXT: [[I_2:%.+]] = trunc i64 [[I_1_ADD0]] to i32
-// OMP50-NEXT: store i32 [[I_2]], i32*
-// OMP50: [[IV2:%.+]] = load i64, i64* [[T1_OMP_IV]]
-// OMP50: [[IV2_1:%.+]] = load i64, i64* [[T1_OMP_IV]]
+// OMP50-NEXT: store i32 [[I_2]], ptr
+// OMP50: [[IV2:%.+]] = load i64, ptr [[T1_OMP_IV]]
+// OMP50: [[IV2_1:%.+]] = load i64, ptr [[T1_OMP_IV]]
 // OMP50-NEXT: [[DIV_1:%.+]] = sdiv i64 [[IV2_1]], 4
 // OMP50-NEXT: [[MUL_1:%.+]] = mul nsw i64 [[DIV_1]], 4
 // OMP50-NEXT: [[J_1:%.+]] = sub nsw i64 [[IV2]], [[MUL_1]]
 // OMP50-NEXT: [[J_2:%.+]] = mul nsw i64 [[J_1]], 2
 // OMP50-NEXT: [[J_2_ADD0:%.+]] = add nsw i64 0, [[J_2]]
-// OMP50-NEXT: store i64 [[J_2_ADD0]], i64*
+// OMP50-NEXT: store i64 [[J_2_ADD0]], ptr
 // simd.for.inc:
-// OMP50: [[IV3:%.+]] = load i64, i64* [[T1_OMP_IV]]
+// OMP50: [[IV3:%.+]] = load i64, ptr [[T1_OMP_IV]]
 // OMP50-NEXT: [[INC:%.+]] = add nsw i64 [[IV3]], 1
-// OMP50-NEXT: store i64 [[INC]], i64*
+// OMP50-NEXT: store i64 [[INC]], ptr
 // OMP50-NEXT: br label {{%.+}}
 // OMP50: [[T1_END]]:
-// OMP50: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}})
+// OMP50: call void @__kmpc_for_static_fini(ptr {{.+}}, i32 %{{.+}})
 // OMP50: ret void
 //
 
@@ -478,34 +478,34 @@ void iter_simple(IterDouble ia, IterDouble ib, IterDouble ic) {
 // CHECK-NEXT: [[DIFF3:%.+]] = add nsw i32 [[DIFF2]], 1
 // CHECK-NEXT: [[DIFF4:%.+]] = sdiv i32 [[DIFF3]], 1
 // CHECK-NEXT: [[DIFF5:%.+]] = sub nsw i32 [[DIFF4]], 1
-// CHECK-NEXT: store i32 [[DIFF5]], i32* [[OMP_LAST_IT:%[^,]+]]{{.+}}
+// CHECK-NEXT: store i32 [[DIFF5]], ptr [[OMP_LAST_IT:%[^,]+]]{{.+}}
   #pragma omp parallel for simd
 
-// CHECK: call void @__kmpc_for_static_init_4(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i32* [[LB:%[^,]+]], i32* [[UB:%[^,]+]], i32* [[STRIDE:%[^,]+]], i32 1, i32 1)
-// CHECK-DAG: [[UB_VAL:%.+]] = load i32, i32* [[UB]],
-// CHECK-DAG: [[OMP_LAST_IT_VAL:%.+]] = load i32, i32* [[OMP_LAST_IT]],
+// CHECK: call void @__kmpc_for_static_init_4(ptr {{[^,]+}}, i32 %{{[^,]+}}, i32 34, ptr %{{[^,]+}}, ptr [[LB:%[^,]+]], ptr [[UB:%[^,]+]], ptr [[STRIDE:%[^,]+]], i32 1, i32 1)
+// CHECK-DAG: [[UB_VAL:%.+]] = load i32, ptr [[UB]],
+// CHECK-DAG: [[OMP_LAST_IT_VAL:%.+]] = load i32, ptr [[OMP_LAST_IT]],
 // CHECK: [[CMP:%.+]] = icmp sgt i32 [[UB_VAL]], [[OMP_LAST_IT_VAL]]
 // CHECK: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]]
 // CHECK: [[TRUE]]:
-// CHECK: [[OMP_LAST_IT_VAL:%.+]] = load i32, i32* [[OMP_LAST_IT]],
+// CHECK: [[OMP_LAST_IT_VAL:%.+]] = load i32, ptr [[OMP_LAST_IT]],
 // CHECK: br label %[[SWITCH:[^,]+]]
 // CHECK: [[FALSE]]:
-// CHECK: [[UB_VAL:%.+]] = load i32, i32* [[UB]],
+// CHECK: [[UB_VAL:%.+]] = load i32, ptr [[UB]],
 // CHECK: br label %[[SWITCH]]
 // CHECK: [[SWITCH]]:
 // CHECK: [[UP:%.+]] = phi i32 [ [[OMP_LAST_IT_VAL]], %[[TRUE]] ], [ [[UB_VAL]], %[[FALSE]] ]
-// CHECK: store i32 [[UP]], i32* [[UB]],
-// CHECK: [[LB_VAL:%.+]] = load i32, i32* [[LB]],
-// CHECK: store i32 [[LB_VAL]], i32* [[IT_OMP_IV:%[^,]+]],
+// CHECK: store i32 [[UP]], ptr [[UB]],
+// CHECK: [[LB_VAL:%.+]] = load i32, ptr [[LB]],
+// CHECK: store i32 [[LB_VAL]], ptr [[IT_OMP_IV:%[^,]+]],
 
-// CHECK: [[IV:%.+]] = load i32, i32* [[IT_OMP_IV]]
-// CHECK-NEXT: [[UB_VAL:%.+]] = load i32, i32* [[UB]]
+// CHECK: [[IV:%.+]] = load i32, ptr [[IT_OMP_IV]]
+// CHECK-NEXT: [[UB_VAL:%.+]] = load i32, ptr [[UB]]
 // CHECK-NEXT: [[CMP:%.+]] = icmp sle i32 [[IV]], [[UB_VAL]]
 // CHECK-NEXT: br i1 [[CMP]], label %[[IT_BODY:[^,]+]], label %[[IT_END:[^,]+]]
   for (IterDouble i = ia; i < ib; ++i) {
 // CHECK: [[IT_BODY]]:
 // Start of body: calculate i from index:
-// CHECK: [[IV1:%.+]] = load i32, i32* [[IT_OMP_IV]]
+// CHECK: [[IV1:%.+]] = load i32, ptr [[IT_OMP_IV]]
 // Call of operator+ (i, IV).
 // CHECK: {{%.+}} = invoke {{.+}} @{{.*}}IterDouble{{.*}}
 // ... loop body ...
@@ -513,16 +513,16 @@ void iter_simple(IterDouble ia, IterDouble ib, IterDouble ic) {
 // Float multiply and save result.
 // CHECK: [[MULR:%.+]] = fmul double {{%.+}}, 5.000000e-01
 // CHECK-NEXT: invoke {{.+}} @{{.*}}IterDouble{{.*}}
-// CHECK: store double [[MULR:%.+]], double* [[RESULT_ADDR:%.+]]
+// CHECK: store double [[MULR:%.+]], ptr [[RESULT_ADDR:%.+]]
    ++ic;
 //
-// CHECK: [[IV2:%.+]] = load i32, i32* [[IT_OMP_IV]]
+// CHECK: [[IV2:%.+]] = load i32, ptr [[IT_OMP_IV]]
 // CHECK-NEXT: [[ADD2:%.+]] = add nsw i32 [[IV2]], 1
-// CHECK-NEXT: store i32 [[ADD2]], i32* [[IT_OMP_IV]]
+// CHECK-NEXT: store i32 [[ADD2]], ptr [[IT_OMP_IV]]
 // br label %{{.*}}, !llvm.loop ![[ITER_LOOP_ID]]
   }
 // CHECK: [[IT_END]]:
-// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}})
+// CHECK: call void @__kmpc_for_static_fini(ptr {{.+}}, i32 %{{.+}})
 // CHECK: ret void
 }
 
@@ -533,25 +533,25 @@ void collapsed(float *a, float *b, float *c, float *d) {
   unsigned j; // middle loop couter, leads to unsigned icmp in loop header.
   // k declared in the loop init below
   short l; // inner loop counter
-// CHECK: call void @__kmpc_for_static_init_4u(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i32* [[LB:%[^,]+]], i32* [[UB:%[^,]+]], i32* [[STRIDE:%[^,]+]], i32 1, i32 1)
-// CHECK: [[UB_VAL:%.+]] = load i32, i32* [[UB]],
+// CHECK: call void @__kmpc_for_static_init_4u(ptr {{[^,]+}}, i32 %{{[^,]+}}, i32 34, ptr %{{[^,]+}}, ptr [[LB:%[^,]+]], ptr [[UB:%[^,]+]], ptr [[STRIDE:%[^,]+]], i32 1, i32 1)
+// CHECK: [[UB_VAL:%.+]] = load i32, ptr [[UB]],
 // CHECK: [[CMP:%.+]] = icmp ugt i32 [[UB_VAL]], 119
 // CHECK: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]]
 // CHECK: [[TRUE]]:
 // CHECK: br label %[[SWITCH:[^,]+]]
 // CHECK: [[FALSE]]:
-// CHECK: [[UB_VAL:%.+]] = load i32, i32* [[UB]],
+// CHECK: [[UB_VAL:%.+]] = load i32, ptr [[UB]],
 // CHECK: br label %[[SWITCH]]
 // CHECK: [[SWITCH]]:
 // CHECK: [[UP:%.+]] = phi i32 [ 119, %[[TRUE]] ], [ [[UB_VAL]], %[[FALSE]] ]
-// CHECK: store i32 [[UP]], i32* [[UB]],
-// CHECK: [[LB_VAL:%.+]] = load i32, i32* [[LB]],
-// CHECK: store i32 [[LB_VAL]], i32* [[OMP_IV:%[^,]+]],
+// CHECK: store i32 [[UP]], ptr [[UB]],
+// CHECK: [[LB_VAL:%.+]] = load i32, ptr [[LB]],
+// CHECK: store i32 [[LB_VAL]], ptr [[OMP_IV:%[^,]+]],
 //
   #pragma omp parallel for simd collapse(4)
 
-// CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]
-// CHECK: [[UB_VAL:%.+]] = load i32, i32* [[UB]]
+// CHECK: [[IV:%.+]] = load i32, ptr [[OMP_IV]]
+// CHECK: [[UB_VAL:%.+]] = load i32, ptr [[UB]]
 // CHECK-NEXT: [[CMP:%.+]] = icmp ule i32 [[IV]], [[UB_VAL]]
 // CHECK-NEXT: br i1 [[CMP]], label %[[COLL1_BODY:[^,]+]], label %[[COLL1_END:[^,]+]]
   for (i = 1; i < 3; i++) // 2 iterations
@@ -561,31 +561,31 @@ void collapsed(float *a, float *b, float *c, float *d) {
         {
 // CHECK: [[COLL1_BODY]]:
 // Start of body: calculate i from index:
-// CHECK: [[IV1:%.+]] = load i32, i32* [[OMP_IV]]
+// CHECK: [[IV1:%.+]] = load i32, ptr [[OMP_IV]]
 // Calculation of the loop counters values.
 // CHECK: [[CALC_I_1:%.+]] = udiv i32 [[IV1]], 60
 // CHECK-NEXT: [[CALC_I_1_MUL1:%.+]] = mul i32 [[CALC_I_1]], 1
 // CHECK-NEXT: [[CALC_I_2:%.+]] = add i32 1, [[CALC_I_1_MUL1]]
-// CHECK-NEXT: store i32 [[CALC_I_2]], i32* [[LC_I:.+]]
+// CHECK-NEXT: store i32 [[CALC_I_2]], ptr [[LC_I:.+]]
 
-// CHECK: [[IV1_2:%.+]] = load i32, i32* [[OMP_IV]]
-// CHECK: [[IV1_2_1:%.+]] = load i32, i32* [[OMP_IV]]
+// CHECK: [[IV1_2:%.+]] = load i32, ptr [[OMP_IV]]
+// CHECK: [[IV1_2_1:%.+]] = load i32, ptr [[OMP_IV]]
 // CHECK-NEXT: [[CALC_J_1:%.+]] = udiv i32 [[IV1_2_1]], 60
 // CHECK-NEXT: [[MUL_1:%.+]] = mul i32 [[CALC_J_1]], 60
 // CHECK-NEXT: [[SUB_3:%.+]] = sub i32 [[IV1_2]], [[MUL_1]]
 // CHECK-NEXT: [[CALC_J_2:%.+]] = udiv i32 [[SUB_3]], 20
 // CHECK-NEXT: [[CALC_J_2_MUL1:%.+]] = mul i32 [[CALC_J_2]], 1
 // CHECK-NEXT: [[CALC_J_3:%.+]] = add i32 2, [[CALC_J_2_MUL1]]
-// CHECK-NEXT: store i32 [[CALC_J_3]], i32* [[LC_J:.+]]
+// CHECK-NEXT: store i32 [[CALC_J_3]], ptr [[LC_J:.+]]
 
-// CHECK: [[IV1_3:%.+]] = load i32, i32* [[OMP_IV]]
-// CHECK: [[IV1_3_1:%.+]] = load i32, i32* [[OMP_IV]]
+// CHECK: [[IV1_3:%.+]] = load i32, ptr [[OMP_IV]]
+// CHECK: [[IV1_3_1:%.+]] = load i32, ptr [[OMP_IV]]
 // CHECK-NEXT: [[DIV_1:%.+]] = udiv i32 [[IV1_3_1]], 60
 // CHECK-NEXT: [[MUL_2:%.+]] = mul i32 [[DIV_1]], 60
 // CHECK-NEXT: [[ADD_3:%.+]] = sub i32 [[IV1_3]], [[MUL_2]]
 
-// CHECK: [[IV1_4:%.+]] = load i32, i32* [[OMP_IV]]
-// CHECK: [[IV1_4_1:%.+]] = load i32, i32* [[OMP_IV]]
+// CHECK: [[IV1_4:%.+]] = load i32, ptr [[OMP_IV]]
+// CHECK: [[IV1_4_1:%.+]] = load i32, ptr [[OMP_IV]]
 // CHECK-NEXT: [[DIV_2:%.+]] = udiv i32 [[IV1_4_1]], 60
 // CHECK-NEXT: [[MUL_3:%.+]] = mul i32 [[DIV_2]], 60
 // CHECK-NEXT: [[SUB_6:%.+]] = sub i32 [[IV1_4]], [[MUL_3]]
@@ -595,16 +595,16 @@ void collapsed(float *a, float *b, float *c, float *d) {
 // CHECK-NEXT: [[DIV_4:%.+]] = udiv i32 [[SUB_7]], 5
 // CHECK-NEXT: [[MUL_5:%.+]] = mul i32 [[DIV_4]], 1
 // CHECK-NEXT: [[ADD_6:%.+]] = add i32 3, [[MUL_5]]
-// CHECK-NEXT: store i32 [[ADD_6]], i32* [[LC_K:.+]]
+// CHECK-NEXT: store i32 [[ADD_6]], ptr [[LC_K:.+]]
 
-// CHECK: [[IV1_5:%.+]] = load i32, i32* [[OMP_IV]]
-// CHECK: [[IV1_5_1:%.+]] = load i32, i32* [[OMP_IV]]
+// CHECK: [[IV1_5:%.+]] = load i32, ptr [[OMP_IV]]
+// CHECK: [[IV1_5_1:%.+]] = load i32, ptr [[OMP_IV]]
 // CHECK-NEXT: [[DIV_5:%.+]] = udiv i32 [[IV1_5_1]], 60
 // CHECK-NEXT: [[MUL_6:%.+]] = mul i32 [[DIV_5]], 60
 // CHECK-NEXT: [[ADD_7:%.+]] = sub i32 [[IV1_5]], [[MUL_6]]
 
-// CHECK: [[IV1_6:%.+]] = load i32, i32* [[OMP_IV]]
-// CHECK: [[IV1_6_1:%.+]] = load i32, i32* [[OMP_IV]]
+// CHECK: [[IV1_6:%.+]] = load i32, ptr [[OMP_IV]]
+// CHECK: [[IV1_6_1:%.+]] = load i32, ptr [[OMP_IV]]
 // CHECK-NEXT: [[DIV_6:%.+]] = udiv i32 [[IV1_6_1]], 60
 // CHECK-NEXT: [[MUL_7:%.+]] = mul i32 [[DIV_6]], 60
 // CHECK-NEXT: [[SUB_10:%.+]] = sub i32 [[IV1_6]], [[MUL_7]]
@@ -612,14 +612,14 @@ void collapsed(float *a, float *b, float *c, float *d) {
 // CHECK-NEXT: [[MUL_8:%.+]] = mul i32 [[DIV_7]], 20
 // CHECK-NEXT: [[ADD_9:%.+]] = sub i32 [[ADD_7]], [[MUL_8]]
 
-// CHECK: [[IV1_7:%.+]] = load i32, i32* [[OMP_IV]]
-// CHECK: [[IV1_7_1:%.+]] = load i32, i32* [[OMP_IV]]
+// CHECK: [[IV1_7:%.+]] = load i32, ptr [[OMP_IV]]
+// CHECK: [[IV1_7_1:%.+]] = load i32, ptr [[OMP_IV]]
 // CHECK-NEXT: [[DIV_8:%.+]] = udiv i32 [[IV1_7_1]], 60
 // CHECK-NEXT: [[MUL_9:%.+]] = mul i32 [[DIV_8]], 60
 // CHECK-NEXT: [[ADD_10:%.+]] = sub i32 [[IV1_7]], [[MUL_9]]
 
-// CHECK: [[IV1_8:%.+]] = load i32, i32* [[OMP_IV]]
-// CHECK: [[IV1_8_1:%.+]] = load i32, i32* [[OMP_IV]]
+// CHECK: [[IV1_8:%.+]] = load i32, ptr [[OMP_IV]]
+// CHECK: [[IV1_8_1:%.+]] = load i32, ptr [[OMP_IV]]
 // CHECK-NEXT: [[DIV_3:%.+]] = udiv i32 [[IV1_8_1]], 60
 // CHECK-NEXT: [[MUL_4:%.+]] = mul i32 [[DIV_3]], 60
 // CHECK-NEXT: [[SUB_7:%.+]] = sub i32 [[IV1_8]], [[MUL_4]]
@@ -632,23 +632,23 @@ void collapsed(float *a, float *b, float *c, float *d) {
 // CHECK-NEXT: [[MUL_6:%.+]] = mul i32 [[SUB_9]], 1
 // CHECK-NEXT: [[CALC_L_2:%.+]] = add i32 4, [[MUL_6]]
 // CHECK-NEXT: [[CALC_L_3:%.+]] = trunc i32 [[CALC_L_2]] to i16
-// CHECK-NEXT: store i16 [[CALC_L_3]], i16* [[LC_L:.+]]
+// CHECK-NEXT: store i16 [[CALC_L_3]], ptr [[LC_L:.+]]
 // ... loop body ...
 // End of body: store into a[i]:
-// CHECK: store float [[RESULT:%.+]], float* [[RESULT_ADDR:%.+]]
+// CHECK: store float [[RESULT:%.+]], ptr [[RESULT_ADDR:%.+]]
     float res = b[j] * c[k];
     a[i] = res * d[l];
-// CHECK: [[IV2:%.+]] = load i32, i32* [[OMP_IV]]
+// CHECK: [[IV2:%.+]] = load i32, ptr [[OMP_IV]]
 // CHECK-NEXT: [[ADD2:%.+]] = add i32 [[IV2]], 1
-// CHECK-NEXT: store i32 [[ADD2]], i32* [[OMP_IV]]
+// CHECK-NEXT: store i32 [[ADD2]], ptr [[OMP_IV]]
 // br label %{{[^,]+}}, !llvm.loop ![[COLL1_LOOP_ID]]
 // CHECK: [[COLL1_END]]:
   }
 // i,j,l are updated; k is not updated.
-// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}})
-// CHECK: store i32 3, i32* [[I:%[^,]+]]
-// CHECK: store i32 5, i32* [[I:%[^,]+]]
-// CHECK: store i16 9, i16* [[I:%[^,]+]]
+// CHECK: call void @__kmpc_for_static_fini(ptr {{.+}}, i32 %{{.+}})
+// CHECK: store i32 3, ptr [[I:%[^,]+]]
+// CHECK: store i32 5, ptr [[I:%[^,]+]]
+// CHECK: store i16 9, ptr [[I:%[^,]+]]
 // CHECK: ret void
 }
 
@@ -661,32 +661,32 @@ void widened(float *a, float *b, float *c, float *d) {
   short j; // inner loop counter
   globalfloat = 1.0;
   int localint = 1;
-// CHECK: store double {{.+}}, double* [[GLOBALFLOAT:@.+]]
+// CHECK: store double {{.+}}, ptr [[GLOBALFLOAT:@.+]]
 // Counter is widened to 64 bits.
 // CHECK:     [[MUL:%.+]] = mul nsw i64 2, %{{.+}}
 // CHECK-NEXT: [[SUB:%.+]] = sub nsw i64 [[MUL]], 1
-// CHECK-NEXT: store i64 [[SUB]], i64* [[OMP_LAST_IT:%[^,]+]],
-// CHECK: call void @__kmpc_for_static_init_8(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 1)
-// CHECK-DAG: [[UB_VAL:%.+]] = load i64, i64* [[UB]],
-// CHECK-DAG: [[OMP_LAST_IT_VAL:%.+]] = load i64, i64* [[OMP_LAST_IT]],
+// CHECK-NEXT: store i64 [[SUB]], ptr [[OMP_LAST_IT:%[^,]+]],
+// CHECK: call void @__kmpc_for_static_init_8(ptr {{[^,]+}}, i32 %{{[^,]+}}, i32 34, ptr %{{[^,]+}}, ptr [[LB:%[^,]+]], ptr [[UB:%[^,]+]], ptr [[STRIDE:%[^,]+]], i64 1, i64 1)
+// CHECK-DAG: [[UB_VAL:%.+]] = load i64, ptr [[UB]],
+// CHECK-DAG: [[OMP_LAST_IT_VAL:%.+]] = load i64, ptr [[OMP_LAST_IT]],
 // CHECK: [[CMP:%.+]] = icmp sgt i64 [[UB_VAL]], [[OMP_LAST_IT_VAL]]
 // CHECK: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]]
 // CHECK: [[TRUE]]:
-// CHECK: [[OMP_LAST_IT_VAL:%.+]] = load i64, i64* [[OMP_LAST_IT]],
+// CHECK: [[OMP_LAST_IT_VAL:%.+]] = load i64, ptr [[OMP_LAST_IT]],
 // CHECK: br label %[[SWITCH:[^,]+]]
 // CHECK: [[FALSE]]:
-// CHECK: [[UB_VAL:%.+]] = load i64, i64* [[UB]],
+// CHECK: [[UB_VAL:%.+]] = load i64, ptr [[UB]],
 // CHECK: br label %[[SWITCH]]
 // CHECK: [[SWITCH]]:
 // CHECK: [[UP:%.+]] = phi i64 [ [[OMP_LAST_IT_VAL]], %[[TRUE]] ], [ [[UB_VAL]], %[[FALSE]] ]
-// CHECK: store i64 [[UP]], i64* [[UB]],
-// CHECK: [[LB_VAL:%.+]] = load i64, i64* [[LB]],
-// CHECK: store i64 [[LB_VAL]], i64* [[OMP_IV:%[^,]+]],
+// CHECK: store i64 [[UP]], ptr [[UB]],
+// CHECK: [[LB_VAL:%.+]] = load i64, ptr [[LB]],
+// CHECK: store i64 [[LB_VAL]], ptr [[OMP_IV:%[^,]+]],
 //
   #pragma omp parallel for simd collapse(2) private(globalfloat, localint)
 
-// CHECK: [[IV:%.+]] = load i64, i64* [[OMP_IV]]
-// CHECK: [[UB_VAL:%.+]] = load i64, i64* [[UB]]
+// CHECK: [[IV:%.+]] = load i64, ptr [[OMP_IV]]
+// CHECK: [[UB_VAL:%.+]] = load i64, ptr [[UB]]
 // CHECK-NEXT: [[CMP:%.+]] = icmp sle i64 [[IV]], [[UB_VAL]]
 // CHECK-NEXT: br i1 [[CMP]], label %[[WIDE1_BODY:[^,]+]], label %[[WIDE1_END:[^,]+]]
   for (i = 1; i < 3; i++) // 2 iterations
@@ -694,36 +694,36 @@ void widened(float *a, float *b, float *c, float *d) {
   {
 // CHECK: [[WIDE1_BODY]]:
 // Start of body: calculate i from index:
-// CHECK: [[IV1:%.+]] = load i64, i64* [[OMP_IV]]
+// CHECK: [[IV1:%.+]] = load i64, ptr [[OMP_IV]]
 // Calculation of the loop counters values...
-// CHECK: store i32 {{[^,]+}}, i32* [[LC_I:.+]]
-// CHECK: [[IV1_2:%.+]] = load i64, i64* [[OMP_IV]]
-// CHECK: store i16 {{[^,]+}}, i16* [[LC_J:.+]]
+// CHECK: store i32 {{[^,]+}}, ptr [[LC_I:.+]]
+// CHECK: [[IV1_2:%.+]] = load i64, ptr [[OMP_IV]]
+// CHECK: store i16 {{[^,]+}}, ptr [[LC_J:.+]]
 // ... loop body ...
 //
 // Here we expect store into private double var, not global
-// CHECK-NOT: store double {{.+}}, double* [[GLOBALFLOAT]]
+// CHECK-NOT: store double {{.+}}, ptr [[GLOBALFLOAT]]
     globalfloat = (float)j/i;
     float res = b[j] * c[j];
 // Store into a[i]:
-// CHECK: store float [[RESULT:%.+]], float* [[RESULT_ADDR:%.+]]
+// CHECK: store float [[RESULT:%.+]], ptr [[RESULT_ADDR:%.+]]
     a[i] = res * d[i];
 // Then there's a store into private var localint:
-// CHECK: store i32 {{.+}}, i32* [[LOCALINT:%[^,]+]]
+// CHECK: store i32 {{.+}}, ptr [[LOCALINT:%[^,]+]]
     localint = (int)j;
-// CHECK: [[IV2:%.+]] = load i64, i64* [[OMP_IV]]
+// CHECK: [[IV2:%.+]] = load i64, ptr [[OMP_IV]]
 // CHECK-NEXT: [[ADD2:%.+]] = add nsw i64 [[IV2]], 1
-// CHECK-NEXT: store i64 [[ADD2]], i64* [[OMP_IV]]
+// CHECK-NEXT: store i64 [[ADD2]], ptr [[OMP_IV]]
 //
 // br label %{{[^,]+}}, !llvm.loop ![[WIDE1_LOOP_ID]]
 // CHECK: [[WIDE1_END]]:
   }
 // i,j are updated.
-// CHECK: store i32 3, i32* [[I:%[^,]+]]
+// CHECK: store i32 3, ptr [[I:%[^,]+]]
 // CHECK: store i16
 //
 // Here we expect store into original localint, not its privatized version.
-// CHECK-NOT: store i32 {{.+}}, i32* [[LOCALINT]]
+// CHECK-NOT: store i32 {{.+}}, ptr [[LOCALINT]]
   localint = (int)j;
 // CHECK: ret void
 }
@@ -744,49 +744,49 @@ for (int i = 0; i < 10; ++i);
 // OMP50: br label {{.+}}, !llvm.loop ![[NOVECT:.+]]
 // CHECK: call void @__kmpc_for_static_fini(
 
-// OMP45: call void @__kmpc_for_static_init_8(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 1)
-// OMP45: [[UB_VAL:%.+]] = load i64, i64* [[UB]],
+// OMP45: call void @__kmpc_for_static_init_8(ptr {{[^,]+}}, i32 %{{[^,]+}}, i32 34, ptr %{{[^,]+}}, ptr [[LB:%[^,]+]], ptr [[UB:%[^,]+]], ptr [[STRIDE:%[^,]+]], i64 1, i64 1)
+// OMP45: [[UB_VAL:%.+]] = load i64, ptr [[UB]],
 // OMP45: [[CMP:%.+]] = icmp sgt i64 [[UB_VAL]], 15
 // OMP45: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]]
 // OMP45: [[TRUE]]:
 // OMP45: br label %[[SWITCH:[^,]+]]
 // OMP45: [[FALSE]]:
-// OMP45: [[UB_VAL:%.+]] = load i64, i64* [[UB]],
+// OMP45: [[UB_VAL:%.+]] = load i64, ptr [[UB]],
 // OMP45: br label %[[SWITCH]]
 // OMP45: [[SWITCH]]:
 // OMP45: [[UP:%.+]] = phi i64 [ 15, %[[TRUE]] ], [ [[UB_VAL]], %[[FALSE]] ]
-// OMP45: store i64 [[UP]], i64* [[UB]],
-// OMP45: [[LB_VAL:%.+]] = load i64, i64* [[LB]],
-// OMP45: store i64 [[LB_VAL]], i64* [[T1_OMP_IV:%[^,]+]],
+// OMP45: store i64 [[UP]], ptr [[UB]],
+// OMP45: [[LB_VAL:%.+]] = load i64, ptr [[LB]],
+// OMP45: store i64 [[LB_VAL]], ptr [[T1_OMP_IV:%[^,]+]],
 
 // ...
-// OMP45: [[IV:%.+]] = load i64, i64* [[T1_OMP_IV]]
-// OMP45-NEXT: [[UB_VAL:%.+]] = load i64, i64* [[UB]]
+// OMP45: [[IV:%.+]] = load i64, ptr [[T1_OMP_IV]]
+// OMP45-NEXT: [[UB_VAL:%.+]] = load i64, ptr [[UB]]
 // OMP45-NEXT: [[CMP1:%.+]] = icmp sle i64 [[IV]], [[UB_VAL]]
 // OMP45-NEXT: br i1 [[CMP1]], label %[[T1_BODY:.+]], label %[[T1_END:[^,]+]]
 // OMP45: [[T1_BODY]]:
 // Loop counters i and j updates:
-// OMP45: [[IV1:%.+]] = load i64, i64* [[T1_OMP_IV]]
+// OMP45: [[IV1:%.+]] = load i64, ptr [[T1_OMP_IV]]
 // OMP45-NEXT: [[I_1:%.+]] = sdiv i64 [[IV1]], 4
 // OMP45-NEXT: [[I_1_MUL1:%.+]] = mul nsw i64 [[I_1]], 1
 // OMP45-NEXT: [[I_1_ADD0:%.+]] = add nsw i64 0, [[I_1_MUL1]]
 // OMP45-NEXT: [[I_2:%.+]] = trunc i64 [[I_1_ADD0]] to i32
-// OMP45-NEXT: store i32 [[I_2]], i32*
-// OMP45: [[IV2:%.+]] = load i64, i64* [[T1_OMP_IV]]
-// OMP45: [[IV2_1:%.+]] = load i64, i64* [[T1_OMP_IV]]
+// OMP45-NEXT: store i32 [[I_2]], ptr
+// OMP45: [[IV2:%.+]] = load i64, ptr [[T1_OMP_IV]]
+// OMP45: [[IV2_1:%.+]] = load i64, ptr [[T1_OMP_IV]]
 // OMP45-NEXT: [[DIV_1:%.+]] = sdiv i64 [[IV2_1]], 4
 // OMP45-NEXT: [[MUL_1:%.+]] = mul nsw i64 [[DIV_1]], 4
 // OMP45-NEXT: [[J_1:%.+]] = sub nsw i64 [[IV2]], [[MUL_1]]
 // OMP45-NEXT: [[J_2:%.+]] = mul nsw i64 [[J_1]], 2
 // OMP45-NEXT: [[J_2_ADD0:%.+]] = add nsw i64 0, [[J_2]]
-// OMP45-NEXT: store i64 [[J_2_ADD0]], i64*
+// OMP45-NEXT: store i64 [[J_2_ADD0]], ptr
 // simd.for.inc:
-// OMP45: [[IV3:%.+]] = load i64, i64* [[T1_OMP_IV]]
+// OMP45: [[IV3:%.+]] = load i64, ptr [[T1_OMP_IV]]
 // OMP45-NEXT: [[INC:%.+]] = add nsw i64 [[IV3]], 1
-// OMP45-NEXT: store i64 [[INC]], i64*
+// OMP45-NEXT: store i64 [[INC]], ptr
 // OMP45-NEXT: br label {{%.+}}
 // OMP45: [[T1_END]]:
-// OMP45: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}})
+// OMP45: call void @__kmpc_for_static_fini(ptr {{.+}}, i32 %{{.+}})
 // OMP45: ret void
 
 // OMP50-LABEL: parallel_simd_atomic
@@ -796,7 +796,7 @@ for (int i = 0; i < 10; ++i);
 // OMP50-LABEL: define internal void @_Z20parallel_simd_atomici.omp_outlined
 // OMP50: call void @__kmpc_for_static_init_4
 // OMP50: omp.inner.for.body:
-// OMP50: %{{[0-9]+}} = atomicrmw add i32* %{{[0-9]+}}, i32 1 monotonic, align 4, !llvm.access.group !47
+// OMP50: %{{[0-9]+}} = atomicrmw add ptr %{{[0-9]+}}, i32 1 monotonic, align 4, !llvm.access.group !47
 
 // OMP45-NOT: !{!"llvm.loop.vectorize.enable", i1 false}
 // OMP45-DAG: ![[VECT]] = distinct !{![[VECT]], ![[PA:.+]], ![[VM:.+]]}
diff --git a/clang/test/OpenMP/simd_codegen.cpp b/clang/test/OpenMP/simd_codegen.cpp
index 22cb4cf7c0c0..9ee95440b16b 100644
--- a/clang/test/OpenMP/simd_codegen.cpp
+++ b/clang/test/OpenMP/simd_codegen.cpp
@@ -1,18 +1,18 @@
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - -fopenmp-version=45 | FileCheck %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s -fopenmp-version=45
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -fopenmp-version=45 | FileCheck %s
-// RUN: %clang_cc1 -no-opaque-pointers -verify -triple x86_64-apple-darwin10 -fopenmp -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - -fopenmp-version=45 | FileCheck %s --check-prefix=TERM_DEBUG
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - -fopenmp-version=50 -DOMP5 | FileCheck %s --check-prefix=CHECK --check-prefix=OMP50 --check-prefix=OMP50RT
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s -fopenmp-version=50 -DOMP5
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -fopenmp-version=50 -DOMP5 | FileCheck %s --check-prefix=CHECK --check-prefix=OMP50 --check-prefix=OMP50RT
-
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - -fopenmp-version=45 | FileCheck %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s -fopenmp-version=45
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -fopenmp-version=45 | FileCheck %s
-// RUN: %clang_cc1 -no-opaque-pointers -verify -triple x86_64-apple-darwin10 -fopenmp-simd -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - -fopenmp-version=45 | FileCheck --check-prefix=TERM_DEBUG %s
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - -fopenmp-version=50 -DOMP5 | FileCheck %s --check-prefix=CHECK --check-prefix=OMP50
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s -fopenmp-version=50 -DOMP5
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -fopenmp-version=50 -DOMP5 | FileCheck %s --check-prefix=CHECK --check-prefix=OMP50
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - -fopenmp-version=45 | FileCheck %s
+// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s -fopenmp-version=45
+// RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -fopenmp-version=45 | FileCheck %s
+// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - -fopenmp-version=45 | FileCheck %s --check-prefix=TERM_DEBUG
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - -fopenmp-version=50 -DOMP5 | FileCheck %s --check-prefix=CHECK --check-prefix=OMP50 --check-prefix=OMP50RT
+// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s -fopenmp-version=50 -DOMP5
+// RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -fopenmp-version=50 -DOMP5 | FileCheck %s --check-prefix=CHECK --check-prefix=OMP50 --check-prefix=OMP50RT
+
+// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - -fopenmp-version=45 | FileCheck %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s -fopenmp-version=45
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -fopenmp-version=45 | FileCheck %s
+// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - -fopenmp-version=45 | FileCheck --check-prefix=TERM_DEBUG %s
+// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - -fopenmp-version=50 -DOMP5 | FileCheck %s --check-prefix=CHECK --check-prefix=OMP50
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s -fopenmp-version=50 -DOMP5
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -fopenmp-version=50 -DOMP5 | FileCheck %s --check-prefix=CHECK --check-prefix=OMP50
 // expected-no-diagnostics
  #ifndef HEADER
  #define HEADER
@@ -33,7 +33,7 @@ struct S {
   int a, b;
 };
 
-// CHECK-LABEL: define {{.*void}} @{{.*}}simple{{.*}}(float* noundef {{.+}}, float* noundef {{.+}}, float* noundef {{.+}}, float* noundef {{.+}})
+// CHECK-LABEL: define {{.*void}} @{{.*}}simple{{.*}}(ptr noundef {{.+}}, ptr noundef {{.+}}, ptr noundef {{.+}}, ptr noundef {{.+}})
 void simple(float *a, float *b, float *c, float *d) {
   S s, *p;
 #ifdef OMP5
@@ -41,34 +41,34 @@ void simple(float *a, float *b, float *c, float *d) {
 #else
   #pragma omp simd
 #endif
-  // CHECK: store i32 0, i32* [[OMP_IV:%[^,]+]]
+  // CHECK: store i32 0, ptr [[OMP_IV:%[^,]+]]
 
-// CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
+// CHECK: [[IV:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[CMP:%.+]] = icmp slt i32 [[IV]], 6
 // CHECK-NEXT: br i1 [[CMP]], label %[[SIMPLE_LOOP1_BODY:.+]], label %[[SIMPLE_LOOP1_END:[^,]+]]
   for (int i = 3; i < 32; i += 5) {
 // CHECK: [[SIMPLE_LOOP1_BODY]]:
 // Start of body: calculate i from IV:
-// CHECK: [[IV1_1:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
+// CHECK: [[IV1_1:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK: [[CALC_I_1:%.+]] = mul nsw i32 [[IV1_1]], 5
 // CHECK-NEXT: [[CALC_I_2:%.+]] = add nsw i32 3, [[CALC_I_1]]
-// CHECK-NEXT: store i32 [[CALC_I_2]], i32* [[LC_I:.+]]{{.*}}!llvm.access.group
+// CHECK-NEXT: store i32 [[CALC_I_2]], ptr [[LC_I:.+]]{{.*}}!llvm.access.group
 // ... loop body ...
 // End of body: store into a[i]:
-// OMP45-NOT: load float*,{{.*}}!nontemporal
+// OMP45-NOT: load ptr,{{.*}}!nontemporal
 // CHECK-NOT: load float,{{.*}}!nontemporal
-// OMP50: load float*,{{.*}}!nontemporal
-// OMP50: load float*,{{.*}}!nontemporal
-// OMP50: load float*,{{.*}}!nontemporal
+// OMP50: load ptr,{{.*}}!nontemporal
+// OMP50: load ptr,{{.*}}!nontemporal
+// OMP50: load ptr,{{.*}}!nontemporal
 // OMP50: load i32,{{.*}}!nontemporal
 // OMP50-NOT: load i32,{{.*}}!nontemporal
-// OMP50: load float*,{{.*}}!nontemporal
+// OMP50: load ptr,{{.*}}!nontemporal
 // CHECK-NOT: load float,{{.*}}!nontemporal
-// CHECK: store float [[RESULT:%.+]], float* {{%.+}}{{.*}}!llvm.access.group
+// CHECK: store float [[RESULT:%.+]], ptr {{%.+}}{{.*}}!llvm.access.group
     a[i] = b[i] * c[i] * d[i] + s.a + p->a;
-// CHECK: [[IV1_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
+// CHECK: [[IV1_2:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[ADD1_2:%.+]] = add nsw i32 [[IV1_2]], 1
-// CHECK-NEXT: store i32 [[ADD1_2]], i32* [[OMP_IV]]{{.*}}!llvm.access.group
+// CHECK-NEXT: store i32 [[ADD1_2]], ptr [[OMP_IV]]{{.*}}!llvm.access.group
 // br label %{{.+}}, !llvm.loop !{{.+}}
   }
 // CHECK: [[SIMPLE_LOOP1_END]]:
@@ -77,138 +77,138 @@ void simple(float *a, float *b, float *c, float *d) {
 
   #pragma omp simd linear(k : 3)
 // CHECK: [[K0:%.+]] = call {{.*}}i64 @{{.*}}get_val
-// CHECK-NEXT: store i64 [[K0]], i64* [[K_VAR:%[^,]+]]
-// CHECK: store i32 0, i32* [[OMP_IV2:%[^,]+]]
-// CHECK: [[K0LOAD:%.+]] = load i64, i64* [[K_VAR]]
-// CHECK-NEXT: store i64 [[K0LOAD]], i64* [[LIN0:%[^,]+]]
+// CHECK-NEXT: store i64 [[K0]], ptr [[K_VAR:%[^,]+]]
+// CHECK: store i32 0, ptr [[OMP_IV2:%[^,]+]]
+// CHECK: [[K0LOAD:%.+]] = load i64, ptr [[K_VAR]]
+// CHECK-NEXT: store i64 [[K0LOAD]], ptr [[LIN0:%[^,]+]]
 
-// CHECK: [[IV2:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.access.group
+// CHECK: [[IV2:%.+]] = load i32, ptr [[OMP_IV2]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[CMP2:%.+]] = icmp slt i32 [[IV2]], 9
 // CHECK-NEXT: br i1 [[CMP2]], label %[[SIMPLE_LOOP2_BODY:.+]], label %[[SIMPLE_LOOP2_END:[^,]+]]
   for (int i = 10; i > 1; i--) {
 // CHECK: [[SIMPLE_LOOP2_BODY]]:
 // Start of body: calculate i from IV:
-// CHECK: [[IV2_0:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.access.group
+// CHECK: [[IV2_0:%.+]] = load i32, ptr [[OMP_IV2]]{{.*}}!llvm.access.group
 // FIXME: It is interesting, why the following "mul 1" was not constant folded?
 // CHECK-NEXT: [[IV2_1:%.+]] = mul nsw i32 [[IV2_0]], 1
 // CHECK-NEXT: [[LC_I_1:%.+]] = sub nsw i32 10, [[IV2_1]]
-// CHECK-NEXT: store i32 [[LC_I_1]], i32* {{.+}}, !llvm.access.group
+// CHECK-NEXT: store i32 [[LC_I_1]], ptr {{.+}}, !llvm.access.group
 //
-// CHECK-NEXT: [[LIN0_1:%.+]] = load i64, i64* [[LIN0]]{{.*}}!llvm.access.group
-// CHECK-NEXT: [[IV2_2:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.access.group
+// CHECK-NEXT: [[LIN0_1:%.+]] = load i64, ptr [[LIN0]]{{.*}}!llvm.access.group
+// CHECK-NEXT: [[IV2_2:%.+]] = load i32, ptr [[OMP_IV2]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[LIN_MUL1:%.+]] = mul nsw i32 [[IV2_2]], 3
 // CHECK-NEXT: [[LIN_EXT1:%.+]] = sext i32 [[LIN_MUL1]] to i64
 // CHECK-NEXT: [[LIN_ADD1:%.+]] = add nsw i64 [[LIN0_1]], [[LIN_EXT1]]
 // Update of the privatized version of linear variable!
-// CHECK-NEXT: store i64 [[LIN_ADD1]], i64* [[K_PRIVATIZED:%[^,]+]]
+// CHECK-NEXT: store i64 [[LIN_ADD1]], ptr [[K_PRIVATIZED:%[^,]+]]
     a[k]++;
     k = k + 3;
-// CHECK: [[IV2_2:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.access.group
+// CHECK: [[IV2_2:%.+]] = load i32, ptr [[OMP_IV2]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[ADD2_2:%.+]] = add nsw i32 [[IV2_2]], 1
-// CHECK-NEXT: store i32 [[ADD2_2]], i32* [[OMP_IV2]]{{.*}}!llvm.access.group
+// CHECK-NEXT: store i32 [[ADD2_2]], ptr [[OMP_IV2]]{{.*}}!llvm.access.group
 // br label {{.+}}, !llvm.loop ![[SIMPLE_LOOP2_ID]]
   }
 // CHECK: [[SIMPLE_LOOP2_END]]:
 //
 // Update linear vars after loop, as the loop was operating on a private version.
-// CHECK: [[LIN0_2:%.+]] = load i64, i64* [[K_PRIVATIZED]],
-// CHECK-NEXT: store i64 [[LIN0_2]], i64* [[K_VAR]]
+// CHECK: [[LIN0_2:%.+]] = load i64, ptr [[K_PRIVATIZED]],
+// CHECK-NEXT: store i64 [[LIN0_2]], ptr [[K_VAR]]
 //
 
   int lin = 12;
   #pragma omp simd linear(lin : get_val()), linear(g_ptr)
 
 // Init linear private var.
-// CHECK: store i32 12, i32* [[LIN_VAR:%[^,]+]]
-// CHECK: store i64 0, i64* [[OMP_IV3:%[^,]+]]
+// CHECK: store i32 12, ptr [[LIN_VAR:%[^,]+]]
+// CHECK: store i64 0, ptr [[OMP_IV3:%[^,]+]]
 
-// CHECK: [[LIN_LOAD:%.+]] = load i32, i32* [[LIN_VAR]]
-// CHECK-NEXT: store i32 [[LIN_LOAD]], i32* [[LIN_START:%[^,]+]]
+// CHECK: [[LIN_LOAD:%.+]] = load i32, ptr [[LIN_VAR]]
+// CHECK-NEXT: store i32 [[LIN_LOAD]], ptr [[LIN_START:%[^,]+]]
 // Remember linear step.
 // CHECK: [[CALL_VAL:%.+]] = invoke
-// CHECK: store i64 [[CALL_VAL]], i64* [[LIN_STEP:%[^,]+]]
+// CHECK: store i64 [[CALL_VAL]], ptr [[LIN_STEP:%[^,]+]]
 
-// CHECK: [[GLIN_LOAD:%.+]] = load double*, double** [[GLIN_VAR:@[^,]+]]
-// CHECK-NEXT: store double* [[GLIN_LOAD]], double** [[GLIN_START:%[^,]+]]
+// CHECK: [[GLIN_LOAD:%.+]] = load ptr, ptr [[GLIN_VAR:@[^,]+]]
+// CHECK-NEXT: store ptr [[GLIN_LOAD]], ptr [[GLIN_START:%[^,]+]]
 
-// CHECK: [[IV3:%.+]] = load i64, i64* [[OMP_IV3]]{{.*}}!llvm.access.group
+// CHECK: [[IV3:%.+]] = load i64, ptr [[OMP_IV3]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[CMP3:%.+]] = icmp ult i64 [[IV3]], 4
 // CHECK-NEXT: br i1 [[CMP3]], label %[[SIMPLE_LOOP3_BODY:.+]], label %[[SIMPLE_LOOP3_END:[^,]+]]
   for (unsigned long long it = 2000; it >= 600; it-=400) {
 // CHECK: [[SIMPLE_LOOP3_BODY]]:
 // Start of body: calculate it from IV:
-// CHECK: [[IV3_0:%.+]] = load i64, i64* [[OMP_IV3]]{{.*}}!llvm.access.group
+// CHECK: [[IV3_0:%.+]] = load i64, ptr [[OMP_IV3]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[LC_IT_1:%.+]] = mul i64 [[IV3_0]], 400
 // CHECK-NEXT: [[LC_IT_2:%.+]] = sub i64 2000, [[LC_IT_1]]
-// CHECK-NEXT: store i64 [[LC_IT_2]], i64* {{.+}}, !llvm.access.group
+// CHECK-NEXT: store i64 [[LC_IT_2]], ptr {{.+}}, !llvm.access.group
 //
 // Linear start and step are used to calculate current value of the linear variable.
-// CHECK: [[LINSTART:.+]] = load i32, i32* [[LIN_START]]{{.*}}!llvm.access.group
-// CHECK: [[LINSTEP:.+]] = load i64, i64* [[LIN_STEP]]{{.*}}!llvm.access.group
-// CHECK-NOT: store i32 {{.+}}, i32* [[LIN_VAR]],{{.*}}!llvm.access.group
-// CHECK: store i32 {{.+}}, i32* [[LIN_PRIV:%.+]],{{.*}}!llvm.access.group
-// CHECK: [[GLINSTART:.+]] = load double*, double** [[GLIN_START]]{{.*}}!llvm.access.group
-// CHECK-NEXT: [[IV3_1:%.+]] = load i64, i64* [[OMP_IV3]]{{.*}}!llvm.access.group
+// CHECK: [[LINSTART:.+]] = load i32, ptr [[LIN_START]]{{.*}}!llvm.access.group
+// CHECK: [[LINSTEP:.+]] = load i64, ptr [[LIN_STEP]]{{.*}}!llvm.access.group
+// CHECK-NOT: store i32 {{.+}}, ptr [[LIN_VAR]],{{.*}}!llvm.access.group
+// CHECK: store i32 {{.+}}, ptr [[LIN_PRIV:%.+]],{{.*}}!llvm.access.group
+// CHECK: [[GLINSTART:.+]] = load ptr, ptr [[GLIN_START]]{{.*}}!llvm.access.group
+// CHECK-NEXT: [[IV3_1:%.+]] = load i64, ptr [[OMP_IV3]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[MUL:%.+]] = mul i64 [[IV3_1]], 1
 // CHECK: [[GEP:%.+]] = getelementptr{{.*}}[[GLINSTART]]
-// CHECK-NEXT: store double* [[GEP]], double** [[G_PTR_CUR:%[^,]+]]{{.*}}!llvm.access.group
+// CHECK-NEXT: store ptr [[GEP]], ptr [[G_PTR_CUR:%[^,]+]]{{.*}}!llvm.access.group
     *g_ptr++ = 0.0;
-// CHECK: [[GEP_VAL:%.+]] = load double{{.*}}[[G_PTR_CUR]]{{.*}}!llvm.access.group
+// CHECK: [[GEP_VAL:%.+]] = load ptr, ptr [[G_PTR_CUR]]{{.*}}!llvm.access.group
 // CHECK: store double{{.*}}[[GEP_VAL]]{{.*}}!llvm.access.group
     a[it + lin]++;
 // CHECK: [[FLT_INC:%.+]] = fadd float
 // CHECK-NEXT: store float [[FLT_INC]],{{.*}}!llvm.access.group
-// CHECK: [[IV3_2:%.+]] = load i64, i64* [[OMP_IV3]]{{.*}}!llvm.access.group
+// CHECK: [[IV3_2:%.+]] = load i64, ptr [[OMP_IV3]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[ADD3_2:%.+]] = add i64 [[IV3_2]], 1
-// CHECK-NEXT: store i64 [[ADD3_2]], i64* [[OMP_IV3]]{{.*}}!llvm.access.group
+// CHECK-NEXT: store i64 [[ADD3_2]], ptr [[OMP_IV3]]{{.*}}!llvm.access.group
   }
 // CHECK: [[SIMPLE_LOOP3_END]]:
 //
 // Linear start and step are used to calculate final value of the linear variables.
-// CHECK: [[LIN:%.+]] = load i32, i32* [[LIN_PRIV]]
-// CHECK-NEXT: store i32 [[LIN]], i32* [[LIN_VAR]],
-// CHECK: [[GLIN:%.+]] = load double*, double** [[G_PTR_CUR]]
-// CHECK-NEXT: store double* [[GLIN]], double** [[GLIN_VAR]]
+// CHECK: [[LIN:%.+]] = load i32, ptr [[LIN_PRIV]]
+// CHECK-NEXT: store i32 [[LIN]], ptr [[LIN_VAR]],
+// CHECK: [[GLIN:%.+]] = load ptr, ptr [[G_PTR_CUR]]
+// CHECK-NEXT: store ptr [[GLIN]], ptr [[GLIN_VAR]]
 
   #pragma omp simd
-// CHECK: store i32 0, i32* [[OMP_IV4:%[^,]+]]
+// CHECK: store i32 0, ptr [[OMP_IV4:%[^,]+]]
 
-// CHECK: [[IV4:%.+]] = load i32, i32* [[OMP_IV4]]{{.*}}!llvm.access.group
+// CHECK: [[IV4:%.+]] = load i32, ptr [[OMP_IV4]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[CMP4:%.+]] = icmp slt i32 [[IV4]], 4
 // CHECK-NEXT: br i1 [[CMP4]], label %[[SIMPLE_LOOP4_BODY:.+]], label %[[SIMPLE_LOOP4_END:[^,]+]]
   for (short it = 6; it <= 20; it-=-4) {
 // CHECK: [[SIMPLE_LOOP4_BODY]]:
 // Start of body: calculate it from IV:
-// CHECK: [[IV4_0:%.+]] = load i32, i32* [[OMP_IV4]]{{.*}}!llvm.access.group
+// CHECK: [[IV4_0:%.+]] = load i32, ptr [[OMP_IV4]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[LC_IT_1:%.+]] = mul nsw i32 [[IV4_0]], 4
 // CHECK-NEXT: [[LC_IT_2:%.+]] = add nsw i32 6, [[LC_IT_1]]
 // CHECK-NEXT: [[LC_IT_3:%.+]] = trunc i32 [[LC_IT_2]] to i16
-// CHECK-NEXT: store i16 [[LC_IT_3]], i16* {{.+}}, !llvm.access.group
+// CHECK-NEXT: store i16 [[LC_IT_3]], ptr {{.+}}, !llvm.access.group
 
-// CHECK: [[IV4_2:%.+]] = load i32, i32* [[OMP_IV4]]{{.*}}!llvm.access.group
+// CHECK: [[IV4_2:%.+]] = load i32, ptr [[OMP_IV4]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[ADD4_2:%.+]] = add nsw i32 [[IV4_2]], 1
-// CHECK-NEXT: store i32 [[ADD4_2]], i32* [[OMP_IV4]]{{.*}}!llvm.access.group
+// CHECK-NEXT: store i32 [[ADD4_2]], ptr [[OMP_IV4]]{{.*}}!llvm.access.group
   }
 // CHECK: [[SIMPLE_LOOP4_END]]:
 
   #pragma omp simd
-// CHECK: store i32 0, i32* [[OMP_IV5:%[^,]+]]
+// CHECK: store i32 0, ptr [[OMP_IV5:%[^,]+]]
 
-// CHECK: [[IV5:%.+]] = load i32, i32* [[OMP_IV5]]{{.*}}!llvm.access.group
+// CHECK: [[IV5:%.+]] = load i32, ptr [[OMP_IV5]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[CMP5:%.+]] = icmp slt i32 [[IV5]], 26
 // CHECK-NEXT: br i1 [[CMP5]], label %[[SIMPLE_LOOP5_BODY:.+]], label %[[SIMPLE_LOOP5_END:[^,]+]]
   for (unsigned char it = 'z'; it >= 'a'; it+=-1) {
 // CHECK: [[SIMPLE_LOOP5_BODY]]:
 // Start of body: calculate it from IV:
-// CHECK: [[IV5_0:%.+]] = load i32, i32* [[OMP_IV5]]{{.*}}!llvm.access.group
+// CHECK: [[IV5_0:%.+]] = load i32, ptr [[OMP_IV5]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[IV5_1:%.+]] = mul nsw i32 [[IV5_0]], 1
 // CHECK-NEXT: [[LC_IT_1:%.+]] = sub nsw i32 122, [[IV5_1]]
 // CHECK-NEXT: [[LC_IT_2:%.+]] = trunc i32 [[LC_IT_1]] to i8
-// CHECK-NEXT: store i8 [[LC_IT_2]], i8* {{.+}}, !llvm.access.group
+// CHECK-NEXT: store i8 [[LC_IT_2]], ptr {{.+}}, !llvm.access.group
 
-// CHECK: [[IV5_2:%.+]] = load i32, i32* [[OMP_IV5]]{{.*}}!llvm.access.group
+// CHECK: [[IV5_2:%.+]] = load i32, ptr [[OMP_IV5]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[ADD5_2:%.+]] = add nsw i32 [[IV5_2]], 1
-// CHECK-NEXT: store i32 [[ADD5_2]], i32* [[OMP_IV5]]{{.*}}!llvm.access.group
+// CHECK-NEXT: store i32 [[ADD5_2]], ptr [[OMP_IV5]]{{.*}}!llvm.access.group
   }
 // CHECK: [[SIMPLE_LOOP5_END]]:
 
@@ -218,112 +218,112 @@ void simple(float *a, float *b, float *c, float *d) {
   }
 
   int A;
-  // CHECK: store i32 -1, i32* [[A:%.+]],
+  // CHECK: store i32 -1, ptr [[A:%.+]],
   A = -1;
   #pragma omp simd lastprivate(CONDITIONAL A)
-// CHECK: store i64 0, i64* [[OMP_IV7:%[^,]+]]
+// CHECK: store i64 0, ptr [[OMP_IV7:%[^,]+]]
 // CHECK: br label %[[SIMD_LOOP7_COND:[^,]+]]
 // CHECK: [[SIMD_LOOP7_COND]]:
-// CHECK-NEXT: [[IV7:%.+]] = load i64, i64* [[OMP_IV7]]{{.*}}!llvm.access.group
+// CHECK-NEXT: [[IV7:%.+]] = load i64, ptr [[OMP_IV7]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[CMP7:%.+]] = icmp slt i64 [[IV7]], 7
 // CHECK-NEXT: br i1 [[CMP7]], label %[[SIMPLE_LOOP7_BODY:.+]], label %[[SIMPLE_LOOP7_END:[^,]+]]
   for (long long i = -10; i < 10; i += 3) {
 // CHECK: [[SIMPLE_LOOP7_BODY]]:
 // Start of body: calculate i from IV:
-// CHECK: [[IV7_0:%.+]] = load i64, i64* [[OMP_IV7]]{{.*}}!llvm.access.group
+// CHECK: [[IV7_0:%.+]] = load i64, ptr [[OMP_IV7]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[LC_IT_1:%.+]] = mul nsw i64 [[IV7_0]], 3
 // CHECK-NEXT: [[LC_IT_2:%.+]] = add nsw i64 -10, [[LC_IT_1]]
-// CHECK-NEXT: store i64 [[LC_IT_2]], i64* [[LC:%[^,]+]],{{.+}}!llvm.access.group
-// CHECK-NEXT: [[LC_VAL:%.+]] = load i64, i64* [[LC]]{{.+}}!llvm.access.group
-// CHECK-NEXT: [[A_VAL:%.+]] = load i32, i32* [[A_PRIV:%[^,]+]],{{.+}}!llvm.access.group
+// CHECK-NEXT: store i64 [[LC_IT_2]], ptr [[LC:%[^,]+]],{{.+}}!llvm.access.group
+// CHECK-NEXT: [[LC_VAL:%.+]] = load i64, ptr [[LC]]{{.+}}!llvm.access.group
+// CHECK-NEXT: [[A_VAL:%.+]] = load i32, ptr [[A_PRIV:%[^,]+]],{{.+}}!llvm.access.group
 // CHECK-NEXT: [[CAST:%.+]] = sext i32 [[A_VAL]] to i64
 // CHECK-NEXT: [[ADD:%.+]] = add nsw i64 [[CAST]], [[LC_VAL]]
 // CHECK-NEXT: [[CONV:%.+]] = trunc i64 [[ADD]] to i32
-// CHECK-NEXT: store i32 [[CONV]], i32* [[A_PRIV]],{{.+}}!llvm.access.group
-// OMP50-NEXT: [[IV:%.+]] = load i64, i64* [[OMP_IV7]],{{.+}}!llvm.access.group
-// OMP50RT:    call void @__kmpc_critical(%struct.ident_t* {{.+}}, i32 [[GTID:%.+]], [8 x i32]* [[A_REGION:@.+]]),{{.+}}!llvm.access.group
-// OMP50-NEXT: [[LAST_IV_VAL:%.+]] = load i64, i64* [[LAST_IV]],{{.+}}!llvm.access.group
+// CHECK-NEXT: store i32 [[CONV]], ptr [[A_PRIV]],{{.+}}!llvm.access.group
+// OMP50-NEXT: [[IV:%.+]] = load i64, ptr [[OMP_IV7]],{{.+}}!llvm.access.group
+// OMP50RT:    call void @__kmpc_critical(ptr {{.+}}, i32 [[GTID:%.+]], ptr [[A_REGION:@.+]]),{{.+}}!llvm.access.group
+// OMP50-NEXT: [[LAST_IV_VAL:%.+]] = load i64, ptr [[LAST_IV]],{{.+}}!llvm.access.group
 // OMP50-NEXT: [[CMP:%.+]] = icmp sle i64 [[LAST_IV_VAL]], [[IV]]
 // OMP50-NEXT: br i1 [[CMP]], label %[[LP_THEN:.+]], label %[[LP_DONE:[^,]+]]
 // OMP50:      [[LP_THEN]]:
-// OMP50-NEXT: store i64 [[IV]], i64* [[LAST_IV]],{{.+}}!llvm.access.group
-// OMP50-NEXT: [[A_VAL:%.+]] = load i32, i32* [[A_PRIV]],{{.+}}!llvm.access.group
-// OMP50-NEXT: store i32 [[A_VAL]], i32* [[LAST_A]],{{.+}}!llvm.access.group
+// OMP50-NEXT: store i64 [[IV]], ptr [[LAST_IV]],{{.+}}!llvm.access.group
+// OMP50-NEXT: [[A_VAL:%.+]] = load i32, ptr [[A_PRIV]],{{.+}}!llvm.access.group
+// OMP50-NEXT: store i32 [[A_VAL]], ptr [[LAST_A]],{{.+}}!llvm.access.group
 // OMP50-NEXT: br label %[[LP_DONE]]
 // OMP50:      [[LP_DONE]]:
-// OMP50RT-NEXT: call void @__kmpc_end_critical(%struct.ident_t* {{.+}}, i32 [[GTID]], [8 x i32]* [[A_REGION]]),{{.+}}!llvm.access.group
+// OMP50RT-NEXT: call void @__kmpc_end_critical(ptr {{.+}}, i32 [[GTID]], ptr [[A_REGION]]),{{.+}}!llvm.access.group
     A += i;
-// CHECK: [[IV7_2:%.+]] = load i64, i64* [[OMP_IV7]]{{.*}}!llvm.access.group
+// CHECK: [[IV7_2:%.+]] = load i64, ptr [[OMP_IV7]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[ADD7_2:%.+]] = add nsw i64 [[IV7_2]], 1
-// CHECK-NEXT: store i64 [[ADD7_2]], i64* [[OMP_IV7]]{{.*}}!llvm.access.group
+// CHECK-NEXT: store i64 [[ADD7_2]], ptr [[OMP_IV7]]{{.*}}!llvm.access.group
   }
 // CHECK: [[SIMPLE_LOOP7_END]]:
-// CHECK-NEXT: store i64 11, i64*
-// OMP50-NEXT: [[LAST_A_VAL:%.+]] = load i32, i32* [[LAST_A]],
-// OMP50-NEXT: store i32 [[LAST_A_VAL]], i32* [[A_PRIV]],
-// CHECK-NEXT: [[A_PRIV_VAL:%.+]] = load i32, i32* [[A_PRIV]],
-// CHECK-NEXT: store i32 [[A_PRIV_VAL]], i32* [[A]],
+// CHECK-NEXT: store i64 11, ptr
+// OMP50-NEXT: [[LAST_A_VAL:%.+]] = load i32, ptr [[LAST_A]],
+// OMP50-NEXT: store i32 [[LAST_A_VAL]], ptr [[A_PRIV]],
+// CHECK-NEXT: [[A_PRIV_VAL:%.+]] = load i32, ptr [[A_PRIV]],
+// CHECK-NEXT: store i32 [[A_PRIV_VAL]], ptr [[A]],
   int R;
-  // CHECK: store i32 -1, i32* [[R:%[^,]+]],
+  // CHECK: store i32 -1, ptr [[R:%[^,]+]],
   R = -1;
-// CHECK: store i64 0, i64* [[OMP_IV8:%[^,]+]],
-// CHECK: store i32 1, i32* [[R_PRIV:%[^,]+]],
+// CHECK: store i64 0, ptr [[OMP_IV8:%[^,]+]],
+// CHECK: store i32 1, ptr [[R_PRIV:%[^,]+]],
 #ifdef OMP5
   #pragma omp simd reduction(*:R) if(A)
 #else
   #pragma omp simd reduction(*:R)
 #endif
-// OMP50:      [[A_VAL:%.+]] = load i32, i32* [[A]],
+// OMP50:      [[A_VAL:%.+]] = load i32, ptr [[A]],
 // OMP50-NEXT: [[COND:%.+]] = icmp ne i32 [[A_VAL]], 0
 // OMP50-NEXT: br i1 [[COND]], label {{%?}}[[THEN:[^,]+]], label {{%?}}[[ELSE:[^,]+]]
 // OMP50:      [[THEN]]:
 
 // CHECK: br label %[[SIMD_LOOP8_COND:[^,]+]]
 // CHECK: [[SIMD_LOOP8_COND]]:
-// CHECK-NEXT: [[IV8:%.+]] = load i64, i64* [[OMP_IV8]]{{.*}}!llvm.access.group
+// CHECK-NEXT: [[IV8:%.+]] = load i64, ptr [[OMP_IV8]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[CMP8:%.+]] = icmp slt i64 [[IV8]], 7
 // CHECK-NEXT: br i1 [[CMP8]], label %[[SIMPLE_LOOP8_BODY:.+]], label %[[SIMPLE_LOOP8_END:[^,]+]]
   for (long long i = -10; i < 10; i += 3) {
 // CHECK: [[SIMPLE_LOOP8_BODY]]:
 // Start of body: calculate i from IV:
-// CHECK: [[IV8_0:%.+]] = load i64, i64* [[OMP_IV8]]{{.*}}!llvm.access.group
+// CHECK: [[IV8_0:%.+]] = load i64, ptr [[OMP_IV8]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[LC_IT_1:%.+]] = mul nsw i64 [[IV8_0]], 3
 // CHECK-NEXT: [[LC_IT_2:%.+]] = add nsw i64 -10, [[LC_IT_1]]
-// CHECK-NEXT: store i64 [[LC_IT_2]], i64* [[LC:%[^,]+]],{{.+}}!llvm.access.group
-// CHECK-NEXT: [[LC_VAL:%.+]] = load i64, i64* [[LC]]{{.+}}!llvm.access.group
-// CHECK: store i32 %{{.+}}, i32* [[R_PRIV]],{{.+}}!llvm.access.group
+// CHECK-NEXT: store i64 [[LC_IT_2]], ptr [[LC:%[^,]+]],{{.+}}!llvm.access.group
+// CHECK-NEXT: [[LC_VAL:%.+]] = load i64, ptr [[LC]]{{.+}}!llvm.access.group
+// CHECK: store i32 %{{.+}}, ptr [[R_PRIV]],{{.+}}!llvm.access.group
     R *= i;
-// CHECK: [[IV8_2:%.+]] = load i64, i64* [[OMP_IV8]]{{.*}}!llvm.access.group
+// CHECK: [[IV8_2:%.+]] = load i64, ptr [[OMP_IV8]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[ADD8_2:%.+]] = add nsw i64 [[IV8_2]], 1
-// CHECK-NEXT: store i64 [[ADD8_2]], i64* [[OMP_IV8]]{{.*}}!llvm.access.group
+// CHECK-NEXT: store i64 [[ADD8_2]], ptr [[OMP_IV8]]{{.*}}!llvm.access.group
   }
 // CHECK: [[SIMPLE_LOOP8_END]]:
 // OMP50: br label {{%?}}[[EXIT:[^,]+]]
 // OMP50: br label %[[SIMD_LOOP8_COND:[^,]+]]
 // OMP50: [[SIMD_LOOP8_COND]]:
-// OMP50-NEXT: [[IV8:%.+]] = load i64, i64* [[OMP_IV8]],{{[^!]*}}
+// OMP50-NEXT: [[IV8:%.+]] = load i64, ptr [[OMP_IV8]],{{[^!]*}}
 // OMP50-NEXT: [[CMP8:%.+]] = icmp slt i64 [[IV8]], 7
 // OMP50-NEXT: br i1 [[CMP8]], label %[[SIMPLE_LOOP8_BODY:.+]], label %[[SIMPLE_LOOP8_END:[^,]+]]
 // OMP50: [[SIMPLE_LOOP8_BODY]]:
 // Start of body: calculate i from IV:
-// OMP50: [[IV8_0:%.+]] = load i64, i64* [[OMP_IV8]],{{[^!]*}}
+// OMP50: [[IV8_0:%.+]] = load i64, ptr [[OMP_IV8]],{{[^!]*}}
 // OMP50-NEXT: [[LC_IT_1:%.+]] = mul nsw i64 [[IV8_0]], 3
 // OMP50-NEXT: [[LC_IT_2:%.+]] = add nsw i64 -10, [[LC_IT_1]]
-// OMP50-NEXT: store i64 [[LC_IT_2]], i64* [[LC:%[^,]+]],{{[^!]*}}
-// OMP50-NEXT: [[LC_VAL:%.+]] = load i64, i64* [[LC]],{{[^!]*}}
-// OMP50: store i32 %{{.+}}, i32* [[R_PRIV]],{{[^!]*}}
-// OMP50: [[IV8_2:%.+]] = load i64, i64* [[OMP_IV8]],{{[^!]*}}
+// OMP50-NEXT: store i64 [[LC_IT_2]], ptr [[LC:%[^,]+]],{{[^!]*}}
+// OMP50-NEXT: [[LC_VAL:%.+]] = load i64, ptr [[LC]],{{[^!]*}}
+// OMP50: store i32 %{{.+}}, ptr [[R_PRIV]],{{[^!]*}}
+// OMP50: [[IV8_2:%.+]] = load i64, ptr [[OMP_IV8]],{{[^!]*}}
 // OMP50-NEXT: [[ADD8_2:%.+]] = add nsw i64 [[IV8_2]], 1
-// OMP50-NEXT: store i64 [[ADD8_2]], i64* [[OMP_IV8]],{{[^!]*}}
+// OMP50-NEXT: store i64 [[ADD8_2]], ptr [[OMP_IV8]],{{[^!]*}}
 // OMP50:      br label {{%?}}[[SIMD_LOOP8_COND]], {{.*}}!llvm.loop ![[DISABLE_VECT:.+]]
 // OMP50: [[SIMPLE_LOOP8_END]]:
 // OMP50: br label {{%?}}[[EXIT]]
 // OMP50: [[EXIT]]:
 
-// CHECK-DAG: [[R_VAL:%.+]] = load i32, i32* [[R]],
-// CHECK-DAG: [[R_PRIV_VAL:%.+]] = load i32, i32* [[R_PRIV]],
+// CHECK-DAG: [[R_VAL:%.+]] = load i32, ptr [[R]],
+// CHECK-DAG: [[R_PRIV_VAL:%.+]] = load i32, ptr [[R_PRIV]],
 // CHECK: [[RED:%.+]] = mul nsw i32 [[R_VAL]], [[R_PRIV_VAL]]
-// CHECK-NEXT: store i32 [[RED]], i32* [[R]],
+// CHECK-NEXT: store i32 [[RED]], ptr [[R]],
 // CHECK-NEXT: ret void
 }
 
@@ -341,32 +341,32 @@ int templ1(T a, T *z) {
 }
 
 // Instatiation templ1<float,2>
-// CHECK-LABEL: define {{.*i32}} @{{.*}}templ1{{.*}}(float noundef {{.+}}, float* noundef {{.+}})
-// CHECK: store i64 0, i64* [[T1_OMP_IV:[^,]+]]
+// CHECK-LABEL: define {{.*i32}} @{{.*}}templ1{{.*}}(float noundef {{.+}}, ptr noundef {{.+}})
+// CHECK: store i64 0, ptr [[T1_OMP_IV:[^,]+]]
 // ...
-// CHECK: [[IV:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group
+// CHECK: [[IV:%.+]] = load i64, ptr [[T1_OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[CMP1:%.+]] = icmp slt i64 [[IV]], 16
 // CHECK-NEXT: br i1 [[CMP1]], label %[[T1_BODY:.+]], label %[[T1_END:[^,]+]]
 // CHECK: [[T1_BODY]]:
 // Loop counters i and j updates:
-// CHECK: [[IV1:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group
+// CHECK: [[IV1:%.+]] = load i64, ptr [[T1_OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[I_1:%.+]] = sdiv i64 [[IV1]], 4
 // CHECK-NEXT: [[I_1_MUL1:%.+]] = mul nsw i64 [[I_1]], 1
 // CHECK-NEXT: [[I_1_ADD0:%.+]] = add nsw i64 0, [[I_1_MUL1]]
 // CHECK-NEXT: [[I_2:%.+]] = trunc i64 [[I_1_ADD0]] to i32
-// CHECK-NEXT: store i32 [[I_2]], i32* {{%.+}}{{.*}}!llvm.access.group
-// CHECK: [[IV2:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group
-// CHECK: [[IV2_1:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group
+// CHECK-NEXT: store i32 [[I_2]], ptr {{%.+}}{{.*}}!llvm.access.group
+// CHECK: [[IV2:%.+]] = load i64, ptr [[T1_OMP_IV]]{{.*}}!llvm.access.group
+// CHECK: [[IV2_1:%.+]] = load i64, ptr [[T1_OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[J_1_DIV1:%.+]] = sdiv i64 [[IV2_1]], 4
 // CHECK-NEXT: [[J_1_MUL1:%.+]] = mul nsw i64 [[J_1_DIV1]], 4
 // CHECK-NEXT: [[J_1_SUB0:%.+]] = sub nsw i64 [[IV2]], [[J_1_MUL1]]
 // CHECK-NEXT: [[J_2:%.+]] = mul nsw i64 [[J_1_SUB0]], 2
 // CHECK-NEXT: [[J_2_ADD0:%.+]] = add nsw i64 0, [[J_2]]
-// CHECK-NEXT: store i64 [[J_2_ADD0]], i64* {{%.+}}{{.*}}!llvm.access.group
+// CHECK-NEXT: store i64 [[J_2_ADD0]], ptr {{%.+}}{{.*}}!llvm.access.group
 // simd.for.inc:
-// CHECK: [[IV3:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group
+// CHECK: [[IV3:%.+]] = load i64, ptr [[T1_OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[INC:%.+]] = add nsw i64 [[IV3]], 1
-// CHECK-NEXT: store i64 [[INC]], i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group
+// CHECK-NEXT: store i64 [[INC]], ptr [[T1_OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: br label {{%.+}}
 // CHECK: [[T1_END]]:
 // CHECK: ret i32 0
@@ -415,19 +415,19 @@ void iter_simple(IterDouble ia, IterDouble ib, IterDouble ic) {
 // CHECK-NEXT: [[DIFF3:%.+]] = add nsw i32 [[DIFF2]], 1
 // CHECK-NEXT: [[DIFF4:%.+]] = sdiv i32 [[DIFF3]], 1
 // CHECK-NEXT: [[DIFF5:%.+]] = sub nsw i32 [[DIFF4]], 1
-// CHECK-NEXT: store i32 [[DIFF5]], i32* [[OMP_LAST_IT:%[^,]+]]{{.+}}
-// CHECK: store i32 0, i32* [[IT_OMP_IV:%[^,]+]]
+// CHECK-NEXT: store i32 [[DIFF5]], ptr [[OMP_LAST_IT:%[^,]+]]{{.+}}
+// CHECK: store i32 0, ptr [[IT_OMP_IV:%[^,]+]]
   #pragma omp simd
 
-// CHECK: [[IV:%.+]] = load i32, i32* [[IT_OMP_IV]]{{.+}} !llvm.access.group
-// CHECK-NEXT: [[LAST_IT:%.+]] = load i32, i32* [[OMP_LAST_IT]]{{.+}}!llvm.access.group
+// CHECK: [[IV:%.+]] = load i32, ptr [[IT_OMP_IV]]{{.+}} !llvm.access.group
+// CHECK-NEXT: [[LAST_IT:%.+]] = load i32, ptr [[OMP_LAST_IT]]{{.+}}!llvm.access.group
 // CHECK-NEXT: [[NUM_IT:%.+]] = add nsw i32 [[LAST_IT]], 1
 // CHECK-NEXT: [[CMP:%.+]] = icmp slt i32 [[IV]], [[NUM_IT]]
 // CHECK-NEXT: br i1 [[CMP]], label %[[IT_BODY:[^,]+]], label %[[IT_END:[^,]+]]
   for (IterDouble i = ia; i < ib; ++i) {
 // CHECK: [[IT_BODY]]:
 // Start of body: calculate i from index:
-// CHECK: [[IV1:%.+]] = load i32, i32* [[IT_OMP_IV]]{{.+}}!llvm.access.group
+// CHECK: [[IV1:%.+]] = load i32, ptr [[IT_OMP_IV]]{{.+}}!llvm.access.group
 // Call of operator+ (i, IV).
 // CHECK: {{%.+}} = invoke {{.+}} @{{.*}}IterDouble{{.*}}
 // ... loop body ...
@@ -435,12 +435,12 @@ void iter_simple(IterDouble ia, IterDouble ib, IterDouble ic) {
 // Float multiply and save result.
 // CHECK: [[MULR:%.+]] = fmul double {{%.+}}, 5.000000e-01
 // CHECK-NEXT: invoke {{.+}} @{{.*}}IterDouble{{.*}}
-// CHECK: store double [[MULR:%.+]], double* [[RESULT_ADDR:%.+]], !llvm.access.group
+// CHECK: store double [[MULR:%.+]], ptr [[RESULT_ADDR:%.+]], !llvm.access.group
    ++ic;
 //
-// CHECK: [[IV2:%.+]] = load i32, i32* [[IT_OMP_IV]]{{.+}}!llvm.access.group
+// CHECK: [[IV2:%.+]] = load i32, ptr [[IT_OMP_IV]]{{.+}}!llvm.access.group
 // CHECK-NEXT: [[ADD2:%.+]] = add nsw i32 [[IV2]], 1
-// CHECK-NEXT: store i32 [[ADD2]], i32* [[IT_OMP_IV]]{{.+}}!llvm.access.group
+// CHECK-NEXT: store i32 [[ADD2]], ptr [[IT_OMP_IV]]{{.+}}!llvm.access.group
 // br label %{{.*}}, !llvm.loop ![[ITER_LOOP_ID]]
   }
 // CHECK: [[IT_END]]:
@@ -454,11 +454,11 @@ void collapsed(float *a, float *b, float *c, float *d) {
   unsigned j; // middle loop couter, leads to unsigned icmp in loop header.
   // k declared in the loop init below
   short l; // inner loop counter
-// CHECK: store i32 0, i32* [[OMP_IV:[^,]+]]
+// CHECK: store i32 0, ptr [[OMP_IV:[^,]+]]
 //
   #pragma omp simd collapse(4)
 
-// CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
+// CHECK: [[IV:%.+]] = load i32, ptr [[OMP_IV]]{{.+}}!llvm.access.group
 // CHECK-NEXT: [[CMP:%.+]] = icmp ult i32 [[IV]], 120
 // CHECK-NEXT: br i1 [[CMP]], label %[[COLL1_BODY:[^,]+]], label %[[COLL1_END:[^,]+]]
   for (i = 1; i < 3; i++) // 2 iterations
@@ -468,31 +468,31 @@ void collapsed(float *a, float *b, float *c, float *d) {
         {
 // CHECK: [[COLL1_BODY]]:
 // Start of body: calculate i from index:
-// CHECK: [[IV1:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
+// CHECK: [[IV1:%.+]] = load i32, ptr [[OMP_IV]]{{.+}}!llvm.access.group
 // Calculation of the loop counters values.
 // CHECK: [[CALC_I_1:%.+]] = udiv i32 [[IV1]], 60
 // CHECK-NEXT: [[CALC_I_1_MUL1:%.+]] = mul i32 [[CALC_I_1]], 1
 // CHECK-NEXT: [[CALC_I_2:%.+]] = add i32 1, [[CALC_I_1_MUL1]]
-// CHECK-NEXT: store i32 [[CALC_I_2]], i32* [[LC_I:.+]]
+// CHECK-NEXT: store i32 [[CALC_I_2]], ptr [[LC_I:.+]]
 
-// CHECK: [[IV1_2:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
-// CHECK: [[IV1_2_1:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
+// CHECK: [[IV1_2:%.+]] = load i32, ptr [[OMP_IV]]{{.+}}!llvm.access.group
+// CHECK: [[IV1_2_1:%.+]] = load i32, ptr [[OMP_IV]]{{.+}}!llvm.access.group
 // CHECK-NEXT: [[CALC_J_1:%.+]] = udiv i32 [[IV1_2_1]], 60
 // CHECK-NEXT: [[MUL_1:%.+]] = mul i32 [[CALC_J_1]], 60
 // CHECK-NEXT: [[SUB_3:%.+]] = sub i32 [[IV1_2]], [[MUL_1]]
 // CHECK-NEXT: [[CALC_J_2:%.+]] = udiv i32 [[SUB_3]], 20
 // CHECK-NEXT: [[CALC_J_2_MUL1:%.+]] = mul i32 [[CALC_J_2]], 1
 // CHECK-NEXT: [[CALC_J_3:%.+]] = add i32 2, [[CALC_J_2_MUL1]]
-// CHECK-NEXT: store i32 [[CALC_J_3]], i32* [[LC_J:.+]]
+// CHECK-NEXT: store i32 [[CALC_J_3]], ptr [[LC_J:.+]]
 
-// CHECK: [[IV1_3:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
-// CHECK: [[IV1_3_1:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
+// CHECK: [[IV1_3:%.+]] = load i32, ptr [[OMP_IV]]{{.+}}!llvm.access.group
+// CHECK: [[IV1_3_1:%.+]] = load i32, ptr [[OMP_IV]]{{.+}}!llvm.access.group
 // CHECK-NEXT: [[DIV_1:%.+]] = udiv i32 [[IV1_3_1]], 60
 // CHECK-NEXT: [[MUL_2:%.+]] = mul i32 [[DIV_1]], 60
 // CHECK-NEXT: [[ADD_3:%.+]] = sub i32 [[IV1_3]], [[MUL_2]]
 
-// CHECK: [[IV1_4:%.+]] = load i32, i32* [[OMP_IV]]
-// CHECK: [[IV1_4_1:%.+]] = load i32, i32* [[OMP_IV]]
+// CHECK: [[IV1_4:%.+]] = load i32, ptr [[OMP_IV]]
+// CHECK: [[IV1_4_1:%.+]] = load i32, ptr [[OMP_IV]]
 // CHECK-NEXT: [[DIV_2:%.+]] = udiv i32 [[IV1_4_1]], 60
 // CHECK-NEXT: [[MUL_3:%.+]] = mul i32 [[DIV_2]], 60
 // CHECK-NEXT: [[SUB_6:%.+]] = sub i32 [[IV1_4]], [[MUL_3]]
@@ -502,16 +502,16 @@ void collapsed(float *a, float *b, float *c, float *d) {
 // CHECK-NEXT: [[DIV_4:%.+]] = udiv i32 [[ADD_5]], 5
 // CHECK-NEXT: [[MUL_5:%.+]] = mul i32 [[DIV_4]], 1
 // CHECK-NEXT: [[ADD_6:%.+]] = add i32 3, [[MUL_5]]
-// CHECK-NEXT: store i32 [[ADD_6]], i32* [[LC_K:.+]]
+// CHECK-NEXT: store i32 [[ADD_6]], ptr [[LC_K:.+]]
 
-// CHECK: [[IV1_5:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
-// CHECK: [[IV1_5_1:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
+// CHECK: [[IV1_5:%.+]] = load i32, ptr [[OMP_IV]]{{.+}}!llvm.access.group
+// CHECK: [[IV1_5_1:%.+]] = load i32, ptr [[OMP_IV]]{{.+}}!llvm.access.group
 // CHECK-NEXT: [[DIV_5:%.+]] = udiv i32 [[IV1_5_1]], 60
 // CHECK-NEXT: [[MUL_6:%.+]] = mul i32 [[DIV_5]], 60
 // CHECK-NEXT: [[SUB_7:%.+]] = sub i32 [[IV1_5]], [[MUL_6]]
 
-// CHECK: [[IV1_6:%.+]] = load i32, i32* [[OMP_IV]]
-// CHECK: [[IV1_6_1:%.+]] = load i32, i32* [[OMP_IV]]
+// CHECK: [[IV1_6:%.+]] = load i32, ptr [[OMP_IV]]
+// CHECK: [[IV1_6_1:%.+]] = load i32, ptr [[OMP_IV]]
 // CHECK-NEXT: [[DIV_6:%.+]] = udiv i32 [[IV1_6_1]], 60
 // CHECK-NEXT: [[MUL_7:%.+]] = mul i32 [[DIV_6]], 60
 // CHECK-NEXT: [[SUB_10:%.+]] = sub i32 [[IV1_6]], [[MUL_7]]
@@ -519,14 +519,14 @@ void collapsed(float *a, float *b, float *c, float *d) {
 // CHECK-NEXT: [[MUL_8:%.+]] = mul i32 [[DIV_7]], 20
 // CHECK-NEXT: [[SUB_11:%.+]] = sub i32 [[SUB_7]], [[MUL_8]]
 
-// CHECK: [[IV1_7:%.+]] = load i32, i32* [[OMP_IV]]
-// CHECK: [[IV1_7_1:%.+]] = load i32, i32* [[OMP_IV]]
+// CHECK: [[IV1_7:%.+]] = load i32, ptr [[OMP_IV]]
+// CHECK: [[IV1_7_1:%.+]] = load i32, ptr [[OMP_IV]]
 // CHECK-NEXT: [[DIV_8:%.+]] = udiv i32 [[IV1_7_1]], 60
 // CHECK-NEXT: [[MUL_9:%.+]] = mul i32 [[DIV_8]], 60
 // CHECK-NEXT: [[SUB_12:%.+]] = sub i32 [[IV1_7]], [[MUL_9]]
 
-// CHECK: [[IV1_8:%.+]] = load i32, i32* [[OMP_IV]]
-// CHECK: [[IV1_8_1:%.+]] = load i32, i32* [[OMP_IV]]
+// CHECK: [[IV1_8:%.+]] = load i32, ptr [[OMP_IV]]
+// CHECK: [[IV1_8_1:%.+]] = load i32, ptr [[OMP_IV]]
 // CHECK-NEXT: [[DIV_3:%.+]] = udiv i32 [[IV1_8_1]], 60
 // CHECK-NEXT: [[MUL_4:%.+]] = mul i32 [[DIV_3]], 60
 // CHECK-NEXT: [[SUB_7:%.+]] = sub i32 [[IV1_8]], [[MUL_4]]
@@ -539,23 +539,23 @@ void collapsed(float *a, float *b, float *c, float *d) {
 // CHECK-NEXT: [[MUL_6:%.+]] = mul i32 [[SUB_9]], 1
 // CHECK-NEXT: [[CALC_L_2:%.+]] = add i32 4, [[MUL_6]]
 // CHECK-NEXT: [[CALC_L_3:%.+]] = trunc i32 [[CALC_L_2]] to i16
-// CHECK-NEXT: store i16 [[CALC_L_3]], i16* [[LC_L:.+]]
+// CHECK-NEXT: store i16 [[CALC_L_3]], ptr [[LC_L:.+]]
 // ... loop body ...
 // End of body: store into a[i]:
-// CHECK: store float [[RESULT:%.+]], float* [[RESULT_ADDR:%.+]]{{.+}}!llvm.access.group
+// CHECK: store float [[RESULT:%.+]], ptr [[RESULT_ADDR:%.+]]{{.+}}!llvm.access.group
     float res = b[j] * c[k];
     a[i] = res * d[l];
-// CHECK: [[IV2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
+// CHECK: [[IV2:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[ADD2:%.+]] = add i32 [[IV2]], 1
-// CHECK-NEXT: store i32 [[ADD2]], i32* [[OMP_IV]]{{.*}}!llvm.access.group
+// CHECK-NEXT: store i32 [[ADD2]], ptr [[OMP_IV]]{{.*}}!llvm.access.group
 // br label %{{[^,]+}}, !llvm.loop ![[COLL1_LOOP_ID]]
 // CHECK: [[COLL1_END]]:
   }
 // i,j,l are updated; k is not updated.
-// CHECK: store i32 3, i32*
-// CHECK-NEXT: store i32 5, i32*
-// CHECK-NEXT: store i32 7, i32*
-// CHECK-NEXT: store i16 9, i16*
+// CHECK: store i32 3, ptr
+// CHECK-NEXT: store i32 5, ptr
+// CHECK-NEXT: store i32 7, ptr
+// CHECK-NEXT: store i16 9, ptr
 // CHECK: ret void
 }
 
@@ -568,14 +568,14 @@ void widened(float *a, float *b, float *c, float *d) {
   short j; // inner loop counter
   globalfloat = 1.0;
   int localint = 1;
-// CHECK: store double {{.+}}, double* [[GLOBALFLOAT:@.+]]
+// CHECK: store double {{.+}}, ptr [[GLOBALFLOAT:@.+]]
 // Counter is widened to 64 bits.
-// CHECK: store i64 0, i64* [[OMP_IV:[^,]+]]
+// CHECK: store i64 0, ptr [[OMP_IV:[^,]+]]
 //
   #pragma omp simd collapse(2) private(globalfloat, localint)
 
-// CHECK: [[IV:%.+]] = load i64, i64* [[OMP_IV]]{{.+}}!llvm.access.group
-// CHECK-NEXT: [[LI:%.+]] = load i64, i64* [[OMP_LI:%[^,]+]]{{.+}}!llvm.access.group
+// CHECK: [[IV:%.+]] = load i64, ptr [[OMP_IV]]{{.+}}!llvm.access.group
+// CHECK-NEXT: [[LI:%.+]] = load i64, ptr [[OMP_LI:%[^,]+]]{{.+}}!llvm.access.group
 // CHECK-NEXT: [[NUMIT:%.+]] = add nsw i64 [[LI]], 1
 // CHECK-NEXT: [[CMP:%.+]] = icmp slt i64 [[IV]], [[NUMIT]]
 // CHECK-NEXT: br i1 [[CMP]], label %[[WIDE1_BODY:[^,]+]], label %[[WIDE1_END:[^,]+]]
@@ -584,171 +584,171 @@ void widened(float *a, float *b, float *c, float *d) {
   {
 // CHECK: [[WIDE1_BODY]]:
 // Start of body: calculate i from index:
-// CHECK: [[IV1:%.+]] = load i64, i64* [[OMP_IV]]{{.+}}!llvm.access.group
+// CHECK: [[IV1:%.+]] = load i64, ptr [[OMP_IV]]{{.+}}!llvm.access.group
 // Calculation of the loop counters values...
-// CHECK: store i32 {{[^,]+}}, i32* [[LC_I:.+]]
-// CHECK: [[IV1_2:%.+]] = load i64, i64* [[OMP_IV]]{{.+}}!llvm.access.group
-// CHECK: store i16 {{[^,]+}}, i16* [[LC_J:.+]]
+// CHECK: store i32 {{[^,]+}}, ptr [[LC_I:.+]]
+// CHECK: [[IV1_2:%.+]] = load i64, ptr [[OMP_IV]]{{.+}}!llvm.access.group
+// CHECK: store i16 {{[^,]+}}, ptr [[LC_J:.+]]
 // ... loop body ...
 //
 // Here we expect store into private double var, not global
-// CHECK-NOT: store double {{.+}}, double* [[GLOBALFLOAT]]
+// CHECK-NOT: store double {{.+}}, ptr [[GLOBALFLOAT]]
     globalfloat = (float)j/i;
     float res = b[j] * c[j];
 // Store into a[i]:
-// CHECK: store float [[RESULT:%.+]], float* [[RESULT_ADDR:%.+]]{{.+}}!llvm.access.group
+// CHECK: store float [[RESULT:%.+]], ptr [[RESULT_ADDR:%.+]]{{.+}}!llvm.access.group
     a[i] = res * d[i];
 // Then there's a store into private var localint:
-// CHECK: store i32 {{.+}}, i32* [[LOCALINT:%[^,]+]]{{.+}}!llvm.access.group
+// CHECK: store i32 {{.+}}, ptr [[LOCALINT:%[^,]+]]{{.+}}!llvm.access.group
     localint = (int)j;
-// CHECK: [[IV2:%.+]] = load i64, i64* [[OMP_IV]]{{.*}}!llvm.access.group
+// CHECK: [[IV2:%.+]] = load i64, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[ADD2:%.+]] = add nsw i64 [[IV2]], 1
-// CHECK-NEXT: store i64 [[ADD2]], i64* [[OMP_IV]]{{.*}}!llvm.access.group
+// CHECK-NEXT: store i64 [[ADD2]], ptr [[OMP_IV]]{{.*}}!llvm.access.group
 //
 // br label %{{[^,]+}}, !llvm.loop ![[WIDE1_LOOP_ID]]
 // CHECK: [[WIDE1_END]]:
   }
 // i,j are updated.
-// CHECK: store i32 3, i32* [[I:%[^,]+]]
+// CHECK: store i32 3, ptr [[I:%[^,]+]]
 // CHECK: store i16
 //
 // Here we expect store into original localint, not its privatized version.
-// CHECK-NOT: store i32 {{.+}}, i32* [[LOCALINT]]
+// CHECK-NOT: store i32 {{.+}}, ptr [[LOCALINT]]
   localint = (int)j;
 // CHECK: ret void
 }
 
-// CHECK-LABEL: define {{.*void}} @{{.*}}linear{{.*}}(float* noundef {{.+}})
+// CHECK-LABEL: define {{.*void}} @{{.*}}linear{{.*}}(ptr noundef {{.+}})
 void linear(float *a) {
   // CHECK: [[VAL_ADDR:%.+]] = alloca i64,
-  // CHECK: [[K_ADDR:%.+]] = alloca i64*,
+  // CHECK: [[K_ADDR:%.+]] = alloca ptr,
   long long val = 0;
   long long &k = val;
 
   #pragma omp simd linear(k : 3)
-// CHECK: store i64* [[VAL_ADDR]], i64** [[K_ADDR]],
-// CHECK: [[VAL_REF:%.+]] = load i64*, i64** [[K_ADDR]],
-// CHECK: store i64* [[VAL_REF]], i64** [[K_ADDR_REF:%.+]],
-// CHECK: store i32 0, i32* [[OMP_IV:%[^,]+]]
-// CHECK: [[K_REF:%.+]] = load i64*, i64** [[K_ADDR_REF]],
-// CHECK: [[K0LOAD:%.+]] = load i64, i64* [[K_REF]]
-// CHECK-NEXT: store i64 [[K0LOAD]], i64* [[LIN0:%[^,]+]]
-
-// CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
+// CHECK: store ptr [[VAL_ADDR]], ptr [[K_ADDR]],
+// CHECK: [[VAL_REF:%.+]] = load ptr, ptr [[K_ADDR]],
+// CHECK: store ptr [[VAL_REF]], ptr [[K_ADDR_REF:%.+]],
+// CHECK: store i32 0, ptr [[OMP_IV:%[^,]+]]
+// CHECK: [[K_REF:%.+]] = load ptr, ptr [[K_ADDR_REF]],
+// CHECK: [[K0LOAD:%.+]] = load i64, ptr [[K_REF]]
+// CHECK-NEXT: store i64 [[K0LOAD]], ptr [[LIN0:%[^,]+]]
+
+// CHECK: [[IV:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[CMP2:%.+]] = icmp slt i32 [[IV]], 9
 // CHECK-NEXT: br i1 [[CMP2]], label %[[SIMPLE_LOOP_BODY:.+]], label %[[SIMPLE_LOOP_END:[^,]+]]
   for (int i = 10; i > 1; i--) {
 // CHECK: [[SIMPLE_LOOP_BODY]]:
 // Start of body: calculate i from IV:
-// CHECK: [[IV_0:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
+// CHECK: [[IV_0:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 // FIXME: It is interesting, why the following "mul 1" was not constant folded?
 // CHECK-NEXT: [[IV_1:%.+]] = mul nsw i32 [[IV_0]], 1
 // CHECK-NEXT: [[LC_I_1:%.+]] = sub nsw i32 10, [[IV_1]]
-// CHECK-NEXT: store i32 [[LC_I_1]], i32* {{.+}}, !llvm.access.group
+// CHECK-NEXT: store i32 [[LC_I_1]], ptr {{.+}}, !llvm.access.group
 //
-// CHECK-NEXT: [[LIN0_1:%.+]] = load i64, i64* [[LIN0]]{{.*}}!llvm.access.group
-// CHECK-NEXT: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
+// CHECK-NEXT: [[LIN0_1:%.+]] = load i64, ptr [[LIN0]]{{.*}}!llvm.access.group
+// CHECK-NEXT: [[IV_2:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[LIN_MUL1:%.+]] = mul nsw i32 [[IV_2]], 3
 // CHECK-NEXT: [[LIN_EXT1:%.+]] = sext i32 [[LIN_MUL1]] to i64
 // CHECK-NEXT: [[LIN_ADD1:%.+]] = add nsw i64 [[LIN0_1]], [[LIN_EXT1]]
 // Update of the privatized version of linear variable!
-// CHECK-NEXT: store i64 [[LIN_ADD1]], i64* [[K_PRIVATIZED:%[^,]+]]
+// CHECK-NEXT: store i64 [[LIN_ADD1]], ptr [[K_PRIVATIZED:%[^,]+]]
     a[k]++;
     k = k + 3;
-// CHECK: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
+// CHECK: [[IV_2:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[ADD2_2:%.+]] = add nsw i32 [[IV_2]], 1
-// CHECK-NEXT: store i32 [[ADD2_2]], i32* [[OMP_IV]]{{.*}}!llvm.access.group
+// CHECK-NEXT: store i32 [[ADD2_2]], ptr [[OMP_IV]]{{.*}}!llvm.access.group
 // br label {{.+}}, !llvm.loop ![[SIMPLE_LOOP_ID]]
   }
 // CHECK: [[SIMPLE_LOOP_END]]:
 //
 // Update linear vars after loop, as the loop was operating on a private version.
-// CHECK: [[K_REF:%.+]] = load i64*, i64** [[K_ADDR_REF]],
-// CHECK: store i64* [[K_REF]], i64** [[K_PRIV_REF:%.+]],
-// CHECK: [[LIN0_2:%.+]] = load i64, i64* [[K_PRIVATIZED]]
-// CHECK-NEXT: [[K_REF:%.+]] = load i64*, i64** [[K_PRIV_REF]],
-// CHECK-NEXT: store i64 [[LIN0_2]], i64* [[K_REF]]
+// CHECK: [[K_REF:%.+]] = load ptr, ptr [[K_ADDR_REF]],
+// CHECK: store ptr [[K_REF]], ptr [[K_PRIV_REF:%.+]],
+// CHECK: [[LIN0_2:%.+]] = load i64, ptr [[K_PRIVATIZED]]
+// CHECK-NEXT: [[K_REF:%.+]] = load ptr, ptr [[K_PRIV_REF]],
+// CHECK-NEXT: store i64 [[LIN0_2]], ptr [[K_REF]]
 //
 
   #pragma omp simd linear(val(k) : 3)
-// CHECK: [[VAL_REF:%.+]] = load i64*, i64** [[K_ADDR]],
-// CHECK: store i64* [[VAL_REF]], i64** [[K_ADDR_REF:%.+]],
-// CHECK: store i32 0, i32* [[OMP_IV:%[^,]+]]
-// CHECK: [[K_REF:%.+]] = load i64*, i64** [[K_ADDR_REF]],
-// CHECK: [[K0LOAD:%.+]] = load i64, i64* [[K_REF]]
-// CHECK-NEXT: store i64 [[K0LOAD]], i64* [[LIN0:%[^,]+]]
-
-// CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
+// CHECK: [[VAL_REF:%.+]] = load ptr, ptr [[K_ADDR]],
+// CHECK: store ptr [[VAL_REF]], ptr [[K_ADDR_REF:%.+]],
+// CHECK: store i32 0, ptr [[OMP_IV:%[^,]+]]
+// CHECK: [[K_REF:%.+]] = load ptr, ptr [[K_ADDR_REF]],
+// CHECK: [[K0LOAD:%.+]] = load i64, ptr [[K_REF]]
+// CHECK-NEXT: store i64 [[K0LOAD]], ptr [[LIN0:%[^,]+]]
+
+// CHECK: [[IV:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[CMP2:%.+]] = icmp slt i32 [[IV]], 9
 // CHECK-NEXT: br i1 [[CMP2]], label %[[SIMPLE_LOOP_BODY:.+]], label %[[SIMPLE_LOOP_END:[^,]+]]
   for (int i = 10; i > 1; i--) {
 // CHECK: [[SIMPLE_LOOP_BODY]]:
 // Start of body: calculate i from IV:
-// CHECK: [[IV_0:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
+// CHECK: [[IV_0:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 // FIXME: It is interesting, why the following "mul 1" was not constant folded?
 // CHECK-NEXT: [[IV_1:%.+]] = mul nsw i32 [[IV_0]], 1
 // CHECK-NEXT: [[LC_I_1:%.+]] = sub nsw i32 10, [[IV_1]]
-// CHECK-NEXT: store i32 [[LC_I_1]], i32* {{.+}}, !llvm.access.group
+// CHECK-NEXT: store i32 [[LC_I_1]], ptr {{.+}}, !llvm.access.group
 //
-// CHECK-NEXT: [[LIN0_1:%.+]] = load i64, i64* [[LIN0]]{{.*}}!llvm.access.group
-// CHECK-NEXT: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
+// CHECK-NEXT: [[LIN0_1:%.+]] = load i64, ptr [[LIN0]]{{.*}}!llvm.access.group
+// CHECK-NEXT: [[IV_2:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[LIN_MUL1:%.+]] = mul nsw i32 [[IV_2]], 3
 // CHECK-NEXT: [[LIN_EXT1:%.+]] = sext i32 [[LIN_MUL1]] to i64
 // CHECK-NEXT: [[LIN_ADD1:%.+]] = add nsw i64 [[LIN0_1]], [[LIN_EXT1]]
 // Update of the privatized version of linear variable!
-// CHECK-NEXT: store i64 [[LIN_ADD1]], i64* [[K_PRIVATIZED:%[^,]+]]
+// CHECK-NEXT: store i64 [[LIN_ADD1]], ptr [[K_PRIVATIZED:%[^,]+]]
     a[k]++;
     k = k + 3;
-// CHECK: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
+// CHECK: [[IV_2:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[ADD2_2:%.+]] = add nsw i32 [[IV_2]], 1
-// CHECK-NEXT: store i32 [[ADD2_2]], i32* [[OMP_IV]]{{.*}}!llvm.access.group
+// CHECK-NEXT: store i32 [[ADD2_2]], ptr [[OMP_IV]]{{.*}}!llvm.access.group
 // br label {{.+}}, !llvm.loop ![[SIMPLE_LOOP_ID]]
   }
 // CHECK: [[SIMPLE_LOOP_END]]:
 //
 // Update linear vars after loop, as the loop was operating on a private version.
-// CHECK: [[K_REF:%.+]] = load i64*, i64** [[K_ADDR_REF]],
-// CHECK: store i64* [[K_REF]], i64** [[K_PRIV_REF:%.+]],
-// CHECK: [[LIN0_2:%.+]] = load i64, i64* [[K_PRIVATIZED]]
-// CHECK-NEXT: [[K_REF:%.+]] = load i64*, i64** [[K_PRIV_REF]],
-// CHECK-NEXT: store i64 [[LIN0_2]], i64* [[K_REF]]
+// CHECK: [[K_REF:%.+]] = load ptr, ptr [[K_ADDR_REF]],
+// CHECK: store ptr [[K_REF]], ptr [[K_PRIV_REF:%.+]],
+// CHECK: [[LIN0_2:%.+]] = load i64, ptr [[K_PRIVATIZED]]
+// CHECK-NEXT: [[K_REF:%.+]] = load ptr, ptr [[K_PRIV_REF]],
+// CHECK-NEXT: store i64 [[LIN0_2]], ptr [[K_REF]]
 //
   #pragma omp simd linear(uval(k) : 3)
-// CHECK: store i32 0, i32* [[OMP_IV:%[^,]+]]
-// CHECK: [[K0LOAD:%.+]] = load i64, i64* [[VAL_ADDR]]
-// CHECK-NEXT: store i64 [[K0LOAD]], i64* [[LIN0:%[^,]+]]
+// CHECK: store i32 0, ptr [[OMP_IV:%[^,]+]]
+// CHECK: [[K0LOAD:%.+]] = load i64, ptr [[VAL_ADDR]]
+// CHECK-NEXT: store i64 [[K0LOAD]], ptr [[LIN0:%[^,]+]]
 
-// CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
+// CHECK: [[IV:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[CMP2:%.+]] = icmp slt i32 [[IV]], 9
 // CHECK-NEXT: br i1 [[CMP2]], label %[[SIMPLE_LOOP_BODY:.+]], label %[[SIMPLE_LOOP_END:[^,]+]]
   for (int i = 10; i > 1; i--) {
 // CHECK: [[SIMPLE_LOOP_BODY]]:
 // Start of body: calculate i from IV:
-// CHECK: [[IV_0:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
+// CHECK: [[IV_0:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 // FIXME: It is interesting, why the following "mul 1" was not constant folded?
 // CHECK-NEXT: [[IV_1:%.+]] = mul nsw i32 [[IV_0]], 1
 // CHECK-NEXT: [[LC_I_1:%.+]] = sub nsw i32 10, [[IV_1]]
-// CHECK-NEXT: store i32 [[LC_I_1]], i32* {{.+}}, !llvm.access.group
+// CHECK-NEXT: store i32 [[LC_I_1]], ptr {{.+}}, !llvm.access.group
 //
-// CHECK-NEXT: [[LIN0_1:%.+]] = load i64, i64* [[LIN0]]{{.*}}!llvm.access.group
-// CHECK-NEXT: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
+// CHECK-NEXT: [[LIN0_1:%.+]] = load i64, ptr [[LIN0]]{{.*}}!llvm.access.group
+// CHECK-NEXT: [[IV_2:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[LIN_MUL1:%.+]] = mul nsw i32 [[IV_2]], 3
 // CHECK-NEXT: [[LIN_EXT1:%.+]] = sext i32 [[LIN_MUL1]] to i64
 // CHECK-NEXT: [[LIN_ADD1:%.+]] = add nsw i64 [[LIN0_1]], [[LIN_EXT1]]
 // Update of the privatized version of linear variable!
-// CHECK-NEXT: store i64 [[LIN_ADD1]], i64* [[K_PRIVATIZED:%[^,]+]]
+// CHECK-NEXT: store i64 [[LIN_ADD1]], ptr [[K_PRIVATIZED:%[^,]+]]
     a[k]++;
     k = k + 3;
-// CHECK: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
+// CHECK: [[IV_2:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[ADD2_2:%.+]] = add nsw i32 [[IV_2]], 1
-// CHECK-NEXT: store i32 [[ADD2_2]], i32* [[OMP_IV]]{{.*}}!llvm.access.group
+// CHECK-NEXT: store i32 [[ADD2_2]], ptr [[OMP_IV]]{{.*}}!llvm.access.group
 // br label {{.+}}, !llvm.loop ![[SIMPLE_LOOP_ID]]
   }
 // CHECK: [[SIMPLE_LOOP_END]]:
 //
 // Update linear vars after loop, as the loop was operating on a private version.
-// CHECK: [[LIN0_2:%.+]] = load i64, i64* [[K_PRIVATIZED]]
-// CHECK-NEXT: store i64 [[LIN0_2]], i64* [[VAL_ADDR]]
+// CHECK: [[LIN0_2:%.+]] = load i64, ptr [[K_PRIVATIZED]]
+// CHECK-NEXT: store i64 [[LIN0_2]], ptr [[VAL_ADDR]]
 //
 }
 
diff --git a/clang/test/OpenMP/target_codegen.cpp b/clang/test/OpenMP/target_codegen.cpp
index af5ad7303de7..1b8260005e30 100644
--- a/clang/test/OpenMP/target_codegen.cpp
+++ b/clang/test/OpenMP/target_codegen.cpp
@@ -1,78 +1,78 @@
 // Test host codegen.
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix OMP45
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix OMP45
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix OMP45
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix OMP45
-
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix OMP50
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix OMP50
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix OMP50
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix OMP50
-
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=51 -D_DOMP51 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix OMP51
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -fopenmp-version=51 -D_DOMP51 -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -fopenmp-version=51 -D_DOMP51 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix OMP51
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=51 -D_DOMP51 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix OMP51
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=51 -D_DOMP51 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=51 -D_DOMP51 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix OMP51
-
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix OMP45
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix OMP45
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix OMP45
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix OMP45
+
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix OMP50
+// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix OMP50
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix OMP50
+// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix OMP50
+
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=51 -D_DOMP51 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix OMP51
+// RUN: %clang_cc1 -fopenmp -x c++ -fopenmp-version=51 -D_DOMP51 -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -x c++ -fopenmp-version=51 -D_DOMP51 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix OMP51
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=51 -D_DOMP51 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix OMP51
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=51 -D_DOMP51 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=51 -D_DOMP51 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix OMP51
+
+// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
 
 // Test target codegen - host bc file has to be created first.
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32
-
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32
-
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32
+
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64
+// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32
+// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32
+
+// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
+// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
+// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
 // SIMD-ONLY1-NOT: {{__kmpc|__tgt}}
 
 // expected-no-diagnostics
 #ifndef HEADER
 #define HEADER
 
-// CHECK-DAG: [[IDENT_T:%.+]] = type { i32, i32, i32, i32, i8* }
+// CHECK-DAG: [[IDENT_T:%.+]] = type { i32, i32, i32, i32, ptr }
 // CHECK-DAG: [[KMP_TASK_T_WITH_PRIVATES:%.+]] = type { [[KMP_TASK_T:%.+]], [[KMP_PRIVATES_T:%.+]] }
-// CHECK-DAG: [[KMP_TASK_T]] = type { i8*, i32 (i32, i8*)*, i32, {{%.+}}, {{%.+}} }
+// CHECK-DAG: [[KMP_TASK_T]] = type { ptr, ptr, i32, {{%.+}}, {{%.+}} }
 // CHECK-DAG: [[TT:%.+]] = type { i64, i8 }
 // CHECK-DAG: [[S1:%.+]] = type { double }
 // CHECK-DAG: [[S2:%.+]] = type { i32, i32, i32 }
-// CHECK-DAG: [[ENTTY:%.+]] = type { i8*, i8*, i[[SZ:32|64]], i32, i32 }
-// CHECK-DAG: [[ANON_T:%.+]] = type { i[[SZ]]*, i32, i32 }
-// CHECK-32-DAG: [[KMP_PRIVATES_T]] = type { [2 x i64], i32*, i32, [2 x i8*], [2 x i8*] }
-// CHECK-64-DAG: [[KMP_PRIVATES_T]] = type { i64*, [2 x i8*], [2 x i8*], [2 x i64], i32 }
+// CHECK-DAG: [[ENTTY:%.+]] = type { ptr, ptr, i[[SZ:32|64]], i32, i32 }
+// CHECK-DAG: [[ANON_T:%.+]] = type { ptr, i32, i32 }
+// CHECK-32-DAG: [[KMP_PRIVATES_T]] = type { [2 x i64], ptr, i32, [2 x ptr], [2 x ptr] }
+// CHECK-64-DAG: [[KMP_PRIVATES_T]] = type { ptr, [2 x ptr], [2 x ptr], [2 x i64], i32 }
 
-// TCHECK: [[ENTTY:%.+]] = type { i8*, i8*, i{{32|64}}, i32, i32 }
+// TCHECK: [[ENTTY:%.+]] = type { ptr, ptr, i{{32|64}}, i32, i32 }
 
 // We have 9 target regions, but only 8 that actually will generate offloading
 // code and have mapped arguments, and only 6 have all-constant map sizes.
@@ -115,7 +115,7 @@
 // TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]]
 
 // Check target registration is registered as a Ctor.
-// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @.omp_offloading.requires_reg, i8* null }]
+// CHECK: appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 0, ptr @.omp_offloading.requires_reg, ptr null }]
 
 
 template<typename tx, typename ty>
@@ -129,9 +129,9 @@ extern int global;
 
 // CHECK: define {{.*}}[[FOO:@.+]](
 int foo(int n) {
-  // CHECK: [[OFFLOADBPTR:%.+]] = alloca [2 x i8*], align
-  // CHECK: [[OFFLOADPTR:%.+]] = alloca [2 x i8*], align
-  // CHECK: [[OFFLOADMAPPER:%.+]] = alloca [2 x i8*], align
+  // CHECK: [[OFFLOADBPTR:%.+]] = alloca [2 x ptr], align
+  // CHECK: [[OFFLOADPTR:%.+]] = alloca [2 x ptr], align
+  // CHECK: [[OFFLOADMAPPER:%.+]] = alloca [2 x ptr], align
   int a = 0;
   short aa = 0;
   float b[10];
@@ -142,10 +142,10 @@ int foo(int n) {
   static long *plocal;
 
 // CHECK:       [[ADD:%.+]] = add nsw i32
-// CHECK:       store i32 [[ADD]], i32* [[DEVICE_CAP:%.+]],
-// CHECK:       [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]],
+// CHECK:       store i32 [[ADD]], ptr [[DEVICE_CAP:%.+]],
+// CHECK:       [[DEV:%.+]] = load i32, ptr [[DEVICE_CAP]],
 // CHECK:       [[DEVICE:%.+]] = sext i32 [[DEV]] to i64
-// CHECK:       [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 [[DEVICE]], i32 {{.+}}, i32 {{.+}}, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
+// CHECK:       [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 [[DEVICE]], i32 {{.+}}, i32 {{.+}}, ptr @.{{.+}}.region_id, ptr [[ARGS:%.+]])
 // CHECK-NEXT:  [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
 // CHECK-NEXT:  br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]
 // CHECK:       [[FAIL]]
@@ -156,38 +156,27 @@ int foo(int n) {
   {
   }
 
-  // CHECK: [[BPRGEP:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[OFFLOADBPTR]], i32 0, i32 0
-  // CHECK: [[PRGEP:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[OFFLOADPTR]], i32 0, i32 0
-  // CHECK: [[BPRGEP:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[OFFLOADBPTR]], i32 0, i32 0
-  // CHECK: [[PRGEP:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[OFFLOADPTR]], i32 0, i32 0
+  // CHECK: [[BPRGEP:%.+]] = getelementptr inbounds [2 x ptr], ptr [[OFFLOADBPTR]], i32 0, i32 0
+  // CHECK: [[PRGEP:%.+]] = getelementptr inbounds [2 x ptr], ptr [[OFFLOADPTR]], i32 0, i32 0
+  // CHECK: [[BPRGEP:%.+]] = getelementptr inbounds [2 x ptr], ptr [[OFFLOADBPTR]], i32 0, i32 0
+  // CHECK: [[PRGEP:%.+]] = getelementptr inbounds [2 x ptr], ptr [[OFFLOADPTR]], i32 0, i32 0
   // CHECK: [[DEVICE:%.+]] = sext i32 {{%.+}} to i64
-  // CHECK-32: [[TASK:%.+]] = call i8* @__kmpc_omp_target_task_alloc([[IDENT_T]]* {{.+}}, i32 %0, i32 1, i32 60, i32 12, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_T_WITH_PRIVATES]]*)* [[OMP_TASK_ENTRY:@.+]] to i32 (i32, i8*)*), i64 [[DEVICE]])
-  // CHECK-64: [[TASK:%.+]] = call i8* @__kmpc_omp_target_task_alloc([[IDENT_T]]* {{.+}}, i32 %0, i32 1, i64 104, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_T_WITH_PRIVATES]]*)* [[OMP_TASK_ENTRY:@.+]] to i32 (i32, i8*)*), i64 [[DEVICE]])
-  // CHECK: [[TASK_WITH_PRIVATES:%.+]] = bitcast i8* [[TASK]] to [[KMP_TASK_T_WITH_PRIVATES]]*
-  // CHECK: [[TASK_WITH_PRIVATES_GEP:%.+]] = getelementptr inbounds [[KMP_TASK_T_WITH_PRIVATES]], [[KMP_TASK_T_WITH_PRIVATES]]* [[TASK_WITH_PRIVATES]], i32 0, i32 1
-  // CHECK-32: [[SIZEGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], [[KMP_PRIVATES_T]]* [[TASK_WITH_PRIVATES_GEP]], i32 0, i32 0
-  // CHECK-32: [[SIZEADDR:%.+]] = bitcast [2 x i64]* [[SIZEGEP]] to i8*
-  // CHECK-32: call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[SIZEADDR]], i8* align 4 bitcast ([2 x i64]* [[SIZET]] to i8*), i32 16, i1 false)
-  // CHECK-32: [[FPBPRGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], [[KMP_PRIVATES_T]]* [[TASK_WITH_PRIVATES_GEP]], i32 0, i32 3
-  // CHECK-32: [[FPBPRCAST:%.+]] = bitcast [2 x i8*]* [[FPBPRGEP]] to i8*
-  // CHECK-32: [[BPRCAST:%.+]] = bitcast i8** [[BPRGEP]] to i8*
-  // CHECK-32: call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[FPBPRCAST]], i8* align 4 [[BPRCAST]], i32 8, i1 false)
-  // CHECK-32: [[FPPRGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], [[KMP_PRIVATES_T]]* [[TASK_WITH_PRIVATES_GEP]], i32 0, i32 4
-  // CHECK-32: [[FPPRCAST:%.+]] = bitcast [2 x i8*]* [[FPPRGEP]] to i8*
-  // CHECK-32: [[PRCAST:%.+]] = bitcast i8** [[PRGEP]] to i8*
-  // CHECK-32: call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[FPPRCAST]], i8* align 4 [[PRCAST]], i32 8, i1 false)
-  // CHECK-64: [[FPBPRGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], [[KMP_PRIVATES_T]]* [[TASK_WITH_PRIVATES_GEP]], i32 0, i32 1
-  // CHECK-64: [[FPBPRCAST:%.+]] = bitcast [2 x i8*]* [[FPBPRGEP]] to i8*
-  // CHECK-64: [[BPR_CAST:%.+]] = bitcast i8** [[BPRGEP]] to i8*
-  // CHECK-64: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[FPBPRCAST]], i8* align 8 [[BPR_CAST]], i64 16, i1 false)
-  // CHECK-64: [[FPPRGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], [[KMP_PRIVATES_T]]* [[TASK_WITH_PRIVATES_GEP]], i32 0, i32 2
-  // CHECK-64: [[FPPRCAST:%.+]] = bitcast [2 x i8*]* [[FPPRGEP]] to i8*
-  // CHECK-64: [[PR_CAST:%.+]] = bitcast i8** [[PRGEP]] to i8*
-  // CHECK-64: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[FPPRCAST]], i8* align 8 [[PR_CAST]], i64 16, i1 false)
-  // CHECK-64: [[SIZEGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], [[KMP_PRIVATES_T]]* [[TASK_WITH_PRIVATES_GEP]], i32 0, i32 3
-  // CHECK-64: [[SIZE_CAST:%.+]] = bitcast [2 x i64]* [[SIZEGEP]] to i8*
-  // CHECK-64: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[SIZE_CAST]], i8* align 8 bitcast ([2 x i64]* [[SIZET]] to i8*), i64 16, i1 false)
-  // CHECK: call i32 @__kmpc_omp_task([[IDENT_T]]* {{.+}}, i32 {{.+}}, i8* [[TASK]])
+  // CHECK-32: [[TASK:%.+]] = call ptr @__kmpc_omp_target_task_alloc(ptr {{.+}}, i32 %0, i32 1, i32 60, i32 12, ptr [[OMP_TASK_ENTRY:@.+]], i64 [[DEVICE]])
+  // CHECK-64: [[TASK:%.+]] = call ptr @__kmpc_omp_target_task_alloc(ptr {{.+}}, i32 %0, i32 1, i64 104, i64 16, ptr [[OMP_TASK_ENTRY:@.+]], i64 [[DEVICE]])
+  // CHECK: [[TASK_WITH_PRIVATES_GEP:%.+]] = getelementptr inbounds [[KMP_TASK_T_WITH_PRIVATES]], ptr [[TASK]], i32 0, i32 1
+  // CHECK-32: [[SIZEGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], ptr [[TASK_WITH_PRIVATES_GEP]], i32 0, i32 0
+  // CHECK-32: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[SIZEGEP]], ptr align 4 [[SIZET]], i32 16, i1 false)
+  // CHECK-32: [[FPBPRGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], ptr [[TASK_WITH_PRIVATES_GEP]], i32 0, i32 3
+  // CHECK-32: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[FPBPRGEP]], ptr align 4 [[BPRGEP]], i32 8, i1 false)
+  // CHECK-32: [[FPPRGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], ptr [[TASK_WITH_PRIVATES_GEP]], i32 0, i32 4
+  // CHECK-32: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[FPPRGEP]], ptr align 4 [[PRGEP]], i32 8, i1 false)
+  // CHECK-64: [[FPBPRGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], ptr [[TASK_WITH_PRIVATES_GEP]], i32 0, i32 1
+  // CHECK-64: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[FPBPRGEP]], ptr align 8 [[BPRGEP]], i64 16, i1 false)
+  // CHECK-64: [[FPPRGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], ptr [[TASK_WITH_PRIVATES_GEP]], i32 0, i32 2
+  // CHECK-64: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[FPPRGEP]], ptr align 8 [[PRGEP]], i64 16, i1 false)
+  // CHECK-64: [[SIZEGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], ptr [[TASK_WITH_PRIVATES_GEP]], i32 0, i32 3
+  // CHECK-64: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[SIZEGEP]], ptr align 8 [[SIZET]], i64 16, i1 false)
+  // CHECK: call i32 @__kmpc_omp_task(ptr {{.+}}, i32 {{.+}}, ptr [[TASK]])
   #pragma omp target device(global + a) nowait
   {
     static int local1;
@@ -201,19 +190,17 @@ int foo(int n) {
     global += 1;
   }
 
-// CHECK-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 {{.+}}, i32 {{.+}}, i32 {{.+}}, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
+// CHECK-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 {{.+}}, i32 {{.+}}, i32 {{.+}}, ptr @.{{.+}}.region_id, ptr [[ARGS:%.+]])
 // CHECK-DAG:   [[BPARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 2
-// CHECK-DAG:   store i8** [[BP:%.+]], i8*** [[BPARG]]
+// CHECK-DAG:   store ptr [[BP:%.+]], ptr [[BPARG]]
 // CHECK-DAG:   [[PARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 3
-// CHECK-DAG:   store i8** [[P:%.+]], i8*** [[PARG]]
-// CHECK-DAG:   [[BP]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BPR:%[^,]+]], i32 0, i32 0
-// CHECK-DAG:   [[P]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[PR:%[^,]+]], i32 0, i32 0
-// CHECK-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BPR]], i32 0, i32 [[IDX0:[0-9]+]]
-// CHECK-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[PR]], i32 0, i32 [[IDX0]]
-// CHECK-DAG:   [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]*
-// CHECK-DAG:   store i[[SZ]] [[BP0:%[^,]+]], i[[SZ]]* [[CBPADDR0]]
-// CHECK-DAG:   store i[[SZ]] [[P0:%[^,]+]], i[[SZ]]* [[CPADDR0]]
+// CHECK-DAG:   store ptr [[P:%.+]], ptr [[PARG]]
+// CHECK-DAG:   [[BP]] = getelementptr inbounds [1 x ptr], ptr [[BPR:%[^,]+]], i32 0, i32 0
+// CHECK-DAG:   [[P]] = getelementptr inbounds [1 x ptr], ptr [[PR:%[^,]+]], i32 0, i32 0
+// CHECK-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[BPR]], i32 0, i32 [[IDX0:[0-9]+]]
+// CHECK-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[PR]], i32 0, i32 [[IDX0]]
+// CHECK-DAG:   store i[[SZ]] [[BP0:%[^,]+]], ptr [[BPADDR0]]
+// CHECK-DAG:   store i[[SZ]] [[P0:%[^,]+]], ptr [[PADDR0]]
 
 // CHECK:       [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
 // CHECK-NEXT:  br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]
@@ -229,27 +216,23 @@ int foo(int n) {
 // CHECK:       [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 10
 // CHECK:       br i1 [[IF]], label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]]
 // CHECK:       [[IFTHEN]]
-// CHECK-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 {{.+}}, i32 {{.+}}, i32 {{.+}}, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
+// CHECK-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 {{.+}}, i32 {{.+}}, i32 {{.+}}, ptr @.{{.+}}.region_id, ptr [[ARGS:%.+]])
 // CHECK-DAG:   [[BPARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 2
-// CHECK-DAG:   store i8** [[BPR:%.+]], i8*** [[BPARG]]
+// CHECK-DAG:   store ptr [[BPR:%.+]], ptr [[BPARG]]
 // CHECK-DAG:   [[PARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 3
-// CHECK-DAG:   store i8** [[PR:%.+]], i8*** [[PARG]]
-// CHECK-DAG:   [[BPR]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP:%[^,]+]], i32 0, i32 0
-// CHECK-DAG:   [[PR]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P:%[^,]+]], i32 0, i32 0
-
-// CHECK-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 0
-// CHECK-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 0
-// CHECK-DAG:   [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]*
-// CHECK-DAG:   store i[[SZ]] [[BP0:%[^,]+]], i[[SZ]]* [[CBPADDR0]]
-// CHECK-DAG:   store i[[SZ]] [[P0:%[^,]+]], i[[SZ]]* [[CPADDR0]]
-
-// CHECK-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 1
-// CHECK-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 1
-// CHECK-DAG:   [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to i[[SZ]]*
-// CHECK-DAG:   store i[[SZ]] [[BP1:%[^,]+]], i[[SZ]]* [[CBPADDR1]]
-// CHECK-DAG:   store i[[SZ]] [[P1:%[^,]+]], i[[SZ]]* [[CPADDR1]]
+// CHECK-DAG:   store ptr [[PR:%.+]], ptr [[PARG]]
+// CHECK-DAG:   [[BPR]] = getelementptr inbounds [2 x ptr], ptr [[BP:%[^,]+]], i32 0, i32 0
+// CHECK-DAG:   [[PR]] = getelementptr inbounds [2 x ptr], ptr [[P:%[^,]+]], i32 0, i32 0
+
+// CHECK-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [2 x ptr], ptr [[BP]], i32 0, i32 0
+// CHECK-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [2 x ptr], ptr [[P]], i32 0, i32 0
+// CHECK-DAG:   store i[[SZ]] [[BP0:%[^,]+]], ptr [[BPADDR0]]
+// CHECK-DAG:   store i[[SZ]] [[P0:%[^,]+]], ptr [[PADDR0]]
+
+// CHECK-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [2 x ptr], ptr [[BP]], i32 0, i32 1
+// CHECK-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [2 x ptr], ptr [[P]], i32 0, i32 1
+// CHECK-DAG:   store i[[SZ]] [[BP1:%[^,]+]], ptr [[BPADDR1]]
+// CHECK-DAG:   store i[[SZ]] [[P1:%[^,]+]], ptr [[PADDR1]]
 // CHECK:       [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
 // CHECK-NEXT:  br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]]
 // CHECK:       [[FAIL]]
@@ -269,14 +252,13 @@ int foo(int n) {
   }
 
   // We capture 3 VLA sizes in this target region
-  // CHECK-64:       [[A_VAL:%.+]] = load i32, i32* %{{.+}},
-  // CHECK-64:       [[A_ADDR:%.+]] = bitcast i[[SZ]]* [[A_CADDR:%.+]] to i32*
-  // CHECK-64:       store i32 [[A_VAL]], i32* [[A_ADDR]],
-  // CHECK-64:       [[A_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CADDR]],
+  // CHECK-64:       [[A_VAL:%.+]] = load i32, ptr %{{.+}},
+  // CHECK-64:       store i32 [[A_VAL]], ptr [[A_CADDR:%.+]],
+  // CHECK-64:       [[A_CVAL:%.+]] = load i[[SZ]], ptr [[A_CADDR]],
 
-  // CHECK-32:       [[A_VAL:%.+]] = load i32, i32* %{{.+}},
-  // CHECK-32:       store i32 [[A_VAL]], i32* [[A_CADDR:%.+]],
-  // CHECK-32:       [[A_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CADDR]],
+  // CHECK-32:       [[A_VAL:%.+]] = load i32, ptr %{{.+}},
+  // CHECK-32:       store i32 [[A_VAL]], ptr [[A_CADDR:%.+]],
+  // CHECK-32:       [[A_CVAL:%.+]] = load i[[SZ]], ptr [[A_CADDR]],
 
   // CHECK:       [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 20
   // CHECK:       br i1 [[IF]], label %[[TRY:[^,]+]], label %[[IFELSE:[^,]+]]
@@ -289,86 +271,68 @@ int foo(int n) {
   // CHECK-32:    [[CNSZSIZE:%.+]] = mul nuw i32 [[CNELEMSIZE2]], 8
   // CHECK-32:    [[CNSIZE:%.+]] = sext i32 [[CNSZSIZE]] to i64
 
-// CHECK-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 {{.+}}, i32 {{.+}}, i32 {{.+}}, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
+// CHECK-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 {{.+}}, i32 {{.+}}, i32 {{.+}}, ptr @.{{.+}}.region_id, ptr [[ARGS:%.+]])
 // CHECK-DAG:   [[BPARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 2
-// CHECK-DAG:   store i8** [[BPR:%.+]], i8*** [[BPARG]]
+// CHECK-DAG:   store ptr [[BPR:%.+]], ptr [[BPARG]]
 // CHECK-DAG:   [[PARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 3
-// CHECK-DAG:   store i8** [[PR:%.+]], i8*** [[PARG]]
+// CHECK-DAG:   store ptr [[PR:%.+]], ptr [[PARG]]
 // CHECK-DAG:   [[SARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 4
-// CHECK-DAG:   store i64* [[SZ4:%.+]], i64** [[SARG]]
-// CHECK-DAG:   [[BPR]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP:%[^,]+]], i32 0, i32 0
-// CHECK-DAG:   [[PR]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P:%[^,]+]], i32 0, i32 0
-// CHECK-DAG:   [[SZ4]] = getelementptr inbounds [9 x i64], [9 x i64]* [[PSZ:%[^,]+]], i32 0, i32 0
-
-// CHECK-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX0:0]]
-// CHECK-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX0]]
-// CHECK-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX1:1]]
-// CHECK-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX1]]
-// CHECK-DAG:   [[BPADDR2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX2:2]]
-// CHECK-DAG:   [[PADDR2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX2]]
-// CHECK-DAG:   [[BPADDR3:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX3:3]]
-// CHECK-DAG:   [[PADDR3:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX3]]
-// CHECK-DAG:   [[PSZ3:%.+]] = getelementptr inbounds [9 x i64], [9 x i64]* [[PSZ]], i32 0, i32 [[IDX3]]
-// CHECK-DAG:   [[BPADDR4:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX4:4]]
-// CHECK-DAG:   [[PADDR4:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX4]]
-// CHECK-DAG:   [[BPADDR5:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX5:5]]
-// CHECK-DAG:   [[PADDR5:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX5]]
-// CHECK-DAG:   [[BPADDR6:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX6:6]]
-// CHECK-DAG:   [[PADDR6:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX6]]
-// CHECK-DAG:   [[BPADDR7:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX7:7]]
-// CHECK-DAG:   [[PADDR7:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX7]]
-// CHECK-DAG:   [[PSZ7:%.+]] = getelementptr inbounds [9 x i64], [9 x i64]* [[PSZ]], i32 0, i32 [[IDX7]]
-// CHECK-DAG:   [[BPADDR8:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX8:8]]
-// CHECK-DAG:   [[PADDR8:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX8]]
+// CHECK-DAG:   store ptr [[SZ4:%.+]], ptr [[SARG]]
+// CHECK-DAG:   [[BPR]] = getelementptr inbounds [9 x ptr], ptr [[BP:%[^,]+]], i32 0, i32 0
+// CHECK-DAG:   [[PR]] = getelementptr inbounds [9 x ptr], ptr [[P:%[^,]+]], i32 0, i32 0
+// CHECK-DAG:   [[SZ4]] = getelementptr inbounds [9 x i64], ptr [[PSZ:%[^,]+]], i32 0, i32 0
+
+// CHECK-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [9 x ptr], ptr [[BP]], i32 0, i32 [[IDX0:0]]
+// CHECK-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [9 x ptr], ptr [[P]], i32 0, i32 [[IDX0]]
+// CHECK-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [9 x ptr], ptr [[BP]], i32 0, i32 [[IDX1:1]]
+// CHECK-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [9 x ptr], ptr [[P]], i32 0, i32 [[IDX1]]
+// CHECK-DAG:   [[BPADDR2:%.+]] = getelementptr inbounds [9 x ptr], ptr [[BP]], i32 0, i32 [[IDX2:2]]
+// CHECK-DAG:   [[PADDR2:%.+]] = getelementptr inbounds [9 x ptr], ptr [[P]], i32 0, i32 [[IDX2]]
+// CHECK-DAG:   [[BPADDR3:%.+]] = getelementptr inbounds [9 x ptr], ptr [[BP]], i32 0, i32 [[IDX3:3]]
+// CHECK-DAG:   [[PADDR3:%.+]] = getelementptr inbounds [9 x ptr], ptr [[P]], i32 0, i32 [[IDX3]]
+// CHECK-DAG:   [[PSZ3:%.+]] = getelementptr inbounds [9 x i64], ptr [[PSZ]], i32 0, i32 [[IDX3]]
+// CHECK-DAG:   [[BPADDR4:%.+]] = getelementptr inbounds [9 x ptr], ptr [[BP]], i32 0, i32 [[IDX4:4]]
+// CHECK-DAG:   [[PADDR4:%.+]] = getelementptr inbounds [9 x ptr], ptr [[P]], i32 0, i32 [[IDX4]]
+// CHECK-DAG:   [[BPADDR5:%.+]] = getelementptr inbounds [9 x ptr], ptr [[BP]], i32 0, i32 [[IDX5:5]]
+// CHECK-DAG:   [[PADDR5:%.+]] = getelementptr inbounds [9 x ptr], ptr [[P]], i32 0, i32 [[IDX5]]
+// CHECK-DAG:   [[BPADDR6:%.+]] = getelementptr inbounds [9 x ptr], ptr [[BP]], i32 0, i32 [[IDX6:6]]
+// CHECK-DAG:   [[PADDR6:%.+]] = getelementptr inbounds [9 x ptr], ptr [[P]], i32 0, i32 [[IDX6]]
+// CHECK-DAG:   [[BPADDR7:%.+]] = getelementptr inbounds [9 x ptr], ptr [[BP]], i32 0, i32 [[IDX7:7]]
+// CHECK-DAG:   [[PADDR7:%.+]] = getelementptr inbounds [9 x ptr], ptr [[P]], i32 0, i32 [[IDX7]]
+// CHECK-DAG:   [[PSZ7:%.+]] = getelementptr inbounds [9 x i64], ptr [[PSZ]], i32 0, i32 [[IDX7]]
+// CHECK-DAG:   [[BPADDR8:%.+]] = getelementptr inbounds [9 x ptr], ptr [[BP]], i32 0, i32 [[IDX8:8]]
+// CHECK-DAG:   [[PADDR8:%.+]] = getelementptr inbounds [9 x ptr], ptr [[P]], i32 0, i32 [[IDX8]]
 
 // The names below are not necessarily consistent with the names used for the
 // addresses above as some are repeated.
-// CHECK-DAG:   [[CBPADDR2:%.+]] = bitcast i8** [[BPADDR2]] to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR2:%.+]] = bitcast i8** [[PADDR2]] to i[[SZ]]*
-// CHECK-DAG:   store i[[SZ]] [[VLA0]], i[[SZ]]* [[CBPADDR2]]
-// CHECK-DAG:   store i[[SZ]] [[VLA0]], i[[SZ]]* [[CPADDR2]]
-
-// CHECK-DAG:   [[CBPADDR6:%.+]] = bitcast i8** [[BPADDR6]] to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR6:%.+]] = bitcast i8** [[PADDR6]] to i[[SZ]]*
-// CHECK-DAG:   store i[[SZ]] [[VLA1]], i[[SZ]]* [[CBPADDR6]]
-// CHECK-DAG:   store i[[SZ]] [[VLA1]], i[[SZ]]* [[CPADDR6]]
-
-// CHECK-DAG:   [[CBPADDR5:%.+]] = bitcast i8** [[BPADDR5]] to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR5:%.+]] = bitcast i8** [[PADDR5]] to i[[SZ]]*
-// CHECK-DAG:   store i[[SZ]] 5, i[[SZ]]* [[CBPADDR5]]
-// CHECK-DAG:   store i[[SZ]] 5, i[[SZ]]* [[CPADDR5]]
-
-// CHECK-DAG:   [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]*
-// CHECK-DAG:   store i[[SZ]] [[A_CVAL]], i[[SZ]]* [[CBPADDR0]]
-// CHECK-DAG:   store i[[SZ]] [[A_CVAL]], i[[SZ]]* [[CPADDR0]]
-
-// CHECK-DAG:   [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to [10 x float]**
-// CHECK-DAG:   [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to [10 x float]**
-// CHECK-DAG:   store [10 x float]* %{{.+}}, [10 x float]** [[CBPADDR1]]
-// CHECK-DAG:   store [10 x float]* %{{.+}}, [10 x float]** [[CPADDR1]]
-
-// CHECK-DAG:   [[CBPADDR3:%.+]] = bitcast i8** [[BPADDR3]] to float**
-// CHECK-DAG:   [[CPADDR3:%.+]] = bitcast i8** [[PADDR3]] to float**
-// CHECK-DAG:   store float* %{{.+}}, float** [[CBPADDR3]]
-// CHECK-DAG:   store float* %{{.+}}, float** [[CPADDR3]]
-// CHECK-DAG:   store i64 [[BNSIZE]], i64* [[PSZ3]]
-
-// CHECK-DAG:   [[CBPADDR4:%.+]] = bitcast i8** [[BPADDR4]] to [5 x [10 x double]]**
-// CHECK-DAG:   [[CPADDR4:%.+]] = bitcast i8** [[PADDR4]] to [5 x [10 x double]]**
-// CHECK-DAG:   store [5 x [10 x double]]* %{{.+}}, [5 x [10 x double]]** [[CBPADDR4]]
-// CHECK-DAG:   store [5 x [10 x double]]* %{{.+}}, [5 x [10 x double]]** [[CPADDR4]]
-
-// CHECK-DAG:   [[CBPADDR7:%.+]] = bitcast i8** [[BPADDR7]] to double**
-// CHECK-DAG:   [[CPADDR7:%.+]] = bitcast i8** [[PADDR7]] to double**
-// CHECK-DAG:   store double* %{{.+}}, double** [[CBPADDR7]]
-// CHECK-DAG:   store double* %{{.+}}, double** [[CPADDR7]]
-// CHECK-DAG:   store i64 [[CNSIZE]], i64* [[PSZ7]]
-
-// CHECK-DAG:   [[CBPADDR8:%.+]] = bitcast i8** [[BPADDR8]] to [[TT]]**
-// CHECK-DAG:   [[CPADDR8:%.+]] = bitcast i8** [[PADDR8]] to [[TT]]**
-// CHECK-DAG:   store [[TT]]* %{{.+}}, [[TT]]** [[CBPADDR8]]
-// CHECK-DAG:   store [[TT]]* %{{.+}}, [[TT]]** [[CPADDR8]]
+// CHECK-DAG:   store i[[SZ]] [[VLA0]], ptr [[BPADDR2]]
+// CHECK-DAG:   store i[[SZ]] [[VLA0]], ptr [[PADDR2]]
+
+// CHECK-DAG:   store i[[SZ]] [[VLA1]], ptr [[BPADDR6]]
+// CHECK-DAG:   store i[[SZ]] [[VLA1]], ptr [[PADDR6]]
+
+// CHECK-DAG:   store i[[SZ]] 5, ptr [[BPADDR5]]
+// CHECK-DAG:   store i[[SZ]] 5, ptr [[PADDR5]]
+
+// CHECK-DAG:   store i[[SZ]] [[A_CVAL]], ptr [[BPADDR0]]
+// CHECK-DAG:   store i[[SZ]] [[A_CVAL]], ptr [[PADDR0]]
+
+// CHECK-DAG:   store ptr %{{.+}}, ptr [[BPADDR1]]
+// CHECK-DAG:   store ptr %{{.+}}, ptr [[PADDR1]]
+
+// CHECK-DAG:   store ptr %{{.+}}, ptr [[BPADDR3]]
+// CHECK-DAG:   store ptr %{{.+}}, ptr [[PADDR3]]
+// CHECK-DAG:   store i64 [[BNSIZE]], ptr [[PSZ3]]
+
+// CHECK-DAG:   store ptr %{{.+}}, ptr [[BPADDR4]]
+// CHECK-DAG:   store ptr %{{.+}}, ptr [[PADDR4]]
+
+// CHECK-DAG:   store ptr %{{.+}}, ptr [[BPADDR7]]
+// CHECK-DAG:   store ptr %{{.+}}, ptr [[PADDR7]]
+// CHECK-DAG:   store i64 [[CNSIZE]], ptr [[PSZ7]]
+
+// CHECK-DAG:   store ptr %{{.+}}, ptr [[BPADDR8]]
+// CHECK-DAG:   store ptr %{{.+}}, ptr [[PADDR8]]
 
 // CHECK:       [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
 // CHECK-NEXT:  br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]]
@@ -401,107 +365,97 @@ int foo(int n) {
 
 // CHECK:       define internal void [[HVT0]]()
 
-// CHECK: define internal void [[HVT0_:@.+]](i[[SZ]]* noundef {{%[^,]+}}, i[[SZ]] noundef {{%[^,]+}})
-// CHECK: define internal {{.*}}i32 [[OMP_TASK_ENTRY]](i32 {{.*}}%0, [[KMP_TASK_T_WITH_PRIVATES]]* noalias noundef %1)
-// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 [[DEVICE:%.+]], i32 {{.+}}, i32 {{.+}}, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
+// CHECK: define internal void [[HVT0_:@.+]](ptr noundef {{%[^,]+}}, i[[SZ]] noundef {{%[^,]+}})
+// CHECK: define internal {{.*}}i32 [[OMP_TASK_ENTRY]](i32 {{.*}}%0, ptr noalias noundef %1)
+// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 [[DEVICE:%.+]], i32 {{.+}}, i32 {{.+}}, ptr @.{{.+}}.region_id, ptr [[ARGS:%.+]])
 // CHECK-DAG: [[BPARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 2
-// CHECK-DAG: store i8** [[BPR:%.+]], i8*** [[BPARG]]
+// CHECK-DAG: store ptr [[BPR:%.+]], ptr [[BPARG]]
 // CHECK-DAG: [[PARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 3
-// CHECK-DAG: store i8** [[PR:%.+]], i8*** [[PARG]]
+// CHECK-DAG: store ptr [[PR:%.+]], ptr [[PARG]]
 // CHECK-DAG: [[SARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 4
-// CHECK-DAG: store i64* [[SIZE:%.+]], i64** [[SARG]]
+// CHECK-DAG: store ptr [[SIZE:%.+]], ptr [[SARG]]
 // CHECK-DAG: [[DEVICE]] = sext i32 [[DEV:%.+]] to i64
-// CHECK-DAG: [[DEV]] = load i32, i32* [[DEVADDR:%.+]], align
-// CHECK-DAG: [[DEVADDR]] = getelementptr inbounds [[ANON_T]], [[ANON_T]]* {{%.+}}, i32 0, i32 2
-// CHECK-DAG: [[BPR]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BPRADDR:%.+]], i[[SZ]] 0, i[[SZ]] 0
-// CHECK-DAG: [[PR]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[PRADDR:%.+]], i[[SZ]] 0, i[[SZ]] 0
-// CHECK-DAG: [[SIZE]] = getelementptr inbounds [2 x i64], [2 x i64]* [[SIZEADDR:%.+]], i[[SZ]] 0, i[[SZ]] 0
-// CHECK-DAG: [[BPRADDR]] = load [2 x i8*]*, [2 x i8*]** [[FPPTR_BPR:%.+]], align
-// CHECK-DAG: [[PRADDR]] = load [2 x i8*]*, [2 x i8*]** [[FPPTR_PR:%.+]], align
-// CHECK-DAG: [[SIZEADDR]] = load [2 x i64]*, [2 x i64]** [[FPPTR_SIZE:%.+]], align
-// CHECK-DAG: [[FN:%.+]] = bitcast void (i8*, ...)* {{%[0-9]+}} to void (i8*,
-// CHECK-DAG: call void [[FN]](i8* {{%[^,]+}}, i[[SZ]]*** [[FPPTR_PLOCAL:%.+]], i32** [[FPPTR_GLOBAL:%.+]], [2 x i8*]** [[FPPTR_BPR]], [2 x i8*]** [[FPPTR_PR]], [2 x i64]** [[FPPTR_SIZE]])
-// CHECK-DAG: [[PLOCALADDR:%.+]] = load i[[SZ]]**, i[[SZ]]*** [[FPPTR_PLOCAL]], align
-// CHECK-DAG: {{%.+}} = load i32*, i32** [[FPPTR_GLOBAL:%.+]], align
+// CHECK-DAG: [[DEV]] = load i32, ptr [[DEVADDR:%.+]], align
+// CHECK-DAG: [[DEVADDR]] = getelementptr inbounds [[ANON_T]], ptr {{%.+}}, i32 0, i32 2
+// CHECK-DAG: [[BPR]] = load ptr, ptr [[FPPTR_BPR:%.+]], align
+// CHECK-DAG: [[PR]] = load ptr, ptr [[FPPTR_PR:%.+]], align
+// CHECK-DAG: [[SIZE]] = load ptr, ptr [[FPPTR_SIZE:%.+]], align
+// CHECK-DAG: call void {{%[0-9]+}}(ptr {{%[^,]+}}, ptr [[FPPTR_PLOCAL:%.+]], ptr [[FPPTR_GLOBAL:%.+]], ptr [[FPPTR_BPR]], ptr [[FPPTR_PR]], ptr [[FPPTR_SIZE]])
+// CHECK-DAG: [[PLOCALADDR:%.+]] = load ptr, ptr [[FPPTR_PLOCAL]], align
+// CHECK-DAG: {{%.+}} = load ptr, ptr [[FPPTR_GLOBAL:%.+]], align
 // CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]
 // CHECK: [[FAIL]]
-// CHECK: [[PLOCAL:%.+]] = load i[[SZ]]*, i[[SZ]]** [[PLOCALADDR]], align
-// CHECK: [[GLOBAL:%.+]] = load i32, i32* {{@.+}}, align
-// CHECK-64: [[CONVI:%.+]] = bitcast i64* [[GLOBALCAST:%.+]] to i32*
-// CHECK-32: store i32 [[GLOBAL]], i32* [[GLOBALCAST:%.+]], align
-// CHECK-64: store i32 [[GLOBAL]], i32* [[CONVI]], align
-// CHECK: [[GLOBAL:%.+]] = load i[[SZ]], i[[SZ]]* [[GLOBALCAST]], align
-// CHECK: call void [[HVT0_]](i[[SZ]]* [[PLOCAL]], i[[SZ]] [[GLOBAL]])
+// CHECK: [[PLOCAL:%.+]] = load ptr, ptr [[PLOCALADDR]], align
+// CHECK: [[GLOBAL:%.+]] = load i32, ptr {{@.+}}, align
+// CHECK-32: store i32 [[GLOBAL]], ptr [[GLOBALCAST:%.+]], align
+// CHECK-64: store i32 [[GLOBAL]], ptr [[GLOBALCAST:%.+]], align
+// CHECK: [[GLOBAL:%.+]] = load i[[SZ]], ptr [[GLOBALCAST]], align
+// CHECK: call void [[HVT0_]](ptr [[PLOCAL]], i[[SZ]] [[GLOBAL]])
 // CHECK-NEXT: br label %[[END]]
 // CHECK: [[END]]
 
 // CHECK:       define internal void [[HVT1]](i[[SZ]] noundef %{{.+}})
 // Create stack storage and store argument in there.
 // CHECK:       [[AA_ADDR:%.+]] = alloca i[[SZ]], align
-// CHECK:       store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align
-// CHECK-64:    [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i32*
-// CHECK-64:    load i32, i32* [[AA_CADDR]], align
-// CHECK-32:    load i32, i32* [[AA_ADDR]], align
+// CHECK:       store i[[SZ]] %{{.+}}, ptr [[AA_ADDR]], align
+// CHECK-64:    load i32, ptr [[AA_ADDR]], align
+// CHECK-32:    load i32, ptr [[AA_ADDR]], align
 
 // CHECK:       define internal void [[HVT2]](i[[SZ]] noundef %{{.+}})
 // Create stack storage and store argument in there.
 // CHECK:       [[AA_ADDR:%.+]] = alloca i[[SZ]], align
-// CHECK:       store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align
-// CHECK:       [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i16*
-// CHECK:       load i16, i16* [[AA_CADDR]], align
+// CHECK:       store i[[SZ]] %{{.+}}, ptr [[AA_ADDR]], align
+// CHECK:       load i16, ptr [[AA_ADDR]], align
 
 // CHECK:       define internal void [[HVT3]]
 // Create stack storage and store argument in there.
 // CHECK:       [[A_ADDR:%.+]] = alloca i[[SZ]], align
 // CHECK:       [[AA_ADDR:%.+]] = alloca i[[SZ]], align
-// CHECK-DAG:   store i[[SZ]] %{{.+}}, i[[SZ]]* [[A_ADDR]], align
-// CHECK-DAG:   store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align
-// CHECK-64-DAG:[[A_CADDR:%.+]] = bitcast i[[SZ]]* [[A_ADDR]] to i32*
-// CHECK-DAG:   [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i16*
-// CHECK-64-DAG:load i32, i32* [[A_CADDR]], align
-// CHECK-32-DAG:load i32, i32* [[A_ADDR]], align
-// CHECK-DAG:   load i16, i16* [[AA_CADDR]], align
+// CHECK-DAG:   store i[[SZ]] %{{.+}}, ptr [[A_ADDR]], align
+// CHECK-DAG:   store i[[SZ]] %{{.+}}, ptr [[AA_ADDR]], align
+// CHECK-64-DAG:load i32, ptr [[A_ADDR]], align
+// CHECK-32-DAG:load i32, ptr [[A_ADDR]], align
+// CHECK-DAG:   load i16, ptr [[AA_ADDR]], align
 
 // CHECK:       define internal void [[HVT4]]
 // Create local storage for each capture.
 // CHECK:       [[LOCAL_A:%.+]] = alloca i[[SZ]]
-// CHECK:       [[LOCAL_B:%.+]] = alloca [10 x float]*
+// CHECK:       [[LOCAL_B:%.+]] = alloca ptr
 // CHECK:       [[LOCAL_VLA1:%.+]] = alloca i[[SZ]]
-// CHECK:       [[LOCAL_BN:%.+]] = alloca float*
-// CHECK:       [[LOCAL_C:%.+]] = alloca [5 x [10 x double]]*
+// CHECK:       [[LOCAL_BN:%.+]] = alloca ptr
+// CHECK:       [[LOCAL_C:%.+]] = alloca ptr
 // CHECK:       [[LOCAL_VLA2:%.+]] = alloca i[[SZ]]
 // CHECK:       [[LOCAL_VLA3:%.+]] = alloca i[[SZ]]
-// CHECK:       [[LOCAL_CN:%.+]] = alloca double*
-// CHECK:       [[LOCAL_D:%.+]] = alloca [[TT]]*
-// CHECK-DAG:   store i[[SZ]] [[ARG_A:%.+]], i[[SZ]]* [[LOCAL_A]]
-// CHECK-DAG:   store [10 x float]* [[ARG_B:%.+]], [10 x float]** [[LOCAL_B]]
-// CHECK-DAG:   store i[[SZ]] [[ARG_VLA1:%.+]], i[[SZ]]* [[LOCAL_VLA1]]
-// CHECK-DAG:   store float* [[ARG_BN:%.+]], float** [[LOCAL_BN]]
-// CHECK-DAG:   store [5 x [10 x double]]* [[ARG_C:%.+]], [5 x [10 x double]]** [[LOCAL_C]]
-// CHECK-DAG:   store i[[SZ]] [[ARG_VLA2:%.+]], i[[SZ]]* [[LOCAL_VLA2]]
-// CHECK-DAG:   store i[[SZ]] [[ARG_VLA3:%.+]], i[[SZ]]* [[LOCAL_VLA3]]
-// CHECK-DAG:   store double* [[ARG_CN:%.+]], double** [[LOCAL_CN]]
-// CHECK-DAG:   store [[TT]]* [[ARG_D:%.+]], [[TT]]** [[LOCAL_D]]
-
-// CHECK-64-DAG:[[REF_A:%.+]] = bitcast i[[SZ]]* [[LOCAL_A]] to i32*
-// CHECK-DAG:   [[REF_B:%.+]] = load [10 x float]*, [10 x float]** [[LOCAL_B]],
-// CHECK-DAG:   [[VAL_VLA1:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_VLA1]],
-// CHECK-DAG:   [[REF_BN:%.+]] = load float*, float** [[LOCAL_BN]],
-// CHECK-DAG:   [[REF_C:%.+]] = load [5 x [10 x double]]*, [5 x [10 x double]]** [[LOCAL_C]],
-// CHECK-DAG:   [[VAL_VLA2:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_VLA2]],
-// CHECK-DAG:   [[VAL_VLA3:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_VLA3]],
-// CHECK-DAG:   [[REF_CN:%.+]] = load double*, double** [[LOCAL_CN]],
-// CHECK-DAG:   [[REF_D:%.+]] = load [[TT]]*, [[TT]]** [[LOCAL_D]],
+// CHECK:       [[LOCAL_CN:%.+]] = alloca ptr
+// CHECK:       [[LOCAL_D:%.+]] = alloca ptr
+// CHECK-DAG:   store i[[SZ]] [[ARG_A:%.+]], ptr [[LOCAL_A]]
+// CHECK-DAG:   store ptr [[ARG_B:%.+]], ptr [[LOCAL_B]]
+// CHECK-DAG:   store i[[SZ]] [[ARG_VLA1:%.+]], ptr [[LOCAL_VLA1]]
+// CHECK-DAG:   store ptr [[ARG_BN:%.+]], ptr [[LOCAL_BN]]
+// CHECK-DAG:   store ptr [[ARG_C:%.+]], ptr [[LOCAL_C]]
+// CHECK-DAG:   store i[[SZ]] [[ARG_VLA2:%.+]], ptr [[LOCAL_VLA2]]
+// CHECK-DAG:   store i[[SZ]] [[ARG_VLA3:%.+]], ptr [[LOCAL_VLA3]]
+// CHECK-DAG:   store ptr [[ARG_CN:%.+]], ptr [[LOCAL_CN]]
+// CHECK-DAG:   store ptr [[ARG_D:%.+]], ptr [[LOCAL_D]]
+
+// CHECK-DAG:   [[REF_B:%.+]] = load ptr, ptr [[LOCAL_B]],
+// CHECK-DAG:   [[VAL_VLA1:%.+]] = load i[[SZ]], ptr [[LOCAL_VLA1]],
+// CHECK-DAG:   [[REF_BN:%.+]] = load ptr, ptr [[LOCAL_BN]],
+// CHECK-DAG:   [[REF_C:%.+]] = load ptr, ptr [[LOCAL_C]],
+// CHECK-DAG:   [[VAL_VLA2:%.+]] = load i[[SZ]], ptr [[LOCAL_VLA2]],
+// CHECK-DAG:   [[VAL_VLA3:%.+]] = load i[[SZ]], ptr [[LOCAL_VLA3]],
+// CHECK-DAG:   [[REF_CN:%.+]] = load ptr, ptr [[LOCAL_CN]],
+// CHECK-DAG:   [[REF_D:%.+]] = load ptr, ptr [[LOCAL_D]],
 
 // Use captures.
-// CHECK-64-DAG:   load i32, i32* [[REF_A]]
-// CHECK-32-DAG:   load i32, i32* [[LOCAL_A]]
-// CHECK-DAG:   getelementptr inbounds [10 x float], [10 x float]* [[REF_B]], i[[SZ]] 0, i[[SZ]] 2
-// CHECK-DAG:   getelementptr inbounds float, float* [[REF_BN]], i[[SZ]] 3
-// CHECK-DAG:   getelementptr inbounds [5 x [10 x double]], [5 x [10 x double]]* [[REF_C]], i[[SZ]] 0, i[[SZ]] 1
-// CHECK-DAG:   getelementptr inbounds double, double* [[REF_CN]], i[[SZ]] %{{.+}}
-// CHECK-DAG:   getelementptr inbounds [[TT]], [[TT]]* [[REF_D]], i32 0, i32 0
+// CHECK-64-DAG:   load i32, ptr [[LOCAL_A]]
+// CHECK-32-DAG:   load i32, ptr [[LOCAL_A]]
+// CHECK-DAG:   getelementptr inbounds [10 x float], ptr [[REF_B]], i[[SZ]] 0, i[[SZ]] 2
+// CHECK-DAG:   getelementptr inbounds float, ptr [[REF_BN]], i[[SZ]] 3
+// CHECK-DAG:   getelementptr inbounds [5 x [10 x double]], ptr [[REF_C]], i[[SZ]] 0, i[[SZ]] 1
+// CHECK-DAG:   getelementptr inbounds double, ptr [[REF_CN]], i[[SZ]] %{{.+}}
+// CHECK-DAG:   getelementptr inbounds [[TT]], ptr [[REF_D]], i32 0, i32 0
 
 template<typename tx>
 tx ftemplate(int n) {
@@ -562,7 +516,7 @@ int bar(int n){
   a += foo(n);
 
   S1 S;
-  // CHECK: call {{.*}}i32 [[FS1:@.+]]([[S1]]* {{.*}}, i32 {{.*}})
+  // CHECK: call {{.*}}i32 [[FS1:@.+]](ptr {{.*}}, i32 {{.*}})
   a += S.r1(n);
 
   // CHECK: call {{.*}}i32 [[FSTATIC:@.+]](i32 {{.*}})
@@ -577,14 +531,13 @@ int bar(int n){
 //
 // CHECK: define {{.*}}[[FS1]]
 //
-// CHECK:          i8* @llvm.stacksave()
-// CHECK-64:       [[B_ADDR:%.+]] = bitcast i[[SZ]]* [[B_CADDR:%.+]] to i32*
-// CHECK-64:       store i32 %{{.+}}, i32* [[B_ADDR]],
-// CHECK-64:       [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_CADDR]],
+// CHECK:          ptr @llvm.stacksave()
+// CHECK-64:       store i32 %{{.+}}, ptr [[B_CADDR:%.+]],
+// CHECK-64:       [[B_CVAL:%.+]] = load i[[SZ]], ptr [[B_CADDR]],
 
-// CHECK-32:       store i32 %{{.+}}, i32* %__vla_expr
-// CHECK-32:       store i32 %{{.+}}, i32* [[B_ADDR:%.+]],
-// CHECK-32:       [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_ADDR]],
+// CHECK-32:       store i32 %{{.+}}, ptr %__vla_expr
+// CHECK-32:       store i32 %{{.+}}, ptr [[B_CADDR:%.+]],
+// CHECK-32:       [[B_CVAL:%.+]] = load i[[SZ]], ptr [[B_CADDR]],
 
 // CHECK:       [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 60
 // CHECK:       br i1 [[IF]], label %[[TRY:[^,]+]], label %[[IFELSE:[^,]+]]
@@ -595,55 +548,45 @@ int bar(int n){
 // CHECK-32:    [[CSZSIZE:%.+]] = mul nuw i32 [[CELEMSIZE2]], 2
 // CHECK-32:    [[CSIZE:%.+]] = sext i32 [[CSZSIZE]] to i64
 
-// CHECK-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 {{.+}}, i32 {{.+}}, i32 {{.+}}, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
+// CHECK-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 {{.+}}, i32 {{.+}}, i32 {{.+}}, ptr @.{{.+}}.region_id, ptr [[ARGS:%.+]])
 // CHECK-DAG:   [[BPARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 2
-// CHECK-DAG:   store i8** [[BPR:%.+]], i8*** [[BPARG]]
+// CHECK-DAG:   store ptr [[BPR:%.+]], ptr [[BPARG]]
 // CHECK-DAG:   [[PARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 3
-// CHECK-DAG:   store i8** [[PR:%.+]], i8*** [[PARG]]
+// CHECK-DAG:   store ptr [[PR:%.+]], ptr [[PARG]]
 // CHECK-DAG:   [[SARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 4
-// CHECK-DAG:   store i64* [[SZ7:%.+]], i64** [[SARG]]
-// CHECK-DAG:   [[BPR]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP:%.+]], i32 0, i32 0
-// CHECK-DAG:   [[PR]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P:%.+]], i32 0, i32 0
-// CHECK-DAG:   [[SZ7]] = getelementptr inbounds [5 x i64], [5 x i64]* [[PSZ:%.+]], i32 0, i32 0
-// CHECK-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 0, i32 [[IDX0:0]]
-// CHECK-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 0, i32 [[IDX0]]
-// CHECK-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 0, i32 [[IDX1:1]]
-// CHECK-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 0, i32 [[IDX1]]
-// CHECK-DAG:   [[BPADDR2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 0, i32 [[IDX2:2]]
-// CHECK-DAG:   [[PADDR2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 0, i32 [[IDX2]]
-// CHECK-DAG:   [[BPADDR3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 0, i32 [[IDX3:3]]
-// CHECK-DAG:   [[PADDR3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 0, i32 [[IDX3]]
-// CHECK-DAG:   [[BPADDR4:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 0, i32 [[IDX4:4]]
-// CHECK-DAG:   [[PADDR4:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 0, i32 [[IDX4]]
-// CHECK-DAG:   [[PSZ4:%.+]] = getelementptr inbounds [5 x i64], [5 x i64]* [[PSZ:%.+]], i32 0, i32 [[IDX4]]
+// CHECK-DAG:   store ptr [[SZ7:%.+]], ptr [[SARG]]
+// CHECK-DAG:   [[BPR]] = getelementptr inbounds [5 x ptr], ptr [[BP:%.+]], i32 0, i32 0
+// CHECK-DAG:   [[PR]] = getelementptr inbounds [5 x ptr], ptr [[P:%.+]], i32 0, i32 0
+// CHECK-DAG:   [[SZ7]] = getelementptr inbounds [5 x i64], ptr [[PSZ:%.+]], i32 0, i32 0
+// CHECK-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [5 x ptr], ptr [[BP]], i32 0, i32 [[IDX0:0]]
+// CHECK-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [5 x ptr], ptr [[P]], i32 0, i32 [[IDX0]]
+// CHECK-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [5 x ptr], ptr [[BP]], i32 0, i32 [[IDX1:1]]
+// CHECK-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [5 x ptr], ptr [[P]], i32 0, i32 [[IDX1]]
+// CHECK-DAG:   [[BPADDR2:%.+]] = getelementptr inbounds [5 x ptr], ptr [[BP]], i32 0, i32 [[IDX2:2]]
+// CHECK-DAG:   [[PADDR2:%.+]] = getelementptr inbounds [5 x ptr], ptr [[P]], i32 0, i32 [[IDX2]]
+// CHECK-DAG:   [[BPADDR3:%.+]] = getelementptr inbounds [5 x ptr], ptr [[BP]], i32 0, i32 [[IDX3:3]]
+// CHECK-DAG:   [[PADDR3:%.+]] = getelementptr inbounds [5 x ptr], ptr [[P]], i32 0, i32 [[IDX3]]
+// CHECK-DAG:   [[BPADDR4:%.+]] = getelementptr inbounds [5 x ptr], ptr [[BP]], i32 0, i32 [[IDX4:4]]
+// CHECK-DAG:   [[PADDR4:%.+]] = getelementptr inbounds [5 x ptr], ptr [[P]], i32 0, i32 [[IDX4]]
+// CHECK-DAG:   [[PSZ4:%.+]] = getelementptr inbounds [5 x i64], ptr [[PSZ:%.+]], i32 0, i32 [[IDX4]]
 
 // The names below are not necessarily consistent with the names used for the
 // addresses above as some are repeated.
-// CHECK-DAG:   [[CBPADDR4:%.+]] = bitcast i8** [[BPADDR4]] to i16**
-// CHECK-DAG:   [[CPADDR4:%.+]] = bitcast i8** [[PADDR4]] to i16**
-// CHECK-DAG:   store i16* %{{.+}}, i16** [[CBPADDR4]]
-// CHECK-DAG:   store i16* %{{.+}}, i16** [[CPADDR4]]
-// CHECK-DAG:   store i64 [[CSIZE]], i64* [[PSZ4]]
-
-// CHECK-DAG:   [[CBPADDR3:%.+]] = bitcast i8** [[BPADDR3]] to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR3:%.+]] = bitcast i8** [[PADDR3]] to i[[SZ]]*
-// CHECK-DAG:   store i[[SZ]] [[VLA0]], i[[SZ]]* [[CBPADDR3]]
-// CHECK-DAG:   store i[[SZ]] [[VLA0]], i[[SZ]]* [[CPADDR3]]
-
-// CHECK-DAG:   [[CBPADDR2:%.+]] = bitcast i8** [[BPADDR2]] to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR2:%.+]] = bitcast i8** [[PADDR2]] to i[[SZ]]*
-// CHECK-DAG:   store i[[SZ]] 2, i[[SZ]]* [[CBPADDR2]]
-// CHECK-DAG:   store i[[SZ]] 2, i[[SZ]]* [[CPADDR2]]
-
-// CHECK-DAG:   [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to i[[SZ]]*
-// CHECK-DAG:   store i[[SZ]] [[B_CVAL]], i[[SZ]]* [[CBPADDR1]]
-// CHECK-DAG:   store i[[SZ]] [[B_CVAL]], i[[SZ]]* [[CPADDR1]]
-
-// CHECK-DAG:   [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to [[S1]]**
-// CHECK-DAG:   [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to double**
-// CHECK-DAG:   store [[S1]]* [[THIS:%.+]], [[S1]]** [[CBPADDR0]]
-// CHECK-DAG:   store double* [[A:%.+]], double** [[CPADDR0]]
+// CHECK-DAG:   store ptr %{{.+}}, ptr [[BPADDR4]]
+// CHECK-DAG:   store ptr %{{.+}}, ptr [[PADDR4]]
+// CHECK-DAG:   store i64 [[CSIZE]], ptr [[PSZ4]]
+
+// CHECK-DAG:   store i[[SZ]] [[VLA0]], ptr [[BPADDR3]]
+// CHECK-DAG:   store i[[SZ]] [[VLA0]], ptr [[PADDR3]]
+
+// CHECK-DAG:   store i[[SZ]] 2, ptr [[BPADDR2]]
+// CHECK-DAG:   store i[[SZ]] 2, ptr [[PADDR2]]
+
+// CHECK-DAG:   store i[[SZ]] [[B_CVAL]], ptr [[BPADDR1]]
+// CHECK-DAG:   store i[[SZ]] [[B_CVAL]], ptr [[PADDR1]]
+
+// CHECK-DAG:   store ptr [[THIS:%.+]], ptr [[BPADDR0]]
+// CHECK-DAG:   store ptr [[A:%.+]], ptr [[PADDR0]]
 
 // CHECK:       [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
 // CHECK-NEXT:  br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]]
@@ -664,41 +607,33 @@ int bar(int n){
 // CHECK:       [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 50
 // CHECK:       br i1 [[IF]], label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]]
 // CHECK:       [[IFTHEN]]
-// CHECK-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 {{.+}}, i32 {{.+}}, i32 {{.+}}, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
+// CHECK-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 {{.+}}, i32 {{.+}}, i32 {{.+}}, ptr @.{{.+}}.region_id, ptr [[ARGS:%.+]])
 // CHECK-DAG:   [[BPARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 2
-// CHECK-DAG:   store i8** [[BPR:%.+]], i8*** [[BPARG]]
+// CHECK-DAG:   store ptr [[BPR:%.+]], ptr [[BPARG]]
 // CHECK-DAG:   [[PARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 3
-// CHECK-DAG:   store i8** [[PR:%.+]], i8*** [[PARG]]
-// CHECK-DAG:   [[BPR]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[BP:%.+]], i32 0, i32 0
-// CHECK-DAG:   [[PR]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[P:%.+]], i32 0, i32 0
-
-// CHECK-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[BP]], i32 0, i32 0
-// CHECK-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[P]], i32 0, i32 0
-// CHECK-DAG:   [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]*
-// CHECK-DAG:   store i[[SZ]] [[VAL0:%[^,]+]], i[[SZ]]* [[CBPADDR0]]
-// CHECK-DAG:   store i[[SZ]] [[VAL0]], i[[SZ]]* [[CPADDR0]]
-
-// CHECK-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[BP]], i32 0, i32 1
-// CHECK-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[P]], i32 0, i32 1
-// CHECK-DAG:   [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to i[[SZ]]*
-// CHECK-DAG:   store i[[SZ]] [[VAL1:%[^,]+]], i[[SZ]]* [[CBPADDR1]]
-// CHECK-DAG:   store i[[SZ]] [[VAL1]], i[[SZ]]* [[CPADDR1]]
-
-// CHECK-DAG:   [[BPADDR2:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[BP]], i32 0, i32 2
-// CHECK-DAG:   [[PADDR2:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[P]], i32 0, i32 2
-// CHECK-DAG:   [[CBPADDR2:%.+]] = bitcast i8** [[BPADDR2]] to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR2:%.+]] = bitcast i8** [[PADDR2]] to i[[SZ]]*
-// CHECK-DAG:   store i[[SZ]] [[VAL2:%[^,]+]], i[[SZ]]* [[CBPADDR2]]
-// CHECK-DAG:   store i[[SZ]] [[VAL2]], i[[SZ]]* [[CPADDR2]]
-
-// CHECK-DAG:   [[BPADDR3:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[BP]], i32 0, i32 3
-// CHECK-DAG:   [[PADDR3:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[P]], i32 0, i32 3
-// CHECK-DAG:   [[CBPADDR3:%.+]] = bitcast i8** [[BPADDR3]] to [10 x i32]**
-// CHECK-DAG:   [[CPADDR3:%.+]] = bitcast i8** [[PADDR3]] to [10 x i32]**
-// CHECK-DAG:   store [10 x i32]* [[VAL3:%[^,]+]], [10 x i32]** [[CBPADDR3]]
-// CHECK-DAG:   store [10 x i32]* [[VAL3]], [10 x i32]** [[CPADDR3]]
+// CHECK-DAG:   store ptr [[PR:%.+]], ptr [[PARG]]
+// CHECK-DAG:   [[BPR]] = getelementptr inbounds [4 x ptr], ptr [[BP:%.+]], i32 0, i32 0
+// CHECK-DAG:   [[PR]] = getelementptr inbounds [4 x ptr], ptr [[P:%.+]], i32 0, i32 0
+
+// CHECK-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [4 x ptr], ptr [[BP]], i32 0, i32 0
+// CHECK-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [4 x ptr], ptr [[P]], i32 0, i32 0
+// CHECK-DAG:   store i[[SZ]] [[VAL0:%[^,]+]], ptr [[BPADDR0]]
+// CHECK-DAG:   store i[[SZ]] [[VAL0]], ptr [[PADDR0]]
+
+// CHECK-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [4 x ptr], ptr [[BP]], i32 0, i32 1
+// CHECK-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [4 x ptr], ptr [[P]], i32 0, i32 1
+// CHECK-DAG:   store i[[SZ]] [[VAL1:%[^,]+]], ptr [[BPADDR1]]
+// CHECK-DAG:   store i[[SZ]] [[VAL1]], ptr [[PADDR1]]
+
+// CHECK-DAG:   [[BPADDR2:%.+]] = getelementptr inbounds [4 x ptr], ptr [[BP]], i32 0, i32 2
+// CHECK-DAG:   [[PADDR2:%.+]] = getelementptr inbounds [4 x ptr], ptr [[P]], i32 0, i32 2
+// CHECK-DAG:   store i[[SZ]] [[VAL2:%[^,]+]], ptr [[BPADDR2]]
+// CHECK-DAG:   store i[[SZ]] [[VAL2]], ptr [[PADDR2]]
+
+// CHECK-DAG:   [[BPADDR3:%.+]] = getelementptr inbounds [4 x ptr], ptr [[BP]], i32 0, i32 3
+// CHECK-DAG:   [[PADDR3:%.+]] = getelementptr inbounds [4 x ptr], ptr [[P]], i32 0, i32 3
+// CHECK-DAG:   store ptr [[VAL3:%[^,]+]], ptr [[BPADDR3]]
+// CHECK-DAG:   store ptr [[VAL3]], ptr [[PADDR3]]
 
 // CHECK:       [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
 // CHECK-NEXT:  br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]]
@@ -719,34 +654,28 @@ int bar(int n){
 // CHECK:       [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 40
 // CHECK:       br i1 [[IF]], label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]]
 // CHECK:       [[IFTHEN]]
-// CHECK-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 {{.+}}, i32 {{.+}}, i32 {{.+}}, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
+// CHECK-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 {{.+}}, i32 {{.+}}, i32 {{.+}}, ptr @.{{.+}}.region_id, ptr [[ARGS:%.+]])
 // CHECK-DAG:   [[BPARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 2
-// CHECK-DAG:   store i8** [[BPR:%.+]], i8*** [[BPARG]]
+// CHECK-DAG:   store ptr [[BPR:%.+]], ptr [[BPARG]]
 // CHECK-DAG:   [[PARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 3
-// CHECK-DAG:   store i8** [[PR:%.+]], i8*** [[PARG]]
-// CHECK-DAG:   [[BPR]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BP:%.+]], i32 0, i32 0
-// CHECK-DAG:   [[PR]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[P:%.+]], i32 0, i32 0
-
-// CHECK-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BP]], i32 0, i32 0
-// CHECK-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[P]], i32 0, i32 0
-// CHECK-DAG:   [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]*
-// CHECK-DAG:   store i[[SZ]] [[VAL0:%[^,]+]], i[[SZ]]* [[CBPADDR0]]
-// CHECK-DAG:   store i[[SZ]] [[VAL0]], i[[SZ]]* [[CPADDR0]]
-
-// CHECK-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BP]], i32 0, i32 1
-// CHECK-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[P]], i32 0, i32 1
-// CHECK-DAG:   [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to i[[SZ]]*
-// CHECK-DAG:   store i[[SZ]] [[VAL1:%[^,]+]], i[[SZ]]* [[CBPADDR1]]
-// CHECK-DAG:   store i[[SZ]] [[VAL1]], i[[SZ]]* [[CPADDR1]]
-
-// CHECK-DAG:   [[BPADDR2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BP]], i32 0, i32 2
-// CHECK-DAG:   [[PADDR2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[P]], i32 0, i32 2
-// CHECK-DAG:   [[CBPADDR2:%.+]] = bitcast i8** [[BPADDR2]] to [10 x i32]**
-// CHECK-DAG:   [[CPADDR2:%.+]] = bitcast i8** [[PADDR2]] to [10 x i32]**
-// CHECK-DAG:   store [10 x i32]* [[VAL2:%[^,]+]], [10 x i32]** [[CBPADDR2]]
-// CHECK-DAG:   store [10 x i32]* [[VAL2]], [10 x i32]** [[CPADDR2]]
+// CHECK-DAG:   store ptr [[PR:%.+]], ptr [[PARG]]
+// CHECK-DAG:   [[BPR]] = getelementptr inbounds [3 x ptr], ptr [[BP:%.+]], i32 0, i32 0
+// CHECK-DAG:   [[PR]] = getelementptr inbounds [3 x ptr], ptr [[P:%.+]], i32 0, i32 0
+
+// CHECK-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [3 x ptr], ptr [[BP]], i32 0, i32 0
+// CHECK-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [3 x ptr], ptr [[P]], i32 0, i32 0
+// CHECK-DAG:   store i[[SZ]] [[VAL0:%[^,]+]], ptr [[BPADDR0]]
+// CHECK-DAG:   store i[[SZ]] [[VAL0]], ptr [[PADDR0]]
+
+// CHECK-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [3 x ptr], ptr [[BP]], i32 0, i32 1
+// CHECK-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [3 x ptr], ptr [[P]], i32 0, i32 1
+// CHECK-DAG:   store i[[SZ]] [[VAL1:%[^,]+]], ptr [[BPADDR1]]
+// CHECK-DAG:   store i[[SZ]] [[VAL1]], ptr [[PADDR1]]
+
+// CHECK-DAG:   [[BPADDR2:%.+]] = getelementptr inbounds [3 x ptr], ptr [[BP]], i32 0, i32 2
+// CHECK-DAG:   [[PADDR2:%.+]] = getelementptr inbounds [3 x ptr], ptr [[P]], i32 0, i32 2
+// CHECK-DAG:   store ptr [[VAL2:%[^,]+]], ptr [[BPADDR2]]
+// CHECK-DAG:   store ptr [[VAL2]], ptr [[PADDR2]]
 
 // CHECK:       [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
 // CHECK-NEXT:  br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]]
@@ -765,29 +694,27 @@ int bar(int n){
 
 // OMP45: define {{.*}}@{{.*}}zee{{.*}}
 
-// OMP45:       [[LOCAL_THIS:%.+]] = alloca [[S2]]*
-// OMP45:       [[BP:%.+]] = alloca [1 x i8*]
-// OMP45:       [[P:%.+]] = alloca [1 x i8*]
-// OMP45:       [[LOCAL_THIS1:%.+]] = load [[S2]]*, [[S2]]** [[LOCAL_THIS]]
+// OMP45:       [[LOCAL_THIS:%.+]] = alloca ptr
+// OMP45:       [[BP:%.+]] = alloca [1 x ptr]
+// OMP45:       [[P:%.+]] = alloca [1 x ptr]
+// OMP45:       [[LOCAL_THIS1:%.+]] = load ptr, ptr [[LOCAL_THIS]]
 
 // OMP45:       call void @__kmpc_critical(
-// OMP45:       [[ARR_IDX:%.+]] = getelementptr inbounds [[S2]], [[S2]]* [[LOCAL_THIS1]], i[[SZ]] 0
-// OMP45:       [[ARR_IDX2:%.+]] = getelementptr inbounds [[S2]], [[S2]]* [[LOCAL_THIS1]], i[[SZ]] 0
-
-// OMP45-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP]], i32 0, i32 0
-// OMP45-DAG:   [[PADDR0:%.+]] =  getelementptr inbounds [1 x i8*], [1 x i8*]* [[P]], i32 0, i32 0
-// OMP45-DAG:   [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to [[S2]]**
-// OMP45-DAG:   [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to [[S2]]**
-// OMP45-DAG:   store [[S2]]* [[ARR_IDX]], [[S2]]** [[CBPADDR0]]
-// OMP45-DAG:   store [[S2]]* [[ARR_IDX2]], [[S2]]** [[CPADDR0]]
-
-// OMP45:       [[BPR:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP]], i32 0, i32 0
-// OMP45:       [[PR:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P]], i32 0, i32 0
-// OMP45:       [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 -1, i32 {{.+}}, i32 {{.+}}, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
+// OMP45:       [[ARR_IDX:%.+]] = getelementptr inbounds [[S2]], ptr [[LOCAL_THIS1]], i[[SZ]] 0
+// OMP45:       [[ARR_IDX2:%.+]] = getelementptr inbounds [[S2]], ptr [[LOCAL_THIS1]], i[[SZ]] 0
+
+// OMP45-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[BP]], i32 0, i32 0
+// OMP45-DAG:   [[PADDR0:%.+]] =  getelementptr inbounds [1 x ptr], ptr [[P]], i32 0, i32 0
+// OMP45-DAG:   store ptr [[ARR_IDX]], ptr [[BPADDR0]]
+// OMP45-DAG:   store ptr [[ARR_IDX2]], ptr [[PADDR0]]
+
+// OMP45:       [[BPR:%.+]] = getelementptr inbounds [1 x ptr], ptr [[BP]], i32 0, i32 0
+// OMP45:       [[PR:%.+]] = getelementptr inbounds [1 x ptr], ptr [[P]], i32 0, i32 0
+// OMP45:       [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 -1, i32 {{.+}}, i32 {{.+}}, ptr @.{{.+}}.region_id, ptr [[ARGS:%.+]])
 // OMP45-NEXT:  [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
 // OMP45-NEXT:  br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]
 // OMP45:       [[FAIL]]
-// OMP45:       call void [[HVT0:@.+]]([[S2]]* [[LOCAL_THIS1]])
+// OMP45:       call void [[HVT0:@.+]](ptr [[LOCAL_THIS1]])
 // OMP45-NEXT:  br label %[[END]]
 // OMP45:       [[END]]
 // OMP45:       call void @__kmpc_end_critical(
@@ -797,27 +724,26 @@ int bar(int n){
 
 // CHECK:       define internal void [[HVT7]]
 // Create local storage for each capture.
-// CHECK:       [[LOCAL_THIS:%.+]] = alloca [[S1]]*
+// CHECK:       [[LOCAL_THIS:%.+]] = alloca ptr
 // CHECK:       [[LOCAL_B:%.+]] = alloca i[[SZ]]
 // CHECK:       [[LOCAL_VLA1:%.+]] = alloca i[[SZ]]
 // CHECK:       [[LOCAL_VLA2:%.+]] = alloca i[[SZ]]
-// CHECK:       [[LOCAL_C:%.+]] = alloca i16*
-// CHECK-DAG:   store [[S1]]* [[ARG_THIS:%.+]], [[S1]]** [[LOCAL_THIS]]
-// CHECK-DAG:   store i[[SZ]] [[ARG_B:%.+]], i[[SZ]]* [[LOCAL_B]]
-// CHECK-DAG:   store i[[SZ]] [[ARG_VLA1:%.+]], i[[SZ]]* [[LOCAL_VLA1]]
-// CHECK-DAG:   store i[[SZ]] [[ARG_VLA2:%.+]], i[[SZ]]* [[LOCAL_VLA2]]
-// CHECK-DAG:   store i16* [[ARG_C:%.+]], i16** [[LOCAL_C]]
+// CHECK:       [[LOCAL_C:%.+]] = alloca ptr
+// CHECK-DAG:   store ptr [[ARG_THIS:%.+]], ptr [[LOCAL_THIS]]
+// CHECK-DAG:   store i[[SZ]] [[ARG_B:%.+]], ptr [[LOCAL_B]]
+// CHECK-DAG:   store i[[SZ]] [[ARG_VLA1:%.+]], ptr [[LOCAL_VLA1]]
+// CHECK-DAG:   store i[[SZ]] [[ARG_VLA2:%.+]], ptr [[LOCAL_VLA2]]
+// CHECK-DAG:   store ptr [[ARG_C:%.+]], ptr [[LOCAL_C]]
 // Store captures in the context.
-// CHECK-DAG:   [[REF_THIS:%.+]] = load [[S1]]*, [[S1]]** [[LOCAL_THIS]],
-// CHECK-64-DAG:[[REF_B:%.+]] = bitcast i[[SZ]]* [[LOCAL_B]] to i32*
-// CHECK-DAG:   [[VAL_VLA1:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_VLA1]],
-// CHECK-DAG:   [[VAL_VLA2:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_VLA2]],
-// CHECK-DAG:   [[REF_C:%.+]] = load i16*, i16** [[LOCAL_C]],
+// CHECK-DAG:   [[REF_THIS:%.+]] = load ptr, ptr [[LOCAL_THIS]],
+// CHECK-DAG:   [[VAL_VLA1:%.+]] = load i[[SZ]], ptr [[LOCAL_VLA1]],
+// CHECK-DAG:   [[VAL_VLA2:%.+]] = load i[[SZ]], ptr [[LOCAL_VLA2]],
+// CHECK-DAG:   [[REF_C:%.+]] = load ptr, ptr [[LOCAL_C]],
 // Use captures.
-// CHECK-DAG:   getelementptr inbounds [[S1]], [[S1]]* [[REF_THIS]], i32 0, i32 0
-// CHECK-64-DAG:load i32, i32* [[REF_B]]
-// CHECK-32-DAG:load i32, i32* [[LOCAL_B]]
-// CHECK-DAG:   getelementptr inbounds i16, i16* [[REF_C]], i[[SZ]] %{{.+}}
+// CHECK-DAG:   getelementptr inbounds [[S1]], ptr [[REF_THIS]], i32 0, i32 0
+// CHECK-64-DAG:load i32, ptr [[LOCAL_B]]
+// CHECK-32-DAG:load i32, ptr [[LOCAL_B]]
+// CHECK-DAG:   getelementptr inbounds i16, ptr [[REF_C]], i[[SZ]] %{{.+}}
 
 
 // CHECK:       define internal void [[HVT6]]
@@ -825,66 +751,59 @@ int bar(int n){
 // CHECK:       [[LOCAL_A:%.+]] = alloca i[[SZ]]
 // CHECK:       [[LOCAL_AA:%.+]] = alloca i[[SZ]]
 // CHECK:       [[LOCAL_AAA:%.+]] = alloca i[[SZ]]
-// CHECK:       [[LOCAL_B:%.+]] = alloca [10 x i32]*
-// CHECK-DAG:   store i[[SZ]] [[ARG_A:%.+]], i[[SZ]]* [[LOCAL_A]]
-// CHECK-DAG:   store i[[SZ]] [[ARG_AA:%.+]], i[[SZ]]* [[LOCAL_AA]]
-// CHECK-DAG:   store i[[SZ]] [[ARG_AAA:%.+]], i[[SZ]]* [[LOCAL_AAA]]
-// CHECK-DAG:   store [10 x i32]* [[ARG_B:%.+]], [10 x i32]** [[LOCAL_B]]
+// CHECK:       [[LOCAL_B:%.+]] = alloca ptr
+// CHECK-DAG:   store i[[SZ]] [[ARG_A:%.+]], ptr [[LOCAL_A]]
+// CHECK-DAG:   store i[[SZ]] [[ARG_AA:%.+]], ptr [[LOCAL_AA]]
+// CHECK-DAG:   store i[[SZ]] [[ARG_AAA:%.+]], ptr [[LOCAL_AAA]]
+// CHECK-DAG:   store ptr [[ARG_B:%.+]], ptr [[LOCAL_B]]
 // Store captures in the context.
-// CHECK-64-DAG:   [[REF_A:%.+]] = bitcast i[[SZ]]* [[LOCAL_A]] to i32*
-// CHECK-DAG:      [[REF_AA:%.+]] = bitcast i[[SZ]]* [[LOCAL_AA]] to i16*
-// CHECK-DAG:      [[REF_AAA:%.+]] = bitcast i[[SZ]]* [[LOCAL_AAA]] to i8*
-// CHECK-DAG:      [[REF_B:%.+]] = load [10 x i32]*, [10 x i32]** [[LOCAL_B]],
+// CHECK-DAG:      [[REF_B:%.+]] = load ptr, ptr [[LOCAL_B]],
 // Use captures.
-// CHECK-64-DAG:   load i32, i32* [[REF_A]]
-// CHECK-DAG:      load i16, i16* [[REF_AA]]
-// CHECK-DAG:      load i8, i8* [[REF_AAA]]
-// CHECK-32-DAG:   load i32, i32* [[LOCAL_A]]
-// CHECK-DAG:      getelementptr inbounds [10 x i32], [10 x i32]* [[REF_B]], i[[SZ]] 0, i[[SZ]] 2
+// CHECK-64-DAG:   load i32, ptr [[LOCAL_A]]
+// CHECK-DAG:      load i16, ptr [[LOCAL_AA]]
+// CHECK-DAG:      load i8, ptr [[LOCAL_AAA]]
+// CHECK-32-DAG:   load i32, ptr [[LOCAL_A]]
+// CHECK-DAG:      getelementptr inbounds [10 x i32], ptr [[REF_B]], i[[SZ]] 0, i[[SZ]] 2
 
 // CHECK:       define internal void [[HVT5]]
 // Create local storage for each capture.
 // CHECK:       [[LOCAL_A:%.+]] = alloca i[[SZ]]
 // CHECK:       [[LOCAL_AA:%.+]] = alloca i[[SZ]]
-// CHECK:       [[LOCAL_B:%.+]] = alloca [10 x i32]*
-// CHECK-DAG:   store i[[SZ]] [[ARG_A:%.+]], i[[SZ]]* [[LOCAL_A]]
-// CHECK-DAG:   store i[[SZ]] [[ARG_AA:%.+]], i[[SZ]]* [[LOCAL_AA]]
-// CHECK-DAG:   store [10 x i32]* [[ARG_B:%.+]], [10 x i32]** [[LOCAL_B]]
+// CHECK:       [[LOCAL_B:%.+]] = alloca ptr
+// CHECK-DAG:   store i[[SZ]] [[ARG_A:%.+]], ptr [[LOCAL_A]]
+// CHECK-DAG:   store i[[SZ]] [[ARG_AA:%.+]], ptr [[LOCAL_AA]]
+// CHECK-DAG:   store ptr [[ARG_B:%.+]], ptr [[LOCAL_B]]
 // Store captures in the context.
-// CHECK-64-DAG:[[REF_A:%.+]] = bitcast i[[SZ]]* [[LOCAL_A]] to i32*
-// CHECK-DAG:   [[REF_AA:%.+]] = bitcast i[[SZ]]* [[LOCAL_AA]] to i16*
-// CHECK-DAG:   [[REF_B:%.+]] = load [10 x i32]*, [10 x i32]** [[LOCAL_B]],
+// CHECK-DAG:   [[REF_B:%.+]] = load ptr, ptr [[LOCAL_B]],
 // Use captures.
-// CHECK-64-DAG:   load i32, i32* [[REF_A]]
-// CHECK-32-DAG:   load i32, i32* [[LOCAL_A]]
-// CHECK-DAG:   load i16, i16* [[REF_AA]]
-// CHECK-DAG:   getelementptr inbounds [10 x i32], [10 x i32]* [[REF_B]], i[[SZ]] 0, i[[SZ]] 2
+// CHECK-64-DAG:   load i32, ptr [[LOCAL_A]]
+// CHECK-32-DAG:   load i32, ptr [[LOCAL_A]]
+// CHECK-DAG:   load i16, ptr [[LOCAL_AA]]
+// CHECK-DAG:   getelementptr inbounds [10 x i32], ptr [[REF_B]], i[[SZ]] 0, i[[SZ]] 2
 
 // OMP50: define internal void @__omp_offloading_{{.+}}_{{.+}}bar{{.+}}_l{{[0-9]+}}(i[[SZ]] noundef %{{.+}})
 
 // OMP50: define {{.*}}@{{.*}}zee{{.*}}
 
-// OMP50:       [[LOCAL_THIS:%.+]] = alloca [[S2]]*
-// OMP50:       [[BP:%.+]] = alloca [1 x i8*]
-// OMP50:       [[P:%.+]] = alloca [1 x i8*]
-// OMP50:       [[LOCAL_THIS1:%.+]] = load [[S2]]*, [[S2]]** [[LOCAL_THIS]]
-// OMP50:       [[ARR_IDX:%.+]] = getelementptr inbounds [[S2]], [[S2]]* [[LOCAL_THIS1]], i[[SZ]] 0
-// OMP50:       [[ARR_IDX2:%.+]] = getelementptr inbounds [[S2]], [[S2]]* [[LOCAL_THIS1]], i[[SZ]] 0
-
-// OMP50-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP]], i32 0, i32 0
-// OMP50-DAG:   [[PADDR0:%.+]] =  getelementptr inbounds [1 x i8*], [1 x i8*]* [[P]], i32 0, i32 0
-// OMP50-DAG:   [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to [[S2]]**
-// OMP50-DAG:   [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to [[S2]]**
-// OMP50-DAG:   store [[S2]]* [[ARR_IDX]], [[S2]]** [[CBPADDR0]]
-// OMP50-DAG:   store [[S2]]* [[ARR_IDX2]], [[S2]]** [[CPADDR0]]
-
-// OMP50:       [[BPR:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP]], i32 0, i32 0
-// OMP50:       [[PR:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P]], i32 0, i32 0
-// OMP50:       [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 -1, i32 {{.+}}, i32 {{.+}}, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
+// OMP50:       [[LOCAL_THIS:%.+]] = alloca ptr
+// OMP50:       [[BP:%.+]] = alloca [1 x ptr]
+// OMP50:       [[P:%.+]] = alloca [1 x ptr]
+// OMP50:       [[LOCAL_THIS1:%.+]] = load ptr, ptr [[LOCAL_THIS]]
+// OMP50:       [[ARR_IDX:%.+]] = getelementptr inbounds [[S2]], ptr [[LOCAL_THIS1]], i[[SZ]] 0
+// OMP50:       [[ARR_IDX2:%.+]] = getelementptr inbounds [[S2]], ptr [[LOCAL_THIS1]], i[[SZ]] 0
+
+// OMP50-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[BP]], i32 0, i32 0
+// OMP50-DAG:   [[PADDR0:%.+]] =  getelementptr inbounds [1 x ptr], ptr [[P]], i32 0, i32 0
+// OMP50-DAG:   store ptr [[ARR_IDX]], ptr [[BPADDR0]]
+// OMP50-DAG:   store ptr [[ARR_IDX2]], ptr [[PADDR0]]
+
+// OMP50:       [[BPR:%.+]] = getelementptr inbounds [1 x ptr], ptr [[BP]], i32 0, i32 0
+// OMP50:       [[PR:%.+]] = getelementptr inbounds [1 x ptr], ptr [[P]], i32 0, i32 0
+// OMP50:       [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 -1, i32 {{.+}}, i32 {{.+}}, ptr @.{{.+}}.region_id, ptr [[ARGS:%.+]])
 // OMP50-NEXT:  [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
 // OMP50-NEXT:  br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]
 // OMP50:       [[FAIL]]
-// OMP50:       call void [[HVT0:@.+]]([[S2]]* [[LOCAL_THIS1]])
+// OMP50:       call void [[HVT0:@.+]](ptr [[LOCAL_THIS1]])
 // OMP50-NEXT:  br label %[[END]]
 // OMP50:       [[END]]
 
diff --git a/clang/test/OpenMP/target_data_codegen.cpp b/clang/test/OpenMP/target_data_codegen.cpp
index 3ea267258fed..a07d7faeeb80 100644
--- a/clang/test/OpenMP/target_data_codegen.cpp
+++ b/clang/test/OpenMP/target_data_codegen.cpp
@@ -3,23 +3,23 @@
 #define HEADER
 
 ///==========================================================================///
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
-
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -DCK1 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64
+// RUN: %clang_cc1 -DCK1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-64
+// RUN: %clang_cc1 -DCK1 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
+// RUN: %clang_cc1 -DCK1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
+
+// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -DCK1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -DCK1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
 #ifdef CK1
 
-// CK1: [[ST:%.+]] = type { i32, double* }
+// CK1: [[ST:%.+]] = type { i32, ptr }
 template <typename T>
 struct ST {
   T a;
@@ -50,24 +50,22 @@ void foo(int arg) {
   float lb[arg];
 
   // Region 00
-  // CK1-DAG: call void @__tgt_target_data_begin_mapper(%struct.ident_t* @{{.+}}, i64 [[DEV:%[^,]+]], i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[SIZE00]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE00]]{{.+}}, i8** null, i8** null)
+  // CK1-DAG: call void @__tgt_target_data_begin_mapper(ptr @{{.+}}, i64 [[DEV:%[^,]+]], i32 1, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[SIZE00]], ptr [[MTYPE00]], ptr null, ptr null)
   // CK1-DAG: [[DEV]] = sext i32 [[DEVi32:%[^,]+]] to i64
-  // CK1-DAG: [[DEVi32]] = load i32, i32* %{{[^,]+}},
+  // CK1-DAG: [[DEVi32]] = load i32, ptr %{{[^,]+}},
   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
   // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
 
   // CK1-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
-  // CK1-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [100 x double]**
-  // CK1-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to [100 x double]**
-  // CK1-DAG: store [100 x double]* @gc, [100 x double]** [[CBP0]]
-  // CK1-DAG: store [100 x double]* @gc, [100 x double]** [[CP0]]
+  // CK1-DAG: store ptr @gc, ptr [[BP0]]
+  // CK1-DAG: store ptr @gc, ptr [[P0]]
 
   // CK1: %{{.+}} = add nsw i32 %{{[^,]+}}, 1
 
-  // CK1-DAG: call void @__tgt_target_data_end_mapper(%struct.ident_t* @{{.+}}, i64 [[DEV:%[^,]+]], i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[SIZE00]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE00]]{{.+}}, i8** null, i8** null)
+  // CK1-DAG: call void @__tgt_target_data_end_mapper(ptr @{{.+}}, i64 [[DEV:%[^,]+]], i32 1, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[SIZE00]], ptr [[MTYPE00]], ptr null, ptr null)
   // CK1-DAG: [[DEV]] = sext i32 [[DEVi32:%[^,]+]] to i64
-  // CK1-DAG: [[DEVi32]] = load i32, i32* %{{[^,]+}},
+  // CK1-DAG: [[DEVi32]] = load i32, ptr %{{[^,]+}},
   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP]]
 // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P]]
   #pragma omp target data if(1+3-5) device(arg) map(from: gc)
@@ -81,16 +79,14 @@ void foo(int arg) {
   // Region 02
   // CK1: br i1 %{{[^,]+}}, label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]]
   // CK1: [[IFTHEN]]
-  // CK1-DAG: call void @__tgt_target_data_begin_mapper(%struct.ident_t* @{{.+}}, i64 4, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[SIZE02]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE02]]{{.+}}, i8** null, i8** null)
+  // CK1-DAG: call void @__tgt_target_data_begin_mapper(ptr @{{.+}}, i64 4, i32 1, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[SIZE02]], ptr [[MTYPE02]], ptr null, ptr null)
   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
   // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
 
   // CK1-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
-  // CK1-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to i32**
-  // CK1-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to i32**
-  // CK1-DAG: store i32* [[VAR0:%.+]], i32** [[CBP0]]
-  // CK1-DAG: store i32* [[VAR0]], i32** [[CP0]]
+  // CK1-DAG: store ptr [[VAR0:%.+]], ptr [[BP0]]
+  // CK1-DAG: store ptr [[VAR0]], ptr [[P0]]
   // CK1: br label %[[IFEND:[^,]+]]
 
   // CK1: [[IFELSE]]
@@ -100,7 +96,7 @@ void foo(int arg) {
   // CK1: br i1 %{{[^,]+}}, label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]]
 
   // CK1: [[IFTHEN]]
-  // CK1-DAG: call void @__tgt_target_data_end_mapper(%struct.ident_t* @{{.+}}, i64 4, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[SIZE02]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE02]]{{.+}}, i8** null, i8** null)
+  // CK1-DAG: call void @__tgt_target_data_end_mapper(ptr @{{.+}}, i64 4, i32 1, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[SIZE02]], ptr [[MTYPE02]], ptr null, ptr null)
   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP]]
   // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P]]
   // CK1: br label %[[IFEND:[^,]+]]
@@ -111,7 +107,7 @@ void foo(int arg) {
   {++arg;}
 
   // Region 03
-  // CK1-DAG: call void @__tgt_target_data_begin_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE03]]{{.+}}, i8** null, i8** null)
+  // CK1-DAG: call void @__tgt_target_data_begin_mapper(ptr @{{.+}}, i64 -1, i32 1, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE03]], ptr null, ptr null)
   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
   // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
   // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
@@ -119,17 +115,15 @@ void foo(int arg) {
   // CK1-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0
-  // CK1-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to float**
-  // CK1-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to float**
-  // CK1-DAG: store float* [[VAR0:%.+]], float** [[CBP0]]
-  // CK1-DAG: store float* [[VAR0]], float** [[CP0]]
-  // CK1-DAG: store i[[sz]] [[CSVAL0:%[^,]+]], i[[sz]]* [[S0]]
+  // CK1-DAG: store ptr [[VAR0:%.+]], ptr [[BP0]]
+  // CK1-DAG: store ptr [[VAR0]], ptr [[P0]]
+  // CK1-DAG: store i[[sz]] [[CSVAL0:%[^,]+]], ptr [[S0]]
   // CK1-64-DAG: [[CSVAL0]] = mul nuw i64 %{{[^,]+}}, 4
   // CK1-32-DAG: [[CSVAL0]] = sext i32 [[CSVAL032:%.+]] to i64
   // CK1-32-DAG: [[CSVAL032]] = mul nuw i32 %{{[^,]+}}, 4
   // CK1: %{{.+}} = add nsw i32 %{{[^,]+}}, 1
 
-  // CK1-DAG: call void @__tgt_target_data_end_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE03]]{{.+}}, i8** null, i8** null)
+  // CK1-DAG: call void @__tgt_target_data_end_mapper(ptr @{{.+}}, i64 -1, i32 1, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE03]], ptr null, ptr null)
   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP]]
   // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P]]
   // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S]]
@@ -140,7 +134,7 @@ void foo(int arg) {
   {++arg;}
 
   // Region 04
-  // CK1-DAG: call void @__tgt_target_data_begin_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%[^,]+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE04]]{{.+}}, i8** null, i8** null)
+  // CK1-DAG: call void @__tgt_target_data_begin_mapper(ptr @{{.+}}, i64 -1, i32 2, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%[^,]+]], ptr [[MTYPE04]], ptr null, ptr null)
   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
   // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
   // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[PSZ:%[^,]+]]
@@ -148,25 +142,21 @@ void foo(int arg) {
   // CK1-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[PS0:%.+]] = getelementptr inbounds {{.+}}[[PSZ]], i{{.+}} 0, i{{.+}} 0
-  // CK1-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[ST]]**
-  // CK1-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to double***
-  // CK1-DAG: store [[ST]]* @gb, [[ST]]** [[CBP0]]
-  // CK1-DAG: store double** getelementptr inbounds ([[ST]], [[ST]]* @gb, i32 0, i32 1), double*** [[CP0]]
-  // CK1-DAG: [[DIV:%.+]] = sdiv exact i64 sub (i64 ptrtoint (double** getelementptr (double*, double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1) to i64)), ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)
-  // CK1-DAG: store i64 [[DIV]], i64* [[PS0]],
+  // CK1-DAG: store ptr @gb, ptr [[BP0]]
+  // CK1-DAG: store ptr getelementptr inbounds ([[ST]], ptr @gb, i32 0, i32 1), ptr [[P0]]
+  // CK1-DAG: [[DIV:%.+]] = sdiv exact i64 sub (i64 ptrtoint (ptr getelementptr (ptr, ptr getelementptr inbounds (%struct.ST, ptr @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (ptr getelementptr inbounds (%struct.ST, ptr @gb, i32 0, i32 1) to i64)), ptrtoint (ptr getelementptr (i8, ptr null, i32 1) to i64)
+  // CK1-DAG: store i64 [[DIV]], ptr [[PS0]],
 
   // CK1-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1
   // CK1-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1
-  // CK1-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to double***
-  // CK1-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to double**
-  // CK1-DAG: store double** getelementptr inbounds ([[ST]], [[ST]]* @gb, i32 0, i32 1), double*** [[CBP1]]
-  // CK1-DAG: store double* [[SEC1:%.+]], double** [[CP1]]
-  // CK1-DAG: [[SEC1]] = getelementptr inbounds {{.+}}double* [[SEC11:%[^,]+]], i{{.+}} 0
-  // CK1-DAG: [[SEC11]] = load double*, double** getelementptr inbounds ([[ST]], [[ST]]* @gb, i32 0, i32 1),
+  // CK1-DAG: store ptr getelementptr inbounds ([[ST]], ptr @gb, i32 0, i32 1), ptr [[BP1]]
+  // CK1-DAG: store ptr [[SEC1:%.+]], ptr [[P1]]
+  // CK1-DAG: [[SEC1]] = getelementptr inbounds {{.+}}ptr [[SEC11:%[^,]+]], i{{.+}} 0
+  // CK1-DAG: [[SEC11]] = load ptr, ptr getelementptr inbounds ([[ST]], ptr @gb, i32 0, i32 1),
 
   // CK1: %{{.+}} = add nsw i32 %{{[^,]+}}, 1
 
-  // CK1-DAG: call void @__tgt_target_data_end_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%[^,]+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE04]]{{.+}}, i8** null, i8** null)
+  // CK1-DAG: call void @__tgt_target_data_end_mapper(ptr @{{.+}}, i64 -1, i32 2, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%[^,]+]], ptr [[MTYPE04]], ptr null, ptr null)
   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP]]
   // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P]]
   // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[PSZ]]
@@ -177,7 +167,7 @@ void foo(int arg) {
   {++arg;}
 
   // Region 05
-  // CK1-DAG: call void @__tgt_target_data_begin_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE05]]{{.+}}, i8** null, i8** null)
+  // CK1-DAG: call void @__tgt_target_data_begin_mapper(ptr @{{.+}}, i64 -1, i32 1, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE05]], ptr null, ptr null)
   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
   // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
   // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
@@ -185,17 +175,15 @@ void foo(int arg) {
   // CK1-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0
-  // CK1-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to float**
-  // CK1-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to float**
-  // CK1-DAG: store float* [[VAR0:%.+]], float** [[CBP0]]
-  // CK1-DAG: store float* [[VAR0]], float** [[CP0]]
-  // CK1-DAG: store i[[sz]] [[CSVAL0:%[^,]+]], i[[sz]]* [[S0]]
+  // CK1-DAG: store ptr [[VAR0:%.+]], ptr [[BP0]]
+  // CK1-DAG: store ptr [[VAR0]], ptr [[P0]]
+  // CK1-DAG: store i[[sz]] [[CSVAL0:%[^,]+]], ptr [[S0]]
   // CK1-64-DAG: [[CSVAL0]] = mul nuw i64 %{{[^,]+}}, 4
   // CK1-32-DAG: [[CSVAL0]] = sext i32 [[CSVAL032:%.+]] to i64
   // CK1-32-DAG: [[CSVAL032]] = mul nuw i32 %{{[^,]+}}, 4
   // CK1: %{{.+}} = add nsw i32 %{{[^,]+}}, 1
 
-  // CK1-DAG: call void @__tgt_target_data_end_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE05]]{{.+}}, i8** null, i8** null)
+  // CK1-DAG: call void @__tgt_target_data_end_mapper(ptr @{{.+}}, i64 -1, i32 1, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE05]], ptr null, ptr null)
   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP]]
   // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P]]
   // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S]]
@@ -206,7 +194,7 @@ void foo(int arg) {
   {++arg;}
 
   // Region 06
-  // CK1-DAG: call void @__tgt_target_data_begin_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE06]]{{.+}}, i8** null, i8** null)
+  // CK1-DAG: call void @__tgt_target_data_begin_mapper(ptr @{{.+}}, i64 -1, i32 1, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE06]], ptr null, ptr null)
   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
   // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
   // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
@@ -214,17 +202,15 @@ void foo(int arg) {
   // CK1-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0
-  // CK1-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to float**
-  // CK1-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to float**
-  // CK1-DAG: store float* [[VAR0:%.+]], float** [[CBP0]]
-  // CK1-DAG: store float* [[VAR0]], float** [[CP0]]
-  // CK1-DAG: store i[[sz]] [[CSVAL0:%[^,]+]], i[[sz]]* [[S0]]
+  // CK1-DAG: store ptr [[VAR0:%.+]], ptr [[BP0]]
+  // CK1-DAG: store ptr [[VAR0]], ptr [[P0]]
+  // CK1-DAG: store i[[sz]] [[CSVAL0:%[^,]+]], ptr [[S0]]
   // CK1-64-DAG: [[CSVAL0]] = mul nuw i64 %{{[^,]+}}, 4
   // CK1-32-DAG: [[CSVAL0]] = sext i32 [[CSVAL032:%.+]] to i64
   // CK1-32-DAG: [[CSVAL032]] = mul nuw i32 %{{[^,]+}}, 4
   // CK1: %{{.+}} = add nsw i32 %{{[^,]+}}, 1
 
-  // CK1-DAG: call void @__tgt_target_data_end_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE06]]{{.+}}, i8** null, i8** null)
+  // CK1-DAG: call void @__tgt_target_data_end_mapper(ptr @{{.+}}, i64 -1, i32 1, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE06]], ptr null, ptr null)
   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP]]
   // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P]]
   // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S]]
@@ -234,23 +220,23 @@ void foo(int arg) {
 }
 #endif
 ///==========================================================================///
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1A -verify -fopenmp -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK1A --check-prefix CK1A-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1A -fopenmp -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1A --check-prefix CK1A-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1A -verify -fopenmp -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK1A --check-prefix CK1A-32
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1A -fopenmp -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1A --check-prefix CK1A-32
-
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1A -verify -fopenmp-simd -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1A -fopenmp-simd -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1A -verify -fopenmp-simd -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1A -fopenmp-simd -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -DCK1A -verify -fopenmp -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK1A --check-prefix CK1A-64
+// RUN: %clang_cc1 -DCK1A -fopenmp -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1A --check-prefix CK1A-64
+// RUN: %clang_cc1 -DCK1A -verify -fopenmp -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK1A --check-prefix CK1A-32
+// RUN: %clang_cc1 -DCK1A -fopenmp -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1A --check-prefix CK1A-32
+
+// RUN: %clang_cc1 -DCK1A -verify -fopenmp-simd -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -DCK1A -fopenmp-simd -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -DCK1A -verify -fopenmp-simd -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -DCK1A -fopenmp-simd -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
 #ifdef CK1A
 
-// CK1A: [[ST:%.+]] = type { i32, double* }
+// CK1A: [[ST:%.+]] = type { i32, ptr }
 template <typename T>
 struct ST {
   T a;
@@ -278,7 +264,7 @@ void foo(int arg) {
   float lb[arg];
 
   // Region 00
-  // CK1A-DAG: call void @__tgt_target_data_begin_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz:32|64]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE00Begin]]{{.+}})
+  // CK1A-DAG: call void @__tgt_target_data_begin_mapper(ptr @{{.+}}, i64 -1, i32 1, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE00Begin]]{{.+}})
   // CK1A-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
   // CK1A-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
   // CK1A-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
@@ -286,17 +272,15 @@ void foo(int arg) {
   // CK1A-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
   // CK1A-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
   // CK1A-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0
-  // CK1A-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to float**
-  // CK1A-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to float**
-  // CK1A-DAG: store float* [[VAR0:%.+]], float** [[CBP0]]
-  // CK1A-DAG: store float* [[VAR0]], float** [[CP0]]
-  // CK1A-DAG: store i[[sz]] [[CSVAL0:%[^,]+]], i[[sz]]* [[S0]]
+  // CK1A-DAG: store ptr [[VAR0:%.+]], ptr [[BP0]]
+  // CK1A-DAG: store ptr [[VAR0]], ptr [[P0]]
+  // CK1A-DAG: store i[[sz:32|64]] [[CSVAL0:%[^,]+]], ptr [[S0]]
   // CK1A-64-DAG: [[CSVAL0]] = mul nuw i64 %{{[^,]+}}, 4
   // CK1A-32-DAG: [[CSVAL0]] = sext i32 [[CSVAL032:%.+]] to i64
   // CK1A-32-DAG: [[CSVAL032]] = mul nuw i32 %{{[^,]+}}, 4
   // CK1A: %{{.+}} = add nsw i32 %{{[^,]+}}, 1
 
-  // CK1A-DAG: call void @__tgt_target_data_end_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE00End]]{{.+}})
+  // CK1A-DAG: call void @__tgt_target_data_end_mapper(ptr @{{.+}}, i64 -1, i32 1, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE00End]]{{.+}})
   // CK1A-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP]]
   // CK1A-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P]]
   // CK1A-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S]]
@@ -304,7 +288,7 @@ void foo(int arg) {
   {++arg;}
 
   // Region 01
-  // CK1A-DAG: call void @__tgt_target_data_begin_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE01Begin]]{{.+}})
+  // CK1A-DAG: call void @__tgt_target_data_begin_mapper(ptr @{{.+}}, i64 -1, i32 1, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE01Begin]]{{.+}})
   // CK1A-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
   // CK1A-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
   // CK1A-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
@@ -312,17 +296,15 @@ void foo(int arg) {
   // CK1A-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
   // CK1A-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
   // CK1A-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0
-  // CK1A-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to float**
-  // CK1A-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to float**
-  // CK1A-DAG: store float* [[VAR0:%.+]], float** [[CBP0]]
-  // CK1A-DAG: store float* [[VAR0]], float** [[CP0]]
-  // CK1A-DAG: store i[[sz]] [[CSVAL0:%[^,]+]], i[[sz]]* [[S0]]
+  // CK1A-DAG: store ptr [[VAR0:%.+]], ptr [[BP0]]
+  // CK1A-DAG: store ptr [[VAR0]], ptr [[P0]]
+  // CK1A-DAG: store i[[sz]] [[CSVAL0:%[^,]+]], ptr [[S0]]
   // CK1A-64-DAG: [[CSVAL0]] = mul nuw i64 %{{[^,]+}}, 4
   // CK1A-32-DAG: [[CSVAL0]] = sext i32 [[CSVAL032:%.+]] to i64
   // CK1A-32-DAG: [[CSVAL032]] = mul nuw i32 %{{[^,]+}}, 4
   // CK1A: %{{.+}} = add nsw i32 %{{[^,]+}}, 1
 
-  // CK1A-DAG: call void @__tgt_target_data_end_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE01End]]{{.+}})
+  // CK1A-DAG: call void @__tgt_target_data_end_mapper(ptr @{{.+}}, i64 -1, i32 1, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE01End]]{{.+}})
   // CK1A-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP]]
   // CK1A-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P]]
   // CK1A-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S]]
@@ -332,23 +314,23 @@ void foo(int arg) {
 }
 #endif
 ///==========================================================================///
-// RUN: %clang_cc1 -no-opaque-pointers -DCK2 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK2 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK2 --check-prefix CK2-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK2 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK2 --check-prefix CK2-32
-// RUN: %clang_cc1 -no-opaque-pointers -DCK2 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK2 --check-prefix CK2-32
-
-// RUN: %clang_cc1 -no-opaque-pointers -DCK2 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK2 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK2 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK2 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -DCK2 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-64
+// RUN: %clang_cc1 -DCK2 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK2 --check-prefix CK2-64
+// RUN: %clang_cc1 -DCK2 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK2 --check-prefix CK2-32
+// RUN: %clang_cc1 -DCK2 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK2 --check-prefix CK2-32
+
+// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -DCK2 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -DCK2 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
 // SIMD-ONLY1-NOT: {{__kmpc|__tgt}}
 #ifdef CK2
 
-// CK2: [[ST:%.+]] = type { i32, double* }
+// CK2: [[ST:%.+]] = type { i32, ptr }
 template <typename T>
 struct ST {
   T a;
@@ -374,32 +356,28 @@ int bar(int arg){
 // Region 00
 // CK2: br i1 %{{[^,]+}}, label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]]
 // CK2: [[IFTHEN]]
-// CK2-DAG: call void @__tgt_target_data_begin_mapper(%struct.ident_t* @{{.+}}, i64 [[DEV:%[^,]+]], i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%[^,]+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE00]]{{.+}}, i8** null, i8** null)
+// CK2-DAG: call void @__tgt_target_data_begin_mapper(ptr @{{.+}}, i64 [[DEV:%[^,]+]], i32 2, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%[^,]+]], ptr [[MTYPE00]], ptr null, ptr null)
 // CK2-DAG: [[DEV]] = sext i32 [[DEVi32:%[^,]+]] to i64
-// CK2-DAG: [[DEVi32]] = load i32, i32* %{{[^,]+}},
-// CK2-DAG: [[GEPBP]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP:%[^,]+]]
-// CK2-DAG: [[GEPP]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P:%[^,]+]]
-// CK2-DAG: [[GEPS]] = getelementptr inbounds [2 x i64], [2 x i64]* [[PS:%[^,]+]]
+// CK2-DAG: [[DEVi32]] = load i32, ptr %{{[^,]+}},
+// CK2-DAG: [[GEPBP]] = getelementptr inbounds [2 x ptr], ptr [[BP:%[^,]+]]
+// CK2-DAG: [[GEPP]] = getelementptr inbounds [2 x ptr], ptr [[P:%[^,]+]]
+// CK2-DAG: [[GEPS]] = getelementptr inbounds [2 x i64], ptr [[PS:%[^,]+]]
 
 // CK2-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
 // CK2-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
-// CK2-DAG: [[PS0:%.+]] = getelementptr inbounds [2 x i64], [2 x i64]* [[PS]], i{{.+}} 0, i{{.+}} 0
-// CK2-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[ST]]**
-// CK2-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to double***
-// CK2-DAG: store [[ST]]* [[VAR0:%.+]], [[ST]]** [[CBP0]]
-// CK2-DAG: store double** [[SEC0:%.+]], double*** [[CP0]]
-// CK2-DAG: store i64 {{%.+}}, i64* [[PS0]],
-// CK2-DAG: [[SEC0]] = getelementptr inbounds {{.*}}[[ST]]* [[VAR0]], i32 0, i32 1
+// CK2-DAG: [[PS0:%.+]] = getelementptr inbounds [2 x i64], ptr [[PS]], i{{.+}} 0, i{{.+}} 0
+// CK2-DAG: store ptr [[VAR0:%.+]], ptr [[BP0]]
+// CK2-DAG: store ptr [[SEC0:%.+]], ptr [[P0]]
+// CK2-DAG: store i64 {{%.+}}, ptr [[PS0]],
+// CK2-DAG: [[SEC0]] = getelementptr inbounds {{.*}}ptr [[VAR0]], i32 0, i32 1
 
 // CK2-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1
 // CK2-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1
-// CK2-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to double***
-// CK2-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to double**
-// CK2-DAG: store double** [[SEC0]], double*** [[CBP1]]
-// CK2-DAG: store double* [[SEC1:%.+]], double** [[CP1]]
-// CK2-DAG: [[SEC1]] = getelementptr inbounds {{.*}}double* [[SEC11:%[^,]+]], i{{.+}} 1
-// CK2-DAG: [[SEC11]] = load double*, double** [[SEC111:%[^,]+]],
-// CK2-DAG: [[SEC111]] = getelementptr inbounds {{.*}}[[ST]]* [[VAR0]], i32 0, i32 1
+// CK2-DAG: store ptr [[SEC0]], ptr [[BP1]]
+// CK2-DAG: store ptr [[SEC1:%.+]], ptr [[P1]]
+// CK2-DAG: [[SEC1]] = getelementptr inbounds {{.*}}ptr [[SEC11:%[^,]+]], i{{.+}} 1
+// CK2-DAG: [[SEC11]] = load ptr, ptr [[SEC111:%[^,]+]],
+// CK2-DAG: [[SEC111]] = getelementptr inbounds {{.*}}ptr [[VAR0]], i32 0, i32 1
 
 // CK2: br label %[[IFEND:[^,]+]]
 
@@ -410,9 +388,9 @@ int bar(int arg){
 // CK2: br i1 %{{[^,]+}}, label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]]
 
 // CK2: [[IFTHEN]]
-// CK2-DAG: call void @__tgt_target_data_end_mapper(%struct.ident_t* @{{.+}}, i64 [[DEV:%[^,]+]], i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%[^,]+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE00]]{{.+}}, i8** null, i8** null)
+// CK2-DAG: call void @__tgt_target_data_end_mapper(ptr @{{.+}}, i64 [[DEV:%[^,]+]], i32 2, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%[^,]+]], ptr [[MTYPE00]], ptr null, ptr null)
 // CK2-DAG: [[DEV]] = sext i32 [[DEVi32:%[^,]+]] to i64
-// CK2-DAG: [[DEVi32]] = load i32, i32* %{{[^,]+}},
+// CK2-DAG: [[DEVi32]] = load i32, ptr %{{[^,]+}},
 // CK2-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP]]
 // CK2-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P]]
 // CK2-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[PS]]
@@ -422,19 +400,19 @@ int bar(int arg){
 // CK2: [[IFEND]]
 #endif
 ///==========================================================================///
-// RUN: %clang_cc1 -no-opaque-pointers -DCK3 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK3 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK3 --check-prefix CK3-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK3 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK3 --check-prefix CK3-32
-// RUN: %clang_cc1 -no-opaque-pointers -DCK3 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK3 --check-prefix CK3-32
-
-// RUN: %clang_cc1 -no-opaque-pointers -DCK3 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK3 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK3 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-64
+// RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK3 --check-prefix CK3-64
+// RUN: %clang_cc1 -DCK3 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK3 --check-prefix CK3-32
+// RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK3 --check-prefix CK3-32
+
+// RUN: %clang_cc1 -DCK3 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK3 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
 // SIMD-ONLY2-NOT: {{__kmpc|__tgt}}
 #ifdef CK3
 
@@ -449,23 +427,23 @@ void no_target_devices(int arg) {
 }
 #endif
 ///==========================================================================///
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK4 --check-prefix CK4-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK4 --check-prefix CK4-32
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK4 --check-prefix CK4-32
-
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -DCK4 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-64
+// RUN: %clang_cc1 -DCK4 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK4 --check-prefix CK4-64
+// RUN: %clang_cc1 -DCK4 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK4 --check-prefix CK4-32
+// RUN: %clang_cc1 -DCK4 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK4 --check-prefix CK4-32
+
+// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -DCK4 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -DCK4 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
 // SIMD-ONLY1-NOT: {{__kmpc|__tgt}}
 #ifdef CK4
 
-// CK4: [[STT:%.+]] = type { i32, double* }
+// CK4: [[STT:%.+]] = type { i32, ptr }
 template <typename T>
 struct STT {
   T a;
@@ -491,32 +469,28 @@ int bar(int arg){
 // Region 00
 // CK4: br i1 %{{[^,]+}}, label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]]
 // CK4: [[IFTHEN]]
-// CK4-DAG: call void @__tgt_target_data_begin_mapper(%struct.ident_t* @{{.+}}, i64 [[DEV:%[^,]+]], i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE00]]{{.+}}, i8** null, i8** null)
+// CK4-DAG: call void @__tgt_target_data_begin_mapper(ptr @{{.+}}, i64 [[DEV:%[^,]+]], i32 2, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE00]], ptr null, ptr null)
 // CK4-DAG: [[DEV]] = sext i32 [[DEVi32:%[^,]+]] to i64
-// CK4-DAG: [[DEVi32]] = load i32, i32* %{{[^,]+}},
-// CK4-DAG: [[GEPBP]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP:%[^,]+]]
-// CK4-DAG: [[GEPP]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P:%[^,]+]]
-// CK4-DAG: [[GEPS]] = getelementptr inbounds [2 x i64], [2 x i64]* [[PS:%[^,]+]]
+// CK4-DAG: [[DEVi32]] = load i32, ptr %{{[^,]+}},
+// CK4-DAG: [[GEPBP]] = getelementptr inbounds [2 x ptr], ptr [[BP:%[^,]+]]
+// CK4-DAG: [[GEPP]] = getelementptr inbounds [2 x ptr], ptr [[P:%[^,]+]]
+// CK4-DAG: [[GEPS]] = getelementptr inbounds [2 x i64], ptr [[PS:%[^,]+]]
 
 // CK4-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
 // CK4-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
 // CK4-DAG: [[PS0:%.+]] = getelementptr inbounds {{.+}}[[PS]], i{{.+}} 0, i{{.+}} 0
-// CK4-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[STT]]**
-// CK4-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to double***
-// CK4-DAG: store [[STT]]* [[VAR0:%.+]], [[STT]]** [[CBP0]]
-// CK4-DAG: store double** [[SEC0:%.+]], double*** [[CP0]]
-// CK4-DAG: store i64 {{%.+}}, i64* [[PS0]],
-// CK4-DAG: [[SEC0]] = getelementptr inbounds {{.*}}[[STT]]* [[VAR0]], i32 0, i32 1
+// CK4-DAG: store ptr [[VAR0:%.+]], ptr [[BP0]]
+// CK4-DAG: store ptr [[SEC0:%.+]], ptr [[P0]]
+// CK4-DAG: store i64 {{%.+}}, ptr [[PS0]],
+// CK4-DAG: [[SEC0]] = getelementptr inbounds {{.*}}ptr [[VAR0]], i32 0, i32 1
 
 // CK4-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1
 // CK4-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1
-// CK4-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to double***
-// CK4-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to double**
-// CK4-DAG: store double** [[SEC0]], double*** [[CBP1]]
-// CK4-DAG: store double* [[SEC1:%.+]], double** [[CP1]]
-// CK4-DAG: [[SEC1]] = getelementptr inbounds {{.*}}double* [[SEC11:%[^,]+]], i{{.+}} 1
-// CK4-DAG: [[SEC11]] = load double*, double** [[SEC111:%[^,]+]],
-// CK4-DAG: [[SEC111]] = getelementptr inbounds {{.*}}[[STT]]* [[VAR0]], i32 0, i32 1
+// CK4-DAG: store ptr [[SEC0]], ptr [[BP1]]
+// CK4-DAG: store ptr [[SEC1:%.+]], ptr [[P1]]
+// CK4-DAG: [[SEC1]] = getelementptr inbounds {{.*}}ptr [[SEC11:%[^,]+]], i{{.+}} 1
+// CK4-DAG: [[SEC11]] = load ptr, ptr [[SEC111:%[^,]+]],
+// CK4-DAG: [[SEC111]] = getelementptr inbounds {{.*}}ptr [[VAR0]], i32 0, i32 1
 
 // CK4: br label %[[IFEND:[^,]+]]
 
@@ -527,9 +501,9 @@ int bar(int arg){
 // CK4: br i1 %{{[^,]+}}, label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]]
 
 // CK4: [[IFTHEN]]
-// CK4-DAG: call void @__tgt_target_data_end_mapper(%struct.ident_t* @{{.+}}, i64 [[DEV:%[^,]+]], i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE00]]{{.+}}, i8** null, i8** null)
+// CK4-DAG: call void @__tgt_target_data_end_mapper(ptr @{{.+}}, i64 [[DEV:%[^,]+]], i32 2, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE00]], ptr null, ptr null)
 // CK4-DAG: [[DEV]] = sext i32 [[DEVi32:%[^,]+]] to i64
-// CK4-DAG: [[DEVi32]] = load i32, i32* %{{[^,]+}},
+// CK4-DAG: [[DEVi32]] = load i32, ptr %{{[^,]+}},
 // CK4-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP]]
 // CK4-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P]]
 // CK4-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[PS]]
@@ -539,19 +513,19 @@ int bar(int arg){
 // CK4: [[IFEND]]
 #endif
 ///==========================================================================///
-// RUN: %clang_cc1 -no-opaque-pointers -DCK5 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK5 --check-prefix CK5-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK5 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK5 --check-prefix CK5-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK5 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK5 --check-prefix CK5-32
-// RUN: %clang_cc1 -no-opaque-pointers -DCK5 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK5 --check-prefix CK5-32
-
-// RUN: %clang_cc1 -no-opaque-pointers -DCK5 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK5 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK5 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK5 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK5 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK5 --check-prefix CK5-64
+// RUN: %clang_cc1 -DCK5 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK5 --check-prefix CK5-64
+// RUN: %clang_cc1 -DCK5 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK5 --check-prefix CK5-32
+// RUN: %clang_cc1 -DCK5 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK5 --check-prefix CK5-32
+
+// RUN: %clang_cc1 -DCK5 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK5 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK5 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK5 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
 // SIMD-ONLY2-NOT: {{__kmpc|__tgt}}#ifdef CK5
 #ifdef CK5
 struct S1 {
@@ -573,19 +547,19 @@ void test_close_modifier(int arg) {
 }
 #endif
 ///==========================================================================///
-// RUN: %clang_cc1 -no-opaque-pointers -DCK6 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK6 --check-prefix CK6-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK6 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK6 --check-prefix CK6-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK6 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK6 --check-prefix CK6-32
-// RUN: %clang_cc1 -no-opaque-pointers -DCK6 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK6 --check-prefix CK6-32
-
-// RUN: %clang_cc1 -no-opaque-pointers -DCK6 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK6 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK6 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK6 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK6 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK6 --check-prefix CK6-64
+// RUN: %clang_cc1 -DCK6 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK6 --check-prefix CK6-64
+// RUN: %clang_cc1 -DCK6 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK6 --check-prefix CK6-32
+// RUN: %clang_cc1 -DCK6 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK6 --check-prefix CK6-32
+
+// RUN: %clang_cc1 -DCK6 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK6 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK6 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK6 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
 // SIMD-ONLY2-NOT: {{__kmpc|__tgt}}
 #ifdef CK6
 void test_close_modifier(int arg) {
@@ -596,13 +570,13 @@ void test_close_modifier(int arg) {
 }
 #endif
 ///==========================================================================///
-// RUN: %clang_cc1 -no-opaque-pointers -DCK7 -verify -fopenmp -fopenmp-targets=x86_64 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK7 --check-prefix CK7-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK7 -fopenmp -fopenmp-targets=x86_64 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=x86_64 -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK7 --check-prefix CK7-64
+// RUN: %clang_cc1 -DCK7 -verify -fopenmp -fopenmp-targets=x86_64 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK7 --check-prefix CK7-64
+// RUN: %clang_cc1 -DCK7 -fopenmp -fopenmp-targets=x86_64 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=x86_64 -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK7 --check-prefix CK7-64
 
-// RUN: %clang_cc1 -no-opaque-pointers -DCK7 -verify -fopenmp-simd -fopenmp-targets=x86_64 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY7 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK7 -fopenmp-simd -fopenmp-targets=x86_64 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=x86_64 -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY7 %s
+// RUN: %clang_cc1 -DCK7 -verify -fopenmp-simd -fopenmp-targets=x86_64 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY7 %s
+// RUN: %clang_cc1 -DCK7 -fopenmp-simd -fopenmp-targets=x86_64 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=x86_64 -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY7 %s
 // SIMD-ONLY7-NOT: {{__kmpc|__tgt}}
 #ifdef CK7
 // CK7: private unnamed_addr constant [2 x i64] [i64 64, i64 64]
@@ -628,19 +602,19 @@ void test_device_ptr_addr(int arg) {
 }
 #endif
 ///==========================================================================///
-// RUN: %clang_cc1 -no-opaque-pointers -DCK8 -verify -fopenmp -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK8 --check-prefix CK8-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK8 -fopenmp -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK8 --check-prefix CK8-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK8 -verify -fopenmp -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK8 --check-prefix CK8-32
-// RUN: %clang_cc1 -no-opaque-pointers -DCK8 -fopenmp -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK8 --check-prefix CK8-32
-
-// RUN: %clang_cc1 -no-opaque-pointers -DCK8 -verify -fopenmp-simd -fopenmp-version=51 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK8 -fopenmp-simd -fopenmp-version=51 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=51 -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK8 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK8 -fopenmp-simd -fopenmp-version=51 -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=51 -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK8 -verify -fopenmp -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK8 --check-prefix CK8-64
+// RUN: %clang_cc1 -DCK8 -fopenmp -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK8 --check-prefix CK8-64
+// RUN: %clang_cc1 -DCK8 -verify -fopenmp -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK8 --check-prefix CK8-32
+// RUN: %clang_cc1 -DCK8 -fopenmp -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK8 --check-prefix CK8-32
+
+// RUN: %clang_cc1 -DCK8 -verify -fopenmp-simd -fopenmp-version=51 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK8 -fopenmp-simd -fopenmp-version=51 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=51 -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK8 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK8 -fopenmp-simd -fopenmp-version=51 -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=51 -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
 // SIMD-ONLY2-NOT: {{__kmpc|__tgt}}#ifdef CK8
 #ifdef CK8
 struct S1 {
@@ -699,19 +673,19 @@ void test_present_modifier(int arg) {
 }
 #endif
 ///==========================================================================///
-// RUN: %clang_cc1 -no-opaque-pointers -DCK9 -verify -fopenmp -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK9 --check-prefix CK9-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK9 -fopenmp -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK9 --check-prefix CK9-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK9 -verify -fopenmp -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK9 --check-prefix CK9-32
-// RUN: %clang_cc1 -no-opaque-pointers -DCK9 -fopenmp -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK9 --check-prefix CK9-32
-
-// RUN: %clang_cc1 -no-opaque-pointers -DCK9 -verify -fopenmp-simd -fopenmp-version=51 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK9 -fopenmp-simd -fopenmp-version=51 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=51 -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK9 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK9 -fopenmp-simd -fopenmp-version=51 -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=51 -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK9 -verify -fopenmp -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK9 --check-prefix CK9-64
+// RUN: %clang_cc1 -DCK9 -fopenmp -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK9 --check-prefix CK9-64
+// RUN: %clang_cc1 -DCK9 -verify -fopenmp -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK9 --check-prefix CK9-32
+// RUN: %clang_cc1 -DCK9 -fopenmp -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK9 --check-prefix CK9-32
+
+// RUN: %clang_cc1 -DCK9 -verify -fopenmp-simd -fopenmp-version=51 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK9 -fopenmp-simd -fopenmp-version=51 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=51 -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK9 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK9 -fopenmp-simd -fopenmp-version=51 -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=51 -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
 // SIMD-ONLY2-NOT: {{__kmpc|__tgt}}
 #ifdef CK9
 void test_present_modifier(int arg) {
diff --git a/clang/test/OpenMP/target_enter_data_codegen.cpp b/clang/test/OpenMP/target_enter_data_codegen.cpp
index 81455cd02e58..7ae7d3d5794e 100644
--- a/clang/test/OpenMP/target_enter_data_codegen.cpp
+++ b/clang/test/OpenMP/target_enter_data_codegen.cpp
@@ -3,23 +3,23 @@
 #define HEADER
 
 ///==========================================================================///
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
-
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -DCK1 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64
+// RUN: %clang_cc1 -DCK1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-64
+// RUN: %clang_cc1 -DCK1 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
+// RUN: %clang_cc1 -DCK1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
+
+// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -DCK1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -DCK1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
 #ifdef CK1
 
-// CK1: [[ST:%.+]] = type { i32, double* }
+// CK1: [[ST:%.+]] = type { i32, ptr }
 template <typename T>
 struct ST {
   T a;
@@ -29,12 +29,12 @@ struct ST {
 ST<int> gb;
 double gc[100];
 
-// CK1: [[IDENT_T:%.+]] = type { i32, i32, i32, i32, i8* }
+// CK1: [[IDENT_T:%.+]] = type { i32, i32, i32, i32, ptr }
 
 // CK1: [[KMP_TASK_T_WITH_PRIVATES:%.+]] = type { [[TASK_T:%[^,]+]], [[KMP_PRIVATES_T:%.+]] }
-// CK1: [[TASK_T]] = type { i8*, i32 (i32, i8*)*, i32, %{{[^,]+}}, %{{[^,]+}} }
-// CK1-32: [[KMP_PRIVATES_T]] = type { [1 x i64], [1 x i8*], [1 x i8*] }
-// CK1-64: [[KMP_PRIVATES_T]] = type { [1 x i8*], [1 x i8*], [1 x i64] }
+// CK1: [[TASK_T]] = type { ptr, ptr, i32, %{{[^,]+}}, %{{[^,]+}} }
+// CK1-32: [[KMP_PRIVATES_T]] = type { [1 x i64], [1 x ptr], [1 x ptr] }
+// CK1-64: [[KMP_PRIVATES_T]] = type { [1 x ptr], [1 x ptr], [1 x i64] }
 
 // CK1: [[SIZE00:@.+]] = {{.+}}constant [1 x i64] [i64 800]
 // CK1: [[MTYPE00:@.+]] = {{.+}}constant [1 x i64] zeroinitializer
@@ -57,34 +57,26 @@ void foo(int arg) {
   float lb[arg];
 
   // Region 00
-  // CK1-DAG:     call i32 @__kmpc_omp_task([[IDENT_T]]* @{{[^,]+}}, i32 %{{[^,]+}}, i8* [[TASK:%.+]])
-  // CK1-DAG:     [[TASK]] = call i8* @__kmpc_omp_target_task_alloc([[IDENT_T]]* @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i{{32|64}} {{36|64}}, i{{32|64}} 4, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_T_WITH_PRIVATES]]*)* [[OMP_TASK_ENTRY:@[^,]+]] to i32 (i32, i8*)*), i64 [[DEV:%.+]])
+  // CK1-DAG:     call i32 @__kmpc_omp_task(ptr @{{[^,]+}}, i32 %{{[^,]+}}, ptr [[TASK:%.+]])
+  // CK1-DAG:     [[TASK]] = call ptr @__kmpc_omp_target_task_alloc(ptr @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i{{32|64}} {{36|64}}, i{{32|64}} 4, ptr [[OMP_TASK_ENTRY:@[^,]+]], i64 [[DEV:%.+]])
   // CK1-DAG:     [[DEV]] = sext i32 [[DEVi32:%[^,]+]] to i64
-  // CK1-DAG:     [[DEVi32]] = load i32, i32* %{{[^,]+}},
-  // CK1-DAG:     [[TASK_CAST:%.+]] = bitcast i8* [[TASK]] to [[KMP_TASK_T_WITH_PRIVATES]]*
-  // CK1-DAG:     [[TASK_WITH_PRIVATES:%.+]] = getelementptr inbounds [[KMP_TASK_T_WITH_PRIVATES]], [[KMP_TASK_T_WITH_PRIVATES]]* [[TASK_CAST]], i32 0, i32 1
-  // CK1-32-DAG:  [[FPBPGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], [[KMP_PRIVATES_T]]* [[TASK_WITH_PRIVATES]], i32 0, i32 1
-  // CK1-64-DAG:  [[FPBPGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], [[KMP_PRIVATES_T]]* [[TASK_WITH_PRIVATES]], i32 0, i32 0
-  // CK1-DAG:     [[FPBPADDR:%.+]] = bitcast [1 x i8*]* [[FPBPGEP]] to i8*
-  // CK1-DAG:     [[BPADDR:%.+]] = bitcast i8** [[BPGEP:%.+]] to i8*
-  // CK1-DAG:     call void @llvm.memcpy.p0i8.p0i8.i{{32|64}}(i8* align {{4|8}} [[FPBPADDR]], i8* align {{4|8}} [[BPADDR]], i{{32|64}} {{4|8}}, i1 false)
-  // CK1-DAG:     [[BPGEP]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP:%.+]], i32 0, i32 0
-  // CK1-DAG:     [[BPGEP:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP]], i32 0, i32 0
-  // CK1-DAG:     [[BPCAST:%.+]] = bitcast i8** [[BPGEP]] to [100 x double]**
-  // CK1-DAG:     store [100 x double]* @gc, [100 x double]** [[BPCAST]], align
-  // CK1-32-DAG:  [[FPPGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], [[KMP_PRIVATES_T]]* [[TASK_WITH_PRIVATES]], i32 0, i32 2
-  // CK1-64-DAG:  [[FPPGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], [[KMP_PRIVATES_T]]* [[TASK_WITH_PRIVATES]], i32 0, i32 1
-  // CK1-DAG:     [[FPPADDR:%.+]] = bitcast [1 x i8*]* [[FPPGEP]] to i8*
-  // CK1-DAG:     [[PADDR:%.+]] = bitcast i8** [[PGEP:%.+]] to i8*
-  // CK1-DAG:     call void @llvm.memcpy.p0i8.p0i8.i{{32|64}}(i8* align {{4|8}} [[FPPADDR]], i8* align {{4|8}} [[PADDR]], i{{32|64}} {{4|8}}, i1 false)
-  // CK1-DAG:     [[PGEP]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P:%.+]], i32 0, i32 0
-  // CK1-DAG:     [[PGEP:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P]], i32 0, i32 0
-  // CK1-DAG:     [[PCAST:%.+]] = bitcast i8** [[PGEP]] to [100 x double]**
-  // CK1-DAG:     store [100 x double]* @gc, [100 x double]** [[PCAST]], align
-  // CK1-32-DAG:  [[FPSZGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], [[KMP_PRIVATES_T]]* [[TASK_WITH_PRIVATES]], i32 0, i32 0
-  // CK1-64-DAG:  [[FPSZGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], [[KMP_PRIVATES_T]]* [[TASK_WITH_PRIVATES]], i32 0, i32 2
-  // CK1-DAG:     [[FPSZADDR:%.+]] = bitcast [1 x i64]* [[FPSZGEP]] to i8*
-  // CK1-DAG:     call void @llvm.memcpy.p0i8.p0i8.i{{32|64}}(i8* align {{4|8}} [[FPSZADDR]], i8* align {{4|8}} bitcast ([1 x i64]* [[SIZE00]] to i8*), i{{32|64}} 8, i1 false)
+  // CK1-DAG:     [[DEVi32]] = load i32, ptr %{{[^,]+}},
+  // CK1-DAG:     [[TASK_WITH_PRIVATES:%.+]] = getelementptr inbounds [[KMP_TASK_T_WITH_PRIVATES]], ptr [[TASK]], i32 0, i32 1
+  // CK1-32-DAG:  [[FPBPGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], ptr [[TASK_WITH_PRIVATES]], i32 0, i32 1
+  // CK1-64-DAG:  [[FPBPGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], ptr [[TASK_WITH_PRIVATES]], i32 0, i32 0
+  // CK1-DAG:     call void @llvm.memcpy.p0.p0.i{{32|64}}(ptr align {{4|8}} [[FPBPGEP]], ptr align {{4|8}} [[BPGEP:%.+]], i{{32|64}} {{4|8}}, i1 false)
+  // CK1-DAG:     [[BPGEP]] = getelementptr inbounds [1 x ptr], ptr [[BP:%.+]], i32 0, i32 0
+  // CK1-DAG:     [[BPGEP:%.+]] = getelementptr inbounds [1 x ptr], ptr [[BP]], i32 0, i32 0
+  // CK1-DAG:     store ptr @gc, ptr [[BPGEP]], align
+  // CK1-32-DAG:  [[FPPGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], ptr [[TASK_WITH_PRIVATES]], i32 0, i32 2
+  // CK1-64-DAG:  [[FPPGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], ptr [[TASK_WITH_PRIVATES]], i32 0, i32 1
+  // CK1-DAG:     call void @llvm.memcpy.p0.p0.i{{32|64}}(ptr align {{4|8}} [[FPPGEP]], ptr align {{4|8}} [[PGEP:%.+]], i{{32|64}} {{4|8}}, i1 false)
+  // CK1-DAG:     [[PGEP]] = getelementptr inbounds [1 x ptr], ptr [[P:%.+]], i32 0, i32 0
+  // CK1-DAG:     [[PGEP:%.+]] = getelementptr inbounds [1 x ptr], ptr [[P]], i32 0, i32 0
+  // CK1-DAG:     store ptr @gc, ptr [[PGEP]], align
+  // CK1-32-DAG:  [[FPSZGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], ptr [[TASK_WITH_PRIVATES]], i32 0, i32 0
+  // CK1-64-DAG:  [[FPSZGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], ptr [[TASK_WITH_PRIVATES]], i32 0, i32 2
+  // CK1-DAG:     call void @llvm.memcpy.p0.p0.i{{32|64}}(ptr align {{4|8}} [[FPSZGEP]], ptr align {{4|8}} [[SIZE00]], i{{32|64}} 8, i1 false)
   // CK1-NOT:     __tgt_target_data_end
   #pragma omp target enter data if(1+3-5) device(arg) map(alloc: gc) nowait
   {++arg;}
@@ -97,16 +89,14 @@ void foo(int arg) {
   // Region 02
   // CK1: br i1 %{{[^,]+}}, label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]]
   // CK1: [[IFTHEN]]
-  // CK1-DAG: call void @__tgt_target_data_begin_mapper(%struct.ident_t* @{{.+}}, i64 4, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[SIZE02]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE02]]{{.+}}, i8** null)
+  // CK1-DAG: call void @__tgt_target_data_begin_mapper(ptr @{{.+}}, i64 4, i32 1, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[SIZE02]], ptr [[MTYPE02]]{{.+}}, ptr null)
   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
   // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
 
   // CK1-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
-  // CK1-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to i32**
-  // CK1-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to i32**
-  // CK1-DAG: store i32* [[VAR0:%.+]], i32** [[CBP0]]
-  // CK1-DAG: store i32* [[VAR0]], i32** [[CP0]]
+  // CK1-DAG: store ptr [[VAR0:%.+]], ptr [[BP0]]
+  // CK1-DAG: store ptr [[VAR0]], ptr [[P0]]
   // CK1: br label %[[IFEND:[^,]+]]
 
   // CK1: [[IFELSE]]
@@ -121,7 +111,7 @@ void foo(int arg) {
   {++arg;}
 
   // Region 03
-  // CK1-DAG: call void @__tgt_target_data_begin_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE03]]{{.+}}, i8** null)
+  // CK1-DAG: call void @__tgt_target_data_begin_mapper(ptr @{{.+}}, i64 -1, i32 1, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE03]]{{.+}}, ptr null)
   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
   // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
   // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
@@ -129,11 +119,9 @@ void foo(int arg) {
   // CK1-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0
-  // CK1-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to float**
-  // CK1-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to float**
-  // CK1-DAG: store float* [[VAR0:%.+]], float** [[CBP0]]
-  // CK1-DAG: store float* [[VAR0]], float** [[CP0]]
-  // CK1-DAG: store i64 [[CSVAL0:%[^,]+]], i64* [[S0]]
+  // CK1-DAG: store ptr [[VAR0:%.+]], ptr [[BP0]]
+  // CK1-DAG: store ptr [[VAR0]], ptr [[P0]]
+  // CK1-DAG: store i64 [[CSVAL0:%[^,]+]], ptr [[S0]]
   // CK1-64-DAG: [[CSVAL0]] = mul nuw i64 %{{[^,]+}}, 4
   // CK1-32-DAG: [[CSVAL0]] = sext i32 [[CSVAL032:%.+]] to i64
   // CK1-32-DAG: [[CSVAL032]] = mul nuw i32 %{{[^,]+}}, 4
@@ -146,7 +134,7 @@ void foo(int arg) {
   {++arg;}
 
   // Region 04
-  // CK1-DAG: call void @__tgt_target_data_begin_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE04]]{{.+}}, i8** null)
+  // CK1-DAG: call void @__tgt_target_data_begin_mapper(ptr @{{.+}}, i64 -1, i32 2, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE04]]{{.+}}, ptr null)
   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
   // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
   // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[PS:%[^,]+]]
@@ -154,22 +142,18 @@ void foo(int arg) {
   // CK1-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[PS]], i{{.+}} 0, i{{.+}} 0
-  // CK1-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[ST]]**
-  // CK1-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to double***
-  // CK1-DAG: store [[ST]]* @gb, [[ST]]** [[CBP0]]
-  // CK1-DAG: store double** getelementptr inbounds ([[ST]], [[ST]]* @gb, i32 0, i32 1), double*** [[CP0]]
-  // CK1-DAG: [[DIV:%.+]] = sdiv exact i64 sub (i64 ptrtoint (double** getelementptr (double*, double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1) to i64)), ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)
-  // CK1-DAG: store i64 [[DIV]], i64* [[S0]],
+  // CK1-DAG: store ptr @gb, ptr [[BP0]]
+  // CK1-DAG: store ptr getelementptr inbounds ([[ST]], ptr @gb, i32 0, i32 1), ptr [[P0]]
+  // CK1-DAG: [[DIV:%.+]] = sdiv exact i64 sub (i64 ptrtoint (ptr getelementptr (ptr, ptr getelementptr inbounds (%struct.ST, ptr @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (ptr getelementptr inbounds (%struct.ST, ptr @gb, i32 0, i32 1) to i64)), ptrtoint (ptr getelementptr (i8, ptr null, i32 1) to i64)
+  // CK1-DAG: store i64 [[DIV]], ptr [[S0]],
 
 
   // CK1-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1
   // CK1-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1
-  // CK1-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to double***
-  // CK1-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to double**
-  // CK1-DAG: store double** getelementptr inbounds ([[ST]], [[ST]]* @gb, i32 0, i32 1), double*** [[CBP1]]
-  // CK1-DAG: store double* [[SEC1:%.+]], double** [[CP1]]
-  // CK1-DAG: [[SEC1]] = getelementptr inbounds {{.+}}double* [[SEC11:%[^,]+]], i{{.+}} 0
-  // CK1-DAG: [[SEC11]] = load double*, double** getelementptr inbounds ([[ST]], [[ST]]* @gb, i32 0, i32 1),
+  // CK1-DAG: store ptr getelementptr inbounds ([[ST]], ptr @gb, i32 0, i32 1), ptr [[BP1]]
+  // CK1-DAG: store ptr [[SEC1:%.+]], ptr [[P1]]
+  // CK1-DAG: [[SEC1]] = getelementptr inbounds {{.+}}ptr [[SEC11:%[^,]+]], i{{.+}} 0
+  // CK1-DAG: [[SEC11]] = load ptr, ptr getelementptr inbounds ([[ST]], ptr @gb, i32 0, i32 1),
 
   // CK1: %{{.+}} = add nsw i32 %{{[^,]+}}, 1
   // CK1-NOT: __tgt_target_data_end
@@ -180,7 +164,7 @@ void foo(int arg) {
   {++arg;}
 
   // Region 05
-  // CK1-DAG: call void @__tgt_target_data_begin_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE05]]{{.+}}, i8** null)
+  // CK1-DAG: call void @__tgt_target_data_begin_mapper(ptr @{{.+}}, i64 -1, i32 1, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE05]]{{.+}}, ptr null)
   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
   // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
   // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
@@ -188,11 +172,9 @@ void foo(int arg) {
   // CK1-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0
-  // CK1-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to float**
-  // CK1-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to float**
-  // CK1-DAG: store float* [[VAR0:%.+]], float** [[CBP0]]
-  // CK1-DAG: store float* [[VAR0]], float** [[CP0]]
-  // CK1-DAG: store i64 [[CSVAL0:%[^,]+]], i64* [[S0]]
+  // CK1-DAG: store ptr [[VAR0:%.+]], ptr [[BP0]]
+  // CK1-DAG: store ptr [[VAR0]], ptr [[P0]]
+  // CK1-DAG: store i64 [[CSVAL0:%[^,]+]], ptr [[S0]]
   // CK1-64-DAG: [[CSVAL0]] = mul nuw i64 %{{[^,]+}}, 4
   // CK1-32-DAG: [[CSVAL0]] = sext i32 [[CSVAL032:%.+]] to i64
   // CK1-32-DAG: [[CSVAL032]] = mul nuw i32 %{{[^,]+}}, 4
@@ -205,7 +187,7 @@ void foo(int arg) {
   {++arg;}
 
   // Region 06
-  // CK1-DAG: call void @__tgt_target_data_begin_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE06]]{{.+}}, i8** null)
+  // CK1-DAG: call void @__tgt_target_data_begin_mapper(ptr @{{.+}}, i64 -1, i32 1, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE06]]{{.+}}, ptr null)
   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
   // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
   // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
@@ -213,11 +195,9 @@ void foo(int arg) {
   // CK1-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0
-  // CK1-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to float**
-  // CK1-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to float**
-  // CK1-DAG: store float* [[VAR0:%.+]], float** [[CBP0]]
-  // CK1-DAG: store float* [[VAR0]], float** [[CP0]]
-  // CK1-DAG: store i64 [[CSVAL0:%[^,]+]], i64* [[S0]]
+  // CK1-DAG: store ptr [[VAR0:%.+]], ptr [[BP0]]
+  // CK1-DAG: store ptr [[VAR0]], ptr [[P0]]
+  // CK1-DAG: store i64 [[CSVAL0:%[^,]+]], ptr [[S0]]
   // CK1-64-DAG: [[CSVAL0]] = mul nuw i64 %{{[^,]+}}, 4
   // CK1-32-DAG: [[CSVAL0]] = sext i32 [[CSVAL032:%.+]] to i64
   // CK1-32-DAG: [[CSVAL032]] = mul nuw i32 %{{[^,]+}}, 4
@@ -228,38 +208,34 @@ void foo(int arg) {
 }
 
 
-// CK1:     define internal {{.*}}i32 [[OMP_TASK_ENTRY]](i32 {{.*}}%0, [[KMP_TASK_T_WITH_PRIVATES]]* noalias noundef %1)
-// CK1-DAG: call void @__tgt_target_data_begin_nowait_mapper(%struct.ident_t* @{{.+}}, i64 %{{[^,]+}}, i32 1, i8** [[BPADDR:%[^,]+]], i8** [[PADDR:%[^,]+]], i64* [[SZADDR:%[^,]+]], i64* getelementptr inbounds ([1 x i64], [1 x i64]* [[MTYPE00]], i32 0, i32 0), i8** null, i8** null)
-// CK1-DAG: [[BPADDR]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[FPBPADDR:%[^,]+]], i{{32|64}} 0, i{{32|64}} 0
-// CK1-DAG: [[PADDR]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[FPPADDR:%[^,]+]], i{{32|64}} 0, i{{32|64}} 0
-// CK1-DAG: [[SZADDR]] = getelementptr inbounds [1 x i64], [1 x i64]* [[FPSZADDR:%[^,]+]], i{{32|64}} 0, i{{32|64}} 0
-// CK1-DAG: [[FPBPADDR]] = load [1 x i8*]*, [1 x i8*]** [[FPBP:%[^,]+]], align
-// CK1-DAG: [[FPPADDR]] = load [1 x i8*]*, [1 x i8*]** [[FPP:%[^,]+]], align
-// CK1-DAG: [[FPSZADDR]] = load [1 x i64]*, [1 x i64]** [[FPSZ:%[^,]+]], align
-// CK1-DAG: [[FN:%.+]] = bitcast void (i8*, ...)* {{%.*}} to void (i8*,
-// CK1-DAG: call void [[FN]](i8* %{{[^,]+}}, [1 x i8*]** [[FPBP]], [1 x i8*]** [[FPP]], [1 x i64]** [[FPSZ]])
+// CK1:     define internal {{.*}}i32 [[OMP_TASK_ENTRY]](i32 {{.*}}%0, ptr noalias noundef %1)
+// CK1-DAG: call void @__tgt_target_data_begin_nowait_mapper(ptr @{{.+}}, i64 %{{[^,]+}}, i32 1, ptr [[BPADDR:%[^,]+]], ptr [[PADDR:%[^,]+]], ptr [[SZADDR:%[^,]+]], ptr [[MTYPE00]], ptr null, ptr null)
+// CK1-DAG: [[BPADDR]] = load ptr, ptr [[FPBP:%[^,]+]], align
+// CK1-DAG: [[PADDR]] = load ptr, ptr [[FPP:%[^,]+]], align
+// CK1-DAG: [[SZADDR]] = load ptr, ptr [[FPSZ:%[^,]+]], align
+// CK1-DAG: call void {{%.*}}(ptr %{{[^,]+}}, ptr [[FPBP]], ptr [[FPP]], ptr [[FPSZ]])
 // CK1:   ret i32 0
 // CK1: }
 
 #endif
 ///==========================================================================///
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1A -verify -fopenmp -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK1A --check-prefix CK1A-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1A -fopenmp -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1A --check-prefix CK1A-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1A -verify -fopenmp -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK1A --check-prefix CK1A-32
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1A -fopenmp -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1A --check-prefix CK1A-32
-
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1A -verify -fopenmp-simd -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1A -fopenmp-simd -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1A -verify -fopenmp-simd -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1A -fopenmp-simd -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -DCK1A -verify -fopenmp -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK1A --check-prefix CK1A-64
+// RUN: %clang_cc1 -DCK1A -fopenmp -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1A --check-prefix CK1A-64
+// RUN: %clang_cc1 -DCK1A -verify -fopenmp -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK1A --check-prefix CK1A-32
+// RUN: %clang_cc1 -DCK1A -fopenmp -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1A --check-prefix CK1A-32
+
+// RUN: %clang_cc1 -DCK1A -verify -fopenmp-simd -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -DCK1A -fopenmp-simd -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=51 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -DCK1A -verify -fopenmp-simd -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -DCK1A -fopenmp-simd -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=51 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
 #ifdef CK1A
 
-// CK1A: [[ST:%.+]] = type { i32, double* }
+// CK1A: [[ST:%.+]] = type { i32, ptr }
 template <typename T>
 struct ST {
   T a;
@@ -281,7 +257,7 @@ void foo(int arg) {
   float lb[arg];
 
   // Region 00
-  // CK1A-DAG: call void @__tgt_target_data_begin_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz:32|64]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE00]]{{.+}})
+  // CK1A-DAG: call void @__tgt_target_data_begin_mapper(ptr @{{.+}}, i64 -1, i32 1, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE00]]{{.+}})
   // CK1A-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
   // CK1A-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
   // CK1A-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
@@ -289,11 +265,9 @@ void foo(int arg) {
   // CK1A-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
   // CK1A-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
   // CK1A-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0
-  // CK1A-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to float**
-  // CK1A-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to float**
-  // CK1A-DAG: store float* [[VAR0:%.+]], float** [[CBP0]]
-  // CK1A-DAG: store float* [[VAR0]], float** [[CP0]]
-  // CK1A-DAG: store i64 [[CSVAL0:%[^,]+]], i64* [[S0]]
+  // CK1A-DAG: store ptr [[VAR0:%.+]], ptr [[BP0]]
+  // CK1A-DAG: store ptr [[VAR0]], ptr [[P0]]
+  // CK1A-DAG: store i64 [[CSVAL0:%[^,]+]], ptr [[S0]]
   // CK1A-64-DAG: [[CSVAL0]] = mul nuw i64 %{{[^,]+}}, 4
   // CK1A-32-DAG: [[CSVAL0]] = sext i32 [[CSVAL032:%.+]] to i64
   // CK1A-32-DAG: [[CSVAL032]] = mul nuw i32 %{{[^,]+}}, 4
@@ -306,7 +280,7 @@ void foo(int arg) {
   {++arg;}
 
   // Region 01
-  // CK1A-DAG: call void @__tgt_target_data_begin_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE01]]{{.+}})
+  // CK1A-DAG: call void @__tgt_target_data_begin_mapper(ptr @{{.+}}, i64 -1, i32 1, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE01]]{{.+}})
   // CK1A-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
   // CK1A-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
   // CK1A-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
@@ -314,11 +288,9 @@ void foo(int arg) {
   // CK1A-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
   // CK1A-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
   // CK1A-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0
-  // CK1A-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to float**
-  // CK1A-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to float**
-  // CK1A-DAG: store float* [[VAR0:%.+]], float** [[CBP0]]
-  // CK1A-DAG: store float* [[VAR0]], float** [[CP0]]
-  // CK1A-DAG: store i64 [[CSVAL0:%[^,]+]], i64* [[S0]]
+  // CK1A-DAG: store ptr [[VAR0:%.+]], ptr [[BP0]]
+  // CK1A-DAG: store ptr [[VAR0]], ptr [[P0]]
+  // CK1A-DAG: store i64 [[CSVAL0:%[^,]+]], ptr [[S0]]
   // CK1A-64-DAG: [[CSVAL0]] = mul nuw i64 %{{[^,]+}}, 4
   // CK1A-32-DAG: [[CSVAL0]] = sext i32 [[CSVAL032:%.+]] to i64
   // CK1A-32-DAG: [[CSVAL032]] = mul nuw i32 %{{[^,]+}}, 4
@@ -329,23 +301,23 @@ void foo(int arg) {
 }
 #endif
 ///==========================================================================///
-// RUN: %clang_cc1 -no-opaque-pointers -DCK2 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK2 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK2 --check-prefix CK2-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK2 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK2 --check-prefix CK2-32
-// RUN: %clang_cc1 -no-opaque-pointers -DCK2 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK2 --check-prefix CK2-32
-
-// RUN: %clang_cc1 -no-opaque-pointers -DCK2 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK2 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK2 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK2 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -DCK2 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-64
+// RUN: %clang_cc1 -DCK2 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK2 --check-prefix CK2-64
+// RUN: %clang_cc1 -DCK2 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK2 --check-prefix CK2-32
+// RUN: %clang_cc1 -DCK2 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK2 --check-prefix CK2-32
+
+// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -DCK2 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -DCK2 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
 // SIMD-ONLY1-NOT: {{__kmpc|__tgt}}
 #ifdef CK2
 
-// CK2: [[ST:%.+]] = type { i32, double* }
+// CK2: [[ST:%.+]] = type { i32, ptr }
 template <typename T>
 struct ST {
   T a;
@@ -371,9 +343,9 @@ int bar(int arg){
 // Region 00
 // CK2: br i1 %{{[^,]+}}, label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]]
 // CK2: [[IFTHEN]]
-// CK2-DAG: call void @__tgt_target_data_begin_mapper(%struct.ident_t* @{{.+}}, i64 [[DEV:%[^,]+]], i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE00]]{{.+}}, i8** null)
+// CK2-DAG: call void @__tgt_target_data_begin_mapper(ptr @{{.+}}, i64 [[DEV:%[^,]+]], i32 2, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE00]]{{.+}}, ptr null)
 // CK2-DAG: [[DEV]] = sext i32 [[DEVi32:%[^,]+]] to i64
-// CK2-DAG: [[DEVi32]] = load i32, i32* %{{[^,]+}},
+// CK2-DAG: [[DEVi32]] = load i32, ptr %{{[^,]+}},
 // CK2-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
 // CK2-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
 // CK2-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[PS:%[^,]+]]
@@ -381,22 +353,18 @@ int bar(int arg){
 // CK2-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
 // CK2-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
 // CK2-DAG: [[PS0:%.+]] = getelementptr inbounds {{.+}}[[PS]], i{{.+}} 0, i{{.+}} 0
-// CK2-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[ST]]**
-// CK2-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to double***
-// CK2-DAG: store [[ST]]* [[VAR0:%.+]], [[ST]]** [[CBP0]]
-// CK2-DAG: store double** [[SEC0:%.+]], double*** [[CP0]]
-// CK2-DAG: store i64 {{%.+}}, i64* [[PS0]],
-// CK2-DAG: [[SEC0]] = getelementptr inbounds {{.*}}[[ST]]* [[VAR0]], i32 0, i32 1
+// CK2-DAG: store ptr [[VAR0:%.+]], ptr [[BP0]]
+// CK2-DAG: store ptr [[SEC0:%.+]], ptr [[P0]]
+// CK2-DAG: store i64 {{%.+}}, ptr [[PS0]],
+// CK2-DAG: [[SEC0]] = getelementptr inbounds {{.*}}ptr [[VAR0]], i32 0, i32 1
 
 // CK2-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1
 // CK2-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1
-// CK2-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to double***
-// CK2-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to double**
-// CK2-DAG: store double** [[SEC0]], double*** [[CBP1]]
-// CK2-DAG: store double* [[SEC1:%.+]], double** [[CP1]]
-// CK2-DAG: [[SEC1]] = getelementptr inbounds {{.*}}double* [[SEC11:%[^,]+]], i{{.+}} 1
-// CK2-DAG: [[SEC11]] = load double*, double** [[SEC111:%[^,]+]],
-// CK2-DAG: [[SEC111]] = getelementptr inbounds {{.*}}[[ST]]* [[VAR0]], i32 0, i32 1
+// CK2-DAG: store ptr [[SEC0]], ptr [[BP1]]
+// CK2-DAG: store ptr [[SEC1:%.+]], ptr [[P1]]
+// CK2-DAG: [[SEC1]] = getelementptr inbounds {{.*}}ptr [[SEC11:%[^,]+]], i{{.+}} 1
+// CK2-DAG: [[SEC11]] = load ptr, ptr [[SEC111:%[^,]+]],
+// CK2-DAG: [[SEC111]] = getelementptr inbounds {{.*}}ptr [[VAR0]], i32 0, i32 1
 
 // CK2: br label %[[IFEND:[^,]+]]
 
@@ -407,19 +375,19 @@ int bar(int arg){
 // CK2-NOT: __tgt_target_data_end
 #endif
 ///==========================================================================///
-// RUN: %clang_cc1 -no-opaque-pointers -DCK3 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK3 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK3 --check-prefix CK3-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK3 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK3 --check-prefix CK3-32
-// RUN: %clang_cc1 -no-opaque-pointers -DCK3 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK3 --check-prefix CK3-32
-
-// RUN: %clang_cc1 -no-opaque-pointers -DCK3 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK3 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK3 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-64
+// RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK3 --check-prefix CK3-64
+// RUN: %clang_cc1 -DCK3 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK3 --check-prefix CK3-32
+// RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK3 --check-prefix CK3-32
+
+// RUN: %clang_cc1 -DCK3 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK3 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
 // SIMD-ONLY2-NOT: {{__kmpc|__tgt}}
 #ifdef CK3
 
@@ -434,38 +402,38 @@ void no_target_devices(int arg) {
 }
 #endif
 ///==========================================================================///
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK4 --check-prefix CK4-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK4 --check-prefix CK4-32
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK4 --check-prefix CK4-32
-
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY3 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY3 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY3 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY3 %s
+// RUN: %clang_cc1 -DCK4 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-64
+// RUN: %clang_cc1 -DCK4 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK4 --check-prefix CK4-64
+// RUN: %clang_cc1 -DCK4 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK4 --check-prefix CK4-32
+// RUN: %clang_cc1 -DCK4 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK4 --check-prefix CK4-32
+
+// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY3 %s
+// RUN: %clang_cc1 -DCK4 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY3 %s
+// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY3 %s
+// RUN: %clang_cc1 -DCK4 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY3 %s
 // SIMD-ONLY3-NOT: {{__kmpc|__tgt}}
 
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCK4 --check-prefix TCK4-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCK4 --check-prefix TCK4-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCK4 --check-prefix TCK4-32
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCK4 --check-prefix TCK4-32
-
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY4 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY4 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY4 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY4 %s
+// RUN: %clang_cc1 -DCK4 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
+// RUN: %clang_cc1 -DCK4 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCK4 --check-prefix TCK4-64
+// RUN: %clang_cc1 -DCK4 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCK4 --check-prefix TCK4-64
+// RUN: %clang_cc1 -DCK4 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
+// RUN: %clang_cc1 -DCK4 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCK4 --check-prefix TCK4-32
+// RUN: %clang_cc1 -DCK4 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCK4 --check-prefix TCK4-32
+
+// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
+// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY4 %s
+// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY4 %s
+// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
+// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY4 %s
+// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY4 %s
 // SIMD-ONLY4-NOT: {{__kmpc|__tgt}}
 #ifdef CK4
 
@@ -480,23 +448,23 @@ void device_side_scan(int arg) {
 }
 #endif
 ///==========================================================================///
-// RUN: %clang_cc1 -no-opaque-pointers -DCK5 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK5 --check-prefix CK5-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK5 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK5 --check-prefix CK5-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK5 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK5 --check-prefix CK5-32
-// RUN: %clang_cc1 -no-opaque-pointers -DCK5 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK5 --check-prefix CK5-32
-
-// RUN: %clang_cc1 -no-opaque-pointers -DCK5 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK5 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK5 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK5 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -DCK5 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK5 --check-prefix CK5-64
+// RUN: %clang_cc1 -DCK5 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK5 --check-prefix CK5-64
+// RUN: %clang_cc1 -DCK5 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK5 --check-prefix CK5-32
+// RUN: %clang_cc1 -DCK5 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK5 --check-prefix CK5-32
+
+// RUN: %clang_cc1 -DCK5 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -DCK5 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -DCK5 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -DCK5 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
 // SIMD-ONLY1-NOT: {{__kmpc|__tgt}}
 #ifdef CK5
 
-// CK5: [[STT:%.+]] = type { i32, double* }
+// CK5: [[STT:%.+]] = type { i32, ptr }
 template <typename T>
 struct STT {
   T a;
@@ -522,9 +490,9 @@ int bar(int arg){
 // Region 00
 // CK5: br i1 %{{[^,]+}}, label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]]
 // CK5: [[IFTHEN]]
-// CK5-DAG: call void @__tgt_target_data_begin_mapper(%struct.ident_t* @{{.+}}, i64 [[DEV:%[^,]+]], i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE00]]{{.+}}, i8** null)
+// CK5-DAG: call void @__tgt_target_data_begin_mapper(ptr @{{.+}}, i64 [[DEV:%[^,]+]], i32 2, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE00]]{{.+}}, ptr null)
 // CK5-DAG: [[DEV]] = sext i32 [[DEVi32:%[^,]+]] to i64
-// CK5-DAG: [[DEVi32]] = load i32, i32* %{{[^,]+}},
+// CK5-DAG: [[DEVi32]] = load i32, ptr %{{[^,]+}},
 // CK5-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
 // CK5-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
 // CK5-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[PS:%[^,]+]]
@@ -532,22 +500,18 @@ int bar(int arg){
 // CK5-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
 // CK5-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
 // CK5-DAG: [[PS0:%.+]] = getelementptr inbounds {{.+}}[[PS]], i{{.+}} 0, i{{.+}} 0
-// CK5-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[STT]]**
-// CK5-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to double***
-// CK5-DAG: store [[STT]]* [[VAR0:%.+]], [[STT]]** [[CBP0]]
-// CK5-DAG: store double** [[SEC0:%.+]], double*** [[CP0]]
-// CK5-DAG: store i64 {{%.+}}, i64* [[PS0]],
-// CK5-DAG: [[SEC0]] = getelementptr inbounds {{.*}}[[STT]]* [[VAR0]], i32 0, i32 1
+// CK5-DAG: store ptr [[VAR0:%.+]], ptr [[BP0]]
+// CK5-DAG: store ptr [[SEC0:%.+]], ptr [[P0]]
+// CK5-DAG: store i64 {{%.+}}, ptr [[PS0]],
+// CK5-DAG: [[SEC0]] = getelementptr inbounds {{.*}}ptr [[VAR0]], i32 0, i32 1
 
 // CK5-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1
 // CK5-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1
-// CK5-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to double***
-// CK5-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to double**
-// CK5-DAG: store double** [[SEC0]], double*** [[CBP1]]
-// CK5-DAG: store double* [[SEC1:%.+]], double** [[CP1]]
-// CK5-DAG: [[SEC1]] = getelementptr inbounds {{.*}}double* [[SEC11:%[^,]+]], i{{.+}} 1
-// CK5-DAG: [[SEC11]] = load double*, double** [[SEC111:%[^,]+]],
-// CK5-DAG: [[SEC111]] = getelementptr inbounds {{.*}}[[STT]]* [[VAR0]], i32 0, i32 1
+// CK5-DAG: store ptr [[SEC0]], ptr [[BP1]]
+// CK5-DAG: store ptr [[SEC1:%.+]], ptr [[P1]]
+// CK5-DAG: [[SEC1]] = getelementptr inbounds {{.*}}ptr [[SEC11:%[^,]+]], i{{.+}} 1
+// CK5-DAG: [[SEC11]] = load ptr, ptr [[SEC111:%[^,]+]],
+// CK5-DAG: [[SEC111]] = getelementptr inbounds {{.*}}ptr [[VAR0]], i32 0, i32 1
 
 // CK5: br label %[[IFEND:[^,]+]]
 
diff --git a/clang/test/OpenMP/target_enter_data_depend_codegen.cpp b/clang/test/OpenMP/target_enter_data_depend_codegen.cpp
index b29bf4f50798..997bca525d47 100644
--- a/clang/test/OpenMP/target_enter_data_depend_codegen.cpp
+++ b/clang/test/OpenMP/target_enter_data_depend_codegen.cpp
@@ -1,20 +1,20 @@
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-64
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-64
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
 
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
 
 // expected-no-diagnostics
-// CK1: [[ST:%.+]] = type { i32, double* }
+// CK1: [[ST:%.+]] = type { i32, ptr }
 // CK1: %struct.kmp_depend_info = type { i[[sz:64|32]],
 // CK1-SAME: i[[sz]], i8 }
 #ifndef HEADER
@@ -51,59 +51,44 @@ void foo(int arg) {
   // CK1: alloca [5 x %struct.kmp_depend_info],
 
   // Region 00
-  // CK1: [[BP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP:%.+]], i32 0, i32 0
-  // CK1: [[BP0_BC:%.+]] = bitcast i8** [[BP0]] to [100 x double]**
-  // CK1: store [100 x double]* @gc, [100 x double]** [[BP0_BC]],
-  // CK1: [[P0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P:%.+]], i32 0, i32 0
-  // CK1: [[P0_BC:%.+]] = bitcast i8** [[P0]] to [100 x double]**
-  // CK1: store [100 x double]* @gc, [100 x double]** [[P0_BC]],
-  // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP]], i32 0, i32 0
-  // CK1: [[GEPP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P]], i32 0, i32 0
-  // CK1: [[CAP_DEVICE:%.+]] = getelementptr inbounds %struct.anon, %struct.anon* [[CAPTURES:%.+]], i32 0, i32 0
-  // CK1: [[DEVICE:%.+]] = load i32, i32* %{{.+}}
-  // CK1: store i32 [[DEVICE]], i32* [[CAP_DEVICE]],
-  // CK1: [[DEV1:%.+]] = load i32, i32* %{{.+}}
+  // CK1: [[BP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[BP:%.+]], i32 0, i32 0
+  // CK1: store ptr @gc, ptr [[BP0]],
+  // CK1: [[P0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[P:%.+]], i32 0, i32 0
+  // CK1: store ptr @gc, ptr [[P0]],
+  // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[BP]], i32 0, i32 0
+  // CK1: [[GEPP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[P]], i32 0, i32 0
+  // CK1: [[CAP_DEVICE:%.+]] = getelementptr inbounds %struct.anon, ptr [[CAPTURES:%.+]], i32 0, i32 0
+  // CK1: [[DEVICE:%.+]] = load i32, ptr %{{.+}}
+  // CK1: store i32 [[DEVICE]], ptr [[CAP_DEVICE]],
+  // CK1: [[DEV1:%.+]] = load i32, ptr %{{.+}}
   // CK1: [[DEV2:%.+]] = sext i32 [[DEV1]] to i64
-  // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_target_task_alloc(%struct.ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{64|36}}, i[[sz]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates*)* [[TASK_ENTRY0:@.+]] to i32 (i32, i8*)*), i64 [[DEV2]])
-  // CK1: [[BC:%.+]] = bitcast i8* [[RES]] to %struct.kmp_task_t_with_privates*
-  // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates, %struct.kmp_task_t_with_privates* [[BC]], i32 0, i32 0
-  // CK1: [[SHAREDS:%.+]] = getelementptr inbounds %struct.kmp_task_t, %struct.kmp_task_t* [[TASK_T]], i32 0, i32 0
-  // CK1: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS]],
-  // CK1: [[BC1:%.+]] = bitcast %struct.anon* [[CAPTURES]] to i8*
-  // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align 4 [[SHAREDS_REF]], i8* align 4 [[BC1]], i[[sz]] 4, i1 false)
-  // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates, %struct.kmp_task_t_with_privates* [[BC]], i32 0, i32 1
-  // CK1-64: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 0
-  // CK1-64: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8*
-  // CK1-64: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-64: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 1
-  // CK1-64: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8*
-  // CK1-64: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-64: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 2
-  // CK1-64: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i64]* [[PRIVS_SIZES]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} bitcast ([1 x i64]* [[SIZE00]] to i8*), i64 {{8|4}}, i1 false)
-  // CK1-32: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 0
-  // CK1-32: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i64]* [[PRIVS_SIZES]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} bitcast ([1 x i64]* [[SIZE00]] to i8*), i[[sz]] {{8|4}}, i1 false)
-  // CK1-32: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 1
-  // CK1-32: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8*
-  // CK1-32: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-32: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 2
-  // CK1-32: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8*
-  // CK1-32: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1: [[BC_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP:%.+]], i[[sz]] 0
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 4, i[[sz]]* [[DEP_SIZE]],
-  // 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 i32 @__kmpc_omp_task_with_deps(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]], i32 1, i8* [[BC]], i32 0, i8* null)
+  // CK1: [[RES:%.+]] = call ptr @__kmpc_omp_target_task_alloc(ptr {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{64|36}}, i[[sz]] 4, ptr [[TASK_ENTRY0:@.+]], i64 [[DEV2]])
+  // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates, ptr [[RES]], i32 0, i32 0
+  // CK1: [[SHAREDS:%.+]] = getelementptr inbounds %struct.kmp_task_t, ptr [[TASK_T]], i32 0, i32 0
+  // CK1: [[SHAREDS_REF:%.+]] = load ptr, ptr [[SHAREDS]],
+  // CK1: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align 4 [[SHAREDS_REF]], ptr align 4 [[CAPTURES]], i[[sz]] 4, i1 false)
+  // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates, ptr [[RES]], i32 0, i32 1
+  // CK1-64: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, ptr [[PRIVS]], i32 0, i32 0
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_BASEPTRS]], ptr align {{8|4}} [[GEPBP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-64: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, ptr [[PRIVS]], i32 0, i32 1
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_PTRS]], ptr align {{8|4}} [[GEPP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-64: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t, ptr [[PRIVS]], i32 0, i32 2
+  // CK1-64: call void @llvm.memcpy.p0.p0.i64(ptr align {{8|4}} [[PRIVS_SIZES]], ptr align {{8|4}} [[SIZE00]], i64 {{8|4}}, i1 false)
+  // CK1-32: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t, ptr [[PRIVS]], i32 0, i32 0
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_SIZES]], ptr align {{8|4}} [[SIZE00]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-32: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, ptr [[PRIVS]], i32 0, i32 1
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_BASEPTRS]], ptr align {{8|4}} [[GEPBP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-32: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, ptr [[PRIVS]], i32 0, i32 2
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_PTRS]], ptr align {{8|4}} [[GEPP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP:%.+]], i[[sz]] 0
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 4, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 1, ptr [[DEP_ATTRS]]
+  // CK1: = call i32 @__kmpc_omp_task_with_deps(ptr @{{.+}}, i32 %{{.+}}, ptr [[RES]], i32 1, ptr [[MAIN_DEP]], i32 0, ptr null)
 
   // CK1: %{{.+}} = add nsw i32 %{{[^,]+}}, 1
   #pragma omp target enter data if(1+3-5) device(arg) map(alloc:gc) nowait depend(in: arg)
@@ -117,77 +102,62 @@ void foo(int arg) {
   // Region 02
   // CK1: br i1 %{{[^,]+}}, label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]]
   // CK1: [[IFTHEN]]
-  // CK1: [[BP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP:%.+]], i32 0, i32 0
-  // CK1: [[BP0_BC:%.+]] = bitcast i8** [[BP0]] to i32**
-  // CK1: store i32* [[ARG:%.+]], i32** [[BP0_BC]],
-  // CK1: [[P0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P:%.+]], i32 0, i32 0
-  // CK1: [[P0_BC:%.+]] = bitcast i8** [[P0]] to i32**
-  // CK1: store i32* [[ARG]], i32** [[P0_BC]],
-  // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP]], i32 0, i32 0
-  // CK1: [[GEPP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P]], i32 0, i32 0
-  // CK1: [[IF_DEVICE:%.+]] = getelementptr inbounds %struct.anon{{.+}}, %struct.anon{{.+}}* [[CAPTURES:%.+]], i32 0, i32 0
-  // CK1: [[IF:%.+]] = load i8, i8* %{{.+}}
+  // CK1: [[BP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[BP:%.+]], i32 0, i32 0
+  // CK1: store ptr [[ARG:%.+]], ptr [[BP0]],
+  // CK1: [[P0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[P:%.+]], i32 0, i32 0
+  // CK1: store ptr [[ARG]], ptr [[P0]],
+  // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[BP]], i32 0, i32 0
+  // CK1: [[GEPP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[P]], i32 0, i32 0
+  // CK1: [[IF_DEVICE:%.+]] = getelementptr inbounds %struct.anon{{.+}}, ptr [[CAPTURES:%.+]], i32 0, i32 0
+  // CK1: [[IF:%.+]] = load i8, ptr %{{.+}}
   // CK1: [[IF_BOOL:%.+]] = trunc i8 [[IF]] to i1
   // CK1: [[IF:%.+]] = zext i1 [[IF_BOOL]] to i8
-  // CK1: store i8 [[IF]], i8* [[IF_DEVICE]],
-  // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{64|36}}, i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY2:@.+]] to i32 (i32, i8*)*))
-  // CK1: [[RES_BC:%.+]] = bitcast i8* [[RES]] to %struct.kmp_task_t_with_privates{{.+}}*
-  // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 0
-  // CK1: [[SHAREDS:%.+]] = getelementptr inbounds %struct.kmp_task_t, %struct.kmp_task_t* [[TASK_T]], i32 0, i32 0
-  // CK1: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS]],
-  // CK1: [[BC1:%.+]] = bitcast %struct.anon{{.+}}* [[CAPTURES]] to i8*
-  // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align 1 [[SHAREDS_REF]], i8* align 1 [[BC1]], i[[sz]] 1, i1 false)
-  // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 1
-  // CK1-64: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0
-  // CK1-64: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8*
-  // CK1-64: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-64: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1
-  // CK1-64: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8*
-  // CK1-64: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-64: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2
-  // CK1-64: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i[[sz]]]* [[PRIVS_SIZES]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} bitcast ([1 x i[[sz]]]* [[SIZE02]] to i8*), i[[sz]] {{8|4}}, i1 false)
-  // CK1-32: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0
-  // CK1-32: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i64]* [[PRIVS_SIZES]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} bitcast ([1 x i64]* [[SIZE02]] to i8*), i[[sz]] {{8|4}}, i1 false)
-  // CK1-32: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1
-  // CK1-32: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8*
-  // CK1-32: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-32: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2
-  // CK1-32: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8*
-  // CK1-32: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1: [[BC_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP:%.+]], i[[sz]] 0
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 4, i[[sz]]* [[DEP_SIZE]],
-  // 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_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 1
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 4, i[[sz]]* [[DEP_SIZE]],
-  // 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: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 2
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] ptrtoint ([100 x double]* @gc to i[[sz]]), i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 800, i[[sz]]* [[DEP_SIZE]],
-  // 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_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]])
+  // CK1: store i8 [[IF]], ptr [[IF_DEVICE]],
+  // CK1: [[RES:%.+]] = call ptr @__kmpc_omp_task_alloc(ptr {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{64|36}}, i[[sz]] 1, ptr [[TASK_ENTRY2:@.+]])
+  // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, ptr [[RES]], i32 0, i32 0
+  // CK1: [[SHAREDS:%.+]] = getelementptr inbounds %struct.kmp_task_t, ptr [[TASK_T]], i32 0, i32 0
+  // CK1: [[SHAREDS_REF:%.+]] = load ptr, ptr [[SHAREDS]],
+  // CK1: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align 1 [[SHAREDS_REF]], ptr align 1 [[CAPTURES]], i[[sz]] 1, i1 false)
+  // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, ptr [[RES]], i32 0, i32 1
+  // CK1-64: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 0
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_BASEPTRS]], ptr align {{8|4}} [[GEPBP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-64: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 1
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_PTRS]], ptr align {{8|4}} [[GEPP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-64: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 2
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_SIZES]], ptr align {{8|4}} [[SIZE02]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-32: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 0
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_SIZES]], ptr align {{8|4}} [[SIZE02]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-32: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 1
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_BASEPTRS]], ptr align {{8|4}} [[GEPBP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-32: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 2
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_PTRS]], ptr align {{8|4}} [[GEPP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP:%.+]], i[[sz]] 0
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 4, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 3, ptr [[DEP_ATTRS]]
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 1
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 4, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 3, ptr [[DEP_ATTRS]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 2
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] ptrtoint (ptr @gc to i[[sz]]), ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 800, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 3, ptr [[DEP_ATTRS]]
+  // CK1: call void @__kmpc_omp_taskwait_deps_51(ptr @{{.+}}, i32 %{{.+}}, i32 3, ptr [[MAIN_DEP]], i32 0, ptr null, i32 0)
+  // CK1: call void @__kmpc_omp_task_begin_if0(ptr @{{.+}}, i32 %{{.+}}, ptr [[RES]])
+  // CK1: = call i32 [[TASK_ENTRY2]](i32 %{{.+}}, ptr [[RES]])
+  // CK1: call void @__kmpc_omp_task_complete_if0(ptr @{{.+}}, i32 %{{.+}}, ptr [[RES]])
 
   // CK1: br label %[[IFEND:[^,]+]]
 
@@ -202,81 +172,65 @@ void foo(int arg) {
   {++arg;}
 
   // Region 03
-  // CK1: [[BP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP:%.+]], i32 0, i32 0
-  // CK1: [[BP0_BC:%.+]] = bitcast i8** [[BP0]] to float**
-  // CK1: store float* [[VLA:%.+]], float** [[BP0_BC]],
-  // CK1: [[P0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P:%.+]], i32 0, i32 0
-  // CK1: [[P0_BC:%.+]] = bitcast i8** [[P0]] to float**
-  // CK1: store float* [[VLA]], float** [[P0_BC]],
-  // CK1: [[S0:%.+]] = getelementptr inbounds [1 x i64], [1 x i64]* [[S:%.+]], i32 0, i32 0
-  // CK1: store i64 {{.+}}, i64* [[S0]],
-  // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP]], i32 0, i32 0
-  // CK1: [[GEPP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P]], i32 0, i32 0
-  // CK1: [[GEPS0:%.+]] = getelementptr inbounds [1 x i64], [1 x i64]* [[S]], i32 0, i32 0
-  // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{64|36}}, i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY3:@.+]] to i32 (i32, i8*)*))
-  // CK1: [[RES_BC:%.+]] = bitcast i8* [[RES]] to %struct.kmp_task_t_with_privates{{.+}}*
-  // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 0
-  // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 1
-  // CK1-64: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0
-  // CK1-64: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8*
-  // CK1-64: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-64: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1
-  // CK1-64: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8*
-  // CK1-64: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-64: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2
-  // CK1-64: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i[[sz]]]* [[PRIVS_SIZES]] to i8*
-  // CK1-64: [[BC_SIZES:%.+]] = bitcast i[[sz]]* [[GEPS0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} [[BC_SIZES]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-32: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0
-  // CK1-32: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i64]* [[PRIVS_SIZES]] to i8*
-  // CK1-32: [[BC_SIZES:%.+]] = bitcast i64* [[GEPS0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} [[BC_SIZES]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-32: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1
-  // CK1-32: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8*
-  // CK1-32: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-32: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2
-  // CK1-32: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8*
-  // CK1-32: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1: [[BC_ADR:%.+]] = ptrtoint float* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP:%.+]], i[[sz]] 0
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] %{{.+}}, i[[sz]]* [[DEP_SIZE]],
-  // 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_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 1
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 4, i[[sz]]* [[DEP_SIZE]],
-  // 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_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 2
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 4, i[[sz]]* [[DEP_SIZE]],
-  // 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: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 3
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] ptrtoint ([100 x double]* @gc to i[[sz]]), i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 800, i[[sz]]* [[DEP_SIZE]],
-  // 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_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]])
+  // CK1: [[BP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[BP:%.+]], i32 0, i32 0
+  // CK1: store ptr [[VLA:%.+]], ptr [[BP0]],
+  // CK1: [[P0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[P:%.+]], i32 0, i32 0
+  // CK1: store ptr [[VLA]], ptr [[P0]],
+  // CK1: [[S0:%.+]] = getelementptr inbounds [1 x i64], ptr [[S:%.+]], i32 0, i32 0
+  // CK1: store i64 {{.+}}, ptr [[S0]],
+  // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[BP]], i32 0, i32 0
+  // CK1: [[GEPP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[P]], i32 0, i32 0
+  // CK1: [[GEPS0:%.+]] = getelementptr inbounds [1 x i64], ptr [[S]], i32 0, i32 0
+  // CK1: [[RES:%.+]] = call ptr @__kmpc_omp_task_alloc(ptr {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{64|36}}, i[[sz]] 1, ptr [[TASK_ENTRY3:@.+]])
+  // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, ptr [[RES]], i32 0, i32 0
+  // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, ptr [[RES]], i32 0, i32 1
+  // CK1-64: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 0
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_BASEPTRS]], ptr align {{8|4}} [[GEPBP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-64: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 1
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_PTRS]], ptr align {{8|4}} [[GEPP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-64: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 2
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_SIZES]], ptr align {{8|4}} [[GEPS0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-32: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 0
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_SIZES]], ptr align {{8|4}} [[GEPS0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-32: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 1
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_BASEPTRS]], ptr align {{8|4}} [[GEPBP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-32: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 2
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_PTRS]], ptr align {{8|4}} [[GEPP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP:%.+]], i[[sz]] 0
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] %{{.+}}, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 3, ptr [[DEP_ATTRS]]
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 1
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 4, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 3, ptr [[DEP_ATTRS]]
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 2
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 4, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 3, ptr [[DEP_ATTRS]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 3
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] ptrtoint (ptr @gc to i[[sz]]), ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 800, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 3, ptr [[DEP_ATTRS]]
+  // CK1: call void @__kmpc_omp_taskwait_deps_51(ptr @{{.+}}, i32 %{{.+}}, i32 4, ptr [[MAIN_DEP]], i32 0, ptr null, i32 0)
+  // CK1: call void @__kmpc_omp_task_begin_if0(ptr @{{.+}}, i32 %{{.+}}, ptr [[RES]])
+  // CK1: = call i32 [[TASK_ENTRY3]](i32 %{{.+}}, ptr [[RES]])
+  // CK1: call void @__kmpc_omp_task_complete_if0(ptr @{{.+}}, i32 %{{.+}}, ptr [[RES]])
   #pragma omp target enter data map(alloc:lb) depend(out: lb, arg, la, gc)
   {++arg;}
 
@@ -284,153 +238,119 @@ void foo(int arg) {
   {++arg;}
 
   // Region 04
-  // CK1: [[DIV:%.+]] = sdiv exact i64 sub (i64 ptrtoint (double** getelementptr (double*, double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1) to i64)), ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)
-  // CK1: [[BP0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP:%.+]], i32 0, i32 0
-  // CK1: [[BP0_BC:%.+]] = bitcast i8** [[BP0]] to %struct.ST**
-  // CK1: store %struct.ST* @gb, %struct.ST** [[BP0_BC]],
-  // CK1: [[P0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P:%.+]], i32 0, i32 0
-  // CK1: [[P0_BC:%.+]] = bitcast i8** [[P0]] to double***
-  // CK1: store double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), double*** [[P0_BC]],
-  // CK1: [[PS0:%.+]] = getelementptr inbounds [2 x i64], [2 x i64]* [[PS:%.+]], i32 0, i32 0
-  // CK1: store i64 [[DIV]], i64* [[PS0]],
-  // CK1: [[BP1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 1
-  // CK1: [[BP1_BC:%.+]] = bitcast i8** [[BP1]] to double***
-  // CK1: store double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), double*** [[BP1_BC]],
-  // CK1: [[P1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 1
-  // CK1: [[P1_BC:%.+]] = bitcast i8** [[P1]] to double**
-  // CK1: store double* %{{.+}}, double** [[P1_BC]],
-  // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 0
-  // CK1: [[GEPP0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 0
-  // CK1: [[GEPS0:%.+]] = getelementptr inbounds [2 x i64], [2 x i64]* [[PS]], i32 0, i32 0
-  // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{88|52}}, i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY4:@.+]] to i32 (i32, i8*)*))
-  // CK1: [[RES_BC:%.+]] = bitcast i8* [[RES]] to %struct.kmp_task_t_with_privates{{.+}}*
-  // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 0
-  // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 1
-  // CK1-64: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0
-  // CK1-64: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [2 x i8*]* [[PRIVS_BASEPTRS]] to i8*
-  // CK1-64: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{16|8}}, i1 false)
-  // CK1-64: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1
-  // CK1-64: [[BC_PRIVS_PTRS:%.+]] = bitcast [2 x i8*]* [[PRIVS_PTRS]] to i8*
-  // CK1-64: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{16|8}}, i1 false)
-  // CK1-64: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2
-  // CK1-64: [[BC_PRIVS_SIZES:%.+]] = bitcast [2 x i[[sz]]]* [[PRIVS_SIZES]] to i8*
-  // CK1-64: [[BC_SIZES:%.+]] = bitcast i64* [[GEPS0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} [[BC_SIZES]], i[[sz]] {{16|8}}, i1 false)
-  // CK1-32: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0
-  // CK1-32: [[BC_PRIVS_SIZES:%.+]] = bitcast [2 x i64]* [[PRIVS_SIZES]] to i8*
-  // CK1-32: [[BC_SIZES:%.+]] = bitcast i64* [[GEPS0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} [[BC_SIZES]], i[[sz]] {{16|8}}, i1 false)
-  // CK1-32: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1
-  // CK1-32: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [2 x i8*]* [[PRIVS_BASEPTRS]] to i8*
-  // CK1-32: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{16|8}}, i1 false)
-  // CK1-32: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2
-  // CK1-32: [[BC_PRIVS_PTRS:%.+]] = bitcast [2 x i8*]* [[PRIVS_PTRS]] to i8*
-  // CK1-32: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{16|8}}, i1 false)
+  // CK1: [[DIV:%.+]] = sdiv exact i64 sub (i64 ptrtoint (ptr getelementptr (ptr, ptr getelementptr inbounds (%struct.ST, ptr @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (ptr getelementptr inbounds (%struct.ST, ptr @gb, i32 0, i32 1) to i64)), ptrtoint (ptr getelementptr (i8, ptr null, i32 1) to i64)
+  // CK1: [[BP0:%.+]] = getelementptr inbounds [2 x ptr], ptr [[BP:%.+]], i32 0, i32 0
+  // CK1: store ptr @gb, ptr [[BP0]],
+  // CK1: [[P0:%.+]] = getelementptr inbounds [2 x ptr], ptr [[P:%.+]], i32 0, i32 0
+  // CK1: store ptr getelementptr inbounds (%struct.ST, ptr @gb, i32 0, i32 1), ptr [[P0]],
+  // CK1: [[PS0:%.+]] = getelementptr inbounds [2 x i64], ptr [[PS:%.+]], i32 0, i32 0
+  // CK1: store i64 [[DIV]], ptr [[PS0]],
+  // CK1: [[BP1:%.+]] = getelementptr inbounds [2 x ptr], ptr [[BP]], i32 0, i32 1
+  // CK1: store ptr getelementptr inbounds (%struct.ST, ptr @gb, i32 0, i32 1), ptr [[BP1]],
+  // CK1: [[P1:%.+]] = getelementptr inbounds [2 x ptr], ptr [[P]], i32 0, i32 1
+  // CK1: store ptr %{{.+}}, ptr [[P1]],
+  // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [2 x ptr], ptr [[BP]], i32 0, i32 0
+  // CK1: [[GEPP0:%.+]] = getelementptr inbounds [2 x ptr], ptr [[P]], i32 0, i32 0
+  // CK1: [[GEPS0:%.+]] = getelementptr inbounds [2 x i64], ptr [[PS]], i32 0, i32 0
+  // CK1: [[RES:%.+]] = call ptr @__kmpc_omp_task_alloc(ptr {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{88|52}}, i[[sz]] 1, ptr [[TASK_ENTRY4:@.+]])
+  // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, ptr [[RES]], i32 0, i32 0
+  // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, ptr [[RES]], i32 0, i32 1
+  // CK1-64: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 0
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_BASEPTRS]], ptr align {{8|4}} [[GEPBP0]], i[[sz]] {{16|8}}, i1 false)
+  // CK1-64: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 1
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_PTRS]], ptr align {{8|4}} [[GEPP0]], i[[sz]] {{16|8}}, i1 false)
+  // CK1-64: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 2
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_SIZES]], ptr align {{8|4}} [[GEPS0]], i[[sz]] {{16|8}}, i1 false)
+  // CK1-32: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 0
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_SIZES]], ptr align {{8|4}} [[GEPS0]], i[[sz]] {{16|8}}, i1 false)
+  // CK1-32: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 1
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_BASEPTRS]], ptr align {{8|4}} [[GEPBP0]], i[[sz]] {{16|8}}, i1 false)
+  // CK1-32: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 2
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_PTRS]], ptr align {{8|4}} [[GEPP0]], i[[sz]] {{16|8}}, i1 false)
   // CK1: %{{.+}} = sub nuw
-  // CK1: [[BC_ADR:%.+]] = ptrtoint double* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP:%.+]], i[[sz]] 0
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] %{{.+}}, i[[sz]]* [[DEP_SIZE]],
-  // 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_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 1
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 4, i[[sz]]* [[DEP_SIZE]],
-  // 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_ADR:%.+]] = ptrtoint float* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 2
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] %{{.+}}, i[[sz]]* [[DEP_SIZE]],
-  // 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: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 3
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] ptrtoint ([100 x double]* @gc to i[[sz]]), i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 800, i[[sz]]* [[DEP_SIZE]],
-  // 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_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 4
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 4, i[[sz]]* [[DEP_SIZE]],
-  // 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_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]])
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP:%.+]], i[[sz]] 0
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] %{{.+}}, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 1, ptr [[DEP_ATTRS]]
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 1
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 4, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 1, ptr [[DEP_ATTRS]]
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 2
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] %{{.+}}, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 1, ptr [[DEP_ATTRS]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 3
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] ptrtoint (ptr @gc to i[[sz]]), ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 800, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 1, ptr [[DEP_ATTRS]]
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 4
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 4, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 1, ptr [[DEP_ATTRS]]
+  // CK1: call void @__kmpc_omp_taskwait_deps_51(ptr @{{.+}}, i32 %{{.+}}, i32 5, ptr [[MAIN_DEP]], i32 0, ptr null, i32 0)
+  // CK1: call void @__kmpc_omp_task_begin_if0(ptr @{{.+}}, i32 %{{.+}}, ptr [[RES]])
+  // CK1: = call i32 [[TASK_ENTRY4]](i32 %{{.+}}, ptr [[RES]])
+  // CK1: call void @__kmpc_omp_task_complete_if0(ptr @{{.+}}, i32 %{{.+}}, ptr [[RES]])
   #pragma omp target enter data map(to:gb.b[:3]) depend(in: gb.b[:3], la, lb, gc, arg)
   {++arg;}
 }
 
-// CK1: define internal{{.*}} i32 [[TASK_ENTRY0]](i32{{.*}}, %struct.kmp_task_t_with_privates* noalias noundef %1)
-// CK1-DAG: call void @__tgt_target_data_begin_nowait_mapper(%struct.ident_t* @{{.+}}, i64 [[DEV:%[^,]+]], i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE00]]{{.+}}, i8** null)
+// CK1: define internal{{.*}} i32 [[TASK_ENTRY0]](i32{{.*}}, ptr noalias noundef %1)
+// CK1-DAG: call void @__tgt_target_data_begin_nowait_mapper(ptr @{{.+}}, i64 [[DEV:%[^,]+]], i32 1, ptr [[BP:%.+]], ptr [[P:%.+]], ptr [[S:%.+]], ptr [[MTYPE00]]{{.+}}, ptr null)
 // CK1-DAG: [[DEV]] = sext i32 [[DEVi32:%[^,]+]] to i64
-// CK1-DAG: [[DEVi32]] = load i32, i32* %{{[^,]+}},
-// CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
-// CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
-// CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
-// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%[^,]+]],
-// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%[^,]+]],
-// CK1-DAG: [[S]] = load [1 x i64]*, [1 x i64]** [[S_PRIV:%[^,]+]],
-// CK1-DAG: [[FN:%.+]] = bitcast void (i8*, ...)* {{%.*}} to void (i8*,
-// CK1-DAG: call void [[FN]](i8* %{{[^,]+}}, [1 x i8*]** [[BP_PRIV]], [1 x i8*]** [[P_PRIV]], [1 x i64]** [[S_PRIV]])
+// CK1-DAG: [[DEVi32]] = load i32, ptr %{{[^,]+}},
+// CK1-DAG: [[BP]] = load ptr, ptr [[BP_PRIV:%[^,]+]],
+// CK1-DAG: [[P]] = load ptr, ptr [[P_PRIV:%[^,]+]],
+// CK1-DAG: [[S]] = load ptr, ptr [[S_PRIV:%[^,]+]],
+// CK1-DAG: call void {{%.*}}(ptr %{{[^,]+}}, ptr [[BP_PRIV]], ptr [[P_PRIV]], ptr [[S_PRIV]])
 // CK1: ret i32 0
 // CK1: }
 
-// CK1: define internal{{.*}} i32 [[TASK_ENTRY2]](i32{{.*}}, %struct.kmp_task_t_with_privates{{.+}}* noalias noundef %1)
-// CK1-DAG: call void @__tgt_target_data_begin_mapper(%struct.ident_t* @{{.+}}, i64 4, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE02]]{{.+}}, i8** null)
-// CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
-// CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
-// CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
-// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%[^,]+]],
-// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%[^,]+]],
-// CK1-DAG: [[S]] = load [1 x i64]*, [1 x i64]** [[S_PRIV:%[^,]+]],
-// CK1-DAG: [[FN:%.+]] = bitcast void (i8*, ...)* {{%.*}} to void (i8*,
-// CK1-DAG: call void [[FN]](i8* %{{[^,]+}}, [1 x i8*]** [[BP_PRIV]], [1 x i8*]** [[P_PRIV]], [1 x i64]** [[S_PRIV]])
+// CK1: define internal{{.*}} i32 [[TASK_ENTRY2]](i32{{.*}}, ptr noalias noundef %1)
+// CK1-DAG: call void @__tgt_target_data_begin_mapper(ptr @{{.+}}, i64 4, i32 1, ptr [[BP:%.+]], ptr [[P:%.+]], ptr [[S:%.+]], ptr [[MTYPE02]]{{.+}}, ptr null)
+// CK1-DAG: [[BP]] = load ptr, ptr [[BP_PRIV:%[^,]+]],
+// CK1-DAG: [[P]] = load ptr, ptr [[P_PRIV:%[^,]+]],
+// CK1-DAG: [[S]] = load ptr, ptr [[S_PRIV:%[^,]+]],
+// CK1-DAG: call void {{%.*}}(ptr %{{[^,]+}}, ptr [[BP_PRIV]], ptr [[P_PRIV]], ptr [[S_PRIV]])
 // CK1: ret i32 0
 // CK1: }
 
-// CK1: define internal{{.*}} i32 [[TASK_ENTRY3]](i32{{.*}}, %struct.kmp_task_t_with_privates{{.+}}* noalias noundef %1)
-// CK1-DAG: call void @__tgt_target_data_begin_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE03]]{{.+}}, i8** null)
-// CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
-// CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
-// CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
-// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%[^,]+]],
-// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%[^,]+]],
-// CK1-DAG: [[S]] = load [1 x i64]*, [1 x i64]** [[S_PRIV:%[^,]+]],
-// CK1-DAG: [[FN:%.+]] = bitcast void (i8*, ...)* {{%.*}} to void (i8*,
-// CK1-DAG: call void [[FN]](i8* %{{[^,]+}}, [1 x i8*]** [[BP_PRIV]], [1 x i8*]** [[P_PRIV]], [1 x i64]** [[S_PRIV]])
+// CK1: define internal{{.*}} i32 [[TASK_ENTRY3]](i32{{.*}}, ptr noalias noundef %1)
+// CK1-DAG: call void @__tgt_target_data_begin_mapper(ptr @{{.+}}, i64 -1, i32 1, ptr [[BP:%.+]], ptr [[P:%.+]], ptr [[S:%.+]], ptr [[MTYPE03]]{{.+}}, ptr null)
+// CK1-DAG: [[BP]] = load ptr, ptr [[BP_PRIV:%[^,]+]],
+// CK1-DAG: [[P]] = load ptr, ptr [[P_PRIV:%[^,]+]],
+// CK1-DAG: [[S]] = load ptr, ptr [[S_PRIV:%[^,]+]],
+// CK1-DAG: call void {{%.*}}(ptr %{{[^,]+}}, ptr [[BP_PRIV]], ptr [[P_PRIV]], ptr [[S_PRIV]])
 // CK1-NOT: __tgt_target_data_end
 // CK1: ret i32 0
 // CK1: }
 
-// CK1: define internal{{.*}} i32 [[TASK_ENTRY4]](i32{{.*}}, %struct.kmp_task_t_with_privates{{.+}}* noalias noundef %1)
-// CK1-DAG: call void @__tgt_target_data_begin_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE04]]{{.+}}, i8** null)
-// CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
-// CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
-// CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
-// CK1-DAG: [[BP]] = load [2 x i8*]*, [2 x i8*]** [[BP_PRIV:%[^,]+]],
-// CK1-DAG: [[P]] = load [2 x i8*]*, [2 x i8*]** [[P_PRIV:%[^,]+]],
-// CK1-DAG: [[S]] = load [2 x i64]*, [2 x i64]** [[S_PRIV:%[^,]+]],
-// CK1-DAG: [[FN:%.+]] = bitcast void (i8*, ...)* {{%.*}} to void (i8*,
-// CK1-DAG: call void [[FN]](i8* %{{[^,]+}}, [2 x i8*]** [[BP_PRIV]], [2 x i8*]** [[P_PRIV]], [2 x i64]** [[S_PRIV]])
+// CK1: define internal{{.*}} i32 [[TASK_ENTRY4]](i32{{.*}}, ptr noalias noundef %1)
+// CK1-DAG: call void @__tgt_target_data_begin_mapper(ptr @{{.+}}, i64 -1, i32 2, ptr [[BP:%.+]], ptr [[P:%.+]], ptr [[S:%.+]], ptr [[MTYPE04]]{{.+}}, ptr null)
+// CK1-DAG: [[BP]] = load ptr, ptr [[BP_PRIV:%[^,]+]],
+// CK1-DAG: [[P]] = load ptr, ptr [[P_PRIV:%[^,]+]],
+// CK1-DAG: [[S]] = load ptr, ptr [[S_PRIV:%[^,]+]],
+// CK1-DAG: call void {{%.*}}(ptr %{{[^,]+}}, ptr [[BP_PRIV]], ptr [[P_PRIV]], ptr [[S_PRIV]])
 // CK1-NOT: __tgt_target_data_end
 // CK1: ret i32 0
 // CK1: }
diff --git a/clang/test/OpenMP/target_exit_data_codegen.cpp b/clang/test/OpenMP/target_exit_data_codegen.cpp
index 77c1ae87ec49..df0d8023371f 100644
--- a/clang/test/OpenMP/target_exit_data_codegen.cpp
+++ b/clang/test/OpenMP/target_exit_data_codegen.cpp
@@ -3,23 +3,23 @@
 #define HEADER
 
 ///==========================================================================///
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
-
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -DCK1 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64
+// RUN: %clang_cc1 -DCK1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-64
+// RUN: %clang_cc1 -DCK1 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
+// RUN: %clang_cc1 -DCK1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
+
+// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -DCK1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -DCK1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
 #ifdef CK1
 
-// CK1: [[ST:%.+]] = type { i32, double* }
+// CK1: [[ST:%.+]] = type { i32, ptr }
 template <typename T>
 struct ST {
   T a;
@@ -29,11 +29,11 @@ struct ST {
 ST<int> gb;
 double gc[100];
 
-// CK1: [[IDENT_T:%.+]] = type { i32, i32, i32, i32, i8* }
+// CK1: [[IDENT_T:%.+]] = type { i32, i32, i32, i32, ptr }
 // CK1: [[KMP_TASK_T_WITH_PRIVATES:%.+]] = type { [[KMP_TASK_T:%[^,]+]], [[KMP_PRIVATES_T:%.+]] }
-// CK1: [[KMP_TASK_T]] = type { i8*, i32 (i32, i8*)*, i32, %{{[^,]+}}, %{{[^,]+}} }
-// CK1-32: [[KMP_PRIVATES_T]] = type { [1 x i64], [1 x i8*], [1 x i8*] }
-// CK1-64: [[KMP_PRIVATES_T]] = type { [1 x i8*], [1 x i8*], [1 x i64] }
+// CK1: [[KMP_TASK_T]] = type { ptr, ptr, i32, %{{[^,]+}}, %{{[^,]+}} }
+// CK1-32: [[KMP_PRIVATES_T]] = type { [1 x i64], [1 x ptr], [1 x ptr] }
+// CK1-64: [[KMP_PRIVATES_T]] = type { [1 x ptr], [1 x ptr], [1 x i64] }
 
 // CK1: [[SIZE00:@.+]] = {{.+}}constant [1 x i64] [i64 800]
 // CK1: [[MTYPE00:@.+]] = {{.+}}constant [1 x i64] [i64 2]
@@ -57,33 +57,25 @@ void foo(int arg) {
 
   // Region 00
   // CK1-NOT: __tgt_target_data_begin
-  // CK1-DAG: call i32 @__kmpc_omp_task([[IDENT_T]]* @{{[^,]+}}, i32 %{{[^,]+}}, i8* [[TASK:%.+]])
-  // CK1-DAG: [[TASK]] = call i8* @__kmpc_omp_target_task_alloc([[IDENT_T]]* @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i[[sz:32|64]] {{36|64}}, i{{32|64}} 4, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_T_WITH_PRIVATES]]*)* [[OMP_TASK_ENTRY:@[^,]+]] to i32 (i32, i8*)*), i64 [[DEV:%.+]])
+  // CK1-DAG: call i32 @__kmpc_omp_task(ptr @{{[^,]+}}, i32 %{{[^,]+}}, ptr [[TASK:%.+]])
+  // CK1-DAG: [[TASK]] = call ptr @__kmpc_omp_target_task_alloc(ptr @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i[[sz:32|64]] {{36|64}}, i{{32|64}} 4, ptr [[OMP_TASK_ENTRY:@[^,]+]], i64 [[DEV:%.+]])
   // CK1-DAG: [[DEV]] = sext i32 [[DEV32:%.+]] to i64
-  // CK1-DAG: [[TASK_WITH_PRIVATES:%.+]] = bitcast i8* [[TASK]] to [[KMP_TASK_T_WITH_PRIVATES]]*
-  // CK1-DAG: [[PRIVATES:%.+]] = getelementptr inbounds [[KMP_TASK_T_WITH_PRIVATES]], [[KMP_TASK_T_WITH_PRIVATES]]* [[TASK_WITH_PRIVATES]], i32 0, i32 1
-  // CK1-32-DAG: [[FPBPGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], [[KMP_PRIVATES_T]]* [[PRIVATES]], i32 0, i32 1
-  // CK1-64-DAG: [[FPBPGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], [[KMP_PRIVATES_T]]* [[PRIVATES]], i32 0, i32 0
-  // CK1-DAG: [[FPBPADDR:%.+]] = bitcast [1 x i8*]* [[FPBPGEP]] to i8*
-  // CK1-DAG: [[BPADDR:%.+]] = bitcast i8** [[BPGEP:%.+]] to i8*
-  // CK1-DAG: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{4|8}} [[FPBPADDR]], i8* align {{4|8}} [[BPADDR]], i[[sz]] {{4|8}}, i1 false)
-  // CK1-DAG: [[BPGEP]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP:%.+]], i32 0, i32 0
-  // CK1-DAG: [[BPGEP:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP]], i32 0, i32 0
-  // CK1-DAG: [[BPADDR:%.+]] = bitcast i8** [[BPGEP]] to [100 x double]**
-  // CK1-DAG: store [100 x double]* [[GC:@[^,]+]], [100 x double]** [[BPADDR]], align
-  // CK1-32-DAG: [[FPPGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], [[KMP_PRIVATES_T]]* [[PRIVATES]], i32 0, i32 2
-  // CK1-64-DAG: [[FPPGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], [[KMP_PRIVATES_T]]* [[PRIVATES]], i32 0, i32 1
-  // CK1-DAG: [[FPPADDR:%.+]] = bitcast [1 x i8*]* [[FPPGEP]] to i8*
-  // CK1-DAG: [[PADDR:%.+]] = bitcast i8** [[PGEP:%.+]] to i8*
-  // CK1-DAG: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{4|8}} [[FPPADDR]], i8* align {{4|8}} [[PADDR]], i[[sz]] {{4|8}}, i1 false)
-  // CK1-DAG: [[PGEP]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P:%.+]], i32 0, i32 0
-  // CK1-DAG: [[PGEP:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P]], i32 0, i32 0
-  // CK1-DAG: [[PADDR:%.+]] = bitcast i8** [[PGEP]] to [100 x double]**
-  // CK1-DAG: store [100 x double]* [[GC]], [100 x double]** [[PADDR]], align
-  // CK1-32-DAG: [[FPSZGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], [[KMP_PRIVATES_T]]* [[PRIVATES]], i32 0, i32 0
-  // CK1-64-DAG: [[FPSZGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], [[KMP_PRIVATES_T]]* [[PRIVATES]], i32 0, i32 2
-  // CK1-DAG: [[FPSZADDR:%.+]] = bitcast [1 x i64]* [[FPSZGEP]] to i8*
-  // CK1-DAG: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{4|8}} [[FPSZADDR]], i8* align {{4|8}} bitcast ([1 x i64]* [[SIZE00]] to i8*), i[[sz]] {{4|8}}, i1 false)
+  // CK1-DAG: [[PRIVATES:%.+]] = getelementptr inbounds [[KMP_TASK_T_WITH_PRIVATES]], ptr [[TASK]], i32 0, i32 1
+  // CK1-32-DAG: [[FPBPGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], ptr [[PRIVATES]], i32 0, i32 1
+  // CK1-64-DAG: [[FPBPGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], ptr [[PRIVATES]], i32 0, i32 0
+  // CK1-DAG: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{4|8}} [[FPBPGEP]], ptr align {{4|8}} [[BPGEP:%.+]], i[[sz]] {{4|8}}, i1 false)
+  // CK1-DAG: [[BPGEP]] = getelementptr inbounds [1 x ptr], ptr [[BP:%.+]], i32 0, i32 0
+  // CK1-DAG: [[BPGEP:%.+]] = getelementptr inbounds [1 x ptr], ptr [[BP]], i32 0, i32 0
+  // CK1-DAG: store ptr [[GC:@[^,]+]], ptr [[BPGEP]], align
+  // CK1-32-DAG: [[FPPGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], ptr [[PRIVATES]], i32 0, i32 2
+  // CK1-64-DAG: [[FPPGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], ptr [[PRIVATES]], i32 0, i32 1
+  // CK1-DAG: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{4|8}} [[FPPGEP]], ptr align {{4|8}} [[PGEP:%.+]], i[[sz]] {{4|8}}, i1 false)
+  // CK1-DAG: [[PGEP]] = getelementptr inbounds [1 x ptr], ptr [[P:%.+]], i32 0, i32 0
+  // CK1-DAG: [[PGEP:%.+]] = getelementptr inbounds [1 x ptr], ptr [[P]], i32 0, i32 0
+  // CK1-DAG: store ptr [[GC]], ptr [[PGEP]], align
+  // CK1-32-DAG: [[FPSZGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], ptr [[PRIVATES]], i32 0, i32 0
+  // CK1-64-DAG: [[FPSZGEP:%.+]] = getelementptr inbounds [[KMP_PRIVATES_T]], ptr [[PRIVATES]], i32 0, i32 2
+  // CK1-DAG: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{4|8}} [[FPSZGEP]], ptr align {{4|8}} [[SIZE00]], i[[sz]] {{4|8}}, i1 false)
   #pragma omp target exit data if(1+3-5) device(arg) map(from: gc) nowait
   {++arg;}
 
@@ -96,16 +88,14 @@ void foo(int arg) {
   // CK1-NOT: __tgt_target_data_begin
   // CK1: br i1 %{{[^,]+}}, label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]]
   // CK1: [[IFTHEN]]
-  // CK1-DAG: call void @__tgt_target_data_end_mapper(%struct.ident_t* @{{.+}}, i64 4, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[SIZE02]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE02]]{{.+}}, i8** null)
+  // CK1-DAG: call void @__tgt_target_data_end_mapper(ptr @{{.+}}, i64 4, i32 1, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[SIZE02]], ptr [[MTYPE02]]{{.+}}, ptr null)
   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
   // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
 
   // CK1-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
-  // CK1-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to i32**
-  // CK1-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to i32**
-  // CK1-DAG: store i32* [[VAL0:%[^,]+]], i32** [[CBP0]]
-  // CK1-DAG: store i32* [[VAL0]], i32** [[CP0]]
+  // CK1-DAG: store ptr [[VAL0:%[^,]+]], ptr [[BP0]]
+  // CK1-DAG: store ptr [[VAL0]], ptr [[P0]]
   // CK1: br label %[[IFEND:[^,]+]]
 
   // CK1: [[IFELSE]]
@@ -120,7 +110,7 @@ void foo(int arg) {
 
   // Region 03
   // CK1-NOT: __tgt_target_data_begin
-  // CK1-DAG: call void @__tgt_target_data_end_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE03]]{{.+}}, i8** null)
+  // CK1-DAG: call void @__tgt_target_data_end_mapper(ptr @{{.+}}, i64 -1, i32 1, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE03]]{{.+}}, ptr null)
   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
   // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
   // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
@@ -128,11 +118,9 @@ void foo(int arg) {
   // CK1-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0
-  // CK1-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to float**
-  // CK1-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to float**
-  // CK1-DAG: store float* [[VAL0:%[^,]+]], float** [[CBP0]]
-  // CK1-DAG: store float* [[VAL0]], float** [[CP0]]
-  // CK1-DAG: store i64 [[CSVAL0:%[^,]+]], i64* [[S0]]
+  // CK1-DAG: store ptr [[VAL0:%[^,]+]], ptr [[BP0]]
+  // CK1-DAG: store ptr [[VAL0]], ptr [[P0]]
+  // CK1-DAG: store i64 [[CSVAL0:%[^,]+]], ptr [[S0]]
   // CK1-64-DAG: [[CSVAL0]] = mul nuw i64 %{{[^,]+}}, 4
   // CK1-32-DAG: [[CSVAL0]] = sext i32 [[CSVAL032:%.+]] to i64
   // CK1-32-DAG: [[CSVAL032]] = mul nuw i32 %{{[^,]+}}, 4
@@ -145,7 +133,7 @@ void foo(int arg) {
 
   // Region 04
   // CK1-NOT: __tgt_target_data_begin
-  // CK1-DAG: call void @__tgt_target_data_end_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE04]]{{.+}}, i8** null)
+  // CK1-DAG: call void @__tgt_target_data_end_mapper(ptr @{{.+}}, i64 -1, i32 2, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE04]]{{.+}}, ptr null)
   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
   // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
   // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[PS:%[^,]+]]
@@ -153,22 +141,18 @@ void foo(int arg) {
   // CK1-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[PS0:%.+]] = getelementptr inbounds {{.+}}[[PS]], i{{.+}} 0, i{{.+}} 0
-  // CK1-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[ST]]**
-  // CK1-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to double***
-  // CK1-DAG: store [[ST]]* @gb, [[ST]]** [[CBP0]]
-  // CK1-DAG: store double** getelementptr inbounds ([[ST]], [[ST]]* @gb, i32 0, i32 1), double*** [[CP0]]
-  // CK1-DAG: [[DIV:%.+]] = sdiv exact i64 sub (i64 ptrtoint (double** getelementptr (double*, double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1) to i64)), ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)
-  // CK1-DAG: store i64 [[DIV]], i64* [[PS0]],
+  // CK1-DAG: store ptr @gb, ptr [[BP0]]
+  // CK1-DAG: store ptr getelementptr inbounds ([[ST]], ptr @gb, i32 0, i32 1), ptr [[P0]]
+  // CK1-DAG: [[DIV:%.+]] = sdiv exact i64 sub (i64 ptrtoint (ptr getelementptr (ptr, ptr getelementptr inbounds (%struct.ST, ptr @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (ptr getelementptr inbounds (%struct.ST, ptr @gb, i32 0, i32 1) to i64)), ptrtoint (ptr getelementptr (i8, ptr null, i32 1) to i64)
+  // CK1-DAG: store i64 [[DIV]], ptr [[PS0]],
 
 
   // CK1-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1
   // CK1-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1
-  // CK1-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to double***
-  // CK1-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to double**
-  // CK1-DAG: store double** getelementptr inbounds ([[ST]], [[ST]]* @gb, i32 0, i32 1), double*** [[CBP1]]
-  // CK1-DAG: store double* [[SEC1:%[^,]+]], double** [[CP1]]
-  // CK1-DAG: [[SEC1]] = getelementptr inbounds {{.+}}double* [[SEC11:%[^,]+]], i{{.+}} 0
-  // CK1-DAG: [[SEC11]] = load double*, double** getelementptr inbounds ([[ST]], [[ST]]* @gb, i32 0, i32 1),
+  // CK1-DAG: store ptr getelementptr inbounds ([[ST]], ptr @gb, i32 0, i32 1), ptr [[BP1]]
+  // CK1-DAG: store ptr [[SEC1:%[^,]+]], ptr [[P1]]
+  // CK1-DAG: [[SEC1]] = getelementptr inbounds {{.+}}ptr [[SEC11:%[^,]+]], i{{.+}} 0
+  // CK1-DAG: [[SEC11]] = load ptr, ptr getelementptr inbounds ([[ST]], ptr @gb, i32 0, i32 1),
 
   // CK1: %{{.+}} = add nsw i32 %{{[^,]+}}, 1
   #pragma omp target exit data map(release: gb.b[:3])
@@ -179,7 +163,7 @@ void foo(int arg) {
 
   // Region 05
   // CK1-NOT: __tgt_target_data_begin
-  // CK1-DAG: call void @__tgt_target_data_end_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE05]]{{.+}}, i8** null)
+  // CK1-DAG: call void @__tgt_target_data_end_mapper(ptr @{{.+}}, i64 -1, i32 1, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE05]]{{.+}}, ptr null)
   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
   // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
   // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
@@ -187,11 +171,9 @@ void foo(int arg) {
   // CK1-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0
-  // CK1-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to float**
-  // CK1-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to float**
-  // CK1-DAG: store float* [[VAL0:%[^,]+]], float** [[CBP0]]
-  // CK1-DAG: store float* [[VAL0]], float** [[CP0]]
-  // CK1-DAG: store i64 [[CSVAL0:%[^,]+]], i64* [[S0]]
+  // CK1-DAG: store ptr [[VAL0:%[^,]+]], ptr [[BP0]]
+  // CK1-DAG: store ptr [[VAL0]], ptr [[P0]]
+  // CK1-DAG: store i64 [[CSVAL0:%[^,]+]], ptr [[S0]]
   // CK1-64-DAG: [[CSVAL0]] = mul nuw i64 %{{[^,]+}}, 4
   // CK1-32-DAG: [[CSVAL0]] = sext i32 [[CSVAL032:%.+]] to i64
   // CK1-32-DAG: [[CSVAL032]] = mul nuw i32 %{{[^,]+}}, 4
@@ -204,7 +186,7 @@ void foo(int arg) {
 
   // Region 06
   // CK1-NOT: __tgt_target_data_begin
-  // CK1-DAG: call void @__tgt_target_data_end_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE06]]{{.+}}, i8** null)
+  // CK1-DAG: call void @__tgt_target_data_end_mapper(ptr @{{.+}}, i64 -1, i32 1, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE06]]{{.+}}, ptr null)
   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
   // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
   // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
@@ -212,11 +194,9 @@ void foo(int arg) {
   // CK1-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
   // CK1-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0
-  // CK1-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to float**
-  // CK1-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to float**
-  // CK1-DAG: store float* [[VAL0:%[^,]+]], float** [[CBP0]]
-  // CK1-DAG: store float* [[VAL0]], float** [[CP0]]
-  // CK1-DAG: store i64 [[CSVAL0:%[^,]+]], i64* [[S0]]
+  // CK1-DAG: store ptr [[VAL0:%[^,]+]], ptr [[BP0]]
+  // CK1-DAG: store ptr [[VAL0]], ptr [[P0]]
+  // CK1-DAG: store i64 [[CSVAL0:%[^,]+]], ptr [[S0]]
   // CK1-64-DAG: [[CSVAL0]] = mul nuw i64 %{{[^,]+}}, 4
   // CK1-32-DAG: [[CSVAL0]] = sext i32 [[CSVAL032:%.+]] to i64
   // CK1-32-DAG: [[CSVAL032]] = mul nuw i32 %{{[^,]+}}, 4
@@ -225,38 +205,34 @@ void foo(int arg) {
   {++arg;}
 }
 
-// CK1:     define internal {{.*}}i32 [[OMP_TASK_ENTRY]](i32 {{.*}}%{{[^,]+}}, [[KMP_TASK_T_WITH_PRIVATES]]* noalias noundef %{{[^,]+}})
-// CK1-DAG: call void @__tgt_target_data_end_nowait_mapper(%struct.ident_t* @{{.+}}, i64 %{{[^,]+}}, i32 1, i8** [[BP:%[^,]+]], i8** [[P:%[^,]+]], i64* [[SZ:%[^,]+]], i64* getelementptr inbounds ([1 x i64], [1 x i64]* [[MTYPE00]], i32 0, i32 0), i8** null, i8** null)
-// CK1-DAG: [[BP]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BPADDR:%[^,]+]], i[[sz]] 0, i[[sz]] 0
-// CK1-DAG: [[P]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[PADDR:%[^,]+]], i[[sz]] 0, i[[sz]] 0
-// CK1-DAG: [[SZ]] = getelementptr inbounds [1 x i64], [1 x i64]* [[SZADDR:%[^,]+]], i[[sz]] 0, i[[sz]] 0
-// CK1-DAG: [[BPADDR]] = load [1 x i8*]*, [1 x i8*]** [[FPBPADDR:%[^,]+]], align
-// CK1-DAG: [[PADDR]] = load [1 x i8*]*, [1 x i8*]** [[FPPADDR:%[^,]+]], align
-// CK1-DAG: [[SZADDR]] = load [1 x i64]*, [1 x i64]** [[FPSZADDR:%[^,]+]], align
-// CK1-DAG: [[FN:%.+]] = bitcast void (i8*, ...)* {{%.*}} to void (i8*,
-// CK1-DAG: call void [[FN]](i8* %{{[^,]+}}, [1 x i8*]** [[FPBPADDR]], [1 x i8*]** [[FPPADDR]], [1 x i64]** [[FPSZADDR]])
+// CK1:     define internal {{.*}}i32 [[OMP_TASK_ENTRY]](i32 {{.*}}%{{[^,]+}}, ptr noalias noundef %{{[^,]+}})
+// CK1-DAG: call void @__tgt_target_data_end_nowait_mapper(ptr @{{.+}}, i64 %{{[^,]+}}, i32 1, ptr [[BP:%[^,]+]], ptr [[P:%[^,]+]], ptr [[SZ:%[^,]+]], ptr [[MTYPE00]], ptr null, ptr null)
+// CK1-DAG: [[BP]] = load ptr, ptr [[FPBPADDR:%[^,]+]], align
+// CK1-DAG: [[P]] = load ptr, ptr [[FPPADDR:%[^,]+]], align
+// CK1-DAG: [[SZ]] = load ptr, ptr [[FPSZADDR:%[^,]+]], align
+// CK1-DAG: call void {{%.*}}(ptr %{{[^,]+}}, ptr [[FPBPADDR]], ptr [[FPPADDR]], ptr [[FPSZADDR]])
 // CK1:     ret i32 0
 // CK1:     }
 
 #endif
 ///==========================================================================///
-// RUN: %clang_cc1 -no-opaque-pointers -DCK2 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK2 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK2 --check-prefix CK2-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK2 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK2 --check-prefix CK2-32
-// RUN: %clang_cc1 -no-opaque-pointers -DCK2 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK2 --check-prefix CK2-32
-
-// RUN: %clang_cc1 -no-opaque-pointers -DCK2 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK2 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK2 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK2 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -DCK2 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-64
+// RUN: %clang_cc1 -DCK2 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK2 --check-prefix CK2-64
+// RUN: %clang_cc1 -DCK2 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK2 --check-prefix CK2-32
+// RUN: %clang_cc1 -DCK2 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK2 --check-prefix CK2-32
+
+// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -DCK2 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -DCK2 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
 // SIMD-ONLY1-NOT: {{__kmpc|__tgt}}
 #ifdef CK2
 
-// CK2: [[ST:%.+]] = type { i32, double* }
+// CK2: [[ST:%.+]] = type { i32, ptr }
 template <typename T>
 struct ST {
   T a;
@@ -283,9 +259,9 @@ int bar(int arg){
 // CK2-NOT: __tgt_target_data_begin
 // CK2: br i1 %{{[^,]+}}, label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]]
 // CK2: [[IFTHEN]]
-// CK2-DAG: call void @__tgt_target_data_end_mapper(%struct.ident_t* @{{.+}}, i64 [[DEV:%[^,]+]], i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE00]]{{.+}}, i8** null)
+// CK2-DAG: call void @__tgt_target_data_end_mapper(ptr @{{.+}}, i64 [[DEV:%[^,]+]], i32 2, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE00]]{{.+}}, ptr null)
 // CK2-DAG: [[DEV]] = sext i32 [[DEVi32:%[^,]+]] to i64
-// CK2-DAG: [[DEVi32]] = load i32, i32* %{{[^,]+}},
+// CK2-DAG: [[DEVi32]] = load i32, ptr %{{[^,]+}},
 // CK2-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
 // CK2-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
 // CK2-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[PS:%[^,]+]]
@@ -293,22 +269,18 @@ int bar(int arg){
 // CK2-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
 // CK2-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
 // CK2-DAG: [[PS0:%.+]] = getelementptr inbounds {{.+}}[[PS]], i{{.+}} 0, i{{.+}} 0
-// CK2-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[ST]]**
-// CK2-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to double***
-// CK2-DAG: store [[ST]]* [[VAR0:%[^,]+]], [[ST]]** [[CBP0]]
-// CK2-DAG: store double** [[SEC0:%[^,]+]], double*** [[CP0]]
-// CK2-DAG: store i64 [[CSVAL0:%[^,]+]], i64* [[PS0]],
-// CK2-DAG: [[SEC0]] = getelementptr inbounds {{.*}}[[ST]]* [[VAR0]], i32 0, i32 1
+// CK2-DAG: store ptr [[VAR0:%[^,]+]], ptr [[BP0]]
+// CK2-DAG: store ptr [[SEC0:%[^,]+]], ptr [[P0]]
+// CK2-DAG: store i64 [[CSVAL0:%[^,]+]], ptr [[PS0]],
+// CK2-DAG: [[SEC0]] = getelementptr inbounds {{.*}}ptr [[VAR0]], i32 0, i32 1
 
 // CK2-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1
 // CK2-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1
-// CK2-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to double***
-// CK2-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to double**
-// CK2-DAG: store double** [[SEC0]], double*** [[CBP1]]
-// CK2-DAG: store double* [[SEC1:%[^,]+]], double** [[CP1]]
-// CK2-DAG: [[SEC1]] = getelementptr inbounds {{.*}}double* [[SEC11:%[^,]+]], i{{.+}} 1
-// CK2-DAG: [[SEC11]] = load double*, double** [[SEC111:%[^,]+]],
-// CK2-DAG: [[SEC111]] = getelementptr inbounds {{.*}}[[ST]]* [[VAR0]], i32 0, i32 1
+// CK2-DAG: store ptr [[SEC0]], ptr [[BP1]]
+// CK2-DAG: store ptr [[SEC1:%[^,]+]], ptr [[P1]]
+// CK2-DAG: [[SEC1]] = getelementptr inbounds {{.*}}ptr [[SEC11:%[^,]+]], i{{.+}} 1
+// CK2-DAG: [[SEC11]] = load ptr, ptr [[SEC111:%[^,]+]],
+// CK2-DAG: [[SEC111]] = getelementptr inbounds {{.*}}ptr [[VAR0]], i32 0, i32 1
 
 // CK2: br label %[[IFEND:[^,]+]]
 
@@ -318,19 +290,19 @@ int bar(int arg){
 // CK2: %{{.+}} = add nsw i32 %{{[^,]+}}, 1
 #endif
 ///==========================================================================///
-// RUN: %clang_cc1 -no-opaque-pointers -DCK3 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK3 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK3 --check-prefix CK3-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK3 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK3 --check-prefix CK3-32
-// RUN: %clang_cc1 -no-opaque-pointers -DCK3 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK3 --check-prefix CK3-32
-
-// RUN: %clang_cc1 -no-opaque-pointers -DCK3 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK3 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK3 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-64
+// RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK3 --check-prefix CK3-64
+// RUN: %clang_cc1 -DCK3 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK3 --check-prefix CK3-32
+// RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK3 --check-prefix CK3-32
+
+// RUN: %clang_cc1 -DCK3 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK3 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
+// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
 // SIMD-ONLY2-NOT: {{__kmpc|__tgt}}
 #ifdef CK3
 
@@ -345,23 +317,23 @@ void no_target_devices(int arg) {
 }
 #endif
 ///==========================================================================///
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK4 --check-prefix CK4-64
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK4 --check-prefix CK4-32
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK4 --check-prefix CK4-32
-
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -DCK4 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -DCK4 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-64
+// RUN: %clang_cc1 -DCK4 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK4 --check-prefix CK4-64
+// RUN: %clang_cc1 -DCK4 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK4 --check-prefix CK4-32
+// RUN: %clang_cc1 -DCK4 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK4 --check-prefix CK4-32
+
+// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -DCK4 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -DCK4 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
 // SIMD-ONLY1-NOT: {{__kmpc|__tgt}}
 #ifdef CK4
 
-// CK4: [[STT:%.+]] = type { i32, double* }
+// CK4: [[STT:%.+]] = type { i32, ptr }
 template <typename T>
 struct STT {
   T a;
@@ -388,9 +360,9 @@ int bar(int arg){
 // CK4-NOT: __tgt_target_data_begin
 // CK4: br i1 %{{[^,]+}}, label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]]
 // CK4: [[IFTHEN]]
-// CK4-DAG: call void @__tgt_target_data_end_mapper(%struct.ident_t* @{{.+}}, i64 [[DEV:%[^,]+]], i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE00]]{{.+}}, i8** null)
+// CK4-DAG: call void @__tgt_target_data_end_mapper(ptr @{{.+}}, i64 [[DEV:%[^,]+]], i32 2, ptr [[GEPBP:%.+]], ptr [[GEPP:%.+]], ptr [[GEPS:%.+]], ptr [[MTYPE00]]{{.+}}, ptr null)
 // CK4-DAG: [[DEV]] = sext i32 [[DEVi32:%[^,]+]] to i64
-// CK4-DAG: [[DEVi32]] = load i32, i32* %{{[^,]+}},
+// CK4-DAG: [[DEVi32]] = load i32, ptr %{{[^,]+}},
 // CK4-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
 // CK4-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
 // CK4-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[PS:%[^,]+]]
@@ -398,22 +370,18 @@ int bar(int arg){
 // CK4-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
 // CK4-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
 // CK4-DAG: [[PS0:%.+]] = getelementptr inbounds {{.+}}[[PS]], i{{.+}} 0, i{{.+}} 0
-// CK4-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[STT]]**
-// CK4-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to double***
-// CK4-DAG: store [[STT]]* [[VAR0:%[^,]+]], [[STT]]** [[CBP0]]
-// CK4-DAG: store double** [[SEC0:%[^,]+]], double*** [[CP0]]
-// CK4-DAG: store i64 [[CSVAL0:%[^,]+]], i64* [[PS0]],
-// CK4-DAG: [[SEC0]] = getelementptr inbounds {{.*}}[[STT]]* [[VAR0]], i32 0, i32 1
+// CK4-DAG: store ptr [[VAR0:%[^,]+]], ptr [[BP0]]
+// CK4-DAG: store ptr [[SEC0:%[^,]+]], ptr [[P0]]
+// CK4-DAG: store i64 [[CSVAL0:%[^,]+]], ptr [[PS0]],
+// CK4-DAG: [[SEC0]] = getelementptr inbounds {{.*}}ptr [[VAR0]], i32 0, i32 1
 
 // CK4-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1
 // CK4-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1
-// CK4-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to double***
-// CK4-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to double**
-// CK4-DAG: store double** [[SEC0]], double*** [[CBP1]]
-// CK4-DAG: store double* [[SEC1:%[^,]+]], double** [[CP1]]
-// CK4-DAG: [[SEC1]] = getelementptr inbounds {{.*}}double* [[SEC11:%[^,]+]], i{{.+}} 1
-// CK4-DAG: [[SEC11]] = load double*, double** [[SEC111:%[^,]+]],
-// CK4-DAG: [[SEC111]] = getelementptr inbounds {{.*}}[[STT]]* [[VAR0]], i32 0, i32 1
+// CK4-DAG: store ptr [[SEC0]], ptr [[BP1]]
+// CK4-DAG: store ptr [[SEC1:%[^,]+]], ptr [[P1]]
+// CK4-DAG: [[SEC1]] = getelementptr inbounds {{.*}}ptr [[SEC11:%[^,]+]], i{{.+}} 1
+// CK4-DAG: [[SEC11]] = load ptr, ptr [[SEC111:%[^,]+]],
+// CK4-DAG: [[SEC111]] = getelementptr inbounds {{.*}}ptr [[VAR0]], i32 0, i32 1
 
 // CK4: br label %[[IFEND:[^,]+]]
 
diff --git a/clang/test/OpenMP/target_exit_data_depend_codegen.cpp b/clang/test/OpenMP/target_exit_data_depend_codegen.cpp
index d562146e7e95..3313398b6a56 100644
--- a/clang/test/OpenMP/target_exit_data_depend_codegen.cpp
+++ b/clang/test/OpenMP/target_exit_data_depend_codegen.cpp
@@ -1,20 +1,20 @@
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-64
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-64
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
 
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
 
 // expected-no-diagnostics
-// CK1: [[ST:%.+]] = type { i32, double* }
+// CK1: [[ST:%.+]] = type { i32, ptr }
 // CK1: %struct.kmp_depend_info = type { i[[sz:64|32]],
 // CK1-SAME: i[[sz]], i8 }
 #ifndef HEADER
@@ -51,59 +51,44 @@ void foo(int arg) {
   // CK1: alloca [5 x %struct.kmp_depend_info],
 
   // Region 00
-  // CK1: [[BP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP:%.+]], i32 0, i32 0
-  // CK1: [[BP0_BC:%.+]] = bitcast i8** [[BP0]] to [100 x double]**
-  // CK1: store [100 x double]* @gc, [100 x double]** [[BP0_BC]],
-  // CK1: [[P0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P:%.+]], i32 0, i32 0
-  // CK1: [[P0_BC:%.+]] = bitcast i8** [[P0]] to [100 x double]**
-  // CK1: store [100 x double]* @gc, [100 x double]** [[P0_BC]],
-  // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP]], i32 0, i32 0
-  // CK1: [[GEPP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P]], i32 0, i32 0
-  // CK1: [[CAP_DEVICE:%.+]] = getelementptr inbounds %struct.anon, %struct.anon* [[CAPTURES:%.+]], i32 0, i32 0
-  // CK1: [[DEVICE:%.+]] = load i32, i32* %{{.+}}
-  // CK1: store i32 [[DEVICE]], i32* [[CAP_DEVICE]],
-  // CK1: [[DEV1:%.+]] = load i32, i32* %{{.+}}
+  // CK1: [[BP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[BP:%.+]], i32 0, i32 0
+  // CK1: store ptr @gc, ptr [[BP0]],
+  // CK1: [[P0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[P:%.+]], i32 0, i32 0
+  // CK1: store ptr @gc, ptr [[P0]],
+  // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[BP]], i32 0, i32 0
+  // CK1: [[GEPP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[P]], i32 0, i32 0
+  // CK1: [[CAP_DEVICE:%.+]] = getelementptr inbounds %struct.anon, ptr [[CAPTURES:%.+]], i32 0, i32 0
+  // CK1: [[DEVICE:%.+]] = load i32, ptr %{{.+}}
+  // CK1: store i32 [[DEVICE]], ptr [[CAP_DEVICE]],
+  // CK1: [[DEV1:%.+]] = load i32, ptr %{{.+}}
   // CK1: [[DEV2:%.+]] = sext i32 [[DEV1]] to i64
-  // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_target_task_alloc(%struct.ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{64|36}}, i[[sz]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates*)* [[TASK_ENTRY0:@.+]] to i32 (i32, i8*)*), i64 [[DEV2]])
-  // CK1: [[BC:%.+]] = bitcast i8* [[RES]] to %struct.kmp_task_t_with_privates*
-  // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates, %struct.kmp_task_t_with_privates* [[BC]], i32 0, i32 0
-  // CK1: [[SHAREDS:%.+]] = getelementptr inbounds %struct.kmp_task_t, %struct.kmp_task_t* [[TASK_T]], i32 0, i32 0
-  // CK1: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS]],
-  // CK1: [[BC1:%.+]] = bitcast %struct.anon* [[CAPTURES]] to i8*
-  // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align 4 [[SHAREDS_REF]], i8* align 4 [[BC1]], i[[sz]] 4, i1 false)
-  // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates, %struct.kmp_task_t_with_privates* [[BC]], i32 0, i32 1
-  // CK1-64: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 0
-  // CK1-64: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8*
-  // CK1-64: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-64: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 1
-  // CK1-64: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8*
-  // CK1-64: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-64: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 2
-  // CK1-64: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i[[sz]]]* [[PRIVS_SIZES]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} bitcast ([1 x i[[sz]]]* [[SIZE00]] to i8*), i[[sz]] {{8|4}}, i1 false)
-  // CK1-32: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 0
-  // CK1-32: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i64]* [[PRIVS_SIZES]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} bitcast ([1 x i64]* [[SIZE00]] to i8*), i[[sz]] {{8|4}}, i1 false)
-  // CK1-32: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 1
-  // CK1-32: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8*
-  // CK1-32: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-32: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 2
-  // CK1-32: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8*
-  // CK1-32: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1: [[BC_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP:%.+]], i[[sz]] 0
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 4, i[[sz]]* [[DEP_SIZE]],
-  // 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 i32 @__kmpc_omp_task_with_deps(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]], i32 1, i8* [[BC]], i32 0, i8* null)
+  // CK1: [[RES:%.+]] = call ptr @__kmpc_omp_target_task_alloc(ptr {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{64|36}}, i[[sz]] 4, ptr [[TASK_ENTRY0:@.+]], i64 [[DEV2]])
+  // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates, ptr [[RES]], i32 0, i32 0
+  // CK1: [[SHAREDS:%.+]] = getelementptr inbounds %struct.kmp_task_t, ptr [[TASK_T]], i32 0, i32 0
+  // CK1: [[SHAREDS_REF:%.+]] = load ptr, ptr [[SHAREDS]],
+  // CK1: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align 4 [[SHAREDS_REF]], ptr align 4 [[CAPTURES]], i[[sz]] 4, i1 false)
+  // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates, ptr [[RES]], i32 0, i32 1
+  // CK1-64: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, ptr [[PRIVS]], i32 0, i32 0
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_BASEPTRS]], ptr align {{8|4}} [[GEPBP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-64: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, ptr [[PRIVS]], i32 0, i32 1
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_PTRS]], ptr align {{8|4}} [[GEPP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-64: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t, ptr [[PRIVS]], i32 0, i32 2
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_SIZES]], ptr align {{8|4}} [[SIZE00]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-32: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t, ptr [[PRIVS]], i32 0, i32 0
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_SIZES]], ptr align {{8|4}} [[SIZE00]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-32: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, ptr [[PRIVS]], i32 0, i32 1
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_BASEPTRS]], ptr align {{8|4}} [[GEPBP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-32: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, ptr [[PRIVS]], i32 0, i32 2
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_PTRS]], ptr align {{8|4}} [[GEPP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP:%.+]], i[[sz]] 0
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 4, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 1, ptr [[DEP_ATTRS]]
+  // CK1: = call i32 @__kmpc_omp_task_with_deps(ptr @{{.+}}, i32 %{{.+}}, ptr [[RES]], i32 1, ptr [[MAIN_DEP]], i32 0, ptr null)
 
   // CK1: %{{.+}} = add nsw i32 %{{[^,]+}}, 1
   #pragma omp target exit data if(1+3-5) device(arg) map(from:gc) nowait depend(in: arg)
@@ -117,77 +102,62 @@ void foo(int arg) {
   // Region 02
   // CK1: br i1 %{{[^,]+}}, label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]]
   // CK1: [[IFTHEN]]
-  // CK1: [[BP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP:%.+]], i32 0, i32 0
-  // CK1: [[BP0_BC:%.+]] = bitcast i8** [[BP0]] to i32**
-  // CK1: store i32* [[ARG:%.+]], i32** [[BP0_BC]],
-  // CK1: [[P0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P:%.+]], i32 0, i32 0
-  // CK1: [[P0_BC:%.+]] = bitcast i8** [[P0]] to i32**
-  // CK1: store i32* [[ARG]], i32** [[P0_BC]],
-  // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP]], i32 0, i32 0
-  // CK1: [[GEPP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P]], i32 0, i32 0
-  // CK1: [[IF_DEVICE:%.+]] = getelementptr inbounds %struct.anon{{.+}}, %struct.anon{{.+}}* [[CAPTURES:%.+]], i32 0, i32 0
-  // CK1: [[IF:%.+]] = load i8, i8* %{{.+}}
+  // CK1: [[BP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[BP:%.+]], i32 0, i32 0
+  // CK1: store ptr [[ARG:%.+]], ptr [[BP0]],
+  // CK1: [[P0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[P:%.+]], i32 0, i32 0
+  // CK1: store ptr [[ARG]], ptr [[P0]],
+  // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[BP]], i32 0, i32 0
+  // CK1: [[GEPP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[P]], i32 0, i32 0
+  // CK1: [[IF_DEVICE:%.+]] = getelementptr inbounds %struct.anon{{.+}}, ptr [[CAPTURES:%.+]], i32 0, i32 0
+  // CK1: [[IF:%.+]] = load i8, ptr %{{.+}}
   // CK1: [[IF_BOOL:%.+]] = trunc i8 [[IF]] to i1
   // CK1: [[IF:%.+]] = zext i1 [[IF_BOOL]] to i8
-  // CK1: store i8 [[IF]], i8* [[IF_DEVICE]],
-  // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{64|36}}, i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY2:@.+]] to i32 (i32, i8*)*))
-  // CK1: [[RES_BC:%.+]] = bitcast i8* [[RES]] to %struct.kmp_task_t_with_privates{{.+}}*
-  // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 0
-  // CK1: [[SHAREDS:%.+]] = getelementptr inbounds %struct.kmp_task_t, %struct.kmp_task_t* [[TASK_T]], i32 0, i32 0
-  // CK1: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS]],
-  // CK1: [[BC1:%.+]] = bitcast %struct.anon{{.+}}* [[CAPTURES]] to i8*
-  // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align 1 [[SHAREDS_REF]], i8* align 1 [[BC1]], i[[sz]] 1, i1 false)
-  // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 1
-  // CK1-64: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0
-  // CK1-64: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8*
-  // CK1-64: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-64: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1
-  // CK1-64: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8*
-  // CK1-64: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-64: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2
-  // CK1-64: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i[[sz]]]* [[PRIVS_SIZES]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} bitcast ([1 x i[[sz]]]* [[SIZE02]] to i8*), i[[sz]] {{8|4}}, i1 false)
-  // CK1-32: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0
-  // CK1-32: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i64]* [[PRIVS_SIZES]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} bitcast ([1 x i64]* [[SIZE02]] to i8*), i[[sz]] {{8|4}}, i1 false)
-  // CK1-32: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1
-  // CK1-32: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8*
-  // CK1-32: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-32: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2
-  // CK1-32: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8*
-  // CK1-32: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1: [[BC_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP:%.+]], i[[sz]] 0
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 4, i[[sz]]* [[DEP_SIZE]],
-  // 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_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 1
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 4, i[[sz]]* [[DEP_SIZE]],
-  // 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: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 2
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] ptrtoint ([100 x double]* @gc to i[[sz]]), i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 800, i[[sz]]* [[DEP_SIZE]],
-  // 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_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]])
+  // CK1: store i8 [[IF]], ptr [[IF_DEVICE]],
+  // CK1: [[RES:%.+]] = call ptr @__kmpc_omp_task_alloc(ptr {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{64|36}}, i[[sz]] 1, ptr [[TASK_ENTRY2:@.+]])
+  // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, ptr [[RES]], i32 0, i32 0
+  // CK1: [[SHAREDS:%.+]] = getelementptr inbounds %struct.kmp_task_t, ptr [[TASK_T]], i32 0, i32 0
+  // CK1: [[SHAREDS_REF:%.+]] = load ptr, ptr [[SHAREDS]],
+  // CK1: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align 1 [[SHAREDS_REF]], ptr align 1 [[CAPTURES]], i[[sz]] 1, i1 false)
+  // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, ptr [[RES]], i32 0, i32 1
+  // CK1-64: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 0
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_BASEPTRS]], ptr align {{8|4}} [[GEPBP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-64: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 1
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_PTRS]], ptr align {{8|4}} [[GEPP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-64: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 2
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_SIZES]], ptr align {{8|4}} [[SIZE02]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-32: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 0
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_SIZES]], ptr align {{8|4}} [[SIZE02]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-32: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 1
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_BASEPTRS]], ptr align {{8|4}} [[GEPBP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-32: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 2
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_PTRS]], ptr align {{8|4}} [[GEPP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP:%.+]], i[[sz]] 0
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 4, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 3, ptr [[DEP_ATTRS]]
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 1
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 4, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 3, ptr [[DEP_ATTRS]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 2
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] ptrtoint (ptr @gc to i[[sz]]), ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 800, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 3, ptr [[DEP_ATTRS]]
+  // CK1: call void @__kmpc_omp_taskwait_deps_51(ptr @{{.+}}, i32 %{{.+}}, i32 3, ptr [[MAIN_DEP]], i32 0, ptr null, i32 0)
+  // CK1: call void @__kmpc_omp_task_begin_if0(ptr @{{.+}}, i32 %{{.+}}, ptr [[RES]])
+  // CK1: = call i32 [[TASK_ENTRY2]](i32 %{{.+}}, ptr [[RES]])
+  // CK1: call void @__kmpc_omp_task_complete_if0(ptr @{{.+}}, i32 %{{.+}}, ptr [[RES]])
 
   // CK1: br label %[[IFEND:[^,]+]]
 
@@ -202,81 +172,65 @@ void foo(int arg) {
   {++arg;}
 
   // Region 03
-  // CK1: [[BP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP:%.+]], i32 0, i32 0
-  // CK1: [[BP0_BC:%.+]] = bitcast i8** [[BP0]] to float**
-  // CK1: store float* [[VLA:%.+]], float** [[BP0_BC]],
-  // CK1: [[P0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P:%.+]], i32 0, i32 0
-  // CK1: [[P0_BC:%.+]] = bitcast i8** [[P0]] to float**
-  // CK1: store float* [[VLA]], float** [[P0_BC]],
-  // CK1: [[S0:%.+]] = getelementptr inbounds [1 x i64], [1 x i64]* [[S:%.+]], i32 0, i32 0
-  // CK1: store i64 {{.+}}, i64* [[S0]],
-  // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP]], i32 0, i32 0
-  // CK1: [[GEPP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P]], i32 0, i32 0
-  // CK1: [[GEPS0:%.+]] = getelementptr inbounds [1 x i64], [1 x i64]* [[S]], i32 0, i32 0
-  // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{64|36}}, i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY3:@.+]] to i32 (i32, i8*)*))
-  // CK1: [[RES_BC:%.+]] = bitcast i8* [[RES]] to %struct.kmp_task_t_with_privates{{.+}}*
-  // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 0
-  // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 1
-  // CK1-64: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0
-  // CK1-64: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8*
-  // CK1-64: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-64: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1
-  // CK1-64: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8*
-  // CK1-64: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-64: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2
-  // CK1-64: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i[[sz]]]* [[PRIVS_SIZES]] to i8*
-  // CK1-64: [[BC_SIZES:%.+]] = bitcast i[[sz]]* [[GEPS0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} [[BC_SIZES]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-32: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0
-  // CK1-32: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i64]* [[PRIVS_SIZES]] to i8*
-  // CK1-32: [[BC_SIZES:%.+]] = bitcast i64* [[GEPS0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} [[BC_SIZES]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-32: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1
-  // CK1-32: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8*
-  // CK1-32: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-32: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2
-  // CK1-32: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8*
-  // CK1-32: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1: [[BC_ADR:%.+]] = ptrtoint float* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP:%.+]], i[[sz]] 0
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] %{{.+}}, i[[sz]]* [[DEP_SIZE]],
-  // 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_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 1
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 4, i[[sz]]* [[DEP_SIZE]],
-  // 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_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 2
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 4, i[[sz]]* [[DEP_SIZE]],
-  // 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: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 3
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] ptrtoint ([100 x double]* @gc to i[[sz]]), i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 800, i[[sz]]* [[DEP_SIZE]],
-  // 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_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]])
+  // CK1: [[BP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[BP:%.+]], i32 0, i32 0
+  // CK1: store ptr [[VLA:%.+]], ptr [[BP0]],
+  // CK1: [[P0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[P:%.+]], i32 0, i32 0
+  // CK1: store ptr [[VLA]], ptr [[P0]],
+  // CK1: [[S0:%.+]] = getelementptr inbounds [1 x i64], ptr [[S:%.+]], i32 0, i32 0
+  // CK1: store i64 {{.+}}, ptr [[S0]],
+  // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[BP]], i32 0, i32 0
+  // CK1: [[GEPP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[P]], i32 0, i32 0
+  // CK1: [[GEPS0:%.+]] = getelementptr inbounds [1 x i64], ptr [[S]], i32 0, i32 0
+  // CK1: [[RES:%.+]] = call ptr @__kmpc_omp_task_alloc(ptr {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{64|36}}, i[[sz]] 1, ptr [[TASK_ENTRY3:@.+]])
+  // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, ptr [[RES]], i32 0, i32 0
+  // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, ptr [[RES]], i32 0, i32 1
+  // CK1-64: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 0
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_BASEPTRS]], ptr align {{8|4}} [[GEPBP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-64: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 1
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_PTRS]], ptr align {{8|4}} [[GEPP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-64: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 2
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_SIZES]], ptr align {{8|4}} [[GEPS0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-32: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 0
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_SIZES]], ptr align {{8|4}} [[GEPS0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-32: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 1
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_BASEPTRS]], ptr align {{8|4}} [[GEPBP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-32: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 2
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_PTRS]], ptr align {{8|4}} [[GEPP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP:%.+]], i[[sz]] 0
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] %{{.+}}, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 3, ptr [[DEP_ATTRS]]
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 1
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 4, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 3, ptr [[DEP_ATTRS]]
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 2
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 4, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 3, ptr [[DEP_ATTRS]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 3
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] ptrtoint (ptr @gc to i[[sz]]), ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 800, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 3, ptr [[DEP_ATTRS]]
+  // CK1: call void @__kmpc_omp_taskwait_deps_51(ptr @{{.+}}, i32 %{{.+}}, i32 4, ptr [[MAIN_DEP]], i32 0, ptr null, i32 0)
+  // CK1: call void @__kmpc_omp_task_begin_if0(ptr @{{.+}}, i32 %{{.+}}, ptr [[RES]])
+  // CK1: = call i32 [[TASK_ENTRY3]](i32 %{{.+}}, ptr [[RES]])
+  // CK1: call void @__kmpc_omp_task_complete_if0(ptr @{{.+}}, i32 %{{.+}}, ptr [[RES]])
   #pragma omp target exit data map(from:lb) depend(out: lb, arg, la, gc)
   {++arg;}
 
@@ -284,153 +238,119 @@ void foo(int arg) {
   {++arg;}
 
   // Region 04
-  // CK1: [[DIV:%.+]] = sdiv exact i64 sub (i64 ptrtoint (double** getelementptr (double*, double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1) to i64)), ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)
-  // CK1: [[BP0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP:%.+]], i32 0, i32 0
-  // CK1: [[BP0_BC:%.+]] = bitcast i8** [[BP0]] to %struct.ST**
-  // CK1: store %struct.ST* @gb, %struct.ST** [[BP0_BC]],
-  // CK1: [[P0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P:%.+]], i32 0, i32 0
-  // CK1: [[P0_BC:%.+]] = bitcast i8** [[P0]] to double***
-  // CK1: store double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), double*** [[P0_BC]],
-  // CK1: [[PS0:%.+]] = getelementptr inbounds [2 x i64], [2 x i64]* [[PS:%.+]], i32 0, i32 0
-  // CK1: store i64 [[DIV]], i64* [[PS0]],
-  // CK1: [[BP1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 1
-  // CK1: [[BP1_BC:%.+]] = bitcast i8** [[BP1]] to double***
-  // CK1: store double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), double*** [[BP1_BC]],
-  // CK1: [[P1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 1
-  // CK1: [[P1_BC:%.+]] = bitcast i8** [[P1]] to double**
-  // CK1: store double* %{{.+}}, double** [[P1_BC]],
-  // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 0
-  // CK1: [[GEPP0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 0
-  // CK1: [[GEPS0:%.+]] = getelementptr inbounds [2 x i64], [2 x i64]* [[PS]], i32 0, i32 0
-  // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{88|52}}, i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY4:@.+]] to i32 (i32, i8*)*))
-  // CK1: [[RES_BC:%.+]] = bitcast i8* [[RES]] to %struct.kmp_task_t_with_privates{{.+}}*
-  // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 0
-  // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 1
-  // CK1-64: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0
-  // CK1-64: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [2 x i8*]* [[PRIVS_BASEPTRS]] to i8*
-  // CK1-64: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{16|8}}, i1 false)
-  // CK1-64: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1
-  // CK1-64: [[BC_PRIVS_PTRS:%.+]] = bitcast [2 x i8*]* [[PRIVS_PTRS]] to i8*
-  // CK1-64: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{16|8}}, i1 false)
-  // CK1-64: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2
-  // CK1-64: [[BC_PRIVS_SIZES:%.+]] = bitcast [2 x i[[sz]]]* [[PRIVS_SIZES]] to i8*
-  // CK1-64: [[BC_SIZES:%.+]] = bitcast i64* [[GEPS0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} [[BC_SIZES]], i[[sz]] {{16|8}}, i1 false)
-  // CK1-32: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0
-  // CK1-32: [[BC_PRIVS_SIZES:%.+]] = bitcast [2 x i64]* [[PRIVS_SIZES]] to i8*
-  // CK1-32: [[BC_SIZES:%.+]] = bitcast i64* [[GEPS0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} [[BC_SIZES]], i[[sz]] {{16|8}}, i1 false)
-  // CK1-32: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1
-  // CK1-32: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [2 x i8*]* [[PRIVS_BASEPTRS]] to i8*
-  // CK1-32: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{16|8}}, i1 false)
-  // CK1-32: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2
-  // CK1-32: [[BC_PRIVS_PTRS:%.+]] = bitcast [2 x i8*]* [[PRIVS_PTRS]] to i8*
-  // CK1-32: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{16|8}}, i1 false)
+  // CK1: [[DIV:%.+]] = sdiv exact i64 sub (i64 ptrtoint (ptr getelementptr (ptr, ptr getelementptr inbounds (%struct.ST, ptr @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (ptr getelementptr inbounds (%struct.ST, ptr @gb, i32 0, i32 1) to i64)), ptrtoint (ptr getelementptr (i8, ptr null, i32 1) to i64)
+  // CK1: [[BP0:%.+]] = getelementptr inbounds [2 x ptr], ptr [[BP:%.+]], i32 0, i32 0
+  // CK1: store ptr @gb, ptr [[BP0]],
+  // CK1: [[P0:%.+]] = getelementptr inbounds [2 x ptr], ptr [[P:%.+]], i32 0, i32 0
+  // CK1: store ptr getelementptr inbounds (%struct.ST, ptr @gb, i32 0, i32 1), ptr [[P0]],
+  // CK1: [[PS0:%.+]] = getelementptr inbounds [2 x i64], ptr [[PS:%.+]], i32 0, i32 0
+  // CK1: store i64 [[DIV]], ptr [[PS0]],
+  // CK1: [[BP1:%.+]] = getelementptr inbounds [2 x ptr], ptr [[BP]], i32 0, i32 1
+  // CK1: store ptr getelementptr inbounds (%struct.ST, ptr @gb, i32 0, i32 1), ptr [[BP1]],
+  // CK1: [[P1:%.+]] = getelementptr inbounds [2 x ptr], ptr [[P]], i32 0, i32 1
+  // CK1: store ptr %{{.+}}, ptr [[P1]],
+  // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [2 x ptr], ptr [[BP]], i32 0, i32 0
+  // CK1: [[GEPP0:%.+]] = getelementptr inbounds [2 x ptr], ptr [[P]], i32 0, i32 0
+  // CK1: [[GEPS0:%.+]] = getelementptr inbounds [2 x i64], ptr [[PS]], i32 0, i32 0
+  // CK1: [[RES:%.+]] = call ptr @__kmpc_omp_task_alloc(ptr {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{88|52}}, i[[sz]] 1, ptr [[TASK_ENTRY4:@.+]])
+  // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, ptr [[RES]], i32 0, i32 0
+  // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, ptr [[RES]], i32 0, i32 1
+  // CK1-64: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 0
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_BASEPTRS]], ptr align {{8|4}} [[GEPBP0]], i[[sz]] {{16|8}}, i1 false)
+  // CK1-64: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 1
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_PTRS]], ptr align {{8|4}} [[GEPP0]], i[[sz]] {{16|8}}, i1 false)
+  // CK1-64: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 2
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_SIZES]], ptr align {{8|4}} [[GEPS0]], i[[sz]] {{16|8}}, i1 false)
+  // CK1-32: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 0
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_SIZES]], ptr align {{8|4}} [[GEPS0]], i[[sz]] {{16|8}}, i1 false)
+  // CK1-32: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 1
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_BASEPTRS]], ptr align {{8|4}} [[GEPBP0]], i[[sz]] {{16|8}}, i1 false)
+  // CK1-32: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 2
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_PTRS]], ptr align {{8|4}} [[GEPP0]], i[[sz]] {{16|8}}, i1 false)
   // CK1: %{{.+}} = sub nuw
-  // CK1: [[BC_ADR:%.+]] = ptrtoint double* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP:%.+]], i[[sz]] 0
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] %{{.+}}, i[[sz]]* [[DEP_SIZE]],
-  // 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_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 1
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 4, i[[sz]]* [[DEP_SIZE]],
-  // 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_ADR:%.+]] = ptrtoint float* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 2
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] %{{.+}}, i[[sz]]* [[DEP_SIZE]],
-  // 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: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 3
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] ptrtoint ([100 x double]* @gc to i[[sz]]), i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 800, i[[sz]]* [[DEP_SIZE]],
-  // 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_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 4
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 4, i[[sz]]* [[DEP_SIZE]],
-  // 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_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]])
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP:%.+]], i[[sz]] 0
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] %{{.+}}, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 1, ptr [[DEP_ATTRS]]
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 1
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 4, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 1, ptr [[DEP_ATTRS]]
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 2
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] %{{.+}}, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 1, ptr [[DEP_ATTRS]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 3
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] ptrtoint (ptr @gc to i[[sz]]), ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 800, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 1, ptr [[DEP_ATTRS]]
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 4
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 4, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 1, ptr [[DEP_ATTRS]]
+  // CK1: call void @__kmpc_omp_taskwait_deps_51(ptr @{{.+}}, i32 %{{.+}}, i32 5, ptr [[MAIN_DEP]], i32 0, ptr null, i32 0)
+  // CK1: call void @__kmpc_omp_task_begin_if0(ptr @{{.+}}, i32 %{{.+}}, ptr [[RES]])
+  // CK1: = call i32 [[TASK_ENTRY4]](i32 %{{.+}}, ptr [[RES]])
+  // CK1: call void @__kmpc_omp_task_complete_if0(ptr @{{.+}}, i32 %{{.+}}, ptr [[RES]])
   #pragma omp target exit data map(from:gb.b[:3]) depend(in: gb.b[:3], la, lb, gc, arg)
   {++arg;}
 }
 
-// CK1: define internal{{.*}} i32 [[TASK_ENTRY0]](i32{{.*}}, %struct.kmp_task_t_with_privates* noalias noundef %1)
-// CK1-DAG: call void @__tgt_target_data_end_nowait_mapper(%struct.ident_t* @{{.+}}, i64 [[DEV:%[^,]+]], i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE00]]{{.+}}, i8** null)
+// CK1: define internal{{.*}} i32 [[TASK_ENTRY0]](i32{{.*}}, ptr noalias noundef %1)
+// CK1-DAG: call void @__tgt_target_data_end_nowait_mapper(ptr @{{.+}}, i64 [[DEV:%[^,]+]], i32 1, ptr [[BP:%.+]], ptr [[P:%.+]], ptr [[S:%.+]], ptr [[MTYPE00]]{{.+}}, ptr null)
 // CK1-DAG: [[DEV]] = sext i32 [[DEVi32:%[^,]+]] to i64
-// CK1-DAG: [[DEVi32]] = load i32, i32* %{{[^,]+}},
-// CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
-// CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
-// CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
-// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%[^,]+]],
-// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%[^,]+]],
-// CK1-DAG: [[S]] = load [1 x i64]*, [1 x i64]** [[S_PRIV:%[^,]+]],
-// CK1-DAG: [[FN:%.+]] = bitcast void (i8*, ...)* {{%.*}} to void (i8*,
-// CK1-DAG: call void [[FN]](i8* %{{[^,]+}}, [1 x i8*]** [[BP_PRIV]], [1 x i8*]** [[P_PRIV]], [1 x i64]** [[S_PRIV]])
+// CK1-DAG: [[DEVi32]] = load i32, ptr %{{[^,]+}},
+// CK1-DAG: [[BP]] = load ptr, ptr [[BP_PRIV:%[^,]+]],
+// CK1-DAG: [[P]] = load ptr, ptr [[P_PRIV:%[^,]+]],
+// CK1-DAG: [[S]] = load ptr, ptr [[S_PRIV:%[^,]+]],
+// CK1-DAG: call void {{%.*}}(ptr %{{[^,]+}}, ptr [[BP_PRIV]], ptr [[P_PRIV]], ptr [[S_PRIV]])
 // CK1: ret i32 0
 // CK1: }
 
-// CK1: define internal{{.*}} i32 [[TASK_ENTRY2]](i32{{.*}}, %struct.kmp_task_t_with_privates{{.+}}* noalias noundef %1)
-// CK1-DAG: call void @__tgt_target_data_end_mapper(%struct.ident_t* @{{.+}}, i64 4, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE02]]{{.+}}, i8** null)
-// CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
-// CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
-// CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
-// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%[^,]+]],
-// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%[^,]+]],
-// CK1-DAG: [[S]] = load [1 x i64]*, [1 x i64]** [[S_PRIV:%[^,]+]],
-// CK1-DAG: [[FN:%.+]] = bitcast void (i8*, ...)* {{%.*}} to void (i8*,
-// CK1-DAG: call void [[FN]](i8* %{{[^,]+}}, [1 x i8*]** [[BP_PRIV]], [1 x i8*]** [[P_PRIV]], [1 x i64]** [[S_PRIV]])
+// CK1: define internal{{.*}} i32 [[TASK_ENTRY2]](i32{{.*}}, ptr noalias noundef %1)
+// CK1-DAG: call void @__tgt_target_data_end_mapper(ptr @{{.+}}, i64 4, i32 1, ptr [[BP:%.+]], ptr [[P:%.+]], ptr [[S:%.+]], ptr [[MTYPE02]]{{.+}}, ptr null)
+// CK1-DAG: [[BP]] = load ptr, ptr [[BP_PRIV:%[^,]+]],
+// CK1-DAG: [[P]] = load ptr, ptr [[P_PRIV:%[^,]+]],
+// CK1-DAG: [[S]] = load ptr, ptr [[S_PRIV:%[^,]+]],
+// CK1-DAG: call void {{%.*}}(ptr %{{[^,]+}}, ptr [[BP_PRIV]], ptr [[P_PRIV]], ptr [[S_PRIV]])
 // CK1: ret i32 0
 // CK1: }
 
-// CK1: define internal{{.*}} i32 [[TASK_ENTRY3]](i32{{.*}}, %struct.kmp_task_t_with_privates{{.+}}* noalias noundef %1)
-// CK1-DAG: call void @__tgt_target_data_end_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE03]]{{.+}}, i8** null)
-// CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
-// CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
-// CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
-// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%[^,]+]],
-// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%[^,]+]],
-// CK1-DAG: [[S]] = load [1 x i64]*, [1 x i64]** [[S_PRIV:%[^,]+]],
-// CK1-DAG: [[FN:%.+]] = bitcast void (i8*, ...)* {{%.*}} to void (i8*,
-// CK1-DAG: call void [[FN]](i8* %{{[^,]+}}, [1 x i8*]** [[BP_PRIV]], [1 x i8*]** [[P_PRIV]], [1 x i64]** [[S_PRIV]])
+// CK1: define internal{{.*}} i32 [[TASK_ENTRY3]](i32{{.*}}, ptr noalias noundef %1)
+// CK1-DAG: call void @__tgt_target_data_end_mapper(ptr @{{.+}}, i64 -1, i32 1, ptr [[BP:%.+]], ptr [[P:%.+]], ptr [[S:%.+]], ptr [[MTYPE03]]{{.+}}, ptr null)
+// CK1-DAG: [[BP]] = load ptr, ptr [[BP_PRIV:%[^,]+]],
+// CK1-DAG: [[P]] = load ptr, ptr [[P_PRIV:%[^,]+]],
+// CK1-DAG: [[S]] = load ptr, ptr [[S_PRIV:%[^,]+]],
+// CK1-DAG: call void {{%.*}}(ptr %{{[^,]+}}, ptr [[BP_PRIV]], ptr [[P_PRIV]], ptr [[S_PRIV]])
 // CK1-NOT: __tgt_target_data_end_mapper
 // CK1: ret i32 0
 // CK1: }
 
-// CK1: define internal{{.*}} i32 [[TASK_ENTRY4]](i32{{.*}}, %struct.kmp_task_t_with_privates{{.+}}* noalias noundef %1)
-// CK1-DAG: call void @__tgt_target_data_end_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE04]]{{.+}}, i8** null)
-// CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
-// CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
-// CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
-// CK1-DAG: [[BP]] = load [2 x i8*]*, [2 x i8*]** [[BP_PRIV:%[^,]+]],
-// CK1-DAG: [[P]] = load [2 x i8*]*, [2 x i8*]** [[P_PRIV:%[^,]+]],
-// CK1-DAG: [[S]] = load [2 x i64]*, [2 x i64]** [[S_PRIV:%[^,]+]],
-// CK1-DAG: [[FN:%.+]] = bitcast void (i8*, ...)* {{%.*}} to void (i8*,
-// CK1-DAG: call void [[FN]](i8* %{{[^,]+}}, [2 x i8*]** [[BP_PRIV]], [2 x i8*]** [[P_PRIV]], [2 x i64]** [[S_PRIV]])
+// CK1: define internal{{.*}} i32 [[TASK_ENTRY4]](i32{{.*}}, ptr noalias noundef %1)
+// CK1-DAG: call void @__tgt_target_data_end_mapper(ptr @{{.+}}, i64 -1, i32 2, ptr [[BP:%.+]], ptr [[P:%.+]], ptr [[S:%.+]], ptr [[MTYPE04]]{{.+}}, ptr null)
+// CK1-DAG: [[BP]] = load ptr, ptr [[BP_PRIV:%[^,]+]],
+// CK1-DAG: [[P]] = load ptr, ptr [[P_PRIV:%[^,]+]],
+// CK1-DAG: [[S]] = load ptr, ptr [[S_PRIV:%[^,]+]],
+// CK1-DAG: call void {{%.*}}(ptr %{{[^,]+}}, ptr [[BP_PRIV]], ptr [[P_PRIV]], ptr [[S_PRIV]])
 // CK1-NOT: __tgt_target_data_end_mapper
 // CK1: ret i32 0
 // CK1: }
diff --git a/clang/test/OpenMP/target_in_reduction_codegen.cpp b/clang/test/OpenMP/target_in_reduction_codegen.cpp
index ea474e65030c..7ace320289be 100644
--- a/clang/test/OpenMP/target_in_reduction_codegen.cpp
+++ b/clang/test/OpenMP/target_in_reduction_codegen.cpp
@@ -1,6 +1,6 @@
-// RUN: %clang_cc1 -no-opaque-pointers -no-enable-noundef-analysis -verify -triple x86_64-apple-darwin10 -fopenmp -x c++ -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK1
-// RUN: %clang_cc1 -no-opaque-pointers -no-enable-noundef-analysis -fopenmp -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -no-enable-noundef-analysis -fopenmp -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK1
+// RUN: %clang_cc1 -no-enable-noundef-analysis -verify -triple x86_64-apple-darwin10 -fopenmp -x c++ -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK1
+// RUN: %clang_cc1 -no-enable-noundef-analysis -fopenmp -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s
+// RUN: %clang_cc1 -no-enable-noundef-analysis -fopenmp -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK1
 // expected-no-diagnostics
 #ifndef HEADER
 #define HEADER
@@ -35,634 +35,588 @@ int main(int argc, char **argv) {
 
 #endif
 // CHECK1-LABEL: define {{[^@]+}}@main
-// CHECK1-SAME: (i32 [[ARGC:%.*]], i8** [[ARGV:%.*]]) #[[ATTR0:[0-9]+]] {
+// CHECK1-SAME: (i32 [[ARGC:%.*]], ptr [[ARGV:%.*]]) #[[ATTR0:[0-9]+]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[ARGC_ADDR:%.*]] = alloca i32, align 4
-// CHECK1-NEXT:    [[ARGV_ADDR:%.*]] = alloca i8**, align 8
+// CHECK1-NEXT:    [[ARGV_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[A:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[B:%.*]] = alloca float, align 4
 // CHECK1-NEXT:    [[C:%.*]] = alloca [5 x %struct.S], align 16
-// CHECK1-NEXT:    [[SAVED_STACK:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    [[SAVED_STACK:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[__VLA_EXPR0:%.*]] = alloca i64, align 8
 // CHECK1-NEXT:    [[DOTRD_INPUT_:%.*]] = alloca [3 x %struct.kmp_taskred_input_t], align 8
-// CHECK1-NEXT:    [[DOTTASK_RED_:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    [[DOTTASK_RED_:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTRD_INPUT_3:%.*]] = alloca [2 x %struct.kmp_taskred_input_t.0], align 8
-// CHECK1-NEXT:    [[DOTTASK_RED_6:%.*]] = alloca i8*, align 8
-// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]])
-// CHECK1-NEXT:    store i32 0, i32* [[RETVAL]], align 4
-// CHECK1-NEXT:    store i32 [[ARGC]], i32* [[ARGC_ADDR]], align 4
-// CHECK1-NEXT:    store i8** [[ARGV]], i8*** [[ARGV_ADDR]], align 8
-// CHECK1-NEXT:    [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [5 x %struct.S], [5 x %struct.S]* [[C]], i32 0, i32 0
-// CHECK1-NEXT:    [[ARRAYCTOR_END:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAY_BEGIN]], i64 5
+// CHECK1-NEXT:    [[DOTTASK_RED_6:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
+// CHECK1-NEXT:    store i32 0, ptr [[RETVAL]], align 4
+// CHECK1-NEXT:    store i32 [[ARGC]], ptr [[ARGC_ADDR]], align 4
+// CHECK1-NEXT:    store ptr [[ARGV]], ptr [[ARGV_ADDR]], align 8
+// CHECK1-NEXT:    [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [5 x %struct.S], ptr [[C]], i32 0, i32 0
+// CHECK1-NEXT:    [[ARRAYCTOR_END:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[ARRAY_BEGIN]], i64 5
 // CHECK1-NEXT:    br label [[ARRAYCTOR_LOOP:%.*]]
 // CHECK1:       arrayctor.loop:
-// CHECK1-NEXT:    [[ARRAYCTOR_CUR:%.*]] = phi %struct.S* [ [[ARRAY_BEGIN]], [[ENTRY:%.*]] ], [ [[ARRAYCTOR_NEXT:%.*]], [[ARRAYCTOR_LOOP]] ]
-// CHECK1-NEXT:    call void @_ZN1SC1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYCTOR_CUR]])
-// CHECK1-NEXT:    [[ARRAYCTOR_NEXT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYCTOR_CUR]], i64 1
-// CHECK1-NEXT:    [[ARRAYCTOR_DONE:%.*]] = icmp eq %struct.S* [[ARRAYCTOR_NEXT]], [[ARRAYCTOR_END]]
+// CHECK1-NEXT:    [[ARRAYCTOR_CUR:%.*]] = phi ptr [ [[ARRAY_BEGIN]], [[ENTRY:%.*]] ], [ [[ARRAYCTOR_NEXT:%.*]], [[ARRAYCTOR_LOOP]] ]
+// CHECK1-NEXT:    call void @_ZN1SC1Ev(ptr nonnull align 4 dereferenceable(4) [[ARRAYCTOR_CUR]])
+// CHECK1-NEXT:    [[ARRAYCTOR_NEXT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYCTOR_CUR]], i64 1
+// CHECK1-NEXT:    [[ARRAYCTOR_DONE:%.*]] = icmp eq ptr [[ARRAYCTOR_NEXT]], [[ARRAYCTOR_END]]
 // CHECK1-NEXT:    br i1 [[ARRAYCTOR_DONE]], label [[ARRAYCTOR_CONT:%.*]], label [[ARRAYCTOR_LOOP]]
 // CHECK1:       arrayctor.cont:
-// CHECK1-NEXT:    [[TMP1:%.*]] = load i32, i32* [[ARGC_ADDR]], align 4
+// CHECK1-NEXT:    [[TMP1:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
 // CHECK1-NEXT:    [[TMP2:%.*]] = zext i32 [[TMP1]] to i64
-// CHECK1-NEXT:    [[TMP3:%.*]] = call i8* @llvm.stacksave()
-// CHECK1-NEXT:    store i8* [[TMP3]], i8** [[SAVED_STACK]], align 8
+// CHECK1-NEXT:    [[TMP3:%.*]] = call ptr @llvm.stacksave()
+// CHECK1-NEXT:    store ptr [[TMP3]], ptr [[SAVED_STACK]], align 8
 // CHECK1-NEXT:    [[VLA:%.*]] = alloca i16, i64 [[TMP2]], align 16
-// CHECK1-NEXT:    store i64 [[TMP2]], i64* [[__VLA_EXPR0]], align 8
-// CHECK1-NEXT:    call void @__kmpc_taskgroup(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-// CHECK1-NEXT:    [[DOTRD_INPUT_GEP_:%.*]] = getelementptr inbounds [3 x %struct.kmp_taskred_input_t], [3 x %struct.kmp_taskred_input_t]* [[DOTRD_INPUT_]], i64 0, i64 0
-// CHECK1-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T:%.*]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 0
-// CHECK1-NEXT:    [[TMP5:%.*]] = bitcast i32* [[A]] to i8*
-// CHECK1-NEXT:    store i8* [[TMP5]], i8** [[TMP4]], align 8
-// CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 1
-// CHECK1-NEXT:    [[TMP7:%.*]] = bitcast i32* [[A]] to i8*
-// CHECK1-NEXT:    store i8* [[TMP7]], i8** [[TMP6]], align 8
-// CHECK1-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 2
-// CHECK1-NEXT:    store i64 4, i64* [[TMP8]], align 8
-// CHECK1-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 3
-// CHECK1-NEXT:    store i8* bitcast (void (i8*, i8*)* @.red_init. to i8*), i8** [[TMP9]], align 8
-// CHECK1-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 4
-// CHECK1-NEXT:    store i8* null, i8** [[TMP10]], align 8
-// CHECK1-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 5
-// CHECK1-NEXT:    store i8* bitcast (void (i8*, i8*)* @.red_comb. to i8*), i8** [[TMP11]], align 8
-// CHECK1-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 6
-// CHECK1-NEXT:    [[TMP13:%.*]] = bitcast i32* [[TMP12]] to i8*
-// CHECK1-NEXT:    call void @llvm.memset.p0i8.i64(i8* align 8 [[TMP13]], i8 0, i64 4, i1 false)
-// CHECK1-NEXT:    [[DOTRD_INPUT_GEP_1:%.*]] = getelementptr inbounds [3 x %struct.kmp_taskred_input_t], [3 x %struct.kmp_taskred_input_t]* [[DOTRD_INPUT_]], i64 0, i64 1
-// CHECK1-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_1]], i32 0, i32 0
-// CHECK1-NEXT:    [[TMP15:%.*]] = bitcast float* [[B]] to i8*
-// CHECK1-NEXT:    store i8* [[TMP15]], i8** [[TMP14]], align 8
-// CHECK1-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_1]], i32 0, i32 1
-// CHECK1-NEXT:    [[TMP17:%.*]] = bitcast float* [[B]] to i8*
-// CHECK1-NEXT:    store i8* [[TMP17]], i8** [[TMP16]], align 8
-// CHECK1-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_1]], i32 0, i32 2
-// CHECK1-NEXT:    store i64 4, i64* [[TMP18]], align 8
-// CHECK1-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_1]], i32 0, i32 3
-// CHECK1-NEXT:    store i8* bitcast (void (i8*, i8*)* @.red_init..1 to i8*), i8** [[TMP19]], align 8
-// CHECK1-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_1]], i32 0, i32 4
-// CHECK1-NEXT:    store i8* null, i8** [[TMP20]], align 8
-// CHECK1-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_1]], i32 0, i32 5
-// CHECK1-NEXT:    store i8* bitcast (void (i8*, i8*)* @.red_comb..2 to i8*), i8** [[TMP21]], align 8
-// CHECK1-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_1]], i32 0, i32 6
-// CHECK1-NEXT:    [[TMP23:%.*]] = bitcast i32* [[TMP22]] to i8*
-// CHECK1-NEXT:    call void @llvm.memset.p0i8.i64(i8* align 8 [[TMP23]], i8 0, i64 4, i1 false)
-// CHECK1-NEXT:    [[DOTRD_INPUT_GEP_2:%.*]] = getelementptr inbounds [3 x %struct.kmp_taskred_input_t], [3 x %struct.kmp_taskred_input_t]* [[DOTRD_INPUT_]], i64 0, i64 2
-// CHECK1-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_2]], i32 0, i32 0
-// CHECK1-NEXT:    [[TMP25:%.*]] = bitcast i32* [[ARGC_ADDR]] to i8*
-// CHECK1-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 8
-// CHECK1-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_2]], i32 0, i32 1
-// CHECK1-NEXT:    [[TMP27:%.*]] = bitcast i32* [[ARGC_ADDR]] to i8*
-// CHECK1-NEXT:    store i8* [[TMP27]], i8** [[TMP26]], align 8
-// CHECK1-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_2]], i32 0, i32 2
-// CHECK1-NEXT:    store i64 4, i64* [[TMP28]], align 8
-// CHECK1-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_2]], i32 0, i32 3
-// CHECK1-NEXT:    store i8* bitcast (void (i8*, i8*)* @.red_init..3 to i8*), i8** [[TMP29]], align 8
-// CHECK1-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_2]], i32 0, i32 4
-// CHECK1-NEXT:    store i8* null, i8** [[TMP30]], align 8
-// CHECK1-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_2]], i32 0, i32 5
-// CHECK1-NEXT:    store i8* bitcast (void (i8*, i8*)* @.red_comb..4 to i8*), i8** [[TMP31]], align 8
-// CHECK1-NEXT:    [[TMP32:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], %struct.kmp_taskred_input_t* [[DOTRD_INPUT_GEP_2]], i32 0, i32 6
-// CHECK1-NEXT:    [[TMP33:%.*]] = bitcast i32* [[TMP32]] to i8*
-// CHECK1-NEXT:    call void @llvm.memset.p0i8.i64(i8* align 8 [[TMP33]], i8 0, i64 4, i1 false)
-// CHECK1-NEXT:    [[TMP34:%.*]] = bitcast [3 x %struct.kmp_taskred_input_t]* [[DOTRD_INPUT_]] to i8*
-// CHECK1-NEXT:    [[TMP35:%.*]] = call i8* @__kmpc_taskred_init(i32 [[TMP0]], i32 3, i8* [[TMP34]])
-// CHECK1-NEXT:    store i8* [[TMP35]], i8** [[DOTTASK_RED_]], align 8
-// CHECK1-NEXT:    call void @__kmpc_taskgroup(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-// CHECK1-NEXT:    [[DOTRD_INPUT_GEP_4:%.*]] = getelementptr inbounds [2 x %struct.kmp_taskred_input_t.0], [2 x %struct.kmp_taskred_input_t.0]* [[DOTRD_INPUT_3]], i64 0, i64 0
-// CHECK1-NEXT:    [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0:%.*]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_4]], i32 0, i32 0
-// CHECK1-NEXT:    [[TMP37:%.*]] = bitcast [5 x %struct.S]* [[C]] to i8*
-// CHECK1-NEXT:    store i8* [[TMP37]], i8** [[TMP36]], align 8
-// CHECK1-NEXT:    [[TMP38:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_4]], i32 0, i32 1
-// CHECK1-NEXT:    [[TMP39:%.*]] = bitcast [5 x %struct.S]* [[C]] to i8*
-// CHECK1-NEXT:    store i8* [[TMP39]], i8** [[TMP38]], align 8
-// CHECK1-NEXT:    [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_4]], i32 0, i32 2
-// CHECK1-NEXT:    store i64 20, i64* [[TMP40]], align 8
-// CHECK1-NEXT:    [[TMP41:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_4]], i32 0, i32 3
-// CHECK1-NEXT:    store i8* bitcast (void (i8*, i8*)* @.red_init..5 to i8*), i8** [[TMP41]], align 8
-// CHECK1-NEXT:    [[TMP42:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_4]], i32 0, i32 4
-// CHECK1-NEXT:    store i8* bitcast (void (i8*)* @.red_fini. to i8*), i8** [[TMP42]], align 8
-// CHECK1-NEXT:    [[TMP43:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_4]], i32 0, i32 5
-// CHECK1-NEXT:    store i8* bitcast (void (i8*, i8*)* @.red_comb..6 to i8*), i8** [[TMP43]], align 8
-// CHECK1-NEXT:    [[TMP44:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_4]], i32 0, i32 6
-// CHECK1-NEXT:    [[TMP45:%.*]] = bitcast i32* [[TMP44]] to i8*
-// CHECK1-NEXT:    call void @llvm.memset.p0i8.i64(i8* align 8 [[TMP45]], i8 0, i64 4, i1 false)
-// CHECK1-NEXT:    [[DOTRD_INPUT_GEP_5:%.*]] = getelementptr inbounds [2 x %struct.kmp_taskred_input_t.0], [2 x %struct.kmp_taskred_input_t.0]* [[DOTRD_INPUT_3]], i64 0, i64 1
-// CHECK1-NEXT:    [[TMP46:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_5]], i32 0, i32 0
-// CHECK1-NEXT:    [[TMP47:%.*]] = bitcast i16* [[VLA]] to i8*
-// CHECK1-NEXT:    store i8* [[TMP47]], i8** [[TMP46]], align 8
-// CHECK1-NEXT:    [[TMP48:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_5]], i32 0, i32 1
-// CHECK1-NEXT:    [[TMP49:%.*]] = bitcast i16* [[VLA]] to i8*
-// CHECK1-NEXT:    store i8* [[TMP49]], i8** [[TMP48]], align 8
+// CHECK1-NEXT:    store i64 [[TMP2]], ptr [[__VLA_EXPR0]], align 8
+// CHECK1-NEXT:    call void @__kmpc_taskgroup(ptr @[[GLOB1]], i32 [[TMP0]])
+// CHECK1-NEXT:    [[DOTRD_INPUT_GEP_:%.*]] = getelementptr inbounds [3 x %struct.kmp_taskred_input_t], ptr [[DOTRD_INPUT_]], i64 0, i64 0
+// CHECK1-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T:%.*]], ptr [[DOTRD_INPUT_GEP_]], i32 0, i32 0
+// CHECK1-NEXT:    store ptr [[A]], ptr [[TMP4]], align 8
+// CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_]], i32 0, i32 1
+// CHECK1-NEXT:    store ptr [[A]], ptr [[TMP6]], align 8
+// CHECK1-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_]], i32 0, i32 2
+// CHECK1-NEXT:    store i64 4, ptr [[TMP8]], align 8
+// CHECK1-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_]], i32 0, i32 3
+// CHECK1-NEXT:    store ptr @.red_init., ptr [[TMP9]], align 8
+// CHECK1-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_]], i32 0, i32 4
+// CHECK1-NEXT:    store ptr null, ptr [[TMP10]], align 8
+// CHECK1-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_]], i32 0, i32 5
+// CHECK1-NEXT:    store ptr @.red_comb., ptr [[TMP11]], align 8
+// CHECK1-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_]], i32 0, i32 6
+// CHECK1-NEXT:    call void @llvm.memset.p0.i64(ptr align 8 [[TMP12]], i8 0, i64 4, i1 false)
+// CHECK1-NEXT:    [[DOTRD_INPUT_GEP_1:%.*]] = getelementptr inbounds [3 x %struct.kmp_taskred_input_t], ptr [[DOTRD_INPUT_]], i64 0, i64 1
+// CHECK1-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_1]], i32 0, i32 0
+// CHECK1-NEXT:    store ptr [[B]], ptr [[TMP14]], align 8
+// CHECK1-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_1]], i32 0, i32 1
+// CHECK1-NEXT:    store ptr [[B]], ptr [[TMP16]], align 8
+// CHECK1-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_1]], i32 0, i32 2
+// CHECK1-NEXT:    store i64 4, ptr [[TMP18]], align 8
+// CHECK1-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_1]], i32 0, i32 3
+// CHECK1-NEXT:    store ptr @.red_init..1, ptr [[TMP19]], align 8
+// CHECK1-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_1]], i32 0, i32 4
+// CHECK1-NEXT:    store ptr null, ptr [[TMP20]], align 8
+// CHECK1-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_1]], i32 0, i32 5
+// CHECK1-NEXT:    store ptr @.red_comb..2, ptr [[TMP21]], align 8
+// CHECK1-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_1]], i32 0, i32 6
+// CHECK1-NEXT:    call void @llvm.memset.p0.i64(ptr align 8 [[TMP22]], i8 0, i64 4, i1 false)
+// CHECK1-NEXT:    [[DOTRD_INPUT_GEP_2:%.*]] = getelementptr inbounds [3 x %struct.kmp_taskred_input_t], ptr [[DOTRD_INPUT_]], i64 0, i64 2
+// CHECK1-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_2]], i32 0, i32 0
+// CHECK1-NEXT:    store ptr [[ARGC_ADDR]], ptr [[TMP24]], align 8
+// CHECK1-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_2]], i32 0, i32 1
+// CHECK1-NEXT:    store ptr [[ARGC_ADDR]], ptr [[TMP26]], align 8
+// CHECK1-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_2]], i32 0, i32 2
+// CHECK1-NEXT:    store i64 4, ptr [[TMP28]], align 8
+// CHECK1-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_2]], i32 0, i32 3
+// CHECK1-NEXT:    store ptr @.red_init..3, ptr [[TMP29]], align 8
+// CHECK1-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_2]], i32 0, i32 4
+// CHECK1-NEXT:    store ptr null, ptr [[TMP30]], align 8
+// CHECK1-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_2]], i32 0, i32 5
+// CHECK1-NEXT:    store ptr @.red_comb..4, ptr [[TMP31]], align 8
+// CHECK1-NEXT:    [[TMP32:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_2]], i32 0, i32 6
+// CHECK1-NEXT:    call void @llvm.memset.p0.i64(ptr align 8 [[TMP32]], i8 0, i64 4, i1 false)
+// CHECK1-NEXT:    [[TMP35:%.*]] = call ptr @__kmpc_taskred_init(i32 [[TMP0]], i32 3, ptr [[DOTRD_INPUT_]])
+// CHECK1-NEXT:    store ptr [[TMP35]], ptr [[DOTTASK_RED_]], align 8
+// CHECK1-NEXT:    call void @__kmpc_taskgroup(ptr @[[GLOB1]], i32 [[TMP0]])
+// CHECK1-NEXT:    [[DOTRD_INPUT_GEP_4:%.*]] = getelementptr inbounds [2 x %struct.kmp_taskred_input_t.0], ptr [[DOTRD_INPUT_3]], i64 0, i64 0
+// CHECK1-NEXT:    [[TMP36:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0:%.*]], ptr [[DOTRD_INPUT_GEP_4]], i32 0, i32 0
+// CHECK1-NEXT:    store ptr [[C]], ptr [[TMP36]], align 8
+// CHECK1-NEXT:    [[TMP38:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_4]], i32 0, i32 1
+// CHECK1-NEXT:    store ptr [[C]], ptr [[TMP38]], align 8
+// CHECK1-NEXT:    [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_4]], i32 0, i32 2
+// CHECK1-NEXT:    store i64 20, ptr [[TMP40]], align 8
+// CHECK1-NEXT:    [[TMP41:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_4]], i32 0, i32 3
+// CHECK1-NEXT:    store ptr @.red_init..5, ptr [[TMP41]], align 8
+// CHECK1-NEXT:    [[TMP42:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_4]], i32 0, i32 4
+// CHECK1-NEXT:    store ptr @.red_fini., ptr [[TMP42]], align 8
+// CHECK1-NEXT:    [[TMP43:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_4]], i32 0, i32 5
+// CHECK1-NEXT:    store ptr @.red_comb..6, ptr [[TMP43]], align 8
+// CHECK1-NEXT:    [[TMP44:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_4]], i32 0, i32 6
+// CHECK1-NEXT:    call void @llvm.memset.p0.i64(ptr align 8 [[TMP44]], i8 0, i64 4, i1 false)
+// CHECK1-NEXT:    [[DOTRD_INPUT_GEP_5:%.*]] = getelementptr inbounds [2 x %struct.kmp_taskred_input_t.0], ptr [[DOTRD_INPUT_3]], i64 0, i64 1
+// CHECK1-NEXT:    [[TMP46:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_5]], i32 0, i32 0
+// CHECK1-NEXT:    store ptr [[VLA]], ptr [[TMP46]], align 8
+// CHECK1-NEXT:    [[TMP48:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_5]], i32 0, i32 1
+// CHECK1-NEXT:    store ptr [[VLA]], ptr [[TMP48]], align 8
 // CHECK1-NEXT:    [[TMP50:%.*]] = mul nuw i64 [[TMP2]], 2
-// CHECK1-NEXT:    [[TMP51:%.*]] = udiv exact i64 [[TMP50]], ptrtoint (i16* getelementptr (i16, i16* null, i32 1) to i64)
-// CHECK1-NEXT:    [[TMP52:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_5]], i32 0, i32 2
-// CHECK1-NEXT:    store i64 [[TMP50]], i64* [[TMP52]], align 8
-// CHECK1-NEXT:    [[TMP53:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_5]], i32 0, i32 3
-// CHECK1-NEXT:    store i8* bitcast (void (i8*, i8*)* @.red_init..7 to i8*), i8** [[TMP53]], align 8
-// CHECK1-NEXT:    [[TMP54:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_5]], i32 0, i32 4
-// CHECK1-NEXT:    store i8* null, i8** [[TMP54]], align 8
-// CHECK1-NEXT:    [[TMP55:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_5]], i32 0, i32 5
-// CHECK1-NEXT:    store i8* bitcast (void (i8*, i8*)* @.red_comb..8 to i8*), i8** [[TMP55]], align 8
-// CHECK1-NEXT:    [[TMP56:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], %struct.kmp_taskred_input_t.0* [[DOTRD_INPUT_GEP_5]], i32 0, i32 6
-// CHECK1-NEXT:    store i32 1, i32* [[TMP56]], align 8
-// CHECK1-NEXT:    [[TMP57:%.*]] = bitcast [2 x %struct.kmp_taskred_input_t.0]* [[DOTRD_INPUT_3]] to i8*
-// CHECK1-NEXT:    [[TMP58:%.*]] = call i8* @__kmpc_taskred_init(i32 [[TMP0]], i32 2, i8* [[TMP57]])
-// CHECK1-NEXT:    store i8* [[TMP58]], i8** [[DOTTASK_RED_6]], align 8
-// CHECK1-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB1]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, i64, i16*, i8**)* @main.omp_outlined to void (i32*, i32*, ...)*), i32* [[A]], i64 [[TMP2]], i16* [[VLA]], i8** [[DOTTASK_RED_]])
-// CHECK1-NEXT:    call void @__kmpc_end_taskgroup(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-// CHECK1-NEXT:    call void @__kmpc_end_taskgroup(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-// CHECK1-NEXT:    store i32 0, i32* [[RETVAL]], align 4
-// CHECK1-NEXT:    [[TMP59:%.*]] = load i8*, i8** [[SAVED_STACK]], align 8
-// CHECK1-NEXT:    call void @llvm.stackrestore(i8* [[TMP59]])
-// CHECK1-NEXT:    [[ARRAY_BEGIN7:%.*]] = getelementptr inbounds [5 x %struct.S], [5 x %struct.S]* [[C]], i32 0, i32 0
-// CHECK1-NEXT:    [[TMP60:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAY_BEGIN7]], i64 5
+// CHECK1-NEXT:    [[TMP51:%.*]] = udiv exact i64 [[TMP50]], ptrtoint (ptr getelementptr (i16, ptr null, i32 1) to i64)
+// CHECK1-NEXT:    [[TMP52:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_5]], i32 0, i32 2
+// CHECK1-NEXT:    store i64 [[TMP50]], ptr [[TMP52]], align 8
+// CHECK1-NEXT:    [[TMP53:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_5]], i32 0, i32 3
+// CHECK1-NEXT:    store ptr @.red_init..7, ptr [[TMP53]], align 8
+// CHECK1-NEXT:    [[TMP54:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_5]], i32 0, i32 4
+// CHECK1-NEXT:    store ptr null, ptr [[TMP54]], align 8
+// CHECK1-NEXT:    [[TMP55:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_5]], i32 0, i32 5
+// CHECK1-NEXT:    store ptr @.red_comb..8, ptr [[TMP55]], align 8
+// CHECK1-NEXT:    [[TMP56:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_5]], i32 0, i32 6
+// CHECK1-NEXT:    store i32 1, ptr [[TMP56]], align 8
+// CHECK1-NEXT:    [[TMP58:%.*]] = call ptr @__kmpc_taskred_init(i32 [[TMP0]], i32 2, ptr [[DOTRD_INPUT_3]])
+// CHECK1-NEXT:    store ptr [[TMP58]], ptr [[DOTTASK_RED_6]], align 8
+// CHECK1-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 4, ptr @main.omp_outlined, ptr [[A]], i64 [[TMP2]], ptr [[VLA]], ptr [[DOTTASK_RED_]])
+// CHECK1-NEXT:    call void @__kmpc_end_taskgroup(ptr @[[GLOB1]], i32 [[TMP0]])
+// CHECK1-NEXT:    call void @__kmpc_end_taskgroup(ptr @[[GLOB1]], i32 [[TMP0]])
+// CHECK1-NEXT:    store i32 0, ptr [[RETVAL]], align 4
+// CHECK1-NEXT:    [[TMP59:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK1-NEXT:    call void @llvm.stackrestore(ptr [[TMP59]])
+// CHECK1-NEXT:    [[ARRAY_BEGIN7:%.*]] = getelementptr inbounds [5 x %struct.S], ptr [[C]], i32 0, i32 0
+// CHECK1-NEXT:    [[TMP60:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN7]], i64 5
 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK1:       arraydestroy.body:
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ [[TMP60]], [[ARRAYCTOR_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
-// CHECK1-NEXT:    call void @_ZN1SD1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3:[0-9]+]]
-// CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN7]]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP60]], [[ARRAYCTOR_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    call void @_ZN1SD1Ev(ptr nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3:[0-9]+]]
+// CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN7]]
 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE8:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK1:       arraydestroy.done8:
-// CHECK1-NEXT:    [[TMP61:%.*]] = load i32, i32* [[RETVAL]], align 4
+// CHECK1-NEXT:    [[TMP61:%.*]] = load i32, ptr [[RETVAL]], align 4
 // CHECK1-NEXT:    ret i32 [[TMP61]]
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SC1Ev
-// CHECK1-SAME: (%struct.S* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1:[0-9]+]] align 2 {
+// CHECK1-SAME: (ptr nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1:[0-9]+]] align 2 {
 // CHECK1-NEXT:  entry:
-// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
-// CHECK1-NEXT:    store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
-// CHECK1-NEXT:    [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
-// CHECK1-NEXT:    call void @_ZN1SC2Ev(%struct.S* nonnull align 4 dereferenceable(4) [[THIS1]])
+// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    call void @_ZN1SC2Ev(ptr nonnull align 4 dereferenceable(4) [[THIS1]])
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.red_init.
-// CHECK1-SAME: (i8* noalias [[TMP0:%.*]], i8* noalias [[TMP1:%.*]]) #[[ATTR5:[0-9]+]] {
+// CHECK1-SAME: (ptr noalias [[TMP0:%.*]], ptr noalias [[TMP1:%.*]]) #[[ATTR5:[0-9]+]] {
 // CHECK1-NEXT:  entry:
-// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
-// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i8*, align 8
-// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
-// CHECK1-NEXT:    store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = bitcast i8** [[DOTADDR]] to i32**
-// CHECK1-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[TMP2]], align 8
-// CHECK1-NEXT:    store i32 0, i32* [[TMP3]], align 4
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    store i32 0, ptr [[TMP3]], align 4
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.red_comb.
-// CHECK1-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR5]] {
+// CHECK1-SAME: (ptr [[TMP0:%.*]], ptr [[TMP1:%.*]]) #[[ATTR5]] {
 // CHECK1-NEXT:  entry:
-// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
-// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i8*, align 8
-// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
-// CHECK1-NEXT:    store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = bitcast i8** [[DOTADDR]] to i32**
-// CHECK1-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[TMP2]], align 8
-// CHECK1-NEXT:    [[TMP4:%.*]] = bitcast i8** [[DOTADDR1]] to i32**
-// CHECK1-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[TMP4]], align 8
-// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP3]], align 4
-// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[TMP5]], align 4
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP3]], align 4
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[TMP5]], align 4
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP6]], [[TMP7]]
-// CHECK1-NEXT:    store i32 [[ADD]], i32* [[TMP3]], align 4
+// CHECK1-NEXT:    store i32 [[ADD]], ptr [[TMP3]], align 4
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.red_init..1
-// CHECK1-SAME: (i8* noalias [[TMP0:%.*]], i8* noalias [[TMP1:%.*]]) #[[ATTR5]] {
+// CHECK1-SAME: (ptr noalias [[TMP0:%.*]], ptr noalias [[TMP1:%.*]]) #[[ATTR5]] {
 // CHECK1-NEXT:  entry:
-// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
-// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i8*, align 8
-// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
-// CHECK1-NEXT:    store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = bitcast i8** [[DOTADDR]] to float**
-// CHECK1-NEXT:    [[TMP3:%.*]] = load float*, float** [[TMP2]], align 8
-// CHECK1-NEXT:    store float 0.000000e+00, float* [[TMP3]], align 4
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    store float 0.000000e+00, ptr [[TMP3]], align 4
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.red_comb..2
-// CHECK1-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR5]] {
+// CHECK1-SAME: (ptr [[TMP0:%.*]], ptr [[TMP1:%.*]]) #[[ATTR5]] {
 // CHECK1-NEXT:  entry:
-// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
-// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i8*, align 8
-// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
-// CHECK1-NEXT:    store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = bitcast i8** [[DOTADDR]] to float**
-// CHECK1-NEXT:    [[TMP3:%.*]] = load float*, float** [[TMP2]], align 8
-// CHECK1-NEXT:    [[TMP4:%.*]] = bitcast i8** [[DOTADDR1]] to float**
-// CHECK1-NEXT:    [[TMP5:%.*]] = load float*, float** [[TMP4]], align 8
-// CHECK1-NEXT:    [[TMP6:%.*]] = load float, float* [[TMP3]], align 4
-// CHECK1-NEXT:    [[TMP7:%.*]] = load float, float* [[TMP5]], align 4
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP6:%.*]] = load float, ptr [[TMP3]], align 4
+// CHECK1-NEXT:    [[TMP7:%.*]] = load float, ptr [[TMP5]], align 4
 // CHECK1-NEXT:    [[ADD:%.*]] = fadd float [[TMP6]], [[TMP7]]
-// CHECK1-NEXT:    store float [[ADD]], float* [[TMP3]], align 4
+// CHECK1-NEXT:    store float [[ADD]], ptr [[TMP3]], align 4
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.red_init..3
-// CHECK1-SAME: (i8* noalias [[TMP0:%.*]], i8* noalias [[TMP1:%.*]]) #[[ATTR5]] {
+// CHECK1-SAME: (ptr noalias [[TMP0:%.*]], ptr noalias [[TMP1:%.*]]) #[[ATTR5]] {
 // CHECK1-NEXT:  entry:
-// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
-// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i8*, align 8
-// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
-// CHECK1-NEXT:    store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = bitcast i8** [[DOTADDR]] to i32**
-// CHECK1-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[TMP2]], align 8
-// CHECK1-NEXT:    store i32 0, i32* [[TMP3]], align 4
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    store i32 0, ptr [[TMP3]], align 4
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.red_comb..4
-// CHECK1-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR5]] {
+// CHECK1-SAME: (ptr [[TMP0:%.*]], ptr [[TMP1:%.*]]) #[[ATTR5]] {
 // CHECK1-NEXT:  entry:
-// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
-// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i8*, align 8
-// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
-// CHECK1-NEXT:    store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = bitcast i8** [[DOTADDR]] to i32**
-// CHECK1-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[TMP2]], align 8
-// CHECK1-NEXT:    [[TMP4:%.*]] = bitcast i8** [[DOTADDR1]] to i32**
-// CHECK1-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[TMP4]], align 8
-// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP3]], align 4
-// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[TMP5]], align 4
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP3]], align 4
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[TMP5]], align 4
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP6]], [[TMP7]]
-// CHECK1-NEXT:    store i32 [[ADD]], i32* [[TMP3]], align 4
+// CHECK1-NEXT:    store i32 [[ADD]], ptr [[TMP3]], align 4
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.red_init..5
-// CHECK1-SAME: (i8* noalias [[TMP0:%.*]], i8* noalias [[TMP1:%.*]]) #[[ATTR5]] {
+// CHECK1-SAME: (ptr noalias [[TMP0:%.*]], ptr noalias [[TMP1:%.*]]) #[[ATTR5]] {
 // CHECK1-NEXT:  entry:
-// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
-// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i8*, align 8
-// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
-// CHECK1-NEXT:    store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = bitcast i8** [[DOTADDR]] to [5 x %struct.S]**
-// CHECK1-NEXT:    [[TMP3:%.*]] = load [5 x %struct.S]*, [5 x %struct.S]** [[TMP2]], align 8
-// CHECK1-NEXT:    [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [5 x %struct.S], [5 x %struct.S]* [[TMP3]], i32 0, i32 0
-// CHECK1-NEXT:    [[TMP4:%.*]] = getelementptr [[STRUCT_S:%.*]], %struct.S* [[ARRAY_BEGIN]], i64 5
-// CHECK1-NEXT:    [[OMP_ARRAYINIT_ISEMPTY:%.*]] = icmp eq %struct.S* [[ARRAY_BEGIN]], [[TMP4]]
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [5 x %struct.S], ptr [[TMP3]], i32 0, i32 0
+// CHECK1-NEXT:    [[TMP4:%.*]] = getelementptr [[STRUCT_S:%.*]], ptr [[ARRAY_BEGIN]], i64 5
+// CHECK1-NEXT:    [[OMP_ARRAYINIT_ISEMPTY:%.*]] = icmp eq ptr [[ARRAY_BEGIN]], [[TMP4]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYINIT_ISEMPTY]], label [[OMP_ARRAYINIT_DONE:%.*]], label [[OMP_ARRAYINIT_BODY:%.*]]
 // CHECK1:       omp.arrayinit.body:
-// CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi %struct.S* [ [[ARRAY_BEGIN]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYINIT_BODY]] ]
-// CHECK1-NEXT:    call void @_ZN1SC1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_DESTELEMENTPAST]])
-// CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr [[STRUCT_S]], %struct.S* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
-// CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq %struct.S* [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP4]]
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi ptr [ [[ARRAY_BEGIN]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYINIT_BODY]] ]
+// CHECK1-NEXT:    call void @_ZN1SC1Ev(ptr nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_DESTELEMENTPAST]])
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr [[STRUCT_S]], ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq ptr [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP4]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYINIT_DONE]], label [[OMP_ARRAYINIT_BODY]]
 // CHECK1:       omp.arrayinit.done:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.red_fini.
-// CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR5]] {
+// CHECK1-SAME: (ptr [[TMP0:%.*]]) #[[ATTR5]] {
 // CHECK1-NEXT:  entry:
-// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
-// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
-// CHECK1-NEXT:    [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = bitcast i8* [[TMP1]] to [5 x %struct.S]*
-// CHECK1-NEXT:    [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [5 x %struct.S], [5 x %struct.S]* [[TMP2]], i32 0, i32 0
-// CHECK1-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAY_BEGIN]], i64 5
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [5 x %struct.S], ptr [[TMP1]], i32 0, i32 0
+// CHECK1-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[ARRAY_BEGIN]], i64 5
 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK1:       arraydestroy.body:
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ [[TMP3]], [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
-// CHECK1-NEXT:    call void @_ZN1SD1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #3
-// CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP3]], [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    call void @_ZN1SD1Ev(ptr nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #3
+// CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK1:       arraydestroy.done1:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SD1Ev
-// CHECK1-SAME: (%struct.S* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
+// CHECK1-SAME: (ptr nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
 // CHECK1-NEXT:  entry:
-// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
-// CHECK1-NEXT:    store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
-// CHECK1-NEXT:    [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
-// CHECK1-NEXT:    call void @_ZN1SD2Ev(%struct.S* nonnull align 4 dereferenceable(4) [[THIS1]]) #3
+// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    call void @_ZN1SD2Ev(ptr nonnull align 4 dereferenceable(4) [[THIS1]]) #3
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.red_comb..6
-// CHECK1-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR5]] {
+// CHECK1-SAME: (ptr [[TMP0:%.*]], ptr [[TMP1:%.*]]) #[[ATTR5]] {
 // CHECK1-NEXT:  entry:
-// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
-// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[REF_TMP:%.*]] = alloca [[STRUCT_S:%.*]], align 4
-// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
-// CHECK1-NEXT:    store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = bitcast i8** [[DOTADDR]] to %struct.S**
-// CHECK1-NEXT:    [[TMP3:%.*]] = load %struct.S*, %struct.S** [[TMP2]], align 8
-// CHECK1-NEXT:    [[TMP4:%.*]] = bitcast i8** [[DOTADDR1]] to %struct.S**
-// CHECK1-NEXT:    [[TMP5:%.*]] = load %struct.S*, %struct.S** [[TMP4]], align 8
-// CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr [[STRUCT_S]], %struct.S* [[TMP3]], i64 5
-// CHECK1-NEXT:    [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq %struct.S* [[TMP3]], [[TMP6]]
+// CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr [[STRUCT_S]], ptr [[TMP3]], i64 5
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq ptr [[TMP3]], [[TMP6]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE2:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
 // CHECK1:       omp.arraycpy.body:
-// CHECK1-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi %struct.S* [ [[TMP5]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
-// CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi %struct.S* [ [[TMP3]], [[ENTRY]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
-// CHECK1-NEXT:    call void @_ZplRK1SS1_(%struct.S* sret([[STRUCT_S]]) align 4 [[REF_TMP]], %struct.S* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_DESTELEMENTPAST]], %struct.S* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_SRCELEMENTPAST]])
-// CHECK1-NEXT:    [[CALL:%.*]] = call nonnull align 4 dereferenceable(4) %struct.S* @_ZN1SaSERKS_(%struct.S* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_DESTELEMENTPAST]], %struct.S* nonnull align 4 dereferenceable(4) [[REF_TMP]])
-// CHECK1-NEXT:    call void @_ZN1SD1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[REF_TMP]]) #3
-// CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr [[STRUCT_S]], %struct.S* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
-// CHECK1-NEXT:    [[OMP_ARRAYCPY_SRC_ELEMENT]] = getelementptr [[STRUCT_S]], %struct.S* [[OMP_ARRAYCPY_SRCELEMENTPAST]], i32 1
-// CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq %struct.S* [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP6]]
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi ptr [ [[TMP5]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi ptr [ [[TMP3]], [[ENTRY]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
+// CHECK1-NEXT:    call void @_ZplRK1SS1_(ptr sret([[STRUCT_S]]) align 4 [[REF_TMP]], ptr nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_DESTELEMENTPAST]], ptr nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_SRCELEMENTPAST]])
+// CHECK1-NEXT:    [[CALL:%.*]] = call nonnull align 4 dereferenceable(4) ptr @_ZN1SaSERKS_(ptr nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_DESTELEMENTPAST]], ptr nonnull align 4 dereferenceable(4) [[REF_TMP]])
+// CHECK1-NEXT:    call void @_ZN1SD1Ev(ptr nonnull align 4 dereferenceable(4) [[REF_TMP]]) #3
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr [[STRUCT_S]], ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_SRC_ELEMENT]] = getelementptr [[STRUCT_S]], ptr [[OMP_ARRAYCPY_SRCELEMENTPAST]], i32 1
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq ptr [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP6]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYCPY_DONE2]], label [[OMP_ARRAYCPY_BODY]]
 // CHECK1:       omp.arraycpy.done2:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZplRK1SS1_
-// CHECK1-SAME: (%struct.S* noalias sret([[STRUCT_S:%.*]]) align 4 [[AGG_RESULT:%.*]], %struct.S* nonnull align 4 dereferenceable(4) [[A:%.*]], %struct.S* nonnull align 4 dereferenceable(4) [[B:%.*]]) #[[ATTR7:[0-9]+]] {
+// CHECK1-SAME: (ptr noalias sret([[STRUCT_S:%.*]]) align 4 [[AGG_RESULT:%.*]], ptr nonnull align 4 dereferenceable(4) [[A:%.*]], ptr nonnull align 4 dereferenceable(4) [[B:%.*]]) #[[ATTR7:[0-9]+]] {
 // CHECK1-NEXT:  entry:
-// CHECK1-NEXT:    [[RESULT_PTR:%.*]] = alloca i8*, align 8
-// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca %struct.S*, align 8
-// CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca %struct.S*, align 8
-// CHECK1-NEXT:    [[TMP0:%.*]] = bitcast %struct.S* [[AGG_RESULT]] to i8*
-// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[RESULT_PTR]], align 8
-// CHECK1-NEXT:    store %struct.S* [[A]], %struct.S** [[A_ADDR]], align 8
-// CHECK1-NEXT:    store %struct.S* [[B]], %struct.S** [[B_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP1:%.*]] = load %struct.S*, %struct.S** [[A_ADDR]], align 8
-// CHECK1-NEXT:    call void @_ZN1SC1ERKS_(%struct.S* nonnull align 4 dereferenceable(4) [[AGG_RESULT]], %struct.S* nonnull align 4 dereferenceable(4) [[TMP1]])
+// CHECK1-NEXT:    [[RESULT_PTR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    store ptr [[AGG_RESULT]], ptr [[RESULT_PTR]], align 8
+// CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
+// CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
+// CHECK1-NEXT:    call void @_ZN1SC1ERKS_(ptr nonnull align 4 dereferenceable(4) [[AGG_RESULT]], ptr nonnull align 4 dereferenceable(4) [[TMP1]])
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SaSERKS_
-// CHECK1-SAME: (%struct.S* nonnull align 4 dereferenceable(4) [[THIS:%.*]], %struct.S* nonnull align 4 dereferenceable(4) [[TMP0:%.*]]) #[[ATTR7]] align 2 {
+// CHECK1-SAME: (ptr nonnull align 4 dereferenceable(4) [[THIS:%.*]], ptr nonnull align 4 dereferenceable(4) [[TMP0:%.*]]) #[[ATTR7]] align 2 {
 // CHECK1-NEXT:  entry:
-// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
-// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca %struct.S*, align 8
-// CHECK1-NEXT:    store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
-// CHECK1-NEXT:    store %struct.S* [[TMP0]], %struct.S** [[DOTADDR]], align 8
-// CHECK1-NEXT:    [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
-// CHECK1-NEXT:    ret %struct.S* [[THIS1]]
+// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    ret ptr [[THIS1]]
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.red_init..7
-// CHECK1-SAME: (i8* noalias [[TMP0:%.*]], i8* noalias [[TMP1:%.*]]) #[[ATTR5]] {
+// CHECK1-SAME: (ptr noalias [[TMP0:%.*]], ptr noalias [[TMP1:%.*]]) #[[ATTR5]] {
 // CHECK1-NEXT:  entry:
-// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
-// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i8*, align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
-// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
-// CHECK1-NEXT:    store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP3:%.*]] = bitcast i8** [[DOTADDR]] to i16**
-// CHECK1-NEXT:    [[TMP4:%.*]] = load i16*, i16** [[TMP3]], align 8
-// CHECK1-NEXT:    [[TMP5:%.*]] = call i8* @__kmpc_threadprivate_cached(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i8* bitcast (i64* @{{reduction_size[.].+[.]}})
-// CHECK1-NEXT:    [[TMP6:%.*]] = bitcast i8* [[TMP5]] to i64*
-// CHECK1-NEXT:    [[TMP7:%.*]] = load i64, i64* [[TMP6]], align 8
-// CHECK1-NEXT:    [[TMP8:%.*]] = getelementptr i16, i16* [[TMP4]], i64 [[TMP7]]
-// CHECK1-NEXT:    [[OMP_ARRAYINIT_ISEMPTY:%.*]] = icmp eq i16* [[TMP4]], [[TMP8]]
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    [[TMP5:%.*]] = call ptr @__kmpc_threadprivate_cached(ptr @[[GLOB1]], i32 [[TMP2]], ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i64, ptr [[TMP5]], align 8
+// CHECK1-NEXT:    [[TMP8:%.*]] = getelementptr i16, ptr [[TMP4]], i64 [[TMP7]]
+// CHECK1-NEXT:    [[OMP_ARRAYINIT_ISEMPTY:%.*]] = icmp eq ptr [[TMP4]], [[TMP8]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYINIT_ISEMPTY]], label [[OMP_ARRAYINIT_DONE:%.*]], label [[OMP_ARRAYINIT_BODY:%.*]]
 // CHECK1:       omp.arrayinit.body:
-// CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi i16* [ [[TMP4]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYINIT_BODY]] ]
-// CHECK1-NEXT:    store i16 0, i16* [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 2
-// CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr i16, i16* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
-// CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq i16* [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP8]]
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi ptr [ [[TMP4]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYINIT_BODY]] ]
+// CHECK1-NEXT:    store i16 0, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 2
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr i16, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq ptr [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP8]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYINIT_DONE]], label [[OMP_ARRAYINIT_BODY]]
 // CHECK1:       omp.arrayinit.done:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.red_comb..8
-// CHECK1-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR5]] {
+// CHECK1-SAME: (ptr [[TMP0:%.*]], ptr [[TMP1:%.*]]) #[[ATTR5]] {
 // CHECK1-NEXT:  entry:
-// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
-// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i8*, align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
-// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
-// CHECK1-NEXT:    store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP3:%.*]] = call i8* @__kmpc_threadprivate_cached(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i8* bitcast (i64* @{{reduction_size[.].+[.]}})
-// CHECK1-NEXT:    [[TMP4:%.*]] = bitcast i8* [[TMP3]] to i64*
-// CHECK1-NEXT:    [[TMP5:%.*]] = load i64, i64* [[TMP4]], align 8
-// CHECK1-NEXT:    [[TMP6:%.*]] = bitcast i8** [[DOTADDR]] to i16**
-// CHECK1-NEXT:    [[TMP7:%.*]] = load i16*, i16** [[TMP6]], align 8
-// CHECK1-NEXT:    [[TMP8:%.*]] = bitcast i8** [[DOTADDR1]] to i16**
-// CHECK1-NEXT:    [[TMP9:%.*]] = load i16*, i16** [[TMP8]], align 8
-// CHECK1-NEXT:    [[TMP10:%.*]] = getelementptr i16, i16* [[TMP7]], i64 [[TMP5]]
-// CHECK1-NEXT:    [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq i16* [[TMP7]], [[TMP10]]
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP3:%.*]] = call ptr @__kmpc_threadprivate_cached(ptr @[[GLOB1]], i32 [[TMP2]], ptr @{{reduction_size[.].+[.]}})
+// CHECK1-NEXT:    [[TMP5:%.*]] = load i64, ptr [[TMP3]], align 8
+// CHECK1-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    [[TMP9:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP10:%.*]] = getelementptr i16, ptr [[TMP7]], i64 [[TMP5]]
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq ptr [[TMP7]], [[TMP10]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE4:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
 // CHECK1:       omp.arraycpy.body:
-// CHECK1-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi i16* [ [[TMP9]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
-// CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi i16* [ [[TMP7]], [[ENTRY]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
-// CHECK1-NEXT:    [[TMP11:%.*]] = load i16, i16* [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 2
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi ptr [ [[TMP9]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi ptr [ [[TMP7]], [[ENTRY]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
+// CHECK1-NEXT:    [[TMP11:%.*]] = load i16, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 2
 // CHECK1-NEXT:    [[CONV:%.*]] = sext i16 [[TMP11]] to i32
-// CHECK1-NEXT:    [[TMP12:%.*]] = load i16, i16* [[OMP_ARRAYCPY_SRCELEMENTPAST]], align 2
+// CHECK1-NEXT:    [[TMP12:%.*]] = load i16, ptr [[OMP_ARRAYCPY_SRCELEMENTPAST]], align 2
 // CHECK1-NEXT:    [[CONV2:%.*]] = sext i16 [[TMP12]] to i32
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[CONV]], [[CONV2]]
 // CHECK1-NEXT:    [[CONV3:%.*]] = trunc i32 [[ADD]] to i16
-// CHECK1-NEXT:    store i16 [[CONV3]], i16* [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 2
-// CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr i16, i16* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
-// CHECK1-NEXT:    [[OMP_ARRAYCPY_SRC_ELEMENT]] = getelementptr i16, i16* [[OMP_ARRAYCPY_SRCELEMENTPAST]], i32 1
-// CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq i16* [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP10]]
+// CHECK1-NEXT:    store i16 [[CONV3]], ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 2
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr i16, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_SRC_ELEMENT]] = getelementptr i16, ptr [[OMP_ARRAYCPY_SRCELEMENTPAST]], i32 1
+// CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq ptr [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP10]]
 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYCPY_DONE4]], label [[OMP_ARRAYCPY_BODY]]
 // CHECK1:       omp.arraycpy.done4:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@main.omp_outlined
-// CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32* nonnull align 4 dereferenceable(4) [[A:%.*]], i64 [[VLA:%.*]], i16* nonnull align 2 dereferenceable(2) [[D:%.*]], i8** nonnull align 8 dereferenceable(8) [[DOTTASK_RED_:%.*]]) #[[ATTR8:[0-9]+]] {
+// CHECK1-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], ptr nonnull align 4 dereferenceable(4) [[A:%.*]], i64 [[VLA:%.*]], ptr nonnull align 2 dereferenceable(2) [[D:%.*]], ptr nonnull align 8 dereferenceable(8) [[DOTTASK_RED_:%.*]]) #[[ATTR8:[0-9]+]] {
 // CHECK1-NEXT:  entry:
-// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca i32*, align 8
+// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[VLA_ADDR:%.*]] = alloca i64, align 8
-// CHECK1-NEXT:    [[D_ADDR:%.*]] = alloca i16*, align 8
-// CHECK1-NEXT:    [[DOTTASK_RED__ADDR:%.*]] = alloca i8**, align 8
+// CHECK1-NEXT:    [[D_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTTASK_RED__ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[AGG_CAPTURED:%.*]] = alloca [[STRUCT_ANON:%.*]], align 8
-// CHECK1-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
-// CHECK1-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
-// CHECK1-NEXT:    store i32* [[A]], i32** [[A_ADDR]], align 8
-// CHECK1-NEXT:    store i64 [[VLA]], i64* [[VLA_ADDR]], align 8
-// CHECK1-NEXT:    store i16* [[D]], i16** [[D_ADDR]], align 8
-// CHECK1-NEXT:    store i8** [[DOTTASK_RED_]], i8*** [[DOTTASK_RED__ADDR]], align 8
-// CHECK1-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP1:%.*]] = load i64, i64* [[VLA_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = load i16*, i16** [[D_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP3:%.*]] = load i8**, i8*** [[DOTTASK_RED__ADDR]], align 8
-// CHECK1-NEXT:    [[TMP4:%.*]] = load i8*, i8** [[TMP3]], align 8
-// CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [[STRUCT_ANON]], %struct.anon* [[AGG_CAPTURED]], i32 0, i32 0
-// CHECK1-NEXT:    store i32* [[TMP0]], i32** [[TMP5]], align 8
-// CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[STRUCT_ANON]], %struct.anon* [[AGG_CAPTURED]], i32 0, i32 1
-// CHECK1-NEXT:    store i64 [[TMP1]], i64* [[TMP6]], align 8
-// CHECK1-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [[STRUCT_ANON]], %struct.anon* [[AGG_CAPTURED]], i32 0, i32 2
-// CHECK1-NEXT:    store i16* [[TMP2]], i16** [[TMP7]], align 8
-// CHECK1-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [[STRUCT_ANON]], %struct.anon* [[AGG_CAPTURED]], i32 0, i32 3
-// CHECK1-NEXT:    [[TMP9:%.*]] = load i8*, i8** [[TMP3]], align 8
-// CHECK1-NEXT:    store i8* [[TMP9]], i8** [[TMP8]], align 8
-// CHECK1-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
-// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
-// CHECK1-NEXT:    [[TMP12:%.*]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* @[[GLOB1]], i32 [[TMP11]], i32 1, i64 48, i64 32, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates*)* @.omp_task_entry. to i32 (i32, i8*)*))
-// CHECK1-NEXT:    [[TMP13:%.*]] = bitcast i8* [[TMP12]] to %struct.kmp_task_t_with_privates*
-// CHECK1-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], %struct.kmp_task_t_with_privates* [[TMP13]], i32 0, i32 0
-// CHECK1-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T:%.*]], %struct.kmp_task_t* [[TMP14]], i32 0, i32 0
-// CHECK1-NEXT:    [[TMP16:%.*]] = load i8*, i8** [[TMP15]], align 8
-// CHECK1-NEXT:    [[TMP17:%.*]] = bitcast %struct.anon* [[AGG_CAPTURED]] to i8*
-// CHECK1-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[TMP16]], i8* align 8 [[TMP17]], i64 32, i1 false)
-// CHECK1-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES]], %struct.kmp_task_t_with_privates* [[TMP13]], i32 0, i32 1
-// CHECK1-NEXT:    [[TMP19:%.*]] = bitcast i8* [[TMP16]] to %struct.anon*
-// CHECK1-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T:%.*]], %struct..kmp_privates.t* [[TMP18]], i32 0, i32 0
-// CHECK1-NEXT:    [[TMP21:%.*]] = load i8*, i8** [[TMP3]], align 8
-// CHECK1-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 8
-// CHECK1-NEXT:    call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @[[GLOB1]], i32 [[TMP11]], i8* [[TMP12]])
-// CHECK1-NEXT:    [[TMP22:%.*]] = call i32 @.omp_task_entry.(i32 [[TMP11]], %struct.kmp_task_t_with_privates* [[TMP13]]) #[[ATTR3]]
-// CHECK1-NEXT:    call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @[[GLOB1]], i32 [[TMP11]], i8* [[TMP12]])
+// CHECK1-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
+// CHECK1-NEXT:    store i64 [[VLA]], ptr [[VLA_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[D]], ptr [[D_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[DOTTASK_RED_]], ptr [[DOTTASK_RED__ADDR]], align 8
+// CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
+// CHECK1-NEXT:    [[TMP1:%.*]] = load i64, ptr [[VLA_ADDR]], align 8
+// CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[D_ADDR]], align 8
+// CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTTASK_RED__ADDR]], align 8
+// CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[TMP3]], align 8
+// CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 0
+// CHECK1-NEXT:    store ptr [[TMP0]], ptr [[TMP5]], align 8
+// CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 1
+// CHECK1-NEXT:    store i64 [[TMP1]], ptr [[TMP6]], align 8
+// CHECK1-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 2
+// CHECK1-NEXT:    store ptr [[TMP2]], ptr [[TMP7]], align 8
+// CHECK1-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 3
+// CHECK1-NEXT:    [[TMP9:%.*]] = load ptr, ptr [[TMP3]], align 8
+// CHECK1-NEXT:    store ptr [[TMP9]], ptr [[TMP8]], align 8
+// CHECK1-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
+// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
+// CHECK1-NEXT:    [[TMP12:%.*]] = call ptr @__kmpc_omp_task_alloc(ptr @[[GLOB1]], i32 [[TMP11]], i32 1, i64 48, i64 32, ptr @.omp_task_entry.)
+// CHECK1-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], ptr [[TMP12]], i32 0, i32 0
+// CHECK1-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T:%.*]], ptr [[TMP14]], i32 0, i32 0
+// CHECK1-NEXT:    [[TMP16:%.*]] = load ptr, ptr [[TMP15]], align 8
+// CHECK1-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[TMP16]], ptr align 8 [[AGG_CAPTURED]], i64 32, i1 false)
+// CHECK1-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES]], ptr [[TMP12]], i32 0, i32 1
+// CHECK1-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T:%.*]], ptr [[TMP18]], i32 0, i32 0
+// CHECK1-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[TMP3]], align 8
+// CHECK1-NEXT:    store ptr [[TMP21]], ptr [[TMP20]], align 8
+// CHECK1-NEXT:    call void @__kmpc_omp_task_begin_if0(ptr @[[GLOB1]], i32 [[TMP11]], ptr [[TMP12]])
+// CHECK1-NEXT:    [[TMP22:%.*]] = call i32 @.omp_task_entry.(i32 [[TMP11]], ptr [[TMP12]]) #[[ATTR3]]
+// CHECK1-NEXT:    call void @__kmpc_omp_task_complete_if0(ptr @[[GLOB1]], i32 [[TMP11]], ptr [[TMP12]])
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@__omp_offloading_{{.*}}_main_l{{[0-9]+}}
-// CHECK1-SAME: (i32* nonnull align 4 dereferenceable(4) [[A:%.*]], i64 [[VLA:%.*]], i16* nonnull align 2 dereferenceable(2) [[D:%.*]], i8* [[DOTTASK_RED_:%.*]]) #[[ATTR9:[0-9]+]] {
+// CHECK1-SAME: (ptr nonnull align 4 dereferenceable(4) [[A:%.*]], i64 [[VLA:%.*]], ptr nonnull align 2 dereferenceable(2) [[D:%.*]], ptr [[DOTTASK_RED_:%.*]]) #[[ATTR9:[0-9]+]] {
 // CHECK1-NEXT:  entry:
-// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca i32*, align 8
+// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[VLA_ADDR:%.*]] = alloca i64, align 8
-// CHECK1-NEXT:    [[D_ADDR:%.*]] = alloca i16*, align 8
-// CHECK1-NEXT:    [[DOTTASK_RED__ADDR:%.*]] = alloca i8*, align 8
+// CHECK1-NEXT:    [[D_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTTASK_RED__ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
-// CHECK1-NEXT:    store i32* [[A]], i32** [[A_ADDR]], align 8
-// CHECK1-NEXT:    store i64 [[VLA]], i64* [[VLA_ADDR]], align 8
-// CHECK1-NEXT:    store i16* [[D]], i16** [[D_ADDR]], align 8
-// CHECK1-NEXT:    store i8* [[DOTTASK_RED_]], i8** [[DOTTASK_RED__ADDR]], align 8
-// CHECK1-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP1:%.*]] = load i64, i64* [[VLA_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = load i16*, i16** [[D_ADDR]], align 8
-// CHECK1-NEXT:    store i32 0, i32* [[I]], align 4
+// CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
+// CHECK1-NEXT:    store i64 [[VLA]], ptr [[VLA_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[D]], ptr [[D_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[DOTTASK_RED_]], ptr [[DOTTASK_RED__ADDR]], align 8
+// CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
+// CHECK1-NEXT:    [[TMP1:%.*]] = load i64, ptr [[VLA_ADDR]], align 8
+// CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[D_ADDR]], align 8
+// CHECK1-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK1-NEXT:    br label [[FOR_COND:%.*]]
 // CHECK1:       for.cond:
-// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, i32* [[I]], align 4
+// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, ptr [[I]], align 4
 // CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP3]], 5
 // CHECK1-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
 // CHECK1:       for.body:
-// CHECK1-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP0]], align 4
+// CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK1-NEXT:    [[IDXPROM_I:%.*]] = sext i32 [[TMP4]] to i64
-// CHECK1-NEXT:    [[ARRAYIDX_I:%.*]] = getelementptr inbounds i16, i16* [[TMP2]], i64 [[IDXPROM_I]]
-// CHECK1-NEXT:    [[TMP5:%.*]] = load i16, i16* [[ARRAYIDX_I]], align 2
+// CHECK1-NEXT:    [[ARRAYIDX_I:%.*]] = getelementptr inbounds i16, ptr [[TMP2]], i64 [[IDXPROM_I]]
+// CHECK1-NEXT:    [[TMP5:%.*]] = load i16, ptr [[ARRAYIDX_I]], align 2
 // CHECK1-NEXT:    [[CONV:%.*]] = sext i16 [[TMP5]] to i32
-// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP0]], align 4
+// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP6]], [[CONV]]
-// CHECK1-NEXT:    store i32 [[ADD]], i32* [[TMP0]], align 4
+// CHECK1-NEXT:    store i32 [[ADD]], ptr [[TMP0]], align 4
 // CHECK1-NEXT:    br label [[FOR_INC:%.*]]
 // CHECK1:       for.inc:
-// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[I]], align 4
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[I]], align 4
 // CHECK1-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP7]], 1
-// CHECK1-NEXT:    store i32 [[INC]], i32* [[I]], align 4
+// CHECK1-NEXT:    store i32 [[INC]], ptr [[I]], align 4
 // CHECK1-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP3:![0-9]+]]
 // CHECK1:       for.end
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.omp_task_privates_map.
-// CHECK1-SAME: (%struct..kmp_privates.t* noalias [[TMP0:%.*]], i8*** noalias [[TMP1:%.*]]) #[[ATTR9:[0-9]+]] {
+// CHECK1-SAME: (ptr noalias [[TMP0:%.*]], ptr noalias [[TMP1:%.*]]) #[[ATTR9:[0-9]+]] {
 // CHECK1-NEXT:  entry:
-// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca %struct..kmp_privates.t*, align 8
-// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i8***, align 8
-// CHECK1-NEXT:    store %struct..kmp_privates.t* [[TMP0]], %struct..kmp_privates.t** [[DOTADDR]], align 8
-// CHECK1-NEXT:    store i8*** [[TMP1]], i8**** [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = load %struct..kmp_privates.t*, %struct..kmp_privates.t** [[DOTADDR]], align 8
-// CHECK1-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T:%.*]], %struct..kmp_privates.t* [[TMP2]], i32 0, i32 0
-// CHECK1-NEXT:    [[TMP4:%.*]] = load i8***, i8**** [[DOTADDR1]], align 8
-// CHECK1-NEXT:    store i8** [[TMP3]], i8*** [[TMP4]], align 8
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T:%.*]], ptr [[TMP2]], i32 0, i32 0
+// CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
+// CHECK1-NEXT:    store ptr [[TMP3]], ptr [[TMP4]], align 8
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.omp_task_entry.
-// CHECK1-SAME: (i32 [[TMP0:%.*]], %struct.kmp_task_t_with_privates* noalias [[TMP1:%.*]]) #[[ATTR5]] {
+// CHECK1-SAME: (i32 [[TMP0:%.*]], ptr noalias [[TMP1:%.*]]) #[[ATTR5]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR_I:%.*]] = alloca i32, align 4
-// CHECK1-NEXT:    [[DOTPART_ID__ADDR_I:%.*]] = alloca i32*, align 8
-// CHECK1-NEXT:    [[DOTPRIVATES__ADDR_I:%.*]] = alloca i8*, align 8
-// CHECK1-NEXT:    [[DOTCOPY_FN__ADDR_I:%.*]] = alloca void (i8*, ...)*, align 8
-// CHECK1-NEXT:    [[DOTTASK_T__ADDR_I:%.*]] = alloca i8*, align 8
-// CHECK1-NEXT:    [[__CONTEXT_ADDR_I:%.*]] = alloca %struct.anon*, align 8
-// CHECK1-NEXT:    [[DOTFIRSTPRIV_PTR_ADDR_I:%.*]] = alloca i8**, align 8
+// CHECK1-NEXT:    [[DOTPART_ID__ADDR_I:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTPRIVATES__ADDR_I:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTCOPY_FN__ADDR_I:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTTASK_T__ADDR_I:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[__CONTEXT_ADDR_I:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTFIRSTPRIV_PTR_ADDR_I:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i32, align 4
-// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca %struct.kmp_task_t_with_privates*, align 8
-// CHECK1-NEXT:    store i32 [[TMP0]], i32* [[DOTADDR]], align 4
-// CHECK1-NEXT:    store %struct.kmp_task_t_with_privates* [[TMP1]], %struct.kmp_task_t_with_privates** [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTADDR]], align 4
-// CHECK1-NEXT:    [[TMP3:%.*]] = load %struct.kmp_task_t_with_privates*, %struct.kmp_task_t_with_privates** [[DOTADDR1]], align 8
-// CHECK1-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], %struct.kmp_task_t_with_privates* [[TMP3]], i32 0, i32 0
-// CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T:%.*]], %struct.kmp_task_t* [[TMP4]], i32 0, i32 2
-// CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T]], %struct.kmp_task_t* [[TMP4]], i32 0, i32 0
-// CHECK1-NEXT:    [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8
-// CHECK1-NEXT:    [[TMP8:%.*]] = bitcast i8* [[TMP7]] to %struct.anon*
-// CHECK1-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES]], %struct.kmp_task_t_with_privates* [[TMP3]], i32 0, i32 1
-// CHECK1-NEXT:    [[TMP10:%.*]] = bitcast %struct..kmp_privates.t* [[TMP9]] to i8*
-// CHECK1-NEXT:    [[TMP11:%.*]] = bitcast %struct.kmp_task_t_with_privates* [[TMP3]] to i8*
+// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    store i32 [[TMP0]], ptr [[DOTADDR]], align 4
+// CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTADDR]], align 4
+// CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
+// CHECK1-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], ptr [[TMP3]], i32 0, i32 0
+// CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T:%.*]], ptr [[TMP4]], i32 0, i32 2
+// CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T]], ptr [[TMP4]], i32 0, i32 0
+// CHECK1-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[TMP6]], align 8
+// CHECK1-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES]], ptr [[TMP3]], i32 0, i32 1
 // CHECK1-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META3:![0-9]+]])
 // CHECK1-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META8:![0-9]+]])
 // CHECK1-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META10:![0-9]+]])
 // CHECK1-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META12:![0-9]+]])
-// CHECK1-NEXT:    store i32 [[TMP2]], i32* [[DOTGLOBAL_TID__ADDR_I]], align 4, !noalias !14
-// CHECK1-NEXT:    store i32* [[TMP5]], i32** [[DOTPART_ID__ADDR_I]], align 8, !noalias !14
-// CHECK1-NEXT:    store i8* [[TMP10]], i8** [[DOTPRIVATES__ADDR_I]], align 8, !noalias !14
-// CHECK1-NEXT:    store void (i8*, ...)* bitcast (void (%struct..kmp_privates.t*, i8***)* @.omp_task_privates_map. to void (i8*, ...)*), void (i8*, ...)** [[DOTCOPY_FN__ADDR_I]], align 8, !noalias !14
-// CHECK1-NEXT:    store i8* [[TMP11]], i8** [[DOTTASK_T__ADDR_I]], align 8, !noalias !14
-// CHECK1-NEXT:    store %struct.anon* [[TMP8]], %struct.anon** [[__CONTEXT_ADDR_I]], align 8, !noalias !14
-// CHECK1-NEXT:    [[TMP12:%.*]] = load %struct.anon*, %struct.anon** [[__CONTEXT_ADDR_I]], align 8, !noalias !14
-// CHECK1-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [[STRUCT_ANON:%.*]], %struct.anon* [[TMP12]], i32 0, i32 1
-// CHECK1-NEXT:    [[TMP14:%.*]] = load i64, i64* [[TMP13]], align 8
-// CHECK1-NEXT:    [[TMP15:%.*]] = load void (i8*, ...)*, void (i8*, ...)** [[DOTCOPY_FN__ADDR_I]], align 8, !noalias !14
-// CHECK1-NEXT:    [[TMP16:%.*]] = load i8*, i8** [[DOTPRIVATES__ADDR_I]], align 8, !noalias !14
-// CHECK1-NEXT:    [[TMP17:%.*]] = bitcast void (i8*, ...)* [[TMP15]] to void (i8*, i8***)*
-// CHECK1-NEXT:    call void [[TMP17]](i8* [[TMP16]], i8*** [[DOTFIRSTPRIV_PTR_ADDR_I]]) #[[ATTR3]]
-// CHECK1-NEXT:    [[TMP18:%.*]] = load i8**, i8*** [[DOTFIRSTPRIV_PTR_ADDR_I]], align 8, !noalias !14
-// CHECK1-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [[STRUCT_ANON]], %struct.anon* [[TMP12]], i32 0, i32 0
-// CHECK1-NEXT:    [[TMP20:%.*]] = load i32*, i32** [[TMP19]], align 8
-// CHECK1-NEXT:    [[TMP21:%.*]] = load i8*, i8** [[TMP18]], align 8
-// CHECK1-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTGLOBAL_TID__ADDR_I]], align 4, !noalias !14
-// CHECK1-NEXT:    [[TMP23:%.*]] = bitcast i32* [[TMP20]] to i8*
-// CHECK1-NEXT:    [[TMP24:%.*]] = call i8* @__kmpc_task_reduction_get_th_data(i32 [[TMP22]], i8* [[TMP21]], i8* [[TMP23]])
-// CHECK1-NEXT:    [[CONV_I:%.*]] = bitcast i8* [[TMP24]] to i32*
-// CHECK1-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [[STRUCT_ANON]], %struct.anon* [[TMP12]], i32 0, i32 0
-// CHECK1-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[TMP25]], align 8
-// CHECK1-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [[STRUCT_ANON]], %struct.anon* [[TMP12]], i32 0, i32 2
-// CHECK1-NEXT:    [[TMP28:%.*]] = load i16*, i16** [[TMP27]], align 8
-// CHECK1-NEXT:    [[TMP29:%.*]] = load i8*, i8** [[TMP18]], align 8
-// CHECK1-NEXT:    call void @__omp_offloading_{{.*}}_main_l{{[0-9]+}}(i32* [[TMP26]], i64 [[TMP14]], i16* [[TMP28]], i8* [[TMP29]]) #[[ATTR3]]
+// CHECK1-NEXT:    store i32 [[TMP2]], ptr [[DOTGLOBAL_TID__ADDR_I]], align 4, !noalias !14
+// CHECK1-NEXT:    store ptr [[TMP5]], ptr [[DOTPART_ID__ADDR_I]], align 8, !noalias !14
+// CHECK1-NEXT:    store ptr [[TMP9]], ptr [[DOTPRIVATES__ADDR_I]], align 8, !noalias !14
+// CHECK1-NEXT:    store ptr @.omp_task_privates_map., ptr [[DOTCOPY_FN__ADDR_I]], align 8, !noalias !14
+// CHECK1-NEXT:    store ptr [[TMP3]], ptr [[DOTTASK_T__ADDR_I]], align 8, !noalias !14
+// CHECK1-NEXT:    store ptr [[TMP7]], ptr [[__CONTEXT_ADDR_I]], align 8, !noalias !14
+// CHECK1-NEXT:    [[TMP12:%.*]] = load ptr, ptr [[__CONTEXT_ADDR_I]], align 8, !noalias !14
+// CHECK1-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [[STRUCT_ANON:%.*]], ptr [[TMP12]], i32 0, i32 1
+// CHECK1-NEXT:    [[TMP14:%.*]] = load i64, ptr [[TMP13]], align 8
+// CHECK1-NEXT:    [[TMP15:%.*]] = load ptr, ptr [[DOTCOPY_FN__ADDR_I]], align 8, !noalias !14
+// CHECK1-NEXT:    [[TMP16:%.*]] = load ptr, ptr [[DOTPRIVATES__ADDR_I]], align 8, !noalias !14
+// CHECK1-NEXT:    call void [[TMP15]](ptr [[TMP16]], ptr [[DOTFIRSTPRIV_PTR_ADDR_I]]) #[[ATTR3]]
+// CHECK1-NEXT:    [[TMP18:%.*]] = load ptr, ptr [[DOTFIRSTPRIV_PTR_ADDR_I]], align 8, !noalias !14
+// CHECK1-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[TMP12]], align 8
+// CHECK1-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[TMP18]], align 8
+// CHECK1-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTGLOBAL_TID__ADDR_I]], align 4, !noalias !14
+// CHECK1-NEXT:    [[TMP24:%.*]] = call ptr @__kmpc_task_reduction_get_th_data(i32 [[TMP22]], ptr [[TMP21]], ptr [[TMP20]])
+// CHECK1-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP12]], align 8
+// CHECK1-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[TMP12]], i32 0, i32 2
+// CHECK1-NEXT:    [[TMP28:%.*]] = load ptr, ptr [[TMP27]], align 8
+// CHECK1-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[TMP18]], align 8
+// CHECK1-NEXT:    call void @__omp_offloading_{{.*}}_main_l{{[0-9]+}}(ptr [[TMP26]], i64 [[TMP14]], ptr [[TMP28]], ptr [[TMP29]]) #[[ATTR3]]
 // CHECK1-NEXT:    ret i32 0
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SC2Ev
-// CHECK1-SAME: (%struct.S* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
+// CHECK1-SAME: (ptr nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
 // CHECK1-NEXT:  entry:
-// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
-// CHECK1-NEXT:    store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
-// CHECK1-NEXT:    [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
-// CHECK1-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[THIS1]], i32 0, i32 0
-// CHECK1-NEXT:    store i32 0, i32* [[A]], align 4
+// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
+// CHECK1-NEXT:    store i32 0, ptr [[A]], align 4
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SD2Ev
-// CHECK1-SAME: (%struct.S* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
+// CHECK1-SAME: (ptr nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
 // CHECK1-NEXT:  entry:
-// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
-// CHECK1-NEXT:    store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
-// CHECK1-NEXT:    [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SC1ERKS_
-// CHECK1-SAME: (%struct.S* nonnull align 4 dereferenceable(4) [[THIS:%.*]], %struct.S* nonnull align 4 dereferenceable(4) [[TMP0:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
+// CHECK1-SAME: (ptr nonnull align 4 dereferenceable(4) [[THIS:%.*]], ptr nonnull align 4 dereferenceable(4) [[TMP0:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
 // CHECK1-NEXT:  entry:
-// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
-// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca %struct.S*, align 8
-// CHECK1-NEXT:    store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
-// CHECK1-NEXT:    store %struct.S* [[TMP0]], %struct.S** [[DOTADDR]], align 8
-// CHECK1-NEXT:    [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP1:%.*]] = load %struct.S*, %struct.S** [[DOTADDR]], align 8
-// CHECK1-NEXT:    call void @_ZN1SC2ERKS_(%struct.S* nonnull align 4 dereferenceable(4) [[THIS1]], %struct.S* nonnull align 4 dereferenceable(4) [[TMP1]])
+// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    call void @_ZN1SC2ERKS_(ptr nonnull align 4 dereferenceable(4) [[THIS1]], ptr nonnull align 4 dereferenceable(4) [[TMP1]])
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SC2ERKS_
-// CHECK1-SAME: (%struct.S* nonnull align 4 dereferenceable(4) [[THIS:%.*]], %struct.S* nonnull align 4 dereferenceable(4) [[TMP0:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
+// CHECK1-SAME: (ptr nonnull align 4 dereferenceable(4) [[THIS:%.*]], ptr nonnull align 4 dereferenceable(4) [[TMP0:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
 // CHECK1-NEXT:  entry:
-// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
-// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca %struct.S*, align 8
-// CHECK1-NEXT:    store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
-// CHECK1-NEXT:    store %struct.S* [[TMP0]], %struct.S** [[DOTADDR]], align 8
-// CHECK1-NEXT:    [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
+// CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
+// CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
+// CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK1-NEXT:    ret void
 //
diff --git a/clang/test/OpenMP/target_private_codegen.cpp b/clang/test/OpenMP/target_private_codegen.cpp
index de060df17af9..2b89034acc24 100644
--- a/clang/test/OpenMP/target_private_codegen.cpp
+++ b/clang/test/OpenMP/target_private_codegen.cpp
@@ -1,22 +1,22 @@
 // Only test codegen on target side, as private clause does not require any action on the host side
 // Test target codegen - host bc file has to be created first.
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32
-
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64
+// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32
+// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32
+
+// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
+// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
+// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
 
 // expected-no-diagnostics
@@ -57,7 +57,7 @@ int foo(int n) {
 
   // TCHECK:  define weak_odr protected void @__omp_offloading_{{.+}}()
   // TCHECK:  [[A:%.+]] = alloca i{{[0-9]+}},
-  // TCHECK:  store i{{[0-9]+}} 1, i{{[0-9]+}}* [[A]],
+  // TCHECK:  store i{{[0-9]+}} 1, ptr [[A]],
   // TCHECK:  ret void
 
 #pragma omp target private(a, aa)
@@ -69,8 +69,8 @@ int foo(int n) {
   // TCHECK:  define weak_odr protected void @__omp_offloading_{{.+}}()
   // TCHECK:  [[A:%.+]] = alloca i{{[0-9]+}},
   // TCHECK:  [[A2:%.+]] = alloca i{{[0-9]+}},
-  // TCHECK:  store i{{[0-9]+}} 1, i{{[0-9]+}}* [[A]],
-  // TCHECK:  store i{{[0-9]+}} 1, i{{[0-9]+}}* [[A2]],
+  // TCHECK:  store i{{[0-9]+}} 1, ptr [[A]],
+  // TCHECK:  store i{{[0-9]+}} 1, ptr [[A2]],
   // TCHECK:  ret void
 
   #pragma omp target private(a, b, bn, c, cn, d)
@@ -91,54 +91,54 @@ int foo(int n) {
   // TCHECK:  [[VLA_ADDR4:%.+]] = alloca i{{[0-9]+}},
   // TCHECK:  [[A:%.+]] = alloca i{{[0-9]+}},
   // TCHECK:  [[B:%.+]] = alloca [10 x float],
-  // TCHECK:  [[SSTACK:%.+]] = alloca i8*,
+  // TCHECK:  [[SSTACK:%.+]] = alloca ptr,
   // TCHECK:  [[C:%.+]] = alloca [5 x [10 x double]],
   // TCHECK:  [[D:%.+]] = alloca [[TT]],
-  // TCHECK:  store i{{[0-9]+}} [[VLA]], i{{[0-9]+}}* [[VLA_ADDR]],
-  // TCHECK:  store i{{[0-9]+}} [[VLA1]], i{{[0-9]+}}* [[VLA_ADDR2]],
-  // TCHECK:  store i{{[0-9]+}} [[VLA3]], i{{[0-9]+}}* [[VLA_ADDR4]],
-  // TCHECK:  [[VLA_ADDR_REF:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR]],
-  // TCHECK:  [[VLA_ADDR_REF2:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR2]],
-  // TCHECK:  [[VLA_ADDR_REF4:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR4]],
-  // TCHECK:  [[RET_STACK:%.+]] = call i8* @llvm.stacksave()
-  // TCHECK:  store i8* [[RET_STACK]], i8** [[SSTACK]],
+  // TCHECK:  store i{{[0-9]+}} [[VLA]], ptr [[VLA_ADDR]],
+  // TCHECK:  store i{{[0-9]+}} [[VLA1]], ptr [[VLA_ADDR2]],
+  // TCHECK:  store i{{[0-9]+}} [[VLA3]], ptr [[VLA_ADDR4]],
+  // TCHECK:  [[VLA_ADDR_REF:%.+]] = load i{{[0-9]+}}, ptr [[VLA_ADDR]],
+  // TCHECK:  [[VLA_ADDR_REF2:%.+]] = load i{{[0-9]+}}, ptr [[VLA_ADDR2]],
+  // TCHECK:  [[VLA_ADDR_REF4:%.+]] = load i{{[0-9]+}}, ptr [[VLA_ADDR4]],
+  // TCHECK:  [[RET_STACK:%.+]] = call ptr @llvm.stacksave()
+  // TCHECK:  store ptr [[RET_STACK]], ptr [[SSTACK]],
   // TCHECK:  [[VLA5:%.+]] = alloca float, i{{[0-9]+}} [[VLA_ADDR_REF]],
   // TCHECK:  [[VLA6_SIZE:%.+]] = mul{{.+}} i{{[0-9]+}} [[VLA_ADDR_REF2]], [[VLA_ADDR_REF4]]
   // TCHECK:  [[VLA6:%.+]] = alloca double, i{{[0-9]+}} [[VLA6_SIZE]],
 
   // a = 1
-  // TCHECK:  store i{{[0-9]+}} 1, i{{[0-9]+}}* [[A]],
+  // TCHECK:  store i{{[0-9]+}} 1, ptr [[A]],
 
   // b[2] = 1.0
-  // TCHECK:  [[B_GEP:%.+]] = getelementptr inbounds [10 x float], [10 x float]* [[B]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
-  // TCHECK:  store float 1.0{{.*}}, float* [[B_GEP]],
+  // TCHECK:  [[B_GEP:%.+]] = getelementptr inbounds [10 x float], ptr [[B]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
+  // TCHECK:  store float 1.0{{.*}}, ptr [[B_GEP]],
 
   // bn[3] = 1.0
-  // TCHECK:  [[BN_GEP:%.+]] = getelementptr inbounds float, float* [[VLA5]], i{{[0-9]+}} 3
-  // TCHECK:  store float 1.0{{.*}}, float* [[BN_GEP]],
+  // TCHECK:  [[BN_GEP:%.+]] = getelementptr inbounds float, ptr [[VLA5]], i{{[0-9]+}} 3
+  // TCHECK:  store float 1.0{{.*}}, ptr [[BN_GEP]],
 
   // c[1][2] = 1.0
-  // TCHECK:  [[C_GEP1:%.+]] = getelementptr inbounds [5 x [10 x double]], [5 x [10 x double]]* [[C]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
-  // TCHECK:  [[C_GEP2:%.+]] = getelementptr inbounds [10 x double], [10 x double]* [[C_GEP1]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
-  // TCHECK:  store double 1.0{{.*}}, double* [[C_GEP2]],
+  // TCHECK:  [[C_GEP1:%.+]] = getelementptr inbounds [5 x [10 x double]], ptr [[C]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
+  // TCHECK:  [[C_GEP2:%.+]] = getelementptr inbounds [10 x double], ptr [[C_GEP1]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
+  // TCHECK:  store double 1.0{{.*}}, ptr [[C_GEP2]],
 
   // cn[1][3] = 1.0
   // TCHECK:  [[CN_IND:%.+]] = mul{{.+}} i{{[0-9]+}} 1, [[VLA_ADDR_REF4]]
-  // TCHECK:  [[CN_GEP_IND:%.+]] = getelementptr inbounds double, double* [[VLA6]], i{{[0-9]+}} [[CN_IND]]
-  // TCHECK:  [[CN_GEP_3:%.+]] = getelementptr inbounds double, double* [[CN_GEP_IND]], i{{[0-9]+}} 3
-  // TCHECK:  store double 1.0{{.*}}, double* [[CN_GEP_3]],
+  // TCHECK:  [[CN_GEP_IND:%.+]] = getelementptr inbounds double, ptr [[VLA6]], i{{[0-9]+}} [[CN_IND]]
+  // TCHECK:  [[CN_GEP_3:%.+]] = getelementptr inbounds double, ptr [[CN_GEP_IND]], i{{[0-9]+}} 3
+  // TCHECK:  store double 1.0{{.*}}, ptr [[CN_GEP_3]],
 
   // d.X = 1
-  // [[X_FIELD:%.+]] = getelementptr inbounds [[TT]] [[TT]]* [[D]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
-  // store i{{[0-9]+}} 1, i{{[0-9]+}}* [[X_FIELD]],
+  // [[X_FIELD:%.+]] = getelementptr inbounds [[TT]] ptr [[D]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
+  // store i{{[0-9]+}} 1, ptr [[X_FIELD]],
 
   // d.Y = 1
-  // [[Y_FIELD:%.+]] = getelementptr inbounds [[TT]] [[TT]]* [[D]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
-  // store i{{[0-9]+}} 1, i{{[0-9]+}}* [[Y_FIELD]],
+  // [[Y_FIELD:%.+]] = getelementptr inbounds [[TT]] ptr [[D]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
+  // store i{{[0-9]+}} 1, ptr [[Y_FIELD]],
 
   // finish
-  // [[RELOAD_SSTACK:%.+]] = load i8*, i8** [[SSTACK]],
-  // call ovid @llvm.stackrestore(i8* [[RELOAD_SSTACK]])
+  // [[RELOAD_SSTACK:%.+]] = load ptr, ptr [[SSTACK]],
+  // call ovid @llvm.stackrestore(ptr [[RELOAD_SSTACK]])
   // ret void
 
   return a;
@@ -184,11 +184,11 @@ int fstatic(int n) {
 // TCHECK:  [[A2:%.+]] = alloca i{{[0-9]+}},
 // TCHECK:  [[A3:%.+]] = alloca i{{[0-9]+}},
 // TCHECK:  [[B:%.+]] = alloca [10 x i{{[0-9]+}}],
-// TCHECK:  store i{{[0-9]+}} 1, i{{[0-9]+}}* [[A]],
-// TCHECK:  store i{{[0-9]+}} 1, i{{[0-9]+}}* [[A2]],
-// TCHECK:  store i{{[0-9]+}} 1, i{{[0-9]+}}* [[A3]],
-// TCHECK:  [[B_GEP:%.+]] = getelementptr inbounds [10 x i{{[0-9]+}}], [10 x i{{[0-9]+}}]* [[B]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
-// TCHECK:  store i{{[0-9]+}} 1, i{{[0-9]+}}* [[B_GEP]],
+// TCHECK:  store i{{[0-9]+}} 1, ptr [[A]],
+// TCHECK:  store i{{[0-9]+}} 1, ptr [[A2]],
+// TCHECK:  store i{{[0-9]+}} 1, ptr [[A3]],
+// TCHECK:  [[B_GEP:%.+]] = getelementptr inbounds [10 x i{{[0-9]+}}], ptr [[B]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
+// TCHECK:  store i{{[0-9]+}} 1, ptr [[B_GEP]],
 // TCHECK:  ret void
 
 struct S1 {
@@ -207,44 +207,44 @@ struct S1 {
     return c[1][1] + (int)b;
   }
 
-  // TCHECK: define weak_odr protected void @__omp_offloading_{{.+}}([[S1]]* noundef [[TH:%.+]], i{{[0-9]+}} noundef [[VLA:%.+]], i{{[0-9]+}} noundef [[VLA1:%.+]])
-  // TCHECK: [[TH_ADDR:%.+]] = alloca [[S1]]*,
+  // TCHECK: define weak_odr protected void @__omp_offloading_{{.+}}(ptr noundef [[TH:%.+]], i{{[0-9]+}} noundef [[VLA:%.+]], i{{[0-9]+}} noundef [[VLA1:%.+]])
+  // TCHECK: [[TH_ADDR:%.+]] = alloca ptr,
   // TCHECK: [[VLA_ADDR:%.+]] = alloca i{{[0-9]+}},
   // TCHECK: [[VLA_ADDR2:%.+]] = alloca i{{[0-9]+}},
   // TCHECK: [[B:%.+]] = alloca i{{[0-9]+}},
-  // TCHECK: [[SSTACK:%.+]] = alloca i8*,
-  // TCHECK: store [[S1]]* [[TH]], [[S1]]** [[TH_ADDR]],
-  // TCHECK: store i{{[0-9]+}} [[VLA]], i{{[0-9]+}}* [[VLA_ADDR]],
-  // TCHECK: store i{{[0-9]+}} [[VLA1]], i{{[0-9]+}}* [[VLA_ADDR2]],
-  // TCHECK: [[TH_ADDR_REF:%.+]] = load [[S1]]*, [[S1]]** [[TH_ADDR]],
-  // TCHECK: [[VLA_ADDR_REF:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR]],
-  // TCHECK: [[VLA_ADDR_REF2:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR2]],
-  // TCHECK: [[RET_STACK:%.+]] = call i8* @llvm.stacksave()
-  // TCHECK: store i8* [[RET_STACK:%.+]], i8** [[SSTACK]],
+  // TCHECK: [[SSTACK:%.+]] = alloca ptr,
+  // TCHECK: store ptr [[TH]], ptr [[TH_ADDR]],
+  // TCHECK: store i{{[0-9]+}} [[VLA]], ptr [[VLA_ADDR]],
+  // TCHECK: store i{{[0-9]+}} [[VLA1]], ptr [[VLA_ADDR2]],
+  // TCHECK: [[TH_ADDR_REF:%.+]] = load ptr, ptr [[TH_ADDR]],
+  // TCHECK: [[VLA_ADDR_REF:%.+]] = load i{{[0-9]+}}, ptr [[VLA_ADDR]],
+  // TCHECK: [[VLA_ADDR_REF2:%.+]] = load i{{[0-9]+}}, ptr [[VLA_ADDR2]],
+  // TCHECK: [[RET_STACK:%.+]] = call ptr @llvm.stacksave()
+  // TCHECK: store ptr [[RET_STACK:%.+]], ptr [[SSTACK]],
 
   // this->a = (double)b + 1.5;
   // TCHECK: [[VLA_IND:%.+]] = mul{{.+}} i{{[0-9]+}} [[VLA_ADDR_REF]], [[VLA_ADDR_REF2]]
   // TCHECK: [[VLA3:%.+]] = alloca i{{[0-9]+}}, i{{[0-9]+}} [[VLA_IND]],
-  // TCHECK: [[B_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[B]],
+  // TCHECK: [[B_VAL:%.+]] = load i{{[0-9]+}}, ptr [[B]],
   // TCHECK: [[B_CONV:%.+]] = sitofp i{{[0-9]+}} [[B_VAL]] to double
   // TCHECK: [[NEW_A_VAL:%.+]] = fadd double [[B_CONV]], 1.5{{.+}}+00
-  // TCHECK: [[A_FIELD:%.+]] = getelementptr inbounds [[S1]], [[S1]]* [[TH_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
-  // TCHECK: store double [[NEW_A_VAL]], double* [[A_FIELD]],
+  // TCHECK: [[A_FIELD:%.+]] = getelementptr inbounds [[S1]], ptr [[TH_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
+  // TCHECK: store double [[NEW_A_VAL]], ptr [[A_FIELD]],
 
   // c[1][1] = ++a;
-  // TCHECK: [[A_FIELD4:%.+]] = getelementptr inbounds [[S1]], [[S1]]* [[TH_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
-  // TCHECK: [[A_FIELD4_VAL:%.+]] = load double, double* [[A_FIELD4]],
+  // TCHECK: [[A_FIELD4:%.+]] = getelementptr inbounds [[S1]], ptr [[TH_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
+  // TCHECK: [[A_FIELD4_VAL:%.+]] = load double, ptr [[A_FIELD4]],
   // TCHECK: [[A_FIELD_INC:%.+]] = fadd double [[A_FIELD4_VAL]], 1.0{{.+}}+00
-  // TCHECK: store double [[A_FIELD_INC]], double* [[A_FIELD4]],
+  // TCHECK: store double [[A_FIELD_INC]], ptr [[A_FIELD4]],
   // TCHECK: [[A_FIELD_INC_CONV:%.+]] = fptosi double [[A_FIELD_INC]] to i{{[0-9]+}}
   // TCHECK: [[C_IND:%.+]] = mul{{.+}} i{{[0-9]+}} 1, [[VLA_ADDR_REF2]]
-  // TCHECK: [[C_1_REF:%.+]] = getelementptr inbounds i{{[0-9]+}}, i{{[0-9]+}}* [[VLA3]], i{{[0-9]+}} [[C_IND]]
-  // TCHECK: [[C_1_1_REF:%.+]] = getelementptr inbounds i{{[0-9]+}}, i{{[0-9]+}}* [[C_1_REF]], i{{[0-9]+}} 1
-  // TCHECK: store i{{[0-9]+}} [[A_FIELD_INC_CONV]], i{{[0-9]+}}* [[C_1_1_REF]],
+  // TCHECK: [[C_1_REF:%.+]] = getelementptr inbounds i{{[0-9]+}}, ptr [[VLA3]], i{{[0-9]+}} [[C_IND]]
+  // TCHECK: [[C_1_1_REF:%.+]] = getelementptr inbounds i{{[0-9]+}}, ptr [[C_1_REF]], i{{[0-9]+}} 1
+  // TCHECK: store i{{[0-9]+}} [[A_FIELD_INC_CONV]], ptr [[C_1_1_REF]],
 
   // finish
-  // TCHECK: [[RELOAD_SSTACK:%.+]] = load i8*, i8** [[SSTACK]],
-  // TCHECK: call void @llvm.stackrestore(i8* [[RELOAD_SSTACK]])
+  // TCHECK: [[RELOAD_SSTACK:%.+]] = load ptr, ptr [[SSTACK]],
+  // TCHECK: call void @llvm.stackrestore(ptr [[RELOAD_SSTACK]])
   // TCHECK: ret void
 };
 
@@ -265,10 +265,10 @@ int bar(int n){
 // TCHECK: [[A:%.+]] = alloca i{{[0-9]+}},
 // TCHECK: [[A2:%.+]] = alloca i{{[0-9]+}},
 // TCHECK: [[B:%.+]] = alloca [10 x i{{[0-9]+}}],
-// TCHECK: store i{{[0-9]+}} 1, i{{[0-9]+}}* [[A]],
-// TCHECK: store i{{[0-9]+}} 1, i{{[0-9]+}}* [[A2]],
-// TCHECK: [[B_GEP:%.+]] = getelementptr inbounds [10 x i{{[0-9]+}}], [10 x i{{[0-9]+}}]* [[B]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
-// TCHECK: store i{{[0-9]+}} 1, i{{[0-9]+}}* [[B_GEP]],
+// TCHECK: store i{{[0-9]+}} 1, ptr [[A]],
+// TCHECK: store i{{[0-9]+}} 1, ptr [[A2]],
+// TCHECK: [[B_GEP:%.+]] = getelementptr inbounds [10 x i{{[0-9]+}}], ptr [[B]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
+// TCHECK: store i{{[0-9]+}} 1, ptr [[B_GEP]],
 // TCHECK: ret void
 
 #endif
diff --git a/clang/test/OpenMP/target_reduction_codegen.cpp b/clang/test/OpenMP/target_reduction_codegen.cpp
index ad20ac8dd68f..c3bbe54093bc 100644
--- a/clang/test/OpenMP/target_reduction_codegen.cpp
+++ b/clang/test/OpenMP/target_reduction_codegen.cpp
@@ -1,22 +1,22 @@
 // Only test codegen on target side, as private clause does not require any action on the host side
 // Test target codegen - host bc file has to be created first.
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32
-
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64
+// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32
+// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32
+
+// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
+// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
+// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
 
 // expected-no-diagnostics
@@ -45,10 +45,10 @@ int foo(int n) {
   {
   }
 
-  // TCHECK: define weak_odr protected void @__omp_offloading_{{.+}}(i32*{{.+}} %{{.+}})
-  // TCHECK: [[A:%.+]] = alloca i{{[0-9]+}}*,
+  // TCHECK: define weak_odr protected void @__omp_offloading_{{.+}}(ptr{{.+}} %{{.+}})
+  // TCHECK: [[A:%.+]] = alloca ptr,
   // TCHECK: store {{.+}}, {{.+}} [[A]],
-  // TCHECK: load i32*, i32** [[A]],
+  // TCHECK: load ptr, ptr [[A]],
   // TCHECK: ret void
 
 #pragma omp target reduction(+:a)
@@ -56,11 +56,11 @@ int foo(int n) {
     a = 1;
   }
 
-  // TCHECK:  define weak_odr protected void @__omp_offloading_{{.+}}(i32*{{.+}} %{{.+}})
-  // TCHECK: [[A:%.+]] = alloca i{{[0-9]+}}*,
+  // TCHECK:  define weak_odr protected void @__omp_offloading_{{.+}}(ptr{{.+}} %{{.+}})
+  // TCHECK: [[A:%.+]] = alloca ptr,
   // TCHECK: store {{.+}}, {{.+}} [[A]],
-  // TCHECK: [[REF:%.+]] = load i32*, i32** [[A]],
-  // TCHECK: store i{{[0-9]+}} 1, i{{[0-9]+}}* [[REF]],
+  // TCHECK: [[REF:%.+]] = load ptr, ptr [[A]],
+  // TCHECK: store i{{[0-9]+}} 1, ptr [[REF]],
   // TCHECK: ret void
 
   #pragma omp target reduction(-:a, aa)
@@ -69,15 +69,15 @@ int foo(int n) {
     aa = 1;
   }
 
-  // TCHECK:  define weak_odr protected void @__omp_offloading_{{.+}}(i32*{{.+}} [[A:%.+]], i16*{{.+}} [[AA:%.+]])
-  // TCHECK:  [[A:%.+]] = alloca i{{[0-9]+}}*,
-  // TCHECK:  [[AA:%.+]] = alloca i{{[0-9]+}}*,
+  // TCHECK:  define weak_odr protected void @__omp_offloading_{{.+}}(ptr{{.+}} [[A:%.+]], ptr{{.+}} [[AA:%.+]])
+  // TCHECK:  [[A:%.+]] = alloca ptr,
+  // TCHECK:  [[AA:%.+]] = alloca ptr,
   // TCHECK: store {{.+}}, {{.+}} [[A]],
   // TCHECK: store {{.+}}, {{.+}} [[AA]],
-  // TCHECK: [[A_REF:%.+]] = load i32*, i32** [[A]],
-  // TCHECK: [[AA_REF:%.+]] = load i16*, i16** [[AA]],
-  // TCHECK:  store i{{[0-9]+}} 1, i{{[0-9]+}}* [[A_REF]],
-  // TCHECK:  store i{{[0-9]+}} 1, i{{[0-9]+}}* [[AA_REF]],
+  // TCHECK: [[A_REF:%.+]] = load ptr, ptr [[A]],
+  // TCHECK: [[AA_REF:%.+]] = load ptr, ptr [[AA]],
+  // TCHECK:  store i{{[0-9]+}} 1, ptr [[A_REF]],
+  // TCHECK:  store i{{[0-9]+}} 1, ptr [[AA_REF]],
   // TCHECK:  ret void
 
   return a;
@@ -118,24 +118,24 @@ int fstatic(int n) {
   return a;
 }
 
-// TCHECK: define weak_odr protected void @__omp_offloading_{{.+}}(i32*{{.+}}, i16*{{.+}}, i8*{{.+}}, [10 x i32]*{{.+}})
-// TCHECK:  [[A:%.+]] = alloca i{{[0-9]+}}*,
-// TCHECK:  [[A2:%.+]] = alloca i{{[0-9]+}}*,
-// TCHECK:  [[A3:%.+]] = alloca i{{[0-9]+}}*,
-// TCHECK:  [[B:%.+]] = alloca [10 x i{{[0-9]+}}]*,
+// TCHECK: define weak_odr protected void @__omp_offloading_{{.+}}(ptr{{.+}}, ptr{{.+}}, ptr{{.+}}, ptr{{.+}})
+// TCHECK:  [[A:%.+]] = alloca ptr,
+// TCHECK:  [[A2:%.+]] = alloca ptr,
+// TCHECK:  [[A3:%.+]] = alloca ptr,
+// TCHECK:  [[B:%.+]] = alloca ptr,
 // TCHECK: store {{.+}}, {{.+}} [[A]],
 // TCHECK: store {{.+}}, {{.+}} [[A2]],
 // TCHECK: store {{.+}}, {{.+}} [[A3]],
 // TCHECK: store {{.+}}, {{.+}} [[B]],
-// TCHECK: [[A_REF:%.+]] = load i32*, i32** [[A]],
-// TCHECK: [[AA_REF:%.+]] = load i16*, i16** [[AA]],
-// TCHECK: [[A3_REF:%.+]] = load i8*, i8** [[A3]],
-// TCHECK: [[B_REF:%.+]] = load {{.+}}*, {{.+}}** [[B]],
-// TCHECK:  store i{{[0-9]+}} 1, i{{[0-9]+}}* [[A_REF]],
-// TCHECK:  store i{{[0-9]+}} 1, i{{[0-9]+}}* [[AA_REF]],
-// TCHECK:  store i{{[0-9]+}} 1, i{{[0-9]+}}* [[A3_REF]],
-// TCHECK:  [[B_GEP:%.+]] = getelementptr inbounds [10 x i{{[0-9]+}}], [10 x i{{[0-9]+}}]* [[B_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
-// TCHECK:  store i{{[0-9]+}} 1, i{{[0-9]+}}* [[B_GEP]],
+// TCHECK: [[A_REF:%.+]] = load ptr, ptr [[A]],
+// TCHECK: [[AA_REF:%.+]] = load ptr, ptr [[AA]],
+// TCHECK: [[A3_REF:%.+]] = load ptr, ptr [[A3]],
+// TCHECK: [[B_REF:%.+]] = load ptr, ptr [[B]],
+// TCHECK:  store i{{[0-9]+}} 1, ptr [[A_REF]],
+// TCHECK:  store i{{[0-9]+}} 1, ptr [[AA_REF]],
+// TCHECK:  store i{{[0-9]+}} 1, ptr [[A3_REF]],
+// TCHECK:  [[B_GEP:%.+]] = getelementptr inbounds [10 x i{{[0-9]+}}], ptr [[B_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
+// TCHECK:  store i{{[0-9]+}} 1, ptr [[B_GEP]],
 // TCHECK:  ret void
 
 struct S1 {
@@ -154,40 +154,40 @@ struct S1 {
     return c[1][1] + (int)b;
   }
 
-  // TCHECK: define weak_odr protected void @__omp_offloading_{{.+}}([[S1]]* noundef [[TH:%.+]], i32*{{.+}}, i{{[0-9]+}} noundef [[VLA:%.+]], i{{[0-9]+}} noundef [[VLA1:%.+]], i16*{{.+}})
-  // TCHECK: [[TH_ADDR:%.+]] = alloca [[S1]]*,
-  // TCHECK: [[B_ADDR:%.+]] = alloca i{{[0-9]+}}*,
+  // TCHECK: define weak_odr protected void @__omp_offloading_{{.+}}(ptr noundef [[TH:%.+]], ptr{{.+}}, i{{[0-9]+}} noundef [[VLA:%.+]], i{{[0-9]+}} noundef [[VLA1:%.+]], ptr{{.+}})
+  // TCHECK: [[TH_ADDR:%.+]] = alloca ptr,
+  // TCHECK: [[B_ADDR:%.+]] = alloca ptr,
   // TCHECK: [[VLA_ADDR:%.+]] = alloca i{{[0-9]+}},
   // TCHECK: [[VLA_ADDR2:%.+]] = alloca i{{[0-9]+}},
-  // TCHECK: [[C_ADDR:%.+]] = alloca i{{[0-9]+}}*,
-  // TCHECK: store [[S1]]* [[TH]], [[S1]]** [[TH_ADDR]],
-  // TCHECK: store i{{[0-9]+}}* {{.+}}, i{{[0-9]+}}** [[B_ADDR]],
-  // TCHECK: store i{{[0-9]+}} [[VLA]], i{{[0-9]+}}* [[VLA_ADDR]],
-  // TCHECK: store i{{[0-9]+}} [[VLA1]], i{{[0-9]+}}* [[VLA_ADDR2]],
-  // TCHECK: store i{{[0-9]+}}* {{.+}}, i{{[0-9]+}}** [[C_ADDR]],
-  // TCHECK: [[TH_ADDR_REF:%.+]] = load [[S1]]*, [[S1]]** [[TH_ADDR]],
-  // TCHECK: [[B_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[B_ADDR]],
-  // TCHECK: [[VLA_ADDR_REF:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR]],
-  // TCHECK: [[VLA_ADDR_REF2:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR2]],
-  // TCHECK: [[C_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[C_ADDR]],
+  // TCHECK: [[C_ADDR:%.+]] = alloca ptr,
+  // TCHECK: store ptr [[TH]], ptr [[TH_ADDR]],
+  // TCHECK: store ptr {{.+}}, ptr [[B_ADDR]],
+  // TCHECK: store i{{[0-9]+}} [[VLA]], ptr [[VLA_ADDR]],
+  // TCHECK: store i{{[0-9]+}} [[VLA1]], ptr [[VLA_ADDR2]],
+  // TCHECK: store ptr {{.+}}, ptr [[C_ADDR]],
+  // TCHECK: [[TH_ADDR_REF:%.+]] = load ptr, ptr [[TH_ADDR]],
+  // TCHECK: [[B_REF:%.+]] = load ptr, ptr [[B_ADDR]],
+  // TCHECK: [[VLA_ADDR_REF:%.+]] = load i{{[0-9]+}}, ptr [[VLA_ADDR]],
+  // TCHECK: [[VLA_ADDR_REF2:%.+]] = load i{{[0-9]+}}, ptr [[VLA_ADDR2]],
+  // TCHECK: [[C_REF:%.+]] = load ptr, ptr [[C_ADDR]],
 
   // this->a = (double)b + 1.5;
-  // TCHECK: [[B_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[B_REF]],
+  // TCHECK: [[B_VAL:%.+]] = load i{{[0-9]+}}, ptr [[B_REF]],
   // TCHECK: [[B_CONV:%.+]] = sitofp i{{[0-9]+}} [[B_VAL]] to double
   // TCHECK: [[NEW_A_VAL:%.+]] = fadd double [[B_CONV]], 1.5{{.+}}+00
-  // TCHECK: [[A_FIELD:%.+]] = getelementptr inbounds [[S1]], [[S1]]* [[TH_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
-  // TCHECK: store double [[NEW_A_VAL]], double* [[A_FIELD]],
+  // TCHECK: [[A_FIELD:%.+]] = getelementptr inbounds [[S1]], ptr [[TH_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
+  // TCHECK: store double [[NEW_A_VAL]], ptr [[A_FIELD]],
 
   // c[1][1] = ++a;
-  // TCHECK: [[A_FIELD4:%.+]] = getelementptr inbounds [[S1]], [[S1]]* [[TH_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
-  // TCHECK: [[A_FIELD4_VAL:%.+]] = load double, double* [[A_FIELD4]],
+  // TCHECK: [[A_FIELD4:%.+]] = getelementptr inbounds [[S1]], ptr [[TH_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
+  // TCHECK: [[A_FIELD4_VAL:%.+]] = load double, ptr [[A_FIELD4]],
   // TCHECK: [[A_FIELD_INC:%.+]] = fadd double [[A_FIELD4_VAL]], 1.0{{.+}}+00
-  // TCHECK: store double [[A_FIELD_INC]], double* [[A_FIELD4]],
+  // TCHECK: store double [[A_FIELD_INC]], ptr [[A_FIELD4]],
   // TCHECK: [[A_FIELD_INC_CONV:%.+]] = fptosi double [[A_FIELD_INC]] to i{{[0-9]+}}
   // TCHECK: [[C_IND:%.+]] = mul{{.+}} i{{[0-9]+}} 1, [[VLA_ADDR_REF2]]
-  // TCHECK: [[C_1_REF:%.+]] = getelementptr inbounds i{{[0-9]+}}, i{{[0-9]+}}* [[C_REF]], i{{[0-9]+}} [[C_IND]]
-  // TCHECK: [[C_1_1_REF:%.+]] = getelementptr inbounds i{{[0-9]+}}, i{{[0-9]+}}* [[C_1_REF]], i{{[0-9]+}} 1
-  // TCHECK: store i{{[0-9]+}} [[A_FIELD_INC_CONV]], i{{[0-9]+}}* [[C_1_1_REF]],
+  // TCHECK: [[C_1_REF:%.+]] = getelementptr inbounds i{{[0-9]+}}, ptr [[C_REF]], i{{[0-9]+}} [[C_IND]]
+  // TCHECK: [[C_1_1_REF:%.+]] = getelementptr inbounds i{{[0-9]+}}, ptr [[C_1_REF]], i{{[0-9]+}} 1
+  // TCHECK: store i{{[0-9]+}} [[A_FIELD_INC_CONV]], ptr [[C_1_1_REF]],
 
   // finish
   // TCHECK: ret void
@@ -206,20 +206,20 @@ int bar(int n){
 }
 
 // template
-// TCHECK: define weak_odr protected void @__omp_offloading_{{.+}}(i{{[0-9]+}}*{{.+}}, i{{[0-9]+}}*{{.+}}, [10 x i32]*{{.+}})
-// TCHECK: [[A:%.+]] = alloca i{{[0-9]+}}*,
-// TCHECK: [[A2:%.+]] = alloca i{{[0-9]+}}*,
-// TCHECK: [[B:%.+]] = alloca [10 x i{{[0-9]+}}]*,
+// TCHECK: define weak_odr protected void @__omp_offloading_{{.+}}(ptr{{.+}}, ptr{{.+}}, ptr{{.+}})
+// TCHECK: [[A:%.+]] = alloca ptr,
+// TCHECK: [[A2:%.+]] = alloca ptr,
+// TCHECK: [[B:%.+]] = alloca ptr,
 // TCHECK: store {{.+}}, {{.+}} [[A]],
 // TCHECK: store {{.+}}, {{.+}} [[A2]],
 // TCHECK: store {{.+}}, {{.+}} [[B]],
-// TCHECK: [[A_REF:%.+]] = load i32*, i32** [[A]],
-// TCHECK: [[AA_REF:%.+]] = load i16*, i16** [[AA]],
-// TCHECK: [[B_REF:%.+]] = load {{.+}}*, {{.+}}** [[B]],
-// TCHECK: store i{{[0-9]+}} 1, i{{[0-9]+}}* [[A_REF]],
-// TCHECK: store i{{[0-9]+}} 1, i{{[0-9]+}}* [[AA_REF]],
-// TCHECK: [[B_GEP:%.+]] = getelementptr inbounds [10 x i{{[0-9]+}}], [10 x i{{[0-9]+}}]* [[B_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
-// TCHECK: store i{{[0-9]+}} 1, i{{[0-9]+}}* [[B_GEP]],
+// TCHECK: [[A_REF:%.+]] = load ptr, ptr [[A]],
+// TCHECK: [[AA_REF:%.+]] = load ptr, ptr [[AA]],
+// TCHECK: [[B_REF:%.+]] = load ptr, ptr [[B]],
+// TCHECK: store i{{[0-9]+}} 1, ptr [[A_REF]],
+// TCHECK: store i{{[0-9]+}} 1, ptr [[AA_REF]],
+// TCHECK: [[B_GEP:%.+]] = getelementptr inbounds [10 x i{{[0-9]+}}], ptr [[B_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
+// TCHECK: store i{{[0-9]+}} 1, ptr [[B_GEP]],
 // TCHECK: ret void
 
 #endif
diff --git a/clang/test/OpenMP/target_simd_codegen.cpp b/clang/test/OpenMP/target_simd_codegen.cpp
index 87debe53dbed..ced8d5c92473 100644
--- a/clang/test/OpenMP/target_simd_codegen.cpp
+++ b/clang/test/OpenMP/target_simd_codegen.cpp
@@ -1,79 +1,79 @@
 // Test host codegen.
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix OMP45
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix OMP45
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix OMP45
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix OMP45
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix OMP50
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix OMP50
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix OMP50
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix OMP50
-
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix OMP45
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix OMP45
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix OMP45
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix OMP45
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix OMP50
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix OMP50
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix OMP50
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix OMP50
+
+// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
 
 // Test target codegen - host bc file has to be created first.
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 --check-prefix TOMP45
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 --check-prefix TOMP45
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 --check-prefix TOMP45
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 --check-prefix TOMP45
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 --check-prefix TOMP50
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 --check-prefix TOMP50
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 --check-prefix TOMP50
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 --check-prefix TOMP50
-
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 --check-prefix TOMP45
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 --check-prefix TOMP45
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 --check-prefix TOMP45
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 --check-prefix TOMP45
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 --check-prefix TOMP50
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 --check-prefix TOMP50
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 --check-prefix TOMP50
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 --check-prefix TOMP50
+
+// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
+// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
+// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
+// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
+// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
 // SIMD-ONLY1-NOT: {{__kmpc|__tgt}}
 
 // expected-no-diagnostics
 #ifndef HEADER
 #define HEADER
 
-// CHECK-DAG: [[IDENT_T:%.+]] = type { i32, i32, i32, i32, i8* }
+// CHECK-DAG: [[IDENT_T:%.+]] = type { i32, i32, i32, i32, ptr }
 // CHECK-DAG: [[KMP_TASK_T_WITH_PRIVATES:%.+]] = type { [[KMP_TASK_T:%.+]] }
-// CHECK-DAG: [[KMP_TASK_T]] = type { i8*, i32 (i32, i8*)*, i32, %{{[^,]+}}, %{{[^,]+}} }
+// CHECK-DAG: [[KMP_TASK_T]] = type { ptr, ptr, i32, %{{[^,]+}}, %{{[^,]+}} }
 // CHECK-DAG: [[TT:%.+]] = type { i64, i8 }
 // CHECK-DAG: [[S1:%.+]] = type { double }
-// CHECK-DAG: [[ENTTY:%.+]] = type { i8*, i8*, i[[SZ:32|64]], i32, i32 }
+// CHECK-DAG: [[ENTTY:%.+]] = type { ptr, ptr, i[[SZ:32|64]], i32, i32 }
 
-// TCHECK: [[ENTTY:%.+]] = type { i8*, i8*, i{{32|64}}, i32, i32 }
+// TCHECK: [[ENTTY:%.+]] = type { ptr, ptr, i{{32|64}}, i32, i32 }
 
 // We have 8 target regions, but only 7 that actually will generate offloading
 // code, only 6 will have mapped arguments, and only 4 have all-constant map
@@ -111,7 +111,7 @@
 // TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]]
 
 // Check target registration is registered as a Ctor.
-// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @.omp_offloading.requires_reg, i8* null }]
+// CHECK: appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 0, ptr @.omp_offloading.requires_reg, ptr null }]
 
 
 template<typename tx, typename ty>
@@ -133,45 +133,39 @@ int foo(int n) {
   double cn[5][n];
   TT<long long, char> d;
 
-  // CHECK-32:    [[TASK:%.+]] = call i8* @__kmpc_omp_target_task_alloc(%struct.ident_t* @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i32 20, i32 1, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_T_WITH_PRIVATES]]*)* [[OMP_TASK_ENTRY:@[^,]+]] to i32 (i32, i8*)*), i64 -1)
-  // CHECK-64:    [[TASK:%.+]] = call i8* @__kmpc_omp_target_task_alloc(%struct.ident_t* @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 40, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_T_WITH_PRIVATES]]*)* [[OMP_TASK_ENTRY:@[^,]+]] to i32 (i32, i8*)*), i64 -1)
-  // CHECK:       call i32 @__kmpc_omp_task(%struct.ident_t* @{{[^,]+}}, i32 %{{[^,]+}}, i8* [[TASK]])
+  // CHECK-32:    [[TASK:%.+]] = call ptr @__kmpc_omp_target_task_alloc(ptr @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i32 20, i32 1, ptr [[OMP_TASK_ENTRY:@[^,]+]], i64 -1)
+  // CHECK-64:    [[TASK:%.+]] = call ptr @__kmpc_omp_target_task_alloc(ptr @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 40, i64 1, ptr [[OMP_TASK_ENTRY:@[^,]+]], i64 -1)
+  // CHECK:       call i32 @__kmpc_omp_task(ptr @{{[^,]+}}, i32 %{{[^,]+}}, ptr [[TASK]])
   #pragma omp target simd nowait
   for (int i = 3; i < 32; i += 5) {
   }
 
-  // CHECK:       call void [[HVT1:@.+]](i[[SZ]] {{[^,]+}}, i{{32|64}}{{[*]*}} {{[^)]+}})
+  // CHECK:       call void [[HVT1:@.+]](i[[SZ]] {{[^,]+}}, {{[^)]+}})
   long long k = get_val();
   #pragma omp target simd if(target: 0) linear(k : 3)
   for (int i = 10; i > 1; i--) {
     a += 1;
   }
 
-  // CHECK-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 [[DEVICE:.+]], i32 1, i32 1, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
+  // CHECK-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 [[DEVICE:.+]], i32 1, i32 1, ptr @.{{.+}}.region_id, ptr [[ARGS:%.+]])
   // CHECK-DAG:   [[BPARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 2
-  // CHECK-DAG:   store i8** [[BP:%.+]], i8*** [[BPARG]]
+  // CHECK-DAG:   store ptr [[BP:%.+]], ptr [[BPARG]]
   // CHECK-DAG:   [[PARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 3
-  // CHECK-DAG:   store i8** [[P:%.+]], i8*** [[PARG]]
-  // CHECK-DAG:   [[BP]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BPR:%[^,]+]], i32 0, i32 0
-  // CHECK-DAG:   [[P]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PR:%[^,]+]], i32 0, i32 0
-  // CHECK-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BPR]], i32 0, i32 0
-  // CHECK-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PR]], i32 0, i32 0
-  // CHECK-DAG:   [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]*
-  // CHECK-DAG:   [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]*
-  // CHECK-DAG:   store i[[SZ]] [[VAL0:%.+]], i[[SZ]]* [[CBPADDR0]],
-  // CHECK-DAG:   store i[[SZ]] [[VAL0]], i[[SZ]]* [[CPADDR0]],
-  // CHECK-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BPR]], i32 0, i32 1
-  // CHECK-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PR]], i32 0, i32 1
-  // CHECK-DAG:   [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to i[[SZ]]*
-  // CHECK-DAG:   [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to i[[SZ]]*
-  // CHECK-DAG:   store i[[SZ]] [[VAL1:%.+]], i[[SZ]]* [[CBPADDR1]],
-  // CHECK-DAG:   store i[[SZ]] [[VAL1]], i[[SZ]]* [[CPADDR1]],
-  // CHECK-DAG:   [[BPADDR2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BPR]], i32 0, i32 2
-  // CHECK-DAG:   [[PADDR2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PR]], i32 0, i32 2
-  // CHECK-DAG:   [[CBPADDR2:%.+]] = bitcast i8** [[BPADDR2]] to i[[SZ]]*
-  // CHECK-DAG:   [[CPADDR2:%.+]] = bitcast i8** [[PADDR2]] to i[[SZ]]*
-  // CHECK-DAG:   store i[[SZ]] [[VAL2:%.+]], i[[SZ]]* [[CBPADDR2]],
-  // CHECK-DAG:   store i[[SZ]] [[VAL2]], i[[SZ]]* [[CPADDR2]],
+  // CHECK-DAG:   store ptr [[P:%.+]], ptr [[PARG]]
+  // CHECK-DAG:   [[BP]] = getelementptr inbounds [3 x ptr], ptr [[BPR:%[^,]+]], i32 0, i32 0
+  // CHECK-DAG:   [[P]] = getelementptr inbounds [3 x ptr], ptr [[PR:%[^,]+]], i32 0, i32 0
+  // CHECK-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [3 x ptr], ptr [[BPR]], i32 0, i32 0
+  // CHECK-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [3 x ptr], ptr [[PR]], i32 0, i32 0
+  // CHECK-DAG:   store i[[SZ]] [[VAL0:%.+]], ptr [[BPADDR0]],
+  // CHECK-DAG:   store i[[SZ]] [[VAL0]], ptr [[PADDR0]],
+  // CHECK-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [3 x ptr], ptr [[BPR]], i32 0, i32 1
+  // CHECK-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [3 x ptr], ptr [[PR]], i32 0, i32 1
+  // CHECK-DAG:   store i[[SZ]] [[VAL1:%.+]], ptr [[BPADDR1]],
+  // CHECK-DAG:   store i[[SZ]] [[VAL1]], ptr [[PADDR1]],
+  // CHECK-DAG:   [[BPADDR2:%.+]] = getelementptr inbounds [3 x ptr], ptr [[BPR]], i32 0, i32 2
+  // CHECK-DAG:   [[PADDR2:%.+]] = getelementptr inbounds [3 x ptr], ptr [[PR]], i32 0, i32 2
+  // CHECK-DAG:   store i[[SZ]] [[VAL2:%.+]], ptr [[BPADDR2]],
+  // CHECK-DAG:   store i[[SZ]] [[VAL2]], ptr [[PADDR2]],
 
   // CHECK-NEXT:  [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
   // CHECK-NEXT:  br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]
@@ -188,27 +182,23 @@ int foo(int n) {
   // CHECK:       [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 10
   // CHECK:       br i1 [[IF]], label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]]
   // CHECK:       [[IFTHEN]]
-  // CHECK-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 [[DEVICE:.+]], i32 1, i32 1, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
+  // CHECK-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 [[DEVICE:.+]], i32 1, i32 1, ptr @.{{.+}}.region_id, ptr [[ARGS:%.+]])
   // CHECK-DAG:   [[BPARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 2
-  // CHECK-DAG:   store i8** [[BPR:%.+]], i8*** [[BPARG]]
+  // CHECK-DAG:   store ptr [[BPR:%.+]], ptr [[BPARG]]
   // CHECK-DAG:   [[PARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 3
-  // CHECK-DAG:   store i8** [[PR:%.+]], i8*** [[PARG]]
-  // CHECK-DAG:   [[BPR]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP:%[^,]+]], i32 0, i32 0
-  // CHECK-DAG:   [[PR]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P:%[^,]+]], i32 0, i32 0
-
-  // CHECK-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 0
-  // CHECK-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 0
-  // CHECK-DAG:   [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]*
-  // CHECK-DAG:   [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]*
-  // CHECK-DAG:   store i[[SZ]] [[VAL0:%.+]], i[[SZ]]* [[CBPADDR0]],
-  // CHECK-DAG:   store i[[SZ]] [[VAL0]], i[[SZ]]* [[CPADDR0]],
-
-  // CHECK-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 1
-  // CHECK-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 1
-  // CHECK-DAG:   [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to i[[SZ]]*
-  // CHECK-DAG:   [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to i[[SZ]]*
-  // CHECK-DAG:   store i[[SZ]] [[VAL1:%.+]], i[[SZ]]* [[CBPADDR1]],
-  // CHECK-DAG:   store i[[SZ]] [[VAL1]], i[[SZ]]* [[CPADDR1]],
+  // CHECK-DAG:   store ptr [[PR:%.+]], ptr [[PARG]]
+  // CHECK-DAG:   [[BPR]] = getelementptr inbounds [2 x ptr], ptr [[BP:%[^,]+]], i32 0, i32 0
+  // CHECK-DAG:   [[PR]] = getelementptr inbounds [2 x ptr], ptr [[P:%[^,]+]], i32 0, i32 0
+
+  // CHECK-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [2 x ptr], ptr [[BP]], i32 0, i32 0
+  // CHECK-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [2 x ptr], ptr [[P]], i32 0, i32 0
+  // CHECK-DAG:   store i[[SZ]] [[VAL0:%.+]], ptr [[BPADDR0]],
+  // CHECK-DAG:   store i[[SZ]] [[VAL0]], ptr [[PADDR0]],
+
+  // CHECK-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [2 x ptr], ptr [[BP]], i32 0, i32 1
+  // CHECK-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [2 x ptr], ptr [[P]], i32 0, i32 1
+  // CHECK-DAG:   store i[[SZ]] [[VAL1:%.+]], ptr [[BPADDR1]],
+  // CHECK-DAG:   store i[[SZ]] [[VAL1]], ptr [[PADDR1]],
   // CHECK:       [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
   // CHECK-NEXT:  br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]]
   // CHECK:       [[FAIL]]
@@ -228,14 +218,13 @@ int foo(int n) {
   }
 
   // We capture 3 VLA sizes in this target region
-  // CHECK-64:       [[A_VAL:%.+]] = load i32, i32* %{{.+}},
-  // CHECK-64:       [[A_ADDR:%.+]] = bitcast i[[SZ]]* [[A_CADDR:%.+]] to i32*
-  // CHECK-64:       store i32 [[A_VAL]], i32* [[A_ADDR]],
-  // CHECK-64:       [[A_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CADDR]],
+  // CHECK-64:       [[A_VAL:%.+]] = load i32, ptr %{{.+}},
+  // CHECK-64:       store i32 [[A_VAL]], ptr [[A_CADDR:%.+]],
+  // CHECK-64:       [[A_CVAL:%.+]] = load i[[SZ]], ptr [[A_CADDR]],
 
-  // CHECK-32:       [[A_VAL:%.+]] = load i32, i32* %{{.+}},
-  // CHECK-32:       store i32 [[A_VAL]], i32* [[A_CADDR:%.+]],
-  // CHECK-32:       [[A_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CADDR]],
+  // CHECK-32:       [[A_VAL:%.+]] = load i32, ptr %{{.+}},
+  // CHECK-32:       store i32 [[A_VAL]], ptr [[A_CADDR:%.+]],
+  // CHECK-32:       [[A_CVAL:%.+]] = load i[[SZ]], ptr [[A_CADDR]],
 
   // CHECK:       [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 20
   // CHECK:       br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]]
@@ -248,86 +237,68 @@ int foo(int n) {
   // CHECK-32:    [[CNSZSIZE:%.+]] = mul nuw i32 [[CNELEMSIZE2]], 8
   // CHECK-32:    [[CNSIZE:%.+]] = sext i32 [[CNSZSIZE]] to i64
 
-// CHECK-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 [[DEVICE:.+]], i32 1, i32 1, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
+// CHECK-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 [[DEVICE:.+]], i32 1, i32 1, ptr @.{{.+}}.region_id, ptr [[ARGS:%.+]])
 // CHECK-DAG:   [[BPARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 2
-// CHECK-DAG:   store i8** [[BPR:%.+]], i8*** [[BPARG]]
+// CHECK-DAG:   store ptr [[BPR:%.+]], ptr [[BPARG]]
 // CHECK-DAG:   [[PARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 3
-// CHECK-DAG:   store i8** [[PR:%.+]], i8*** [[PARG]]
+// CHECK-DAG:   store ptr [[PR:%.+]], ptr [[PARG]]
 // CHECK-DAG:   [[SARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 4
-// CHECK-DAG:   store i64* [[SR:%.+]], i64** [[SARG]]
-// CHECK-DAG:   [[BPR]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP:%[^,]+]], i32 0, i32 0
-// CHECK-DAG:   [[PR]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P:%[^,]+]], i32 0, i32 0
-// CHECK-DAG:   [[SR]] = getelementptr inbounds [9 x i64], [9 x i64]* [[S:%[^,]+]], i32 0, i32 0
-
-// CHECK-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX0:[0-9]+]]
-// CHECK-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX0]]
-// CHECK-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX1:[0-9]+]]
-// CHECK-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX1]]
-// CHECK-DAG:   [[BPADDR2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX2:[0-9]+]]
-// CHECK-DAG:   [[PADDR2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX2]]
-// CHECK-DAG:   [[SADDR3:%.+]] = getelementptr inbounds [9 x i64], [9 x i64]* [[S]], i32 0, i32 [[IDX3:[0-9]+]]
-// CHECK-DAG:   [[BPADDR3:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX3]]
-// CHECK-DAG:   [[PADDR3:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX3]]
-// CHECK-DAG:   [[BPADDR4:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX4:[0-9]+]]
-// CHECK-DAG:   [[PADDR4:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX4]]
-// CHECK-DAG:   [[BPADDR5:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX5:[0-9]+]]
-// CHECK-DAG:   [[PADDR5:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX5]]
-// CHECK-DAG:   [[BPADDR6:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX6:[0-9]+]]
-// CHECK-DAG:   [[PADDR6:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX6]]
-// CHECK-DAG:   [[SADDR7:%.+]] = getelementptr inbounds [9 x i64], [9 x i64]* [[S]], i32 0, i32 [[IDX7:[0-9]+]]
-// CHECK-DAG:   [[BPADDR7:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX7]]
-// CHECK-DAG:   [[PADDR7:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX7]]
-// CHECK-DAG:   [[BPADDR8:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX8:[0-9]+]]
-// CHECK-DAG:   [[PADDR8:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX8]]
+// CHECK-DAG:   store ptr [[SR:%.+]], ptr [[SARG]]
+// CHECK-DAG:   [[BPR]] = getelementptr inbounds [9 x ptr], ptr [[BP:%[^,]+]], i32 0, i32 0
+// CHECK-DAG:   [[PR]] = getelementptr inbounds [9 x ptr], ptr [[P:%[^,]+]], i32 0, i32 0
+// CHECK-DAG:   [[SR]] = getelementptr inbounds [9 x i64], ptr [[S:%[^,]+]], i32 0, i32 0
+
+// CHECK-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [9 x ptr], ptr [[BP]], i32 0, i32 [[IDX0:[0-9]+]]
+// CHECK-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [9 x ptr], ptr [[P]], i32 0, i32 [[IDX0]]
+// CHECK-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [9 x ptr], ptr [[BP]], i32 0, i32 [[IDX1:[0-9]+]]
+// CHECK-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [9 x ptr], ptr [[P]], i32 0, i32 [[IDX1]]
+// CHECK-DAG:   [[BPADDR2:%.+]] = getelementptr inbounds [9 x ptr], ptr [[BP]], i32 0, i32 [[IDX2:[0-9]+]]
+// CHECK-DAG:   [[PADDR2:%.+]] = getelementptr inbounds [9 x ptr], ptr [[P]], i32 0, i32 [[IDX2]]
+// CHECK-DAG:   [[SADDR3:%.+]] = getelementptr inbounds [9 x i64], ptr [[S]], i32 0, i32 [[IDX3:[0-9]+]]
+// CHECK-DAG:   [[BPADDR3:%.+]] = getelementptr inbounds [9 x ptr], ptr [[BP]], i32 0, i32 [[IDX3]]
+// CHECK-DAG:   [[PADDR3:%.+]] = getelementptr inbounds [9 x ptr], ptr [[P]], i32 0, i32 [[IDX3]]
+// CHECK-DAG:   [[BPADDR4:%.+]] = getelementptr inbounds [9 x ptr], ptr [[BP]], i32 0, i32 [[IDX4:[0-9]+]]
+// CHECK-DAG:   [[PADDR4:%.+]] = getelementptr inbounds [9 x ptr], ptr [[P]], i32 0, i32 [[IDX4]]
+// CHECK-DAG:   [[BPADDR5:%.+]] = getelementptr inbounds [9 x ptr], ptr [[BP]], i32 0, i32 [[IDX5:[0-9]+]]
+// CHECK-DAG:   [[PADDR5:%.+]] = getelementptr inbounds [9 x ptr], ptr [[P]], i32 0, i32 [[IDX5]]
+// CHECK-DAG:   [[BPADDR6:%.+]] = getelementptr inbounds [9 x ptr], ptr [[BP]], i32 0, i32 [[IDX6:[0-9]+]]
+// CHECK-DAG:   [[PADDR6:%.+]] = getelementptr inbounds [9 x ptr], ptr [[P]], i32 0, i32 [[IDX6]]
+// CHECK-DAG:   [[SADDR7:%.+]] = getelementptr inbounds [9 x i64], ptr [[S]], i32 0, i32 [[IDX7:[0-9]+]]
+// CHECK-DAG:   [[BPADDR7:%.+]] = getelementptr inbounds [9 x ptr], ptr [[BP]], i32 0, i32 [[IDX7]]
+// CHECK-DAG:   [[PADDR7:%.+]] = getelementptr inbounds [9 x ptr], ptr [[P]], i32 0, i32 [[IDX7]]
+// CHECK-DAG:   [[BPADDR8:%.+]] = getelementptr inbounds [9 x ptr], ptr [[BP]], i32 0, i32 [[IDX8:[0-9]+]]
+// CHECK-DAG:   [[PADDR8:%.+]] = getelementptr inbounds [9 x ptr], ptr [[P]], i32 0, i32 [[IDX8]]
 
 // The names below are not necessarily consistent with the names used for the
 // addresses above as some are repeated.
-// CHECK-DAG:   store i[[SZ]] [[A_CVAL]], i[[SZ]]* [[CBPADDR0:%.+]],
-// CHECK-DAG:   store i[[SZ]] [[A_CVAL]], i[[SZ]]* [[CPADDR0:%.+]],
-// CHECK-DAG:   [[CBPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]*
-
-// CHECK-DAG:   store [10 x float]* %{{.+}}, [10 x float]** [[CBPADDR1:%.+]],
-// CHECK-DAG:   store [10 x float]* %{{.+}}, [10 x float]** [[CPADDR1:%.+]],
-// CHECK-DAG:   [[CBPADDR1]] = bitcast i8** {{%[^,]+}} to [10 x float]**
-// CHECK-DAG:   [[CPADDR1]] = bitcast i8** {{%[^,]+}} to [10 x float]**
-
-// CHECK-DAG:   store i[[SZ]] %{{.+}}, i[[SZ]]* [[CBPADDR2:%.+]],
-// CHECK-DAG:   store i[[SZ]] %{{.+}}, i[[SZ]]* [[CPADDR2:%.+]],
-// CHECK-DAG:   [[CBPADDR2]] = bitcast i8** {{%[^,]+}} to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR2]] = bitcast i8** {{%[^,]+}} to i[[SZ]]*
-
-// CHECK-DAG:   store float* %{{.+}}, float** [[CBPADDR3:%.+]],
-// CHECK-DAG:   store float* %{{.+}}, float** [[CPADDR3:%.+]],
-// CHECK-DAG:   [[CBPADDR3]] = bitcast i8** {{%[^,]+}} to float**
-// CHECK-DAG:   [[CPADDR3]] = bitcast i8** {{%[^,]+}} to float**
-// CHECK-DAG:   store i64 [[BNSIZE]], i64* {{%[^,]+}}
-
-// CHECK-DAG:   store [5 x [10 x double]]* %{{.+}}, [5 x [10 x double]]** [[CBPADDR4:%.+]],
-// CHECK-DAG:   store [5 x [10 x double]]* %{{.+}}, [5 x [10 x double]]** [[CPADDR4:%.+]],
-// CHECK-DAG:   [[CBPADDR4]] = bitcast i8** {{%[^,]+}} to [5 x [10 x double]]**
-// CHECK-DAG:   [[CPADDR4]] = bitcast i8** {{%[^,]+}} to [5 x [10 x double]]**
-
-// CHECK-DAG:   store i[[SZ]] 5, i[[SZ]]* [[CBPADDR5:%.+]],
-// CHECK-DAG:   store i[[SZ]] 5, i[[SZ]]* [[CPADDR5:%.+]],
-// CHECK-DAG:   [[CBPADDR5]] = bitcast i8** {{%[^,]+}} to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR5]] = bitcast i8** {{%[^,]+}} to i[[SZ]]*
-
-// CHECK-DAG:   store i[[SZ]] %{{.+}}, i[[SZ]]* [[CBPADDR6:%.+]],
-// CHECK-DAG:   store i[[SZ]] %{{.+}}, i[[SZ]]* [[CPADDR6:%.+]],
-// CHECK-DAG:   [[CBPADDR6]] = bitcast i8** {{%[^,]+}} to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR6]] = bitcast i8** {{%[^,]+}} to i[[SZ]]*
-
-// CHECK-DAG:   store double* %{{.+}}, double** [[CBPADDR7:%.+]],
-// CHECK-DAG:   store double* %{{.+}}, double** [[CPADDR7:%.+]],
-// CHECK-DAG:   [[CBPADDR7]] = bitcast i8** {{%[^,]+}} to double**
-// CHECK-DAG:   [[CPADDR7]] = bitcast i8** {{%[^,]+}} to double**
-// CHECK-DAG:   store i64 [[CNSIZE]], i64* {{%[^,]+}}
-
-// CHECK-DAG:   store [[TT]]* %{{.+}}, [[TT]]** [[CBPADDR8:%.+]],
-// CHECK-DAG:   store [[TT]]* %{{.+}}, [[TT]]** [[CPADDR8:%.+]],
-// CHECK-DAG:   [[CBPADDR8]] = bitcast i8** {{%[^,]+}} to [[TT]]**
-// CHECK-DAG:   [[CPADDR8]] = bitcast i8** {{%[^,]+}} to [[TT]]**
+// CHECK-DAG:   store i[[SZ]] [[A_CVAL]], ptr {{%[^,]+}},
+// CHECK-DAG:   store i[[SZ]] [[A_CVAL]], ptr {{%[^,]+}},
+
+// CHECK-DAG:   store ptr %{{.+}}, ptr {{%[^,]+}},
+// CHECK-DAG:   store ptr %{{.+}}, ptr {{%[^,]+}},
+
+// CHECK-DAG:   store i[[SZ]] %{{.+}}, ptr {{%[^,]+}},
+// CHECK-DAG:   store i[[SZ]] %{{.+}}, ptr {{%[^,]+}},
+
+// CHECK-DAG:   store ptr %{{.+}}, ptr {{%[^,]+}},
+// CHECK-DAG:   store ptr %{{.+}}, ptr {{%[^,]+}},
+// CHECK-DAG:   store i64 [[BNSIZE]], ptr {{%[^,]+}}
+
+// CHECK-DAG:   store ptr %{{.+}}, ptr {{%[^,]+}},
+// CHECK-DAG:   store ptr %{{.+}}, ptr {{%[^,]+}},
+
+// CHECK-DAG:   store i[[SZ]] 5, ptr {{%[^,]+}},
+// CHECK-DAG:   store i[[SZ]] 5, ptr {{%[^,]+}},
+
+// CHECK-DAG:   store i[[SZ]] %{{.+}}, ptr {{%[^,]+}},
+// CHECK-DAG:   store i[[SZ]] %{{.+}}, ptr {{%[^,]+}},
+
+// CHECK-DAG:   store ptr %{{.+}}, ptr {{%[^,]+}},
+// CHECK-DAG:   store ptr %{{.+}}, ptr {{%[^,]+}},
+// CHECK-DAG:   store i64 [[CNSIZE]], ptr {{%[^,]+}}
+
+// CHECK-DAG:   store ptr %{{.+}}, ptr {{%[^,]+}},
+// CHECK-DAG:   store ptr %{{.+}}, ptr {{%[^,]+}},
 
 // CHECK-NEXT:  [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
 // CHECK-NEXT:  br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]
@@ -359,8 +330,8 @@ int foo(int n) {
 // CHECK:       ret void
 // CHECK-NEXT:  }
 
-// CHECK:       define internal {{.*}}i32 [[OMP_TASK_ENTRY]](i32 {{.*}}%0, [[KMP_TASK_T_WITH_PRIVATES]]* noalias noundef %1)
-// CHECK:       [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 [[DEVICE:.+]], i32 1, i32 1, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
+// CHECK:       define internal {{.*}}i32 [[OMP_TASK_ENTRY]](i32 {{.*}}%0, ptr noalias noundef %1)
+// CHECK:       [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 [[DEVICE:.+]], i32 1, i32 1, ptr @.{{.+}}.region_id, ptr [[ARGS:%.+]])
 // CHECK-NEXT:  [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
 // CHECK-NEXT:  br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]
 // CHECK:       [[FAIL]]
@@ -368,12 +339,11 @@ int foo(int n) {
 // CHECK-NEXT:  br label %[[END]]
 // CHECK:       [[END]]
 
-// CHECK:       define internal void [[HVT1]](i[[SZ]] noundef %{{.+}}, i{{32|64}}{{[*]*.*}} %{{.+}})
+// CHECK:       define internal void [[HVT1]](i[[SZ]] noundef %{{.+}}, {{.+}})
 // CHECK:       [[AA_ADDR:%.+]] = alloca i[[SZ]], align
-// CHECK:       store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align
-// CHECK-64:    [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i32*
-// CHECK-64:    [[AA:%.+]] = load i32, i32* [[AA_CADDR]], align
-// CHECK-32:    [[AA:%.+]] = load i32, i32* [[AA_ADDR]], align
+// CHECK:       store i[[SZ]] %{{.+}}, ptr [[AA_ADDR]], align
+// CHECK-64:    [[AA:%.+]] = load i32, ptr [[AA_ADDR]], align
+// CHECK-32:    [[AA:%.+]] = load i32, ptr [[AA_ADDR]], align
 // CHECK:       !llvm.access.group
 // CHECK:       !llvm.loop
 // CHECK:       ret void
@@ -381,9 +351,8 @@ int foo(int n) {
 
 // CHECK:       define internal void [[HVT2]](i[[SZ]] noundef %{{.+}}, i[[SZ]] noundef %{{.+}}, i[[SZ]] noundef %{{.+}})
 // CHECK:       [[AA_ADDR:%.+]] = alloca i[[SZ]], align
-// CHECK:       store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align
-// CHECK:       [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i16*
-// CHECK:       [[AA:%.+]] = load i16, i16* [[AA_CADDR]], align
+// CHECK:       store i[[SZ]] %{{.+}}, ptr [[AA_ADDR]], align
+// CHECK:       [[AA:%.+]] = load i16, ptr [[AA_ADDR]], align
 // CHECK:       !llvm.loop
 // CHECK:       ret void
 // CHECK-NEXT:  }
@@ -391,10 +360,8 @@ int foo(int n) {
 // CHECK:       define internal void [[HVT3]]
 // CHECK:       [[A_ADDR:%.+]] = alloca i[[SZ]], align
 // CHECK:       [[AA_ADDR:%.+]] = alloca i[[SZ]], align
-// CHECK-DAG:   store i[[SZ]] %{{.+}}, i[[SZ]]* [[A_ADDR]], align
-// CHECK-DAG:   store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align
-// CHECK-64-DAG:[[A_CADDR:%.+]] = bitcast i[[SZ]]* [[A_ADDR]] to i32*
-// CHECK-DAG:   [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i16*
+// CHECK-DAG:   store i[[SZ]] %{{.+}}, ptr [[A_ADDR]], align
+// CHECK-DAG:   store i[[SZ]] %{{.+}}, ptr [[AA_ADDR]], align
 // CHECK:       !llvm.loop
 // CHECK:       ret void
 // CHECK-NEXT:  }
@@ -402,33 +369,32 @@ int foo(int n) {
 // CHECK:       define internal void [[HVT4]]
 // Create local storage for each capture.
 // CHECK:       [[LOCAL_A:%.+]] = alloca i[[SZ]]
-// CHECK:       [[LOCAL_B:%.+]] = alloca [10 x float]*
+// CHECK:       [[LOCAL_B:%.+]] = alloca ptr
 // CHECK:       [[LOCAL_VLA1:%.+]] = alloca i[[SZ]]
-// CHECK:       [[LOCAL_BN:%.+]] = alloca float*
-// CHECK:       [[LOCAL_C:%.+]] = alloca [5 x [10 x double]]*
+// CHECK:       [[LOCAL_BN:%.+]] = alloca ptr
+// CHECK:       [[LOCAL_C:%.+]] = alloca ptr
 // CHECK:       [[LOCAL_VLA2:%.+]] = alloca i[[SZ]]
 // CHECK:       [[LOCAL_VLA3:%.+]] = alloca i[[SZ]]
-// CHECK:       [[LOCAL_CN:%.+]] = alloca double*
-// CHECK:       [[LOCAL_D:%.+]] = alloca [[TT]]*
-// CHECK-DAG:   store i[[SZ]] [[ARG_A:%.+]], i[[SZ]]* [[LOCAL_A]]
-// CHECK-DAG:   store [10 x float]* [[ARG_B:%.+]], [10 x float]** [[LOCAL_B]]
-// CHECK-DAG:   store i[[SZ]] [[ARG_VLA1:%.+]], i[[SZ]]* [[LOCAL_VLA1]]
-// CHECK-DAG:   store float* [[ARG_BN:%.+]], float** [[LOCAL_BN]]
-// CHECK-DAG:   store [5 x [10 x double]]* [[ARG_C:%.+]], [5 x [10 x double]]** [[LOCAL_C]]
-// CHECK-DAG:   store i[[SZ]] [[ARG_VLA2:%.+]], i[[SZ]]* [[LOCAL_VLA2]]
-// CHECK-DAG:   store i[[SZ]] [[ARG_VLA3:%.+]], i[[SZ]]* [[LOCAL_VLA3]]
-// CHECK-DAG:   store double* [[ARG_CN:%.+]], double** [[LOCAL_CN]]
-// CHECK-DAG:   store [[TT]]* [[ARG_D:%.+]], [[TT]]** [[LOCAL_D]]
-
-// CHECK-64-DAG:[[CONV_AP:%.+]] = bitcast i[[SZ]]* [[LOCAL_A]] to i32*
-// CHECK-DAG:   [[REF_B:%.+]] = load [10 x float]*, [10 x float]** [[LOCAL_B]],
-// CHECK-DAG:   [[VAL_VLA1:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_VLA1]],
-// CHECK-DAG:   [[REF_BN:%.+]] = load float*, float** [[LOCAL_BN]],
-// CHECK-DAG:   [[REF_C:%.+]] = load [5 x [10 x double]]*, [5 x [10 x double]]** [[LOCAL_C]],
-// CHECK-DAG:   [[VAL_VLA2:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_VLA2]],
-// CHECK-DAG:   [[VAL_VLA3:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_VLA3]],
-// CHECK-DAG:   [[REF_CN:%.+]] = load double*, double** [[LOCAL_CN]],
-// CHECK-DAG:   [[REF_D:%.+]] = load [[TT]]*, [[TT]]** [[LOCAL_D]],
+// CHECK:       [[LOCAL_CN:%.+]] = alloca ptr
+// CHECK:       [[LOCAL_D:%.+]] = alloca ptr
+// CHECK-DAG:   store i[[SZ]] [[ARG_A:%.+]], ptr [[LOCAL_A]]
+// CHECK-DAG:   store ptr [[ARG_B:%.+]], ptr [[LOCAL_B]]
+// CHECK-DAG:   store i[[SZ]] [[ARG_VLA1:%.+]], ptr [[LOCAL_VLA1]]
+// CHECK-DAG:   store ptr [[ARG_BN:%.+]], ptr [[LOCAL_BN]]
+// CHECK-DAG:   store ptr [[ARG_C:%.+]], ptr [[LOCAL_C]]
+// CHECK-DAG:   store i[[SZ]] [[ARG_VLA2:%.+]], ptr [[LOCAL_VLA2]]
+// CHECK-DAG:   store i[[SZ]] [[ARG_VLA3:%.+]], ptr [[LOCAL_VLA3]]
+// CHECK-DAG:   store ptr [[ARG_CN:%.+]], ptr [[LOCAL_CN]]
+// CHECK-DAG:   store ptr [[ARG_D:%.+]], ptr [[LOCAL_D]]
+
+// CHECK-DAG:   [[REF_B:%.+]] = load ptr, ptr [[LOCAL_B]],
+// CHECK-DAG:   [[VAL_VLA1:%.+]] = load i[[SZ]], ptr [[LOCAL_VLA1]],
+// CHECK-DAG:   [[REF_BN:%.+]] = load ptr, ptr [[LOCAL_BN]],
+// CHECK-DAG:   [[REF_C:%.+]] = load ptr, ptr [[LOCAL_C]],
+// CHECK-DAG:   [[VAL_VLA2:%.+]] = load i[[SZ]], ptr [[LOCAL_VLA2]],
+// CHECK-DAG:   [[VAL_VLA3:%.+]] = load i[[SZ]], ptr [[LOCAL_VLA3]],
+// CHECK-DAG:   [[REF_CN:%.+]] = load ptr, ptr [[LOCAL_CN]],
+// CHECK-DAG:   [[REF_D:%.+]] = load ptr, ptr [[LOCAL_D]],
 
 
 template<typename tx>
@@ -494,7 +460,7 @@ int bar(int n){
   a += foo(n);
 
   S1 S;
-  // CHECK: call {{.*}}i32 [[FS1:@.+]]([[S1]]* {{.*}}, i32 {{.*}})
+  // CHECK: call {{.*}}i32 [[FS1:@.+]](ptr {{.*}}, i32 {{.*}})
   a += S.r1(n);
 
   // CHECK: call {{.*}}i32 [[FSTATIC:@.+]](i32 {{.*}})
@@ -509,22 +475,20 @@ int bar(int n){
 //
 // CHECK: define {{.*}}[[FS1]]
 //
-// CHECK:          i8* @llvm.stacksave()
-// CHECK-32:       store i32 %{{.+}}, i32* %__vla_expr
+// CHECK:          ptr @llvm.stacksave()
+// CHECK-32:       store i32 %{{.+}}, ptr %__vla_expr
 // OMP50:       [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 60
-// CHECK-64:       [[B_ADDR:%.+]] = bitcast i[[SZ]]* [[B_CADDR:%.+]] to i32*
-// CHECK-64:       store i32 %{{.+}}, i32* [[B_ADDR]],
-// CHECK-64:       [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_CADDR]],
+// CHECK-64:       store i32 %{{.+}}, ptr [[B_ADDR:%.+]],
+// CHECK-64:       [[B_CVAL:%.+]] = load i[[SZ]], ptr [[B_ADDR]],
 
-// CHECK-32:       store i32 %{{.+}}, i32* [[B_ADDR:%.+]],
-// CHECK-32:       [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_ADDR]],
+// CHECK-32:       store i32 %{{.+}}, ptr [[B_ADDR:%.+]],
+// CHECK-32:       [[B_CVAL:%.+]] = load i[[SZ]], ptr [[B_ADDR]],
 
 // OMP45:       [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 60
 // OMP50:          [[TOBOOL:%.+]] = trunc i8 %{{.+}} to i1
-// OMP50:          [[CONV:%.+]] = bitcast i[[SZ]]* [[CAP:%.+]] to i8*
 // OMP50:          [[FROMBOOL:%.+]] = zext i1 [[TOBOOL]] to i8
-// OMP50:          store i8 [[FROMBOOL]], i8* [[CONV]],
-// OMP50:          [[SIMD_COND:%.+]] = load i[[SZ]], i[[SZ]]* [[CAP]],
+// OMP50:          store i8 [[FROMBOOL]], ptr [[CAP:%.+]],
+// OMP50:          [[SIMD_COND:%.+]] = load i[[SZ]], ptr [[CAP]],
 // OMP50:          [[IF:%.+]] = trunc i8 %{{.+}} to i1
 // CHECK:       br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]]
 // CHECK:       [[TRY]]
@@ -534,77 +498,65 @@ int bar(int n){
 // CHECK-32:    [[CSZSIZE:%.+]] = mul nuw i32 [[CELEMSIZE2]], 2
 // CHECK-32:    [[CSIZE:%.+]] = sext i32 [[CSZSIZE]] to i64
 
-// OMP45-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 [[DEVICE:.+]], i32 1, i32 1, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
+// OMP45-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 [[DEVICE:.+]], i32 1, i32 1, ptr @.{{.+}}.region_id, ptr [[ARGS:%.+]])
 // OMP45-DAG:   [[BPARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 2
-// OMP45-DAG:   store i8** [[BPR:%.+]], i8*** [[BPARG]]
+// OMP45-DAG:   store ptr [[BPR:%.+]], ptr [[BPARG]]
 // OMP45-DAG:   [[PARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 3
-// OMP45-DAG:   store i8** [[PR:%.+]], i8*** [[PARG]]
-// OMP45-DAG:   [[BPR]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP:%.+]], i32 0, i32 0
-// OMP45-DAG:   [[PR]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P:%.+]], i32 0, i32 0
-// OMP45-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX0:[0-9]+]]
-// OMP45-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX0]]
-// OMP45-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX1:[0-9]+]]
-// OMP45-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX1]]
-// OMP45-DAG:   [[BPADDR2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX2:[0-9]+]]
-// OMP45-DAG:   [[PADDR2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX2]]
-// OMP45-DAG:   [[BPADDR3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX3:[0-9]+]]
-// OMP45-DAG:   [[PADDR3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX3]]
-// OMP45-DAG:   [[SADDR4:%.+]] = getelementptr inbounds [5 x i64], [5 x i64]* [[S]], i32 [[IDX4:[0-9]+]]
-// OMP45-DAG:   [[BPADDR4:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX4]]
-// OMP45-DAG:   [[PADDR4:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX4]]
-// OMP50-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 [[DEVICE:.+]], i32 1, i32 1, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
+// OMP45-DAG:   store ptr [[PR:%.+]], ptr [[PARG]]
+// OMP45-DAG:   [[BPR]] = getelementptr inbounds [5 x ptr], ptr [[BP:%.+]], i32 0, i32 0
+// OMP45-DAG:   [[PR]] = getelementptr inbounds [5 x ptr], ptr [[P:%.+]], i32 0, i32 0
+// OMP45-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [5 x ptr], ptr [[BP]], i32 [[IDX0:[0-9]+]]
+// OMP45-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [5 x ptr], ptr [[P]], i32 [[IDX0]]
+// OMP45-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [5 x ptr], ptr [[BP]], i32 [[IDX1:[0-9]+]]
+// OMP45-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [5 x ptr], ptr [[P]], i32 [[IDX1]]
+// OMP45-DAG:   [[BPADDR2:%.+]] = getelementptr inbounds [5 x ptr], ptr [[BP]], i32 [[IDX2:[0-9]+]]
+// OMP45-DAG:   [[PADDR2:%.+]] = getelementptr inbounds [5 x ptr], ptr [[P]], i32 [[IDX2]]
+// OMP45-DAG:   [[BPADDR3:%.+]] = getelementptr inbounds [5 x ptr], ptr [[BP]], i32 [[IDX3:[0-9]+]]
+// OMP45-DAG:   [[PADDR3:%.+]] = getelementptr inbounds [5 x ptr], ptr [[P]], i32 [[IDX3]]
+// OMP45-DAG:   [[SADDR4:%.+]] = getelementptr inbounds [5 x i64], ptr [[S]], i32 [[IDX4:[0-9]+]]
+// OMP45-DAG:   [[BPADDR4:%.+]] = getelementptr inbounds [5 x ptr], ptr [[BP]], i32 [[IDX4]]
+// OMP45-DAG:   [[PADDR4:%.+]] = getelementptr inbounds [5 x ptr], ptr [[P]], i32 [[IDX4]]
+// OMP50-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 [[DEVICE:.+]], i32 1, i32 1, ptr @.{{.+}}.region_id, ptr [[ARGS:%.+]])
 // OMP50-DAG:   [[BPARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 2
-// OMP50-DAG:   store i8** [[BPR:%.+]], i8*** [[BPARG]]
+// OMP50-DAG:   store ptr [[BPR:%.+]], ptr [[BPARG]]
 // OMP50-DAG:   [[PARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 3
-// OMP50-DAG:   store i8** [[PR:%.+]], i8*** [[PARG]]
-// OMP50-DAG:   [[BPR]] = getelementptr inbounds [6 x  i8*], [6 x  i8*]* [[BP:%.+]], i32 0, i32 0
-// OMP50-DAG:   [[PR]] = getelementptr inbounds [6 x  i8*], [6 x  i8*]* [[P:%.+]], i32 0, i32 0
-// OMP50-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [6 x  i8*], [6 x  i8*]* [[BP]], i32 [[IDX0:[0-9]+]]
-// OMP50-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [6 x  i8*], [6 x  i8*]* [[P]], i32 [[IDX0]]
-// OMP50-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [6 x  i8*], [6 x  i8*]* [[BP]], i32 [[IDX1:[0-9]+]]
-// OMP50-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [6 x  i8*], [6 x  i8*]* [[P]], i32 [[IDX1]]
-// OMP50-DAG:   [[BPADDR2:%.+]] = getelementptr inbounds [6 x  i8*], [6 x  i8*]* [[BP]], i32 [[IDX2:[0-9]+]]
-// OMP50-DAG:   [[PADDR2:%.+]] = getelementptr inbounds [6 x  i8*], [6 x  i8*]* [[P]], i32 [[IDX2]]
-// OMP50-DAG:   [[BPADDR3:%.+]] = getelementptr inbounds [6 x  i8*], [6 x  i8*]* [[BP]], i32 [[IDX3:[0-9]+]]
-// OMP50-DAG:   [[PADDR3:%.+]] = getelementptr inbounds [6 x  i8*], [6 x  i8*]* [[P]], i32 [[IDX3]]
-// OMP50-DAG:   [[SADDR4:%.+]] = getelementptr inbounds [6 x  i64], [6 x  i64]* [[S]], i32 [[IDX4:[0-9]+]]
-// OMP50-DAG:   [[BPADDR4:%.+]] = getelementptr inbounds [6 x  i8*], [6 x  i8*]* [[BP]], i32 [[IDX4]]
-// OMP50-DAG:   [[PADDR4:%.+]] = getelementptr inbounds [6 x  i8*], [6 x  i8*]* [[P]], i32 [[IDX4]]
-// OMP50-DAG:   [[BPADDR5:%.+]] = getelementptr inbounds [6 x  i8*], [6 x  i8*]* [[BP]], i32 [[IDX5:[0-9]+]]
-// OMP50-DAG:   [[PADDR5:%.+]] = getelementptr inbounds [6 x  i8*], [6 x  i8*]* [[P]], i32 [[IDX5]]
+// OMP50-DAG:   store ptr [[PR:%.+]], ptr [[PARG]]
+// OMP50-DAG:   [[BPR]] = getelementptr inbounds [6 x  ptr], ptr [[BP:%.+]], i32 0, i32 0
+// OMP50-DAG:   [[PR]] = getelementptr inbounds [6 x  ptr], ptr [[P:%.+]], i32 0, i32 0
+// OMP50-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [6 x  ptr], ptr [[BP]], i32 [[IDX0:[0-9]+]]
+// OMP50-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [6 x  ptr], ptr [[P]], i32 [[IDX0]]
+// OMP50-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [6 x  ptr], ptr [[BP]], i32 [[IDX1:[0-9]+]]
+// OMP50-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [6 x  ptr], ptr [[P]], i32 [[IDX1]]
+// OMP50-DAG:   [[BPADDR2:%.+]] = getelementptr inbounds [6 x  ptr], ptr [[BP]], i32 [[IDX2:[0-9]+]]
+// OMP50-DAG:   [[PADDR2:%.+]] = getelementptr inbounds [6 x  ptr], ptr [[P]], i32 [[IDX2]]
+// OMP50-DAG:   [[BPADDR3:%.+]] = getelementptr inbounds [6 x  ptr], ptr [[BP]], i32 [[IDX3:[0-9]+]]
+// OMP50-DAG:   [[PADDR3:%.+]] = getelementptr inbounds [6 x  ptr], ptr [[P]], i32 [[IDX3]]
+// OMP50-DAG:   [[SADDR4:%.+]] = getelementptr inbounds [6 x  i64], ptr [[S]], i32 [[IDX4:[0-9]+]]
+// OMP50-DAG:   [[BPADDR4:%.+]] = getelementptr inbounds [6 x  ptr], ptr [[BP]], i32 [[IDX4]]
+// OMP50-DAG:   [[PADDR4:%.+]] = getelementptr inbounds [6 x  ptr], ptr [[P]], i32 [[IDX4]]
+// OMP50-DAG:   [[BPADDR5:%.+]] = getelementptr inbounds [6 x  ptr], ptr [[BP]], i32 [[IDX5:[0-9]+]]
+// OMP50-DAG:   [[PADDR5:%.+]] = getelementptr inbounds [6 x  ptr], ptr [[P]], i32 [[IDX5]]
 
 // The names below are not necessarily consistent with the names used for the
 // addresses above as some are repeated.
-// CHECK-DAG:   store [[S1]]* %{{.+}}, [[S1]]** [[CBPADDR0:%.+]],
-// CHECK-DAG:   store [[S1]]* %{{.+}}, [[S1]]** [[CPADDR0:%.+]],
-// CHECK-DAG:   [[CBPADDR0]] = bitcast i8** {{%[^,]+}} to [[S1]]**
-// CHECK-DAG:   [[CPADDR0]] = bitcast i8** {{%[^,]+}} to [[S1]]**
-
-// CHECK-DAG:   store i[[SZ]] [[B_CVAL]], i[[SZ]]* [[CBPADDR1:%.+]],
-// CHECK-DAG:   store i[[SZ]] [[B_CVAL]], i[[SZ]]* [[CPADDR1:%.+]],
-// CHECK-DAG:   [[CBPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]*
-
-// CHECK-DAG:   store i[[SZ]] 2, i[[SZ]]* [[CBPADDR2:%.+]],
-// CHECK-DAG:   store i[[SZ]] 2, i[[SZ]]* [[CPADDR2:%.+]],
-// CHECK-DAG:   [[CBPADDR2]] = bitcast i8** {{%[^,]+}} to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR2]] = bitcast i8** {{%[^,]+}} to i[[SZ]]*
-
-// CHECK-DAG:   store i[[SZ]] %{{.+}}, i[[SZ]]* [[CBPADDR3:%.+]],
-// CHECK-DAG:   store i[[SZ]] %{{.+}}, i[[SZ]]* [[CPADDR3:%.+]],
-// CHECK-DAG:   [[CBPADDR3]] = bitcast i8** {{%[^,]+}} to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR3]] = bitcast i8** {{%[^,]+}} to i[[SZ]]*
-
-// CHECK-DAG:   store i16* %{{.+}}, i16** [[CBPADDR4:%.+]],
-// CHECK-DAG:   store i16* %{{.+}}, i16** [[CPADDR4:%.+]],
-// CHECK-DAG:   [[CBPADDR4]] = bitcast i8** {{%[^,]+}} to i16**
-// CHECK-DAG:   [[CPADDR4]] = bitcast i8** {{%[^,]+}} to i16**
-// CHECK-DAG:   store i64 [[CSIZE]], i64* {{%[^,]+}}
-
-// OMP50-DAG:   store i[[SZ]] [[SIMD_COND]], i[[SZ]]* [[CBPADDR5:%.+]],
-// OMP50-DAG:   store i[[SZ]] [[SIMD_COND]], i[[SZ]]* [[CPADDR5:%.+]],
-// OMP50-DAG:   [[CBPADDR5]] = bitcast i8** {{%[^,]+}} to i[[SZ]]*
-// OMP50-DAG:   [[CPADDR5]] = bitcast i8** {{%[^,]+}} to i[[SZ]]*
+// CHECK-DAG:   store ptr %{{.+}}, ptr {{%[^,]+}},
+// CHECK-DAG:   store ptr %{{.+}}, ptr {{%[^,]+}},
+
+// CHECK-DAG:   store i[[SZ]] [[B_CVAL]], ptr {{%[^,]+}},
+// CHECK-DAG:   store i[[SZ]] [[B_CVAL]], ptr {{%[^,]+}},
+
+// CHECK-DAG:   store i[[SZ]] 2, ptr {{%[^,]+}},
+// CHECK-DAG:   store i[[SZ]] 2, ptr {{%[^,]+}},
+
+// CHECK-DAG:   store i[[SZ]] %{{.+}}, ptr {{%[^,]+}},
+// CHECK-DAG:   store i[[SZ]] %{{.+}}, ptr {{%[^,]+}},
+
+// CHECK-DAG:   store ptr %{{.+}}, ptr {{%[^,]+}},
+// CHECK-DAG:   store ptr %{{.+}}, ptr {{%[^,]+}},
+// CHECK-DAG:   store i64 [[CSIZE]], ptr {{%[^,]+}}
+
+// OMP50-DAG:   store i[[SZ]] [[SIMD_COND]], ptr {{%[^,]+}}
+// OMP50-DAG:   store i[[SZ]] [[SIMD_COND]], ptr {{%[^,]+}}
 
 // CHECK-NEXT:  [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
 // CHECK-NEXT:  br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]
@@ -621,41 +573,33 @@ int bar(int n){
 // CHECK:       [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 50
 // CHECK:       br i1 [[IF]], label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]]
 // CHECK:       [[IFTHEN]]
-// CHECK-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 [[DEVICE:.+]], i32 1, i32 1, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
+// CHECK-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 [[DEVICE:.+]], i32 1, i32 1, ptr @.{{.+}}.region_id, ptr [[ARGS:%.+]])
 // CHECK-DAG:   [[BPARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 2
-// CHECK-DAG:   store i8** [[BPR:%.+]], i8*** [[BPARG]]
+// CHECK-DAG:   store ptr [[BPR:%.+]], ptr [[BPARG]]
 // CHECK-DAG:   [[PARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 3
-// CHECK-DAG:   store i8** [[PR:%.+]], i8*** [[PARG]]
-// CHECK-DAG:   [[BPR]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[BP:%.+]], i32 0, i32 0
-// CHECK-DAG:   [[PR]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[P:%.+]], i32 0, i32 0
-
-// CHECK-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[BP]], i32 0, i32 0
-// CHECK-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[P]], i32 0, i32 0
-// CHECK-DAG:   [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]*
-// CHECK-DAG:   store i[[SZ]] [[VAL0:%.+]], i[[SZ]]* [[CBPADDR0]],
-// CHECK-DAG:   store i[[SZ]] [[VAL0]], i[[SZ]]* [[CPADDR0]],
-
-// CHECK-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[BP]], i32 0, i32 1
-// CHECK-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[P]], i32 0, i32 1
-// CHECK-DAG:   [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to i[[SZ]]*
-// CHECK-DAG:   store i[[SZ]] [[VAL1:%.+]], i[[SZ]]* [[CBPADDR1]],
-// CHECK-DAG:   store i[[SZ]] [[VAL1]], i[[SZ]]* [[CPADDR1]],
-
-// CHECK-DAG:   [[BPADDR2:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[BP]], i32 0, i32 2
-// CHECK-DAG:   [[PADDR2:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[P]], i32 0, i32 2
-// CHECK-DAG:   [[CBPADDR2:%.+]] = bitcast i8** [[BPADDR2]] to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR2:%.+]] = bitcast i8** [[PADDR2]] to i[[SZ]]*
-// CHECK-DAG:   store i[[SZ]] [[VAL2:%.+]], i[[SZ]]* [[CBPADDR2]],
-// CHECK-DAG:   store i[[SZ]] [[VAL2]], i[[SZ]]* [[CPADDR2]],
-
-// CHECK-DAG:   [[BPADDR3:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[BP]], i32 0, i32 3
-// CHECK-DAG:   [[PADDR3:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[P]], i32 0, i32 3
-// CHECK-DAG:   [[CBPADDR3:%.+]] = bitcast i8** [[BPADDR3]] to [10 x i32]**
-// CHECK-DAG:   [[CPADDR3:%.+]] = bitcast i8** [[PADDR3]] to [10 x i32]**
-// CHECK-DAG:   store [10 x i32]* [[VAL3:%.+]], [10 x i32]** [[CBPADDR3]],
-// CHECK-DAG:   store [10 x i32]* [[VAL3]], [10 x i32]** [[CPADDR3]],
+// CHECK-DAG:   store ptr [[PR:%.+]], ptr [[PARG]]
+// CHECK-DAG:   [[BPR]] = getelementptr inbounds [4 x ptr], ptr [[BP:%.+]], i32 0, i32 0
+// CHECK-DAG:   [[PR]] = getelementptr inbounds [4 x ptr], ptr [[P:%.+]], i32 0, i32 0
+
+// CHECK-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [4 x ptr], ptr [[BP]], i32 0, i32 0
+// CHECK-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [4 x ptr], ptr [[P]], i32 0, i32 0
+// CHECK-DAG:   store i[[SZ]] [[VAL0:%.+]], ptr [[BPADDR0]],
+// CHECK-DAG:   store i[[SZ]] [[VAL0]], ptr [[PADDR0]],
+
+// CHECK-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [4 x ptr], ptr [[BP]], i32 0, i32 1
+// CHECK-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [4 x ptr], ptr [[P]], i32 0, i32 1
+// CHECK-DAG:   store i[[SZ]] [[VAL1:%.+]], ptr [[BPADDR1]],
+// CHECK-DAG:   store i[[SZ]] [[VAL1]], ptr [[PADDR1]],
+
+// CHECK-DAG:   [[BPADDR2:%.+]] = getelementptr inbounds [4 x ptr], ptr [[BP]], i32 0, i32 2
+// CHECK-DAG:   [[PADDR2:%.+]] = getelementptr inbounds [4 x ptr], ptr [[P]], i32 0, i32 2
+// CHECK-DAG:   store i[[SZ]] [[VAL2:%.+]], ptr [[BPADDR2]],
+// CHECK-DAG:   store i[[SZ]] [[VAL2]], ptr [[PADDR2]],
+
+// CHECK-DAG:   [[BPADDR3:%.+]] = getelementptr inbounds [4 x ptr], ptr [[BP]], i32 0, i32 3
+// CHECK-DAG:   [[PADDR3:%.+]] = getelementptr inbounds [4 x ptr], ptr [[P]], i32 0, i32 3
+// CHECK-DAG:   store ptr [[VAL3:%.+]], ptr [[BPADDR3]],
+// CHECK-DAG:   store ptr [[VAL3]], ptr [[PADDR3]],
 
 // CHECK:       [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
 // CHECK-NEXT:  br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]]
@@ -675,34 +619,28 @@ int bar(int n){
 // CHECK:       [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 40
 // CHECK:       br i1 [[IF]], label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]]
 // CHECK:       [[IFTHEN]]
-// CHECK-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(%struct.ident_t* @{{.+}}, i64 [[DEVICE:.+]], i32 1, i32 1, i8* @.{{.+}}.region_id, %struct.__tgt_kernel_arguments* [[ARGS:%.+]])
+// CHECK-DAG:   [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 [[DEVICE:.+]], i32 1, i32 1, ptr @.{{.+}}.region_id, ptr [[ARGS:%.+]])
 // CHECK-DAG:   [[BPARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 2
-// CHECK-DAG:   store i8** [[BPR:%.+]], i8*** [[BPARG]]
+// CHECK-DAG:   store ptr [[BPR:%.+]], ptr [[BPARG]]
 // CHECK-DAG:   [[PARG:%.+]] = getelementptr inbounds {{.+}}[[ARGS]], i32 0, i32 3
-// CHECK-DAG:   store i8** [[PR:%.+]], i8*** [[PARG]]
-// CHECK-DAG:   [[BPR]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BP:%.+]], i32 0, i32 0
-// CHECK-DAG:   [[PR]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[P:%.+]], i32 0, i32 0
-
-// CHECK-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BP]], i32 0, i32 0
-// CHECK-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[P]], i32 0, i32 0
-// CHECK-DAG:   [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]*
-// CHECK-DAG:   store i[[SZ]] [[VAL0:%.+]], i[[SZ]]* [[CBPADDR0]],
-// CHECK-DAG:   store i[[SZ]] [[VAL0]], i[[SZ]]* [[CPADDR0]],
-
-// CHECK-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BP]], i32 0, i32 1
-// CHECK-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[P]], i32 0, i32 1
-// CHECK-DAG:   [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to i[[SZ]]*
-// CHECK-DAG:   [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to i[[SZ]]*
-// CHECK-DAG:   store i[[SZ]] [[VAL1:%.+]], i[[SZ]]* [[CBPADDR1]],
-// CHECK-DAG:   store i[[SZ]] [[VAL1]], i[[SZ]]* [[CPADDR1]],
-
-// CHECK-DAG:   [[BPADDR2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BP]], i32 0, i32 2
-// CHECK-DAG:   [[PADDR2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[P]], i32 0, i32 2
-// CHECK-DAG:   [[CBPADDR2:%.+]] = bitcast i8** [[BPADDR2]] to [10 x i32]**
-// CHECK-DAG:   [[CPADDR2:%.+]] = bitcast i8** [[PADDR2]] to [10 x i32]**
-// CHECK-DAG:   store [10 x i32]* [[VAL2:%.+]], [10 x i32]** [[CBPADDR2]],
-// CHECK-DAG:   store [10 x i32]* [[VAL2]], [10 x i32]** [[CPADDR2]],
+// CHECK-DAG:   store ptr [[PR:%.+]], ptr [[PARG]]
+// CHECK-DAG:   [[BPR]] = getelementptr inbounds [3 x ptr], ptr [[BP:%.+]], i32 0, i32 0
+// CHECK-DAG:   [[PR]] = getelementptr inbounds [3 x ptr], ptr [[P:%.+]], i32 0, i32 0
+
+// CHECK-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [3 x ptr], ptr [[BP]], i32 0, i32 0
+// CHECK-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [3 x ptr], ptr [[P]], i32 0, i32 0
+// CHECK-DAG:   store i[[SZ]] [[VAL0:%.+]], ptr [[BPADDR0]],
+// CHECK-DAG:   store i[[SZ]] [[VAL0]], ptr [[PADDR0]],
+
+// CHECK-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [3 x ptr], ptr [[BP]], i32 0, i32 1
+// CHECK-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [3 x ptr], ptr [[P]], i32 0, i32 1
+// CHECK-DAG:   store i[[SZ]] [[VAL1:%.+]], ptr [[BPADDR1]],
+// CHECK-DAG:   store i[[SZ]] [[VAL1]], ptr [[PADDR1]],
+
+// CHECK-DAG:   [[BPADDR2:%.+]] = getelementptr inbounds [3 x ptr], ptr [[BP]], i32 0, i32 2
+// CHECK-DAG:   [[PADDR2:%.+]] = getelementptr inbounds [3 x ptr], ptr [[P]], i32 0, i32 2
+// CHECK-DAG:   store ptr [[VAL2:%.+]], ptr [[BPADDR2]],
+// CHECK-DAG:   store ptr [[VAL2]], ptr [[PADDR2]],
 
 // CHECK:       [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
 // CHECK-NEXT:  br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]]
@@ -721,25 +659,23 @@ int bar(int n){
 
 // CHECK:       define internal void [[HVT7]]
 // Create local storage for each capture.
-// CHECK:       [[LOCAL_THIS:%.+]] = alloca [[S1]]*
+// CHECK:       [[LOCAL_THIS:%.+]] = alloca ptr
 // CHECK:       [[LOCAL_B:%.+]] = alloca i[[SZ]]
 // CHECK:       [[LOCAL_VLA1:%.+]] = alloca i[[SZ]]
 // CHECK:       [[LOCAL_VLA2:%.+]] = alloca i[[SZ]]
-// CHECK:       [[LOCAL_C:%.+]] = alloca i16*
+// CHECK:       [[LOCAL_C:%.+]] = alloca ptr
 // OMP50:       [[LOCAL_SIMD_COND_CASTED:%.+]] = alloca i[[SZ]],
-// CHECK-DAG:   store [[S1]]* [[ARG_THIS:%.+]], [[S1]]** [[LOCAL_THIS]]
-// CHECK-DAG:   store i[[SZ]] [[ARG_B:%.+]], i[[SZ]]* [[LOCAL_B]]
-// CHECK-DAG:   store i[[SZ]] [[ARG_VLA1:%.+]], i[[SZ]]* [[LOCAL_VLA1]]
-// CHECK-DAG:   store i[[SZ]] [[ARG_VLA2:%.+]], i[[SZ]]* [[LOCAL_VLA2]]
-// CHECK-DAG:   store i16* [[ARG_C:%.+]], i16** [[LOCAL_C]]
+// CHECK-DAG:   store ptr [[ARG_THIS:%.+]], ptr [[LOCAL_THIS]]
+// CHECK-DAG:   store i[[SZ]] [[ARG_B:%.+]], ptr [[LOCAL_B]]
+// CHECK-DAG:   store i[[SZ]] [[ARG_VLA1:%.+]], ptr [[LOCAL_VLA1]]
+// CHECK-DAG:   store i[[SZ]] [[ARG_VLA2:%.+]], ptr [[LOCAL_VLA2]]
+// CHECK-DAG:   store ptr [[ARG_C:%.+]], ptr [[LOCAL_C]]
 // Store captures in the context.
-// CHECK-DAG:   [[REF_THIS:%.+]] = load [[S1]]*, [[S1]]** [[LOCAL_THIS]],
-// CHECK-64-DAG:[[CONV_BP:%.+]] = bitcast i[[SZ]]* [[LOCAL_B]] to i32*
-// CHECK-DAG:   [[VAL_VLA1:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_VLA1]],
-// CHECK-DAG:   [[VAL_VLA2:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_VLA2]],
-// CHECK-DAG:   [[REF_C:%.+]] = load i16*, i16** [[LOCAL_C]],
-// OMP50-DAG:   [[CONV_COND:%.+]] = bitcast i[[SZ]]* [[LOCAL_SIMD_COND_CASTED]] to i8*
-// OMP50-DAG:   [[SIMD_COND:%.+]] = load i8, i8* [[CONV_COND]],
+// CHECK-DAG:   [[REF_THIS:%.+]] = load ptr, ptr [[LOCAL_THIS]],
+// CHECK-DAG:   [[VAL_VLA1:%.+]] = load i[[SZ]], ptr [[LOCAL_VLA1]],
+// CHECK-DAG:   [[VAL_VLA2:%.+]] = load i[[SZ]], ptr [[LOCAL_VLA2]],
+// CHECK-DAG:   [[REF_C:%.+]] = load ptr, ptr [[LOCAL_C]],
+// OMP50-DAG:   [[SIMD_COND:%.+]] = load i8, ptr [[LOCAL_SIMD_COND_CASTED]],
 // OMP50-DAG:   trunc i8 [[SIMD_COND]] to i1
 // OMP45-NOT:   !nontemporal
 // OMP50:       store double {{.*}}!nontemporal
@@ -751,29 +687,24 @@ int bar(int n){
 // CHECK:       [[LOCAL_A:%.+]] = alloca i[[SZ]]
 // CHECK:       [[LOCAL_AA:%.+]] = alloca i[[SZ]]
 // CHECK:       [[LOCAL_AAA:%.+]] = alloca i[[SZ]]
-// CHECK:       [[LOCAL_B:%.+]] = alloca [10 x i32]*
-// CHECK-DAG:   store i[[SZ]] [[ARG_A:%.+]], i[[SZ]]* [[LOCAL_A]]
-// CHECK-DAG:   store i[[SZ]] [[ARG_AA:%.+]], i[[SZ]]* [[LOCAL_AA]]
-// CHECK-DAG:   store i[[SZ]] [[ARG_AAA:%.+]], i[[SZ]]* [[LOCAL_AAA]]
-// CHECK-DAG:   store [10 x i32]* [[ARG_B:%.+]], [10 x i32]** [[LOCAL_B]]
+// CHECK:       [[LOCAL_B:%.+]] = alloca ptr
+// CHECK-DAG:   store i[[SZ]] [[ARG_A:%.+]], ptr [[LOCAL_A]]
+// CHECK-DAG:   store i[[SZ]] [[ARG_AA:%.+]], ptr [[LOCAL_AA]]
+// CHECK-DAG:   store i[[SZ]] [[ARG_AAA:%.+]], ptr [[LOCAL_AAA]]
+// CHECK-DAG:   store ptr [[ARG_B:%.+]], ptr [[LOCAL_B]]
 // Store captures in the context.
-// CHECK-64-DAG:[[CONV_AP:%.+]] = bitcast i[[SZ]]* [[LOCAL_A]] to i32*
-// CHECK-DAG:   [[CONV_AAP:%.+]] = bitcast i[[SZ]]* [[LOCAL_AA]] to i16*
-// CHECK-DAG:   [[CONV_AAAP:%.+]] = bitcast i[[SZ]]* [[LOCAL_AAA]] to i8*
-// CHECK-DAG:   [[REF_B:%.+]] = load [10 x i32]*, [10 x i32]** [[LOCAL_B]],
+// CHECK-DAG:   [[REF_B:%.+]] = load ptr, ptr [[LOCAL_B]],
 
 // CHECK:       define internal void [[HVT5]]
 // Create local storage for each capture.
 // CHECK:       [[LOCAL_A:%.+]] = alloca i[[SZ]]
 // CHECK:       [[LOCAL_AA:%.+]] = alloca i[[SZ]]
-// CHECK:       [[LOCAL_B:%.+]] = alloca [10 x i32]*
-// CHECK-DAG:   store i[[SZ]] [[ARG_A:%.+]], i[[SZ]]* [[LOCAL_A]]
-// CHECK-DAG:   store i[[SZ]] [[ARG_AA:%.+]], i[[SZ]]* [[LOCAL_AA]]
-// CHECK-DAG:   store [10 x i32]* [[ARG_B:%.+]], [10 x i32]** [[LOCAL_B]]
+// CHECK:       [[LOCAL_B:%.+]] = alloca ptr
+// CHECK-DAG:   store i[[SZ]] [[ARG_A:%.+]], ptr [[LOCAL_A]]
+// CHECK-DAG:   store i[[SZ]] [[ARG_AA:%.+]], ptr [[LOCAL_AA]]
+// CHECK-DAG:   store ptr [[ARG_B:%.+]], ptr [[LOCAL_B]]
 // Store captures in the context.
-// CHECK-64-DAG:[[CONV_AP:%.+]] = bitcast i[[SZ]]* [[LOCAL_A]] to i32*
-// CHECK-DAG:   [[CONV_AAP:%.+]] = bitcast i[[SZ]]* [[LOCAL_AA]] to i16*
-// CHECK-DAG:   [[REF_B:%.+]] = load [10 x i32]*, [10 x i32]** [[LOCAL_B]],
+// CHECK-DAG:   [[REF_B:%.+]] = load ptr, ptr [[LOCAL_B]],
 
 // OMP45-NOT: !{!"llvm.loop.vectorize.enable", i1 false}
 // TOMP45-NOT: !{!"llvm.loop.vectorize.enable", i1 false}
diff --git a/clang/test/OpenMP/target_update_depend_codegen.cpp b/clang/test/OpenMP/target_update_depend_codegen.cpp
index da0207017cd7..a19ff8fe3acc 100644
--- a/clang/test/OpenMP/target_update_depend_codegen.cpp
+++ b/clang/test/OpenMP/target_update_depend_codegen.cpp
@@ -1,20 +1,20 @@
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-64
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-64
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
 
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
-// RUN: %clang_cc1 -no-opaque-pointers -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
 
 // expected-no-diagnostics
-// CK1: [[ST:%.+]] = type { i32, double* }
+// CK1: [[ST:%.+]] = type { i32, ptr }
 // CK1: %struct.kmp_depend_info = type { i[[sz:64|32]],
 // CK1-SAME: i[[sz]], i8 }
 #ifndef HEADER
@@ -51,59 +51,44 @@ void foo(int arg) {
   // CK1: alloca [5 x %struct.kmp_depend_info],
 
   // Region 00
-  // CK1: [[BP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP:%.+]], i32 0, i32 0
-  // CK1: [[BP0_BC:%.+]] = bitcast i8** [[BP0]] to [100 x double]**
-  // CK1: store [100 x double]* @gc, [100 x double]** [[BP0_BC]],
-  // CK1: [[P0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P:%.+]], i32 0, i32 0
-  // CK1: [[P0_BC:%.+]] = bitcast i8** [[P0]] to [100 x double]**
-  // CK1: store [100 x double]* @gc, [100 x double]** [[P0_BC]],
-  // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP]], i32 0, i32 0
-  // CK1: [[GEPP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P]], i32 0, i32 0
-  // CK1: [[CAP_DEVICE:%.+]] = getelementptr inbounds %struct.anon, %struct.anon* [[CAPTURES:%.+]], i32 0, i32 0
-  // CK1: [[DEVICE:%.+]] = load i32, i32* %{{.+}}
-  // CK1: store i32 [[DEVICE]], i32* [[CAP_DEVICE]],
-  // CK1: [[DEV1:%.+]] = load i32, i32* %{{.+}}
+  // CK1: [[BP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[BP:%.+]], i32 0, i32 0
+  // CK1: store ptr @gc, ptr [[BP0]],
+  // CK1: [[P0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[P:%.+]], i32 0, i32 0
+  // CK1: store ptr @gc, ptr [[P0]],
+  // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[BP]], i32 0, i32 0
+  // CK1: [[GEPP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[P]], i32 0, i32 0
+  // CK1: [[CAP_DEVICE:%.+]] = getelementptr inbounds %struct.anon, ptr [[CAPTURES:%.+]], i32 0, i32 0
+  // CK1: [[DEVICE:%.+]] = load i32, ptr %{{.+}}
+  // CK1: store i32 [[DEVICE]], ptr [[CAP_DEVICE]],
+  // CK1: [[DEV1:%.+]] = load i32, ptr %{{.+}}
   // CK1: [[DEV2:%.+]] = sext i32 [[DEV1]] to i64
-  // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_target_task_alloc(%struct.ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{64|36}}, i[[sz]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates*)* [[TASK_ENTRY0:@.+]] to i32 (i32, i8*)*), i64 [[DEV2]])
-  // CK1: [[BC:%.+]] = bitcast i8* [[RES]] to %struct.kmp_task_t_with_privates*
-  // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates, %struct.kmp_task_t_with_privates* [[BC]], i32 0, i32 0
-  // CK1: [[SHAREDS:%.+]] = getelementptr inbounds %struct.kmp_task_t, %struct.kmp_task_t* [[TASK_T]], i32 0, i32 0
-  // CK1: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS]],
-  // CK1: [[BC1:%.+]] = bitcast %struct.anon* [[CAPTURES]] to i8*
-  // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align 4 [[SHAREDS_REF]], i8* align 4 [[BC1]], i[[sz]] 4, i1 false)
-  // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates, %struct.kmp_task_t_with_privates* [[BC]], i32 0, i32 1
-  // CK1-64: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 0
-  // CK1-64: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8*
-  // CK1-64: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align 8 [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-64: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 1
-  // CK1-64: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8*
-  // CK1-64: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align 8 [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-64: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 2
-  // CK1-64: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i64]* [[PRIVS_SIZES]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align 8 [[BC_PRIVS_SIZES]], i8* align {{8|4}} bitcast ([1 x i64]* [[SIZE00]] to i8*), i[[sz]] 8, i1 false)
-  // CK1-32: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 0
-  // CK1-32: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i64]* [[PRIVS_SIZES]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align 4 [[BC_PRIVS_SIZES]], i8* align 4 bitcast ([1 x i64]* [[SIZE00]] to i8*), i[[sz]] 8, i1 false)
-  // CK1-32: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 1
-  // CK1-32: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8*
-  // CK1-32: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-32: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 2
-  // CK1-32: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8*
-  // CK1-32: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1: [[BC_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP:%.+]], i[[sz]] 0
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 4, i[[sz]]* [[DEP_SIZE]],
-  // 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 i32 @__kmpc_omp_task_with_deps(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]], i32 1, i8* [[BC]], i32 0, i8* null)
+  // CK1: [[RES:%.+]] = call ptr @__kmpc_omp_target_task_alloc(ptr {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{64|36}}, i[[sz]] 4, ptr [[TASK_ENTRY0:@.+]], i64 [[DEV2]])
+  // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates, ptr [[RES]], i32 0, i32 0
+  // CK1: [[SHAREDS:%.+]] = getelementptr inbounds %struct.kmp_task_t, ptr [[TASK_T]], i32 0, i32 0
+  // CK1: [[SHAREDS_REF:%.+]] = load ptr, ptr [[SHAREDS]],
+  // CK1: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align 4 [[SHAREDS_REF]], ptr align 4 [[CAPTURES]], i[[sz]] 4, i1 false)
+  // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates, ptr [[RES]], i32 0, i32 1
+  // CK1-64: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, ptr [[PRIVS]], i32 0, i32 0
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align 8 [[PRIVS_BASEPTRS]], ptr align {{8|4}} [[GEPBP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-64: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, ptr [[PRIVS]], i32 0, i32 1
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align 8 [[PRIVS_PTRS]], ptr align {{8|4}} [[GEPP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-64: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t, ptr [[PRIVS]], i32 0, i32 2
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align 8 [[PRIVS_SIZES]], ptr align {{8|4}} [[SIZE00]], i[[sz]] 8, i1 false)
+  // CK1-32: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t, ptr [[PRIVS]], i32 0, i32 0
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align 4 [[PRIVS_SIZES]], ptr align 4 [[SIZE00]], i[[sz]] 8, i1 false)
+  // CK1-32: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, ptr [[PRIVS]], i32 0, i32 1
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_BASEPTRS]], ptr align {{8|4}} [[GEPBP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-32: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, ptr [[PRIVS]], i32 0, i32 2
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_PTRS]], ptr align {{8|4}} [[GEPP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP:%.+]], i[[sz]] 0
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 4, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 1, ptr [[DEP_ATTRS]]
+  // CK1: = call i32 @__kmpc_omp_task_with_deps(ptr @{{.+}}, i32 %{{.+}}, ptr [[RES]], i32 1, ptr [[MAIN_DEP]], i32 0, ptr null)
 
   // CK1: %{{.+}} = add nsw i32 %{{[^,]+}}, 1
   #pragma omp target update if(1+3-5) device(arg) from(gc) nowait depend(in: arg)
@@ -117,77 +102,62 @@ void foo(int arg) {
   // Region 02
   // CK1: br i1 %{{[^,]+}}, label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]]
   // CK1: [[IFTHEN]]
-  // CK1: [[BP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP:%.+]], i32 0, i32 0
-  // CK1: [[BP0_BC:%.+]] = bitcast i8** [[BP0]] to i32**
-  // CK1: store i32* [[ARG:%.+]], i32** [[BP0_BC]],
-  // CK1: [[P0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P:%.+]], i32 0, i32 0
-  // CK1: [[P0_BC:%.+]] = bitcast i8** [[P0]] to i32**
-  // CK1: store i32* [[ARG]], i32** [[P0_BC]],
-  // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP]], i32 0, i32 0
-  // CK1: [[GEPP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P]], i32 0, i32 0
-  // CK1: [[IF_DEVICE:%.+]] = getelementptr inbounds %struct.anon{{.+}}, %struct.anon{{.+}}* [[CAPTURES:%.+]], i32 0, i32 0
-  // CK1: [[IF:%.+]] = load i8, i8* %{{.+}}
+  // CK1: [[BP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[BP:%.+]], i32 0, i32 0
+  // CK1: store ptr [[ARG:%.+]], ptr [[BP0]],
+  // CK1: [[P0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[P:%.+]], i32 0, i32 0
+  // CK1: store ptr [[ARG]], ptr [[P0]],
+  // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[BP]], i32 0, i32 0
+  // CK1: [[GEPP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[P]], i32 0, i32 0
+  // CK1: [[IF_DEVICE:%.+]] = getelementptr inbounds %struct.anon{{.+}}, ptr [[CAPTURES:%.+]], i32 0, i32 0
+  // CK1: [[IF:%.+]] = load i8, ptr %{{.+}}
   // CK1: [[IF_BOOL:%.+]] = trunc i8 [[IF]] to i1
   // CK1: [[IF:%.+]] = zext i1 [[IF_BOOL]] to i8
-  // CK1: store i8 [[IF]], i8* [[IF_DEVICE]],
-  // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{64|36}}, i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY2:@.+]] to i32 (i32, i8*)*))
-  // CK1: [[RES_BC:%.+]] = bitcast i8* [[RES]] to %struct.kmp_task_t_with_privates{{.+}}*
-  // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 0
-  // CK1: [[SHAREDS:%.+]] = getelementptr inbounds %struct.kmp_task_t, %struct.kmp_task_t* [[TASK_T]], i32 0, i32 0
-  // CK1: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS]],
-  // CK1: [[BC1:%.+]] = bitcast %struct.anon{{.+}}* [[CAPTURES]] to i8*
-  // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align 1 [[SHAREDS_REF]], i8* align 1 [[BC1]], i[[sz]] 1, i1 false)
-  // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 1
-  // CK1-64: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0
-  // CK1-64: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8*
-  // CK1-64: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-64: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1
-  // CK1-64: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8*
-  // CK1-64: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-64: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2
-  // CK1-64: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i[[sz]]]* [[PRIVS_SIZES]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} bitcast ([1 x i[[sz]]]* [[SIZE02]] to i8*), i[[sz]] {{8|4}}, i1 false)
-  // CK1-32: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0
-  // CK1-32: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i64]* [[PRIVS_SIZES]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} bitcast ([1 x i64]* [[SIZE02]] to i8*), i[[sz]] {{8|4}}, i1 false)
-  // CK1-32: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1
-  // CK1-32: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8*
-  // CK1-32: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-32: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2
-  // CK1-32: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8*
-  // CK1-32: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1: [[BC_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP:%.+]], i[[sz]] 0
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 4, i[[sz]]* [[DEP_SIZE]],
-  // 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_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 1
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 4, i[[sz]]* [[DEP_SIZE]],
-  // 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: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 2
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] ptrtoint ([100 x double]* @gc to i[[sz]]), i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 800, i[[sz]]* [[DEP_SIZE]],
-  // 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_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]])
+  // CK1: store i8 [[IF]], ptr [[IF_DEVICE]],
+  // CK1: [[RES:%.+]] = call ptr @__kmpc_omp_task_alloc(ptr {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{64|36}}, i[[sz]] 1, ptr [[TASK_ENTRY2:@.+]])
+  // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, ptr [[RES]], i32 0, i32 0
+  // CK1: [[SHAREDS:%.+]] = getelementptr inbounds %struct.kmp_task_t, ptr [[TASK_T]], i32 0, i32 0
+  // CK1: [[SHAREDS_REF:%.+]] = load ptr, ptr [[SHAREDS]],
+  // CK1: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align 1 [[SHAREDS_REF]], ptr align 1 [[CAPTURES]], i[[sz]] 1, i1 false)
+  // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, ptr [[RES]], i32 0, i32 1
+  // CK1-64: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 0
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_BASEPTRS]], ptr align {{8|4}} [[GEPBP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-64: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 1
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_PTRS]], ptr align {{8|4}} [[GEPP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-64: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 2
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_SIZES]], ptr align {{8|4}} [[SIZE02]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-32: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 0
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_SIZES]], ptr align {{8|4}} [[SIZE02]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-32: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 1
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_BASEPTRS]], ptr align {{8|4}} [[GEPBP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-32: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 2
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_PTRS]], ptr align {{8|4}} [[GEPP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP:%.+]], i[[sz]] 0
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 4, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 3, ptr [[DEP_ATTRS]]
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 1
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 4, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 3, ptr [[DEP_ATTRS]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 2
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] ptrtoint (ptr @gc to i[[sz]]), ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 800, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 3, ptr [[DEP_ATTRS]]
+  // CK1: call void @__kmpc_omp_taskwait_deps_51(ptr @{{.+}}, i32 %{{.+}}, i32 3, ptr [[MAIN_DEP]], i32 0, ptr null, i32 0)
+  // CK1: call void @__kmpc_omp_task_begin_if0(ptr @{{.+}}, i32 %{{.+}}, ptr [[RES]])
+  // CK1: = call i32 [[TASK_ENTRY2]](i32 %{{.+}}, ptr [[RES]])
+  // CK1: call void @__kmpc_omp_task_complete_if0(ptr @{{.+}}, i32 %{{.+}}, ptr [[RES]])
 
   // CK1: br label %[[IFEND:[^,]+]]
 
@@ -202,81 +172,65 @@ void foo(int arg) {
   {++arg;}
 
   // Region 03
-  // CK1: [[BP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP:%.+]], i32 0, i32 0
-  // CK1: [[BP0_BC:%.+]] = bitcast i8** [[BP0]] to float**
-  // CK1: store float* [[VLA:%.+]], float** [[BP0_BC]],
-  // CK1: [[P0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P:%.+]], i32 0, i32 0
-  // CK1: [[P0_BC:%.+]] = bitcast i8** [[P0]] to float**
-  // CK1: store float* [[VLA]], float** [[P0_BC]],
-  // CK1: [[S0:%.+]] = getelementptr inbounds [1 x i64], [1 x i64]* [[S:%.+]], i32 0, i32 0
-  // CK1: store i64 {{.+}}, i64* [[S0]],
-  // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP]], i32 0, i32 0
-  // CK1: [[GEPP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P]], i32 0, i32 0
-  // CK1: [[GEPS0:%.+]] = getelementptr inbounds [1 x i64], [1 x i64]* [[S]], i32 0, i32 0
-  // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{64|36}}, i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY3:@.+]] to i32 (i32, i8*)*))
-  // CK1: [[RES_BC:%.+]] = bitcast i8* [[RES]] to %struct.kmp_task_t_with_privates{{.+}}*
-  // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 0
-  // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 1
-  // CK1-64: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0
-  // CK1-64: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8*
-  // CK1-64: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-64: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1
-  // CK1-64: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8*
-  // CK1-64: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-64: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2
-  // CK1-64: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i[[sz]]]* [[PRIVS_SIZES]] to i8*
-  // CK1-64: [[BC_SIZES:%.+]] = bitcast i[[sz]]* [[GEPS0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} [[BC_SIZES]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-32: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0
-  // CK1-32: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i64]* [[PRIVS_SIZES]] to i8*
-  // CK1-32: [[BC_SIZES:%.+]] = bitcast i64* [[GEPS0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} [[BC_SIZES]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-32: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1
-  // CK1-32: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8*
-  // CK1-32: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1-32: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2
-  // CK1-32: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8*
-  // CK1-32: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false)
-  // CK1: [[BC_ADR:%.+]] = ptrtoint float* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP:%.+]], i[[sz]] 0
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] %{{.+}}, i[[sz]]* [[DEP_SIZE]],
-  // 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_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 1
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 4, i[[sz]]* [[DEP_SIZE]],
-  // 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_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 2
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 4, i[[sz]]* [[DEP_SIZE]],
-  // 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: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 3
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] ptrtoint ([100 x double]* @gc to i[[sz]]), i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 800, i[[sz]]* [[DEP_SIZE]],
-  // 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_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]])
+  // CK1: [[BP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[BP:%.+]], i32 0, i32 0
+  // CK1: store ptr [[VLA:%.+]], ptr [[BP0]],
+  // CK1: [[P0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[P:%.+]], i32 0, i32 0
+  // CK1: store ptr [[VLA]], ptr [[P0]],
+  // CK1: [[S0:%.+]] = getelementptr inbounds [1 x i64], ptr [[S:%.+]], i32 0, i32 0
+  // CK1: store i64 {{.+}}, ptr [[S0]],
+  // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[BP]], i32 0, i32 0
+  // CK1: [[GEPP0:%.+]] = getelementptr inbounds [1 x ptr], ptr [[P]], i32 0, i32 0
+  // CK1: [[GEPS0:%.+]] = getelementptr inbounds [1 x i64], ptr [[S]], i32 0, i32 0
+  // CK1: [[RES:%.+]] = call ptr @__kmpc_omp_task_alloc(ptr {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{64|36}}, i[[sz]] 1, ptr [[TASK_ENTRY3:@.+]])
+  // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, ptr [[RES]], i32 0, i32 0
+  // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, ptr [[RES]], i32 0, i32 1
+  // CK1-64: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 0
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_BASEPTRS]], ptr align {{8|4}} [[GEPBP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-64: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 1
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_PTRS]], ptr align {{8|4}} [[GEPP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-64: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 2
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_SIZES]], ptr align {{8|4}} [[GEPS0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-32: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 0
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_SIZES]], ptr align {{8|4}} [[GEPS0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-32: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 1
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_BASEPTRS]], ptr align {{8|4}} [[GEPBP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1-32: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 2
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_PTRS]], ptr align {{8|4}} [[GEPP0]], i[[sz]] {{8|4}}, i1 false)
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP:%.+]], i[[sz]] 0
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] %{{.+}}, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 3, ptr [[DEP_ATTRS]]
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 1
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 4, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 3, ptr [[DEP_ATTRS]]
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 2
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 4, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 3, ptr [[DEP_ATTRS]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 3
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] ptrtoint (ptr @gc to i[[sz]]), ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 800, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 3, ptr [[DEP_ATTRS]]
+  // CK1: call void @__kmpc_omp_taskwait_deps_51(ptr @{{.+}}, i32 %{{.+}}, i32 4, ptr [[MAIN_DEP]], i32 0, ptr null, i32 0)
+  // CK1: call void @__kmpc_omp_task_begin_if0(ptr @{{.+}}, i32 %{{.+}}, ptr [[RES]])
+  // CK1: = call i32 [[TASK_ENTRY3]](i32 %{{.+}}, ptr [[RES]])
+  // CK1: call void @__kmpc_omp_task_complete_if0(ptr @{{.+}}, i32 %{{.+}}, ptr [[RES]])
   #pragma omp target update from(lb) depend(out: lb, arg, la, gc)
   {++arg;}
 
@@ -284,153 +238,119 @@ void foo(int arg) {
   {++arg;}
 
   // Region 04
-  // CK1: [[DIV:%.+]] = sdiv exact i64 sub (i64 ptrtoint (double** getelementptr (double*, double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1) to i64)), ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)
-  // CK1: [[BP0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP:%.+]], i32 0, i32 0
-  // CK1: [[BP0_BC:%.+]] = bitcast i8** [[BP0]] to %struct.ST**
-  // CK1: store %struct.ST* @gb, %struct.ST** [[BP0_BC]],
-  // CK1: [[P0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P:%.+]], i32 0, i32 0
-  // CK1: [[P0_BC:%.+]] = bitcast i8** [[P0]] to double***
-  // CK1: store double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), double*** [[P0_BC]],
-  // CK1: [[PS0:%.+]] = getelementptr inbounds [2 x i64], [2 x i64]* [[PS:%.+]], i32 0, i32 0
-  // CK1: store i64 [[DIV]], i64* [[PS0]],
-  // CK1: [[BP1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 1
-  // CK1: [[BP1_BC:%.+]] = bitcast i8** [[BP1]] to double***
-  // CK1: store double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), double*** [[BP1_BC]],
-  // CK1: [[P1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 1
-  // CK1: [[P1_BC:%.+]] = bitcast i8** [[P1]] to double**
-  // CK1: store double* %{{.+}}, double** [[P1_BC]],
-  // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 0
-  // CK1: [[GEPP0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 0
-  // CK1: [[GEPS0:%.+]] = getelementptr inbounds [2 x i64], [2 x i64]* [[PS]], i32 0, i32 0
-  // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{88|52}}, i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY4:@.+]] to i32 (i32, i8*)*))
-  // CK1: [[RES_BC:%.+]] = bitcast i8* [[RES]] to %struct.kmp_task_t_with_privates{{.+}}*
-  // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 0
-  // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 1
-  // CK1-64: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0
-  // CK1-64: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [2 x i8*]* [[PRIVS_BASEPTRS]] to i8*
-  // CK1-64: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{16|8}}, i1 false)
-  // CK1-64: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1
-  // CK1-64: [[BC_PRIVS_PTRS:%.+]] = bitcast [2 x i8*]* [[PRIVS_PTRS]] to i8*
-  // CK1-64: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{16|8}}, i1 false)
-  // CK1-64: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2
-  // CK1-64: [[BC_PRIVS_SIZES:%.+]] = bitcast [2 x i[[sz]]]* [[PRIVS_SIZES]] to i8*
-  // CK1-64: [[BC_SIZES:%.+]] = bitcast i64* [[GEPS0]] to i8*
-  // CK1-64: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} [[BC_SIZES]], i[[sz]] {{16|8}}, i1 false)
-  // CK1-32: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0
-  // CK1-32: [[BC_PRIVS_SIZES:%.+]] = bitcast [2 x i64]* [[PRIVS_SIZES]] to i8*
-  // CK1-32: [[BC_SIZES:%.+]] = bitcast i64* [[GEPS0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} [[BC_SIZES]], i[[sz]] {{16|8}}, i1 false)
-  // CK1-32: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1
-  // CK1-32: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [2 x i8*]* [[PRIVS_BASEPTRS]] to i8*
-  // CK1-32: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{16|8}}, i1 false)
-  // CK1-32: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2
-  // CK1-32: [[BC_PRIVS_PTRS:%.+]] = bitcast [2 x i8*]* [[PRIVS_PTRS]] to i8*
-  // CK1-32: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8*
-  // CK1-32: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{16|8}}, i1 false)
+  // CK1: [[DIV:%.+]] = sdiv exact i64 sub (i64 ptrtoint (ptr getelementptr (ptr, ptr getelementptr inbounds (%struct.ST, ptr @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (ptr getelementptr inbounds (%struct.ST, ptr @gb, i32 0, i32 1) to i64)), ptrtoint (ptr getelementptr (i8, ptr null, i32 1) to i64)
+  // CK1: [[BP0:%.+]] = getelementptr inbounds [2 x ptr], ptr [[BP:%.+]], i32 0, i32 0
+  // CK1: store ptr @gb, ptr [[BP0]],
+  // CK1: [[P0:%.+]] = getelementptr inbounds [2 x ptr], ptr [[P:%.+]], i32 0, i32 0
+  // CK1: store ptr getelementptr inbounds (%struct.ST, ptr @gb, i32 0, i32 1), ptr [[P0]],
+  // CK1: [[PS0:%.+]] = getelementptr inbounds [2 x i64], ptr [[PS:%.+]], i32 0, i32 0
+  // CK1: store i64 [[DIV]], ptr [[PS0]],
+  // CK1: [[BP1:%.+]] = getelementptr inbounds [2 x ptr], ptr [[BP]], i32 0, i32 1
+  // CK1: store ptr getelementptr inbounds (%struct.ST, ptr @gb, i32 0, i32 1), ptr [[BP1]],
+  // CK1: [[P1:%.+]] = getelementptr inbounds [2 x ptr], ptr [[P]], i32 0, i32 1
+  // CK1: store ptr %{{.+}}, ptr [[P1]],
+  // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [2 x ptr], ptr [[BP]], i32 0, i32 0
+  // CK1: [[GEPP0:%.+]] = getelementptr inbounds [2 x ptr], ptr [[P]], i32 0, i32 0
+  // CK1: [[GEPS0:%.+]] = getelementptr inbounds [2 x i64], ptr [[PS]], i32 0, i32 0
+  // CK1: [[RES:%.+]] = call ptr @__kmpc_omp_task_alloc(ptr {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{88|52}}, i[[sz]] 1, ptr [[TASK_ENTRY4:@.+]])
+  // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, ptr [[RES]], i32 0, i32 0
+  // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, ptr [[RES]], i32 0, i32 1
+  // CK1-64: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 0
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_BASEPTRS]], ptr align {{8|4}} [[GEPBP0]], i[[sz]] {{16|8}}, i1 false)
+  // CK1-64: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 1
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_PTRS]], ptr align {{8|4}} [[GEPP0]], i[[sz]] {{16|8}}, i1 false)
+  // CK1-64: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 2
+  // CK1-64: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_SIZES]], ptr align {{8|4}} [[GEPS0]], i[[sz]] {{16|8}}, i1 false)
+  // CK1-32: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 0
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_SIZES]], ptr align {{8|4}} [[GEPS0]], i[[sz]] {{16|8}}, i1 false)
+  // CK1-32: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 1
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_BASEPTRS]], ptr align {{8|4}} [[GEPBP0]], i[[sz]] {{16|8}}, i1 false)
+  // CK1-32: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, ptr [[PRIVS]], i32 0, i32 2
+  // CK1-32: call void @llvm.memcpy.p0.p0.i[[sz]](ptr align {{8|4}} [[PRIVS_PTRS]], ptr align {{8|4}} [[GEPP0]], i[[sz]] {{16|8}}, i1 false)
   // CK1: %{{.+}} = sub nuw
-  // CK1: [[BC_ADR:%.+]] = ptrtoint double* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP:%.+]], i[[sz]] 0
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] %{{.+}}, i[[sz]]* [[DEP_SIZE]],
-  // 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_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 1
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 4, i[[sz]]* [[DEP_SIZE]],
-  // 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_ADR:%.+]] = ptrtoint float* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 2
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] %{{.+}}, i[[sz]]* [[DEP_SIZE]],
-  // 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: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 3
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] ptrtoint ([100 x double]* @gc to i[[sz]]), i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 800, i[[sz]]* [[DEP_SIZE]],
-  // 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_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]]
-  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, %struct.kmp_depend_info* [[MAIN_DEP]], i[[sz]] 4
-  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0
-  // CK1: store i[[sz]] [[BC_ADR]], i[[sz]]* [[DEP_ADR]],
-  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 1
-  // CK1: store i[[sz]] 4, i[[sz]]* [[DEP_SIZE]],
-  // 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_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]])
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP:%.+]], i[[sz]] 0
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] %{{.+}}, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 1, ptr [[DEP_ATTRS]]
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 1
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 4, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 1, ptr [[DEP_ATTRS]]
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 2
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] %{{.+}}, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 1, ptr [[DEP_ATTRS]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 3
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] ptrtoint (ptr @gc to i[[sz]]), ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 800, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 1, ptr [[DEP_ATTRS]]
+  // CK1: [[BC_ADR:%.+]] = ptrtoint ptr %{{.+}} to i[[sz]]
+  // CK1: [[DEP:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[MAIN_DEP]], i[[sz]] 4
+  // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 0
+  // CK1: store i[[sz]] [[BC_ADR]], ptr [[DEP_ADR]],
+  // CK1: [[DEP_SIZE:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 1
+  // CK1: store i[[sz]] 4, ptr [[DEP_SIZE]],
+  // CK1: [[DEP_ATTRS:%.+]] = getelementptr inbounds %struct.kmp_depend_info, ptr [[DEP]], i32 0, i32 2
+  // CK1: store i8 1, ptr [[DEP_ATTRS]]
+  // CK1: call void @__kmpc_omp_taskwait_deps_51(ptr @{{.+}}, i32 %{{.+}}, i32 5, ptr [[MAIN_DEP]], i32 0, ptr null, i32 0)
+  // CK1: call void @__kmpc_omp_task_begin_if0(ptr @{{.+}}, i32 %{{.+}}, ptr [[RES]])
+  // CK1: = call i32 [[TASK_ENTRY4]](i32 %{{.+}}, ptr [[RES]])
+  // CK1: call void @__kmpc_omp_task_complete_if0(ptr @{{.+}}, i32 %{{.+}}, ptr [[RES]])
   #pragma omp target update to(gb.b[:3]) depend(in: gb.b[:3], la, lb, gc, arg)
   {++arg;}
 }
 
-// CK1: define internal{{.*}} i32 [[TASK_ENTRY0]](i32{{.*}}, %struct.kmp_task_t_with_privates* noalias noundef %1)
-// CK1-DAG: call void @__tgt_target_data_update_nowait_mapper(%struct.ident_t* @{{.+}}, i64 [[DEV:%[^,]+]], i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE00]]{{.+}}, i8** null)
+// CK1: define internal{{.*}} i32 [[TASK_ENTRY0]](i32{{.*}}, ptr noalias noundef %1)
+// CK1-DAG: call void @__tgt_target_data_update_nowait_mapper(ptr @{{.+}}, i64 [[DEV:%[^,]+]], i32 1, ptr [[BP:%.+]], ptr [[P:%.+]], ptr [[S:%.+]], ptr [[MTYPE00]]{{.+}}, ptr null)
 // CK1-DAG: [[DEV]] = sext i32 [[DEVi32:%[^,]+]] to i64
-// CK1-DAG: [[DEVi32]] = load i32, i32* %{{[^,]+}},
-// CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
-// CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
-// CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
-// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%[^,]+]],
-// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%[^,]+]],
-// CK1-DAG: [[S]] = load [1 x i64]*, [1 x i64]** [[S_PRIV:%[^,]+]],
-// CK1-DAG: [[FN:%.+]] = bitcast void (i8*, ...)* {{%.*}} to void (i8*,
-// CK1-DAG: call void [[FN]](i8* %{{[^,]+}}, [1 x i8*]** [[BP_PRIV]], [1 x i8*]** [[P_PRIV]], [1 x i64]** [[S_PRIV]])
+// CK1-DAG: [[DEVi32]] = load i32, ptr %{{[^,]+}},
+// CK1-DAG: [[BP]] = load ptr, ptr [[BP_PRIV:%[^,]+]],
+// CK1-DAG: [[P]] = load ptr, ptr [[P_PRIV:%[^,]+]],
+// CK1-DAG: [[S]] = load ptr, ptr [[S_PRIV:%[^,]+]],
+// CK1-DAG: call void {{%.*}}(ptr %{{[^,]+}}, ptr [[BP_PRIV]], ptr [[P_PRIV]], ptr [[S_PRIV]])
 // CK1: ret i32 0
 // CK1: }
 
-// CK1: define internal{{.*}} i32 [[TASK_ENTRY2]](i32{{.*}}, %struct.kmp_task_t_with_privates{{.+}}* noalias noundef %1)
-// CK1-DAG: call void @__tgt_target_data_update_mapper(%struct.ident_t* @{{.+}}, i64 4, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE02]]{{.+}}, i8** null)
-// CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
-// CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
-// CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
-// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%[^,]+]],
-// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%[^,]+]],
-// CK1-DAG: [[S]] = load [1 x i64]*, [1 x i64]** [[S_PRIV:%[^,]+]],
-// CK1-DAG: [[FN:%.+]] = bitcast void (i8*, ...)* {{%.*}} to void (i8*,
-// CK1-DAG: call void [[FN]](i8* %{{[^,]+}}, [1 x i8*]** [[BP_PRIV]], [1 x i8*]** [[P_PRIV]], [1 x i64]** [[S_PRIV]])
+// CK1: define internal{{.*}} i32 [[TASK_ENTRY2]](i32{{.*}}, ptr noalias noundef %1)
+// CK1-DAG: call void @__tgt_target_data_update_mapper(ptr @{{.+}}, i64 4, i32 1, ptr [[BP:%.+]], ptr [[P:%.+]], ptr [[S:%.+]], ptr [[MTYPE02]]{{.+}}, ptr null)
+// CK1-DAG: [[BP]] = load ptr, ptr [[BP_PRIV:%[^,]+]],
+// CK1-DAG: [[P]] = load ptr, ptr [[P_PRIV:%[^,]+]],
+// CK1-DAG: [[S]] = load ptr, ptr [[S_PRIV:%[^,]+]],
+// CK1-DAG: call void {{%.*}}(ptr %{{[^,]+}}, ptr [[BP_PRIV]], ptr [[P_PRIV]], ptr [[S_PRIV]])
 // CK1: ret i32 0
 // CK1: }
 
-// CK1: define internal{{.*}} i32 [[TASK_ENTRY3]](i32{{.*}}, %struct.kmp_task_t_with_privates{{.+}}* noalias noundef %1)
-// CK1-DAG: call void @__tgt_target_data_update_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE03]]{{.+}}, i8** null)
-// CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
-// CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
-// CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
-// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%[^,]+]],
-// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%[^,]+]],
-// CK1-DAG: [[S]] = load [1 x i64]*, [1 x i64]** [[S_PRIV:%[^,]+]],
-// CK1-DAG: [[FN:%.+]] = bitcast void (i8*, ...)* {{%.*}} to void (i8*,
-// CK1-DAG: call void [[FN]](i8* %{{[^,]+}}, [1 x i8*]** [[BP_PRIV]], [1 x i8*]** [[P_PRIV]], [1 x i64]** [[S_PRIV]])
+// CK1: define internal{{.*}} i32 [[TASK_ENTRY3]](i32{{.*}}, ptr noalias noundef %1)
+// CK1-DAG: call void @__tgt_target_data_update_mapper(ptr @{{.+}}, i64 -1, i32 1, ptr [[BP:%.+]], ptr [[P:%.+]], ptr [[S:%.+]], ptr [[MTYPE03]]{{.+}}, ptr null)
+// CK1-DAG: [[BP]] = load ptr, ptr [[BP_PRIV:%[^,]+]],
+// CK1-DAG: [[P]] = load ptr, ptr [[P_PRIV:%[^,]+]],
+// CK1-DAG: [[S]] = load ptr, ptr [[S_PRIV:%[^,]+]],
+// CK1-DAG: call void {{%.*}}(ptr %{{[^,]+}}, ptr [[BP_PRIV]], ptr [[P_PRIV]], ptr [[S_PRIV]])
 // CK1-NOT: __tgt_target_data_end
 // CK1: ret i32 0
 // CK1: }
 
-// CK1: define internal{{.*}} i32 [[TASK_ENTRY4]](i32{{.*}}, %struct.kmp_task_t_with_privates{{.+}}* noalias noundef %1)
-// CK1-DAG: call void @__tgt_target_data_update_mapper(%struct.ident_t* @{{.+}}, i64 -1, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i64* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE04]]{{.+}}, i8** null)
-// CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
-// CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
-// CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
-// CK1-DAG: [[BP]] = load [2 x i8*]*, [2 x i8*]** [[BP_PRIV:%[^,]+]],
-// CK1-DAG: [[P]] = load [2 x i8*]*, [2 x i8*]** [[P_PRIV:%[^,]+]],
-// CK1-DAG: [[S]] = load [2 x i64]*, [2 x i64]** [[S_PRIV:%[^,]+]],
-// CK1-DAG: [[FN:%.+]] = bitcast void (i8*, ...)* {{%.*}} to void (i8*,
-// CK1-DAG: call void [[FN]](i8* %{{[^,]+}}, [2 x i8*]** [[BP_PRIV]], [2 x i8*]** [[P_PRIV]], [2 x i64]** [[S_PRIV]])
+// CK1: define internal{{.*}} i32 [[TASK_ENTRY4]](i32{{.*}}, ptr noalias noundef %1)
+// CK1-DAG: call void @__tgt_target_data_update_mapper(ptr @{{.+}}, i64 -1, i32 2, ptr [[BP:%.+]], ptr [[P:%.+]], ptr [[S:%.+]], ptr [[MTYPE04]]{{.+}}, ptr null)
+// CK1-DAG: [[BP]] = load ptr, ptr [[BP_PRIV:%[^,]+]],
+// CK1-DAG: [[P]] = load ptr, ptr [[P_PRIV:%[^,]+]],
+// CK1-DAG: [[S]] = load ptr, ptr [[S_PRIV:%[^,]+]],
+// CK1-DAG: call void {{%.*}}(ptr %{{[^,]+}}, ptr [[BP_PRIV]], ptr [[P_PRIV]], ptr [[S_PRIV]])
 // CK1-NOT: __tgt_target_data_end
 // CK1: ret i32 0
 // CK1: }