[ConstantFold] Remove unnecessary bounded index restriction

[nikic]

Summary

The fold for merging a GEP of GEP into a single GEP currently bails if doing so would result in notional overindexing. The justification given in the comment above this check is dangerously incorrect: GEPs with notional overindexing are perfectly fine, and if some code treats them incorrectly, then that code is broken, not the GEP. Such a GEP might legally appear in source IR, so only preventing its creation cannot be sufficient. (The constant folder also ends up canonicalizing the GEP to remove the notional overindexing, but that's neither here nor there.)

This check dates back to https://github.com/llvm/llvm-project/commit/bd4fef4a8939db18f39b108e19097b25e2c7c47a, and as far as I can tell the original issue this was trying to patch around has since been resolved.

[nikic]

Update some clang tests. Will update the rest if this is generally fine.

[nlopes]

LGTM. I agree there's no issue in combining OOB geps.
The code seems to handle well the mix of inbounds and non-bounds geps, by dropping inbounds.

I glanced through the OMP test changes, as they looked scary, but they seem fine. They produce a +1 pointer, which is ok.

clang/test/CodeGen/clear_cache.c

 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
 // RUN: %clang_cc1 -emit-llvm %s -o - | FileCheck %s
 
 char buffer[32] = "This is a largely unused buffer";
 
 // __builtin___clear_cache always maps to @llvm.clear_cache, but what
 // each back-end produces is different, and this is tested in LLVM
 
 // CHECK-LABEL: @main(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
 // CHECK-NEXT:    store i32 0, i32* [[RETVAL]], align 4
-// CHECK-NEXT:    call void @llvm.clear_cache(i8* getelementptr inbounds ([32 x i8], [32 x i8]* @buffer, i64 0, i64 0), i8* getelementptr inbounds (i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @buffer, i64 0, i64 0), i64 32))
+// CHECK-NEXT:    call void @llvm.clear_cache(i8* getelementptr inbounds ([32 x i8], [32 x i8]* @buffer, i64 0, i64 0), i8* getelementptr inbounds ([32 x i8], [32 x i8]* @buffer, i64 1, i64 0))
 // CHECK-NEXT:    ret i32 0
 //
 int main() {
   __builtin___clear_cache(buffer, buffer+32);
   return 0;
 }

clang/test/CodeGenCXX/cxx11-initializer-aggregate.cpp

 
 // CHECK-LABEL: define {{.*}}@__cxx_global_var_init(
 //
 // CHECK: call {{.*}}@_ZN14NonTrivialInit1AC1Ev(
 // CHECK: getelementptr inbounds {{.*}}, i64 1
 // CHECK: br i1
 //
 // CHECK: getelementptr inbounds {{.*}}, i64 1
-// CHECK: icmp eq {{.*}}, i64 30
+// CHECK: icmp eq {{.*}}, getelementptr inbounds {{.*}}, i64 1, i64 0
 // CHECK: br i1
 //
 // CHECK: call i32 @__cxa_atexit(
 namespace NonTrivialInit {
   struct A { A(); A(const A&) = delete; ~A(); };
   struct B { A a[20]; };
   // NB, this must be large enough to be worth memsetting for this test to be
   // meaningful.

clang/test/CodeGenCXX/for-range.cpp

 // CHECK-NEXT:    [[A:%.*]] = alloca [[STRUCT_A:%.*]], align 1
 // CHECK-NEXT:    [[__RANGE1:%.*]] = alloca [5 x %struct.B]*, align 8
 // CHECK-NEXT:    [[__BEGIN1:%.*]] = alloca %struct.B*, align 8
 // CHECK-NEXT:    [[__END1:%.*]] = alloca %struct.B*, align 8
 // CHECK-NEXT:    [[B:%.*]] = alloca [[STRUCT_B:%.*]], align 1
 // CHECK-NEXT:    call void @_ZN1AC1Ev(%struct.A* nonnull align 1 dereferenceable(1) [[A]])
 // CHECK-NEXT:    store [5 x %struct.B]* @array, [5 x %struct.B]** [[__RANGE1]], align 8
 // CHECK-NEXT:    store %struct.B* getelementptr inbounds ([5 x %struct.B], [5 x %struct.B]* @array, i64 0, i64 0), %struct.B** [[__BEGIN1]], align 8
-// CHECK-NEXT:    store %struct.B* getelementptr inbounds ([[STRUCT_B]], %struct.B* getelementptr inbounds ([5 x %struct.B], [5 x %struct.B]* @array, i64 0, i64 0), i64 5), %struct.B** [[__END1]], align 8
+// CHECK-NEXT:    store %struct.B* getelementptr inbounds ([5 x %struct.B], [5 x %struct.B]* @array, i64 1, i64 0), %struct.B** [[__END1]], align 8
 // CHECK-NEXT:    br label [[FOR_COND:%.*]]
 // CHECK:       for.cond:
 // CHECK-NEXT:    [[TMP0:%.*]] = load %struct.B*, %struct.B** [[__BEGIN1]], align 8
 // CHECK-NEXT:    [[TMP1:%.*]] = load %struct.B*, %struct.B** [[__END1]], align 8
 // CHECK-NEXT:    [[CMP:%.*]] = icmp ne %struct.B* [[TMP0]], [[TMP1]]
 // CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
 // CHECK:       for.body:
 // CHECK-NEXT:    [[TMP2:%.*]] = load %struct.B*, %struct.B** [[__BEGIN1]], align 8

clang/test/CodeGenCXX/global-array-destruction.cpp

 struct T {
   double d;
   int n;
   ~T();
 };
 T t[2][3] = { 1.0, 2, 3.0, 4, 5.0, 6, 7.0, 8, 9.0, 10, 11.0, 12 };
 
 // CHECK: call {{.*}} @__cxa_atexit
-// CHECK: getelementptr inbounds ({{.*}} getelementptr inbounds ([2 x [3 x {{.*}}]], [2 x [3 x {{.*}}]]* @t, i32 0, i32 0, i32 0), i64 6)
+// CHECK: getelementptr inbounds ([2 x [3 x {{.*}}]], [2 x [3 x {{.*}}]]* @t, i64 1, i64 0, i64 0)
 // CHECK: call void @_ZN1TD1Ev
 // CHECK: icmp eq {{.*}} @t
 // CHECK: br i1 {{.*}}
 
 static T t2[2][3] = { 1.0, 2, 3.0, 4, 5.0, 6, 7.0, 8, 9.0, 10, 11.0, 12 };
 
 // CHECK: call {{.*}} @__cxa_atexit
-// CHECK: getelementptr inbounds ({{.*}} getelementptr inbounds ([2 x [3 x {{.*}}]], [2 x [3 x {{.*}}]]* @_ZL2t2, i32 0, i32 0, i32 0), i64 6)
+// CHECK: getelementptr inbounds ([2 x [3 x {{.*}}]], [2 x [3 x {{.*}}]]* @_ZL2t2, i64 1, i64 0, i64 0)
 // CHECK: call void @_ZN1TD1Ev
 // CHECK: icmp eq {{.*}} @_ZL2t2
 // CHECK: br i1 {{.*}}
 
 using U = T[2][3];
 U &&u = U{ {{1.0, 2}, {3.0, 4}, {5.0, 6}}, {{7.0, 8}, {9.0, 10}, {11.0, 12}} };
 
 // CHECK: call {{.*}} @__cxa_atexit
-// CHECK: getelementptr inbounds ({{.*}}* getelementptr inbounds ([2 x [3 x {{.*}}]], [2 x [3 x {{.*}}]]* @_ZGR1u_, i32 0, i32 0, i32 0), i64 6)
+// CHECK: getelementptr inbounds ([2 x [3 x {{.*}}]], [2 x [3 x {{.*}}]]* @_ZGR1u_, i64 1, i64 0, i64 0)
 // CHECK: call void @_ZN1TD1Ev
 // CHECK: icmp eq {{.*}} @_ZGR1u_
 // CHECK: br i1 {{.*}}

clang/test/CodeGenCXX/template-param-objects.cpp

 // CHECK-SAME: = linkonce_odr constant { <{ [11 x i8], [21 x i8] }> } { <{ [11 x i8], [21 x i8] }> <{ [11 x i8] c"hello world", [21 x i8] zeroinitializer }> }, comdat
 
 // ITANIUM: @p
 // MSABI: @"?p@@3PEBDEB"
 // CHECK-SAME: global i8* getelementptr inbounds ({{.*}}* [[HELLO]], i32 0, i32 0, i32 0, i32 0)
 const char *p = begin<S{"hello world"}>();
 // ITANIUM: @q
 // MSABI: @"?q@@3PEBDEB"
-// CHECK-SAME: global i8* getelementptr (i8, i8* getelementptr inbounds ({{.*}}* [[HELLO]], i32 0, i32 0, i32 0, i32 0), i64 11)
+// CHECK-SAME: global i8* getelementptr ({{.*}}* [[HELLO]], i32 0, i32 0, i32 0, i64 11)
 const char *q = end<S{"hello world"}>();

clang/test/OpenMP/for_firstprivate_codegen.cpp

 //
 // CHECK1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK1:       arraydestroy.body:
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK1-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK1:       arraydestroy.done1:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK2:       arraydestroy.body:
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK2-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK2:       arraydestroy.done1:
 // CHECK2-NEXT:    ret void
 //
 //
 // CHECK2-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK3-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK3-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK3:       arraydestroy.body:
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK3-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK3-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK3-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK3:       arraydestroy.done1:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK4-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
 // CHECK4-NEXT:  entry:
 // CHECK4-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK4-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK4-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK4:       arraydestroy.body:
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK4-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK4-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK4-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK4:       arraydestroy.done1:
 // CHECK4-NEXT:    ret void
 //
 //
 // CHECK4-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 // CHECK4-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_for_firstprivate_codegen.cpp
 // CHECK4-SAME: () #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
 // CHECK4-NEXT:  entry:
 // CHECK4-NEXT:    call void @__cxx_global_var_init()
 // CHECK4-NEXT:    call void @__cxx_global_var_init.1()
 // CHECK4-NEXT:    call void @__cxx_global_var_init.2()
 // CHECK4-NEXT:    ret void
 //
-//
 

clang/test/OpenMP/parallel_copyin_codegen.cpp

 // CHECK1-NEXT:    [[REF_TMP:%.*]] = alloca [[STRUCT_S]], align 4
 // CHECK1-NEXT:    store i32 0, i32* [[RETVAL]], align 4
 // CHECK1-NEXT:    call void @_ZN1SIfEC1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[TEST]])
 // CHECK1-NEXT:    call void @_ZN1SIfEC1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[REF_TMP]])
 // CHECK1-NEXT:    [[CALL:%.*]] = call nonnull align 4 dereferenceable(4) %struct.S* @_ZN1SIfEaSERKS0_(%struct.S* nonnull align 4 dereferenceable(4) [[TEST]], %struct.S* nonnull align 4 dereferenceable(4) [[REF_TMP]])
 // CHECK1-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[REF_TMP]]) #[[ATTR3:[0-9]+]]
 // CHECK1-NEXT:    [[TMP0:%.*]] = load atomic i8, i8* bitcast (i64* @_ZGVZ4mainE5s_arr to i8*) acquire, align 8
 // CHECK1-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0
-// CHECK1-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF2:![0-9]+]]
+// CHECK1-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF3:![0-9]+]]
 // CHECK1:       init.check:
 // CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_guard_acquire(i64* @_ZGVZ4mainE5s_arr) #[[ATTR3]]
 // CHECK1-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP1]], 0
 // CHECK1-NEXT:    br i1 [[TOBOOL]], label [[INIT:%.*]], label [[INIT_END]]
 // CHECK1:       init:
 // CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]])
 // CHECK1-NEXT:    call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB1]], i8* bitcast ([2 x %struct.S]* @_ZZ4mainE5s_arr to i8*), i8* (i8*)* @.__kmpc_global_ctor_., i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_.)
 // CHECK1-NEXT:    call void @_ZN1SIfEC1Ef(%struct.S* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i64 0, i64 0), float 1.000000e+00)
 // CHECK1-NEXT:    call void @_ZN1SIfEC1Ef(%struct.S* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i64 0, i64 1), float 2.000000e+00)
 // CHECK1-NEXT:    [[TMP3:%.*]] = call i32 @__cxa_atexit(void (i8*)* @__cxx_global_array_dtor, i8* null, i8* @__dso_handle) #[[ATTR3]]
 // CHECK1-NEXT:    call void @__cxa_guard_release(i64* @_ZGVZ4mainE5s_arr) #[[ATTR3]]
 // CHECK1-NEXT:    br label [[INIT_END]]
 // CHECK1:       init.end:
 // CHECK1-NEXT:    [[TMP4:%.*]] = load atomic i8, i8* bitcast (i64* @_ZGVZ4mainE3var to i8*) acquire, align 8
 // CHECK1-NEXT:    [[GUARD_UNINITIALIZED1:%.*]] = icmp eq i8 [[TMP4]], 0
-// CHECK1-NEXT:    br i1 [[GUARD_UNINITIALIZED1]], label [[INIT_CHECK2:%.*]], label [[INIT_END5:%.*]], !prof [[PROF2]]
+// CHECK1-NEXT:    br i1 [[GUARD_UNINITIALIZED1]], label [[INIT_CHECK2:%.*]], label [[INIT_END5:%.*]], !prof [[PROF3]]
 // CHECK1:       init.check2:
 // CHECK1-NEXT:    [[TMP5:%.*]] = call i32 @__cxa_guard_acquire(i64* @_ZGVZ4mainE3var) #[[ATTR3]]
 // CHECK1-NEXT:    [[TOBOOL3:%.*]] = icmp ne i32 [[TMP5]], 0
 // CHECK1-NEXT:    br i1 [[TOBOOL3]], label [[INIT4:%.*]], label [[INIT_END5]]
 // CHECK1:       init4:
 // CHECK1-NEXT:    [[TMP6:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK1-NEXT:    call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB1]], i8* bitcast (%struct.S* @_ZZ4mainE3var to i8*), i8* (i8*)* @.__kmpc_global_ctor_..1, i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_..2)
 // CHECK1-NEXT:    call void @_ZN1SIfEC1Ef(%struct.S* nonnull align 4 dereferenceable(4) @_ZZ4mainE3var, float 3.000000e+00)

 //
 // CHECK1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR4]] section ".text.startup" {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK1:       arraydestroy.body:
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK1-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
 // CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i32 0, i32 0)
 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK1:       arraydestroy.done1:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.__kmpc_global_ctor_..1

 // CHECK1-NEXT:    [[TEST:%.*]] = alloca [[STRUCT_S_0:%.*]], align 4
 // CHECK1-NEXT:    [[REF_TMP:%.*]] = alloca [[STRUCT_S_0]], align 4
 // CHECK1-NEXT:    call void @_ZN1SIiEC1Ev(%struct.S.0* nonnull align 4 dereferenceable(4) [[TEST]])
 // CHECK1-NEXT:    call void @_ZN1SIiEC1Ev(%struct.S.0* nonnull align 4 dereferenceable(4) [[REF_TMP]])
 // CHECK1-NEXT:    [[CALL:%.*]] = call nonnull align 4 dereferenceable(4) %struct.S.0* @_ZN1SIiEaSERKS0_(%struct.S.0* nonnull align 4 dereferenceable(4) [[TEST]], %struct.S.0* nonnull align 4 dereferenceable(4) [[REF_TMP]])
 // CHECK1-NEXT:    call void @_ZN1SIiED1Ev(%struct.S.0* nonnull align 4 dereferenceable(4) [[REF_TMP]]) #[[ATTR3]]
 // CHECK1-NEXT:    [[TMP0:%.*]] = load atomic i8, i8* bitcast (i64* @_ZGVZ5tmainIiET_vE5s_arr to i8*) acquire, align 8
 // CHECK1-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0
-// CHECK1-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF2]]
+// CHECK1-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF3]]
 // CHECK1:       init.check:
 // CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_guard_acquire(i64* @_ZGVZ5tmainIiET_vE5s_arr) #[[ATTR3]]
 // CHECK1-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP1]], 0
 // CHECK1-NEXT:    br i1 [[TOBOOL]], label [[INIT:%.*]], label [[INIT_END]]
 // CHECK1:       init:
 // CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK1-NEXT:    call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB1]], i8* bitcast ([2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr to i8*), i8* (i8*)* @.__kmpc_global_ctor_..4, i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_..5)
 // CHECK1-NEXT:    call void @_ZN1SIiEC1Ei(%struct.S.0* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i64 0, i64 0), i32 1)
 // CHECK1-NEXT:    call void @_ZN1SIiEC1Ei(%struct.S.0* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i64 0, i64 1), i32 2)
 // CHECK1-NEXT:    [[TMP3:%.*]] = call i32 @__cxa_atexit(void (i8*)* @__cxx_global_array_dtor.6, i8* null, i8* @__dso_handle) #[[ATTR3]]
 // CHECK1-NEXT:    call void @__cxa_guard_release(i64* @_ZGVZ5tmainIiET_vE5s_arr) #[[ATTR3]]
 // CHECK1-NEXT:    br label [[INIT_END]]
 // CHECK1:       init.end:
 // CHECK1-NEXT:    [[TMP4:%.*]] = load atomic i8, i8* bitcast (i64* @_ZGVZ5tmainIiET_vE3var to i8*) acquire, align 8
 // CHECK1-NEXT:    [[GUARD_UNINITIALIZED1:%.*]] = icmp eq i8 [[TMP4]], 0
-// CHECK1-NEXT:    br i1 [[GUARD_UNINITIALIZED1]], label [[INIT_CHECK2:%.*]], label [[INIT_END5:%.*]], !prof [[PROF2]]
+// CHECK1-NEXT:    br i1 [[GUARD_UNINITIALIZED1]], label [[INIT_CHECK2:%.*]], label [[INIT_END5:%.*]], !prof [[PROF3]]
 // CHECK1:       init.check2:
 // CHECK1-NEXT:    [[TMP5:%.*]] = call i32 @__cxa_guard_acquire(i64* @_ZGVZ5tmainIiET_vE3var) #[[ATTR3]]
 // CHECK1-NEXT:    [[TOBOOL3:%.*]] = icmp ne i32 [[TMP5]], 0
 // CHECK1-NEXT:    br i1 [[TOBOOL3]], label [[INIT4:%.*]], label [[INIT_END5]]
 // CHECK1:       init4:
 // CHECK1-NEXT:    [[TMP6:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK1-NEXT:    call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB1]], i8* bitcast (%struct.S.0* @_ZZ5tmainIiET_vE3var to i8*), i8* (i8*)* @.__kmpc_global_ctor_..7, i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_..8)
 // CHECK1-NEXT:    call void @_ZN1SIiEC1Ei(%struct.S.0* nonnull align 4 dereferenceable(4) @_ZZ5tmainIiET_vE3var, i32 3)

 //
 // CHECK1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor.6
 // CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR4]] section ".text.startup" {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK1:       arraydestroy.body:
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S.0* [ getelementptr inbounds ([[STRUCT_S_0:%.*]], %struct.S.0* getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], %struct.S.0* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S.0* [ getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0:%.*]], %struct.S.0* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK1-NEXT:    call void @_ZN1SIiED1Ev(%struct.S.0* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
 // CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S.0* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i32 0, i32 0)
 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK1:       arraydestroy.done1:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@.__kmpc_global_ctor_..7

 // CHECK2-NEXT:    [[REF_TMP:%.*]] = alloca [[STRUCT_S]], align 4
 // CHECK2-NEXT:    store i32 0, i32* [[RETVAL]], align 4
 // CHECK2-NEXT:    call void @_ZN1SIfEC1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[TEST]])
 // CHECK2-NEXT:    call void @_ZN1SIfEC1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[REF_TMP]])
 // CHECK2-NEXT:    [[CALL:%.*]] = call nonnull align 4 dereferenceable(4) %struct.S* @_ZN1SIfEaSERKS0_(%struct.S* nonnull align 4 dereferenceable(4) [[TEST]], %struct.S* nonnull align 4 dereferenceable(4) [[REF_TMP]])
 // CHECK2-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[REF_TMP]]) #[[ATTR3:[0-9]+]]
 // CHECK2-NEXT:    [[TMP0:%.*]] = load atomic i8, i8* bitcast (i64* @_ZGVZ4mainE5s_arr to i8*) acquire, align 8
 // CHECK2-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0
-// CHECK2-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF2:![0-9]+]]
+// CHECK2-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF3:![0-9]+]]
 // CHECK2:       init.check:
 // CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_guard_acquire(i64* @_ZGVZ4mainE5s_arr) #[[ATTR3]]
 // CHECK2-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP1]], 0
 // CHECK2-NEXT:    br i1 [[TOBOOL]], label [[INIT:%.*]], label [[INIT_END]]
 // CHECK2:       init:
 // CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]])
 // CHECK2-NEXT:    call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB1]], i8* bitcast ([2 x %struct.S]* @_ZZ4mainE5s_arr to i8*), i8* (i8*)* @.__kmpc_global_ctor_., i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_.)
 // CHECK2-NEXT:    call void @_ZN1SIfEC1Ef(%struct.S* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i64 0, i64 0), float 1.000000e+00)
 // CHECK2-NEXT:    call void @_ZN1SIfEC1Ef(%struct.S* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i64 0, i64 1), float 2.000000e+00)
 // CHECK2-NEXT:    [[TMP3:%.*]] = call i32 @__cxa_atexit(void (i8*)* @__cxx_global_array_dtor, i8* null, i8* @__dso_handle) #[[ATTR3]]
 // CHECK2-NEXT:    call void @__cxa_guard_release(i64* @_ZGVZ4mainE5s_arr) #[[ATTR3]]
 // CHECK2-NEXT:    br label [[INIT_END]]
 // CHECK2:       init.end:
 // CHECK2-NEXT:    [[TMP4:%.*]] = load atomic i8, i8* bitcast (i64* @_ZGVZ4mainE3var to i8*) acquire, align 8
 // CHECK2-NEXT:    [[GUARD_UNINITIALIZED1:%.*]] = icmp eq i8 [[TMP4]], 0
-// CHECK2-NEXT:    br i1 [[GUARD_UNINITIALIZED1]], label [[INIT_CHECK2:%.*]], label [[INIT_END5:%.*]], !prof [[PROF2]]
+// CHECK2-NEXT:    br i1 [[GUARD_UNINITIALIZED1]], label [[INIT_CHECK2:%.*]], label [[INIT_END5:%.*]], !prof [[PROF3]]
 // CHECK2:       init.check2:
 // CHECK2-NEXT:    [[TMP5:%.*]] = call i32 @__cxa_guard_acquire(i64* @_ZGVZ4mainE3var) #[[ATTR3]]
 // CHECK2-NEXT:    [[TOBOOL3:%.*]] = icmp ne i32 [[TMP5]], 0
 // CHECK2-NEXT:    br i1 [[TOBOOL3]], label [[INIT4:%.*]], label [[INIT_END5]]
 // CHECK2:       init4:
 // CHECK2-NEXT:    [[TMP6:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK2-NEXT:    call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB1]], i8* bitcast (%struct.S* @_ZZ4mainE3var to i8*), i8* (i8*)* @.__kmpc_global_ctor_..1, i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_..2)
 // CHECK2-NEXT:    call void @_ZN1SIfEC1Ef(%struct.S* nonnull align 4 dereferenceable(4) @_ZZ4mainE3var, float 3.000000e+00)

 //
 // CHECK2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR4]] section ".text.startup" {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK2:       arraydestroy.body:
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK2-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
 // CHECK2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i32 0, i32 0)
 // CHECK2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK2:       arraydestroy.done1:
 // CHECK2-NEXT:    ret void
 //
 //
 // CHECK2-LABEL: define {{[^@]+}}@.__kmpc_global_ctor_..1

 // CHECK2-NEXT:    [[TEST:%.*]] = alloca [[STRUCT_S_0:%.*]], align 4
 // CHECK2-NEXT:    [[REF_TMP:%.*]] = alloca [[STRUCT_S_0]], align 4
 // CHECK2-NEXT:    call void @_ZN1SIiEC1Ev(%struct.S.0* nonnull align 4 dereferenceable(4) [[TEST]])
 // CHECK2-NEXT:    call void @_ZN1SIiEC1Ev(%struct.S.0* nonnull align 4 dereferenceable(4) [[REF_TMP]])
 // CHECK2-NEXT:    [[CALL:%.*]] = call nonnull align 4 dereferenceable(4) %struct.S.0* @_ZN1SIiEaSERKS0_(%struct.S.0* nonnull align 4 dereferenceable(4) [[TEST]], %struct.S.0* nonnull align 4 dereferenceable(4) [[REF_TMP]])
 // CHECK2-NEXT:    call void @_ZN1SIiED1Ev(%struct.S.0* nonnull align 4 dereferenceable(4) [[REF_TMP]]) #[[ATTR3]]
 // CHECK2-NEXT:    [[TMP0:%.*]] = load atomic i8, i8* bitcast (i64* @_ZGVZ5tmainIiET_vE5s_arr to i8*) acquire, align 8
 // CHECK2-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0
-// CHECK2-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF2]]
+// CHECK2-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF3]]
 // CHECK2:       init.check:
 // CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_guard_acquire(i64* @_ZGVZ5tmainIiET_vE5s_arr) #[[ATTR3]]
 // CHECK2-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP1]], 0
 // CHECK2-NEXT:    br i1 [[TOBOOL]], label [[INIT:%.*]], label [[INIT_END]]
 // CHECK2:       init:
 // CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK2-NEXT:    call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB1]], i8* bitcast ([2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr to i8*), i8* (i8*)* @.__kmpc_global_ctor_..4, i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_..5)
 // CHECK2-NEXT:    call void @_ZN1SIiEC1Ei(%struct.S.0* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i64 0, i64 0), i32 1)
 // CHECK2-NEXT:    call void @_ZN1SIiEC1Ei(%struct.S.0* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i64 0, i64 1), i32 2)
 // CHECK2-NEXT:    [[TMP3:%.*]] = call i32 @__cxa_atexit(void (i8*)* @__cxx_global_array_dtor.6, i8* null, i8* @__dso_handle) #[[ATTR3]]
 // CHECK2-NEXT:    call void @__cxa_guard_release(i64* @_ZGVZ5tmainIiET_vE5s_arr) #[[ATTR3]]
 // CHECK2-NEXT:    br label [[INIT_END]]
 // CHECK2:       init.end:
 // CHECK2-NEXT:    [[TMP4:%.*]] = load atomic i8, i8* bitcast (i64* @_ZGVZ5tmainIiET_vE3var to i8*) acquire, align 8
 // CHECK2-NEXT:    [[GUARD_UNINITIALIZED1:%.*]] = icmp eq i8 [[TMP4]], 0
-// CHECK2-NEXT:    br i1 [[GUARD_UNINITIALIZED1]], label [[INIT_CHECK2:%.*]], label [[INIT_END5:%.*]], !prof [[PROF2]]
+// CHECK2-NEXT:    br i1 [[GUARD_UNINITIALIZED1]], label [[INIT_CHECK2:%.*]], label [[INIT_END5:%.*]], !prof [[PROF3]]
 // CHECK2:       init.check2:
 // CHECK2-NEXT:    [[TMP5:%.*]] = call i32 @__cxa_guard_acquire(i64* @_ZGVZ5tmainIiET_vE3var) #[[ATTR3]]
 // CHECK2-NEXT:    [[TOBOOL3:%.*]] = icmp ne i32 [[TMP5]], 0
 // CHECK2-NEXT:    br i1 [[TOBOOL3]], label [[INIT4:%.*]], label [[INIT_END5]]
 // CHECK2:       init4:
 // CHECK2-NEXT:    [[TMP6:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK2-NEXT:    call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB1]], i8* bitcast (%struct.S.0* @_ZZ5tmainIiET_vE3var to i8*), i8* (i8*)* @.__kmpc_global_ctor_..7, i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_..8)
 // CHECK2-NEXT:    call void @_ZN1SIiEC1Ei(%struct.S.0* nonnull align 4 dereferenceable(4) @_ZZ5tmainIiET_vE3var, i32 3)

 //
 // CHECK2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor.6
 // CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR4]] section ".text.startup" {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK2:       arraydestroy.body:
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S.0* [ getelementptr inbounds ([[STRUCT_S_0:%.*]], %struct.S.0* getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], %struct.S.0* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S.0* [ getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0:%.*]], %struct.S.0* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK2-NEXT:    call void @_ZN1SIiED1Ev(%struct.S.0* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
 // CHECK2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S.0* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i32 0, i32 0)
 // CHECK2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK2:       arraydestroy.done1:
 // CHECK2-NEXT:    ret void
 //
 //
 // CHECK2-LABEL: define {{[^@]+}}@.__kmpc_global_ctor_..7

 // CHECK4-NEXT:    ret void
 //
 //
 // CHECK5-LABEL: define {{[^@]+}}@_Z10array_funcv
 // CHECK5-SAME: () #[[ATTR0:[0-9]+]] {
 // CHECK5-NEXT:  entry:
 // CHECK5-NEXT:    [[TMP0:%.*]] = load atomic i8, i8* bitcast (i64* @_ZGVZ10array_funcvE1s to i8*) acquire, align 8
 // CHECK5-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0
-// CHECK5-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF2:![0-9]+]]
+// CHECK5-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF3:![0-9]+]]
 // CHECK5:       init.check:
 // CHECK5-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_guard_acquire(i64* @_ZGVZ10array_funcvE1s) #[[ATTR1:[0-9]+]]
 // CHECK5-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP1]], 0
 // CHECK5-NEXT:    br i1 [[TOBOOL]], label [[INIT:%.*]], label [[INIT_END]]
 // CHECK5:       init:
 // CHECK5-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]])
 // CHECK5-NEXT:    call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB1]], i8* bitcast ([2 x %struct.St]* @_ZZ10array_funcvE1s to i8*), i8* (i8*)* @.__kmpc_global_ctor_., i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_.)
 // CHECK5-NEXT:    br label [[ARRAYCTOR_LOOP:%.*]]
 // CHECK5:       arrayctor.loop:
 // CHECK5-NEXT:    [[ARRAYCTOR_CUR:%.*]] = phi %struct.St* [ getelementptr inbounds ([2 x %struct.St], [2 x %struct.St]* @_ZZ10array_funcvE1s, i32 0, i32 0), [[INIT]] ], [ [[ARRAYCTOR_NEXT:%.*]], [[ARRAYCTOR_LOOP]] ]
 // CHECK5-NEXT:    call void @_ZN2StC1Ev(%struct.St* nonnull align 4 dereferenceable(8) [[ARRAYCTOR_CUR]])
 // CHECK5-NEXT:    [[ARRAYCTOR_NEXT]] = getelementptr inbounds [[STRUCT_ST:%.*]], %struct.St* [[ARRAYCTOR_CUR]], i64 1
-// CHECK5-NEXT:    [[ARRAYCTOR_DONE:%.*]] = icmp eq %struct.St* [[ARRAYCTOR_NEXT]], getelementptr inbounds ([[STRUCT_ST]], %struct.St* getelementptr inbounds ([2 x %struct.St], [2 x %struct.St]* @_ZZ10array_funcvE1s, i32 0, i32 0), i64 2)
+// CHECK5-NEXT:    [[ARRAYCTOR_DONE:%.*]] = icmp eq %struct.St* [[ARRAYCTOR_NEXT]], getelementptr inbounds ([2 x %struct.St], [2 x %struct.St]* @_ZZ10array_funcvE1s, i64 1, i64 0)
 // CHECK5-NEXT:    br i1 [[ARRAYCTOR_DONE]], label [[ARRAYCTOR_CONT:%.*]], label [[ARRAYCTOR_LOOP]]
 // CHECK5:       arrayctor.cont:
 // CHECK5-NEXT:    [[TMP3:%.*]] = call i32 @__cxa_atexit(void (i8*)* @__cxx_global_array_dtor, i8* null, i8* @__dso_handle) #[[ATTR1]]
 // CHECK5-NEXT:    call void @__cxa_guard_release(i64* @_ZGVZ10array_funcvE1s) #[[ATTR1]]
 // CHECK5-NEXT:    br label [[INIT_END]]
 // CHECK5:       init.end:
 // CHECK5-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB1]], i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* @.omp_outlined. to void (i32*, i32*, ...)*))
 // CHECK5-NEXT:    ret void

 //
 // CHECK5-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK5-SAME: (i8* [[TMP0:%.*]]) #[[ATTR2]] section "__TEXT,__StaticInit,regular,pure_instructions" {
 // CHECK5-NEXT:  entry:
 // CHECK5-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK5-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK5-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK5:       arraydestroy.body:
-// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.St* [ getelementptr inbounds ([[STRUCT_ST:%.*]], %struct.St* getelementptr inbounds ([2 x %struct.St], [2 x %struct.St]* @_ZZ10array_funcvE1s, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_ST]], %struct.St* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.St* [ getelementptr inbounds ([2 x %struct.St], [2 x %struct.St]* @_ZZ10array_funcvE1s, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_ST:%.*]], %struct.St* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK5-NEXT:    call void @_ZN2StD1Ev(%struct.St* nonnull align 4 dereferenceable(8) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR1]]
 // CHECK5-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.St* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.St], [2 x %struct.St]* @_ZZ10array_funcvE1s, i32 0, i32 0)
 // CHECK5-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK5:       arraydestroy.done1:
 // CHECK5-NEXT:    ret void
 //
 //
 // CHECK5-LABEL: define {{[^@]+}}@.omp_outlined.

 // CHECK11-NEXT:    [[REF_TMP:%.*]] = alloca [[STRUCT_S]], align 4
 // CHECK11-NEXT:    store i32 0, i32* [[RETVAL]], align 4
 // CHECK11-NEXT:    call void @_ZN1SIfEC1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[TEST]])
 // CHECK11-NEXT:    call void @_ZN1SIfEC1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[REF_TMP]])
 // CHECK11-NEXT:    [[CALL:%.*]] = call nonnull align 4 dereferenceable(4) %struct.S* @_ZN1SIfEaSERKS0_(%struct.S* nonnull align 4 dereferenceable(4) [[TEST]], %struct.S* nonnull align 4 dereferenceable(4) [[REF_TMP]])
 // CHECK11-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[REF_TMP]]) #[[ATTR4:[0-9]+]]
 // CHECK11-NEXT:    [[TMP0:%.*]] = load i8, i8* @_ZGVZ4mainE5s_arr, align 1
 // CHECK11-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0
-// CHECK11-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF2:![0-9]+]]
+// CHECK11-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF3:![0-9]+]]
 // CHECK11:       init.check:
 // CHECK11-NEXT:    call void @_ZN1SIfEC1Ef(%struct.S* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i64 0, i64 0), float 1.000000e+00)
 // CHECK11-NEXT:    call void @_ZN1SIfEC1Ef(%struct.S* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i64 0, i64 1), float 2.000000e+00)
 // CHECK11-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_thread_atexit(void (i8*)* @__cxx_global_array_dtor, i8* null, i8* @__dso_handle) #[[ATTR4]]
 // CHECK11-NEXT:    store i8 1, i8* @_ZGVZ4mainE5s_arr, align 1
 // CHECK11-NEXT:    br label [[INIT_END]]
 // CHECK11:       init.end:
 // CHECK11-NEXT:    [[TMP2:%.*]] = load i8, i8* @_ZGVZ4mainE3var, align 1
 // CHECK11-NEXT:    [[GUARD_UNINITIALIZED1:%.*]] = icmp eq i8 [[TMP2]], 0
-// CHECK11-NEXT:    br i1 [[GUARD_UNINITIALIZED1]], label [[INIT_CHECK2:%.*]], label [[INIT_END3:%.*]], !prof [[PROF2]]
+// CHECK11-NEXT:    br i1 [[GUARD_UNINITIALIZED1]], label [[INIT_CHECK2:%.*]], label [[INIT_END3:%.*]], !prof [[PROF3]]
 // CHECK11:       init.check2:
 // CHECK11-NEXT:    call void @_ZN1SIfEC1Ef(%struct.S* nonnull align 4 dereferenceable(4) @_ZZ4mainE3var, float 3.000000e+00)
 // CHECK11-NEXT:    [[TMP3:%.*]] = call i32 @__cxa_thread_atexit(void (i8*)* bitcast (void (%struct.S*)* @_ZN1SIfED1Ev to void (i8*)*), i8* bitcast (%struct.S* @_ZZ4mainE3var to i8*), i8* @__dso_handle) #[[ATTR4]]
 // CHECK11-NEXT:    store i8 1, i8* @_ZGVZ4mainE3var, align 1
 // CHECK11-NEXT:    br label [[INIT_END3]]
 // CHECK11:       init.end3:
 // CHECK11-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, [2 x i32]*, [2 x %struct.S]*, %struct.S*)* @.omp_outlined. to void (i32*, i32*, ...)*), i32* @_ZZ4mainE5t_var, [2 x i32]* @_ZZ4mainE3vec, [2 x %struct.S]* @_ZZ4mainE5s_arr, %struct.S* @_ZZ4mainE3var)
 // CHECK11-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB2]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*)* @.omp_outlined..1 to void (i32*, i32*, ...)*), i32* @_ZZ4mainE5t_var)

 //
 // CHECK11-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK11-SAME: (i8* [[TMP0:%.*]]) #[[ATTR3:[0-9]+]] section ".text.startup" {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK11-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK11-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK11:       arraydestroy.body:
-// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK11-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
 // CHECK11-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i32 0, i32 0)
 // CHECK11-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK11:       arraydestroy.done1:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@.omp_outlined.

 // CHECK11-NEXT:    [[TMP5:%.*]] = icmp ne i64 [[TMP4]], ptrtoint (i32* @_ZZ4mainE5t_var to i64)
 // CHECK11-NEXT:    br i1 [[TMP5]], label [[COPYIN_NOT_MASTER:%.*]], label [[COPYIN_NOT_MASTER_END:%.*]]
 // CHECK11:       copyin.not.master:
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP6]], i32* @_ZZ4mainE5t_var, align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = bitcast [2 x i32]* [[TMP1]] to i8*
 // CHECK11-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 bitcast ([2 x i32]* @_ZZ4mainE3vec to i8*), i8* align 4 [[TMP7]], i64 8, i1 false)
 // CHECK11-NEXT:    [[TMP8:%.*]] = bitcast [2 x %struct.S]* [[TMP2]] to %struct.S*
-// CHECK11-NEXT:    br i1 icmp eq (%struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i32 0, i32 0), %struct.S* getelementptr ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i32 0, i32 0), i64 2)), label [[OMP_ARRAYCPY_DONE1:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
+// CHECK11-NEXT:    br i1 icmp eq (%struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i32 0, i32 0), %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i64 1, i64 0)), label [[OMP_ARRAYCPY_DONE1:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
 // CHECK11:       omp.arraycpy.body:
 // CHECK11-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi %struct.S* [ [[TMP8]], [[COPYIN_NOT_MASTER]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
 // CHECK11-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i32 0, i32 0), [[COPYIN_NOT_MASTER]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
 // CHECK11-NEXT:    [[CALL:%.*]] = call nonnull align 4 dereferenceable(4) %struct.S* @_ZN1SIfEaSERKS0_(%struct.S* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_DESTELEMENTPAST]], %struct.S* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_SRCELEMENTPAST]])
-// CHECK11-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr [[STRUCT_S]], %struct.S* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
+// CHECK11-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr [[STRUCT_S:%.*]], %struct.S* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
 // CHECK11-NEXT:    [[OMP_ARRAYCPY_SRC_ELEMENT]] = getelementptr [[STRUCT_S]], %struct.S* [[OMP_ARRAYCPY_SRCELEMENTPAST]], i32 1
-// CHECK11-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq %struct.S* [[OMP_ARRAYCPY_DEST_ELEMENT]], getelementptr ([[STRUCT_S]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i32 0, i32 0), i64 2)
+// CHECK11-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq %struct.S* [[OMP_ARRAYCPY_DEST_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i64 1, i64 0)
 // CHECK11-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYCPY_DONE1]], label [[OMP_ARRAYCPY_BODY]]
 // CHECK11:       omp.arraycpy.done1:
 // CHECK11-NEXT:    [[CALL2:%.*]] = call nonnull align 4 dereferenceable(4) %struct.S* @_ZN1SIfEaSERKS0_(%struct.S* nonnull align 4 dereferenceable(4) @_ZZ4mainE3var, %struct.S* nonnull align 4 dereferenceable(4) [[TMP3]])
 // CHECK11-NEXT:    br label [[COPYIN_NOT_MASTER_END]]
 // CHECK11:       copyin.not.master.end:
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK11-NEXT:    [[TMP10:%.*]] = load i32, i32* [[TMP9]], align 4
 // CHECK11-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB1:[0-9]+]], i32 [[TMP10]])

 // CHECK11-NEXT:    [[TEST:%.*]] = alloca [[STRUCT_S_0:%.*]], align 4
 // CHECK11-NEXT:    [[REF_TMP:%.*]] = alloca [[STRUCT_S_0]], align 4
 // CHECK11-NEXT:    call void @_ZN1SIiEC1Ev(%struct.S.0* nonnull align 4 dereferenceable(4) [[TEST]])
 // CHECK11-NEXT:    call void @_ZN1SIiEC1Ev(%struct.S.0* nonnull align 4 dereferenceable(4) [[REF_TMP]])
 // CHECK11-NEXT:    [[CALL:%.*]] = call nonnull align 4 dereferenceable(4) %struct.S.0* @_ZN1SIiEaSERKS0_(%struct.S.0* nonnull align 4 dereferenceable(4) [[TEST]], %struct.S.0* nonnull align 4 dereferenceable(4) [[REF_TMP]])
 // CHECK11-NEXT:    call void @_ZN1SIiED1Ev(%struct.S.0* nonnull align 4 dereferenceable(4) [[REF_TMP]]) #[[ATTR4]]
 // CHECK11-NEXT:    [[TMP0:%.*]] = load i8, i8* bitcast (i64* @_ZGVZ5tmainIiET_vE5s_arr to i8*), align 8
 // CHECK11-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0
-// CHECK11-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF2]]
+// CHECK11-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF3]]
 // CHECK11:       init.check:
 // CHECK11-NEXT:    call void @_ZN1SIiEC1Ei(%struct.S.0* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i64 0, i64 0), i32 1)
 // CHECK11-NEXT:    call void @_ZN1SIiEC1Ei(%struct.S.0* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i64 0, i64 1), i32 2)
 // CHECK11-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_thread_atexit(void (i8*)* @__cxx_global_array_dtor.2, i8* null, i8* @__dso_handle) #[[ATTR4]]
 // CHECK11-NEXT:    store i8 1, i8* bitcast (i64* @_ZGVZ5tmainIiET_vE5s_arr to i8*), align 8
 // CHECK11-NEXT:    br label [[INIT_END]]
 // CHECK11:       init.end:
 // CHECK11-NEXT:    [[TMP2:%.*]] = load i8, i8* bitcast (i64* @_ZGVZ5tmainIiET_vE3var to i8*), align 8
 // CHECK11-NEXT:    [[GUARD_UNINITIALIZED1:%.*]] = icmp eq i8 [[TMP2]], 0
-// CHECK11-NEXT:    br i1 [[GUARD_UNINITIALIZED1]], label [[INIT_CHECK2:%.*]], label [[INIT_END3:%.*]], !prof [[PROF2]]
+// CHECK11-NEXT:    br i1 [[GUARD_UNINITIALIZED1]], label [[INIT_CHECK2:%.*]], label [[INIT_END3:%.*]], !prof [[PROF3]]
 // CHECK11:       init.check2:
 // CHECK11-NEXT:    call void @_ZN1SIiEC1Ei(%struct.S.0* nonnull align 4 dereferenceable(4) @_ZZ5tmainIiET_vE3var, i32 3)
 // CHECK11-NEXT:    [[TMP3:%.*]] = call i32 @__cxa_thread_atexit(void (i8*)* bitcast (void (%struct.S.0*)* @_ZN1SIiED1Ev to void (i8*)*), i8* bitcast (%struct.S.0* @_ZZ5tmainIiET_vE3var to i8*), i8* @__dso_handle) #[[ATTR4]]
 // CHECK11-NEXT:    store i8 1, i8* bitcast (i64* @_ZGVZ5tmainIiET_vE3var to i8*), align 8
 // CHECK11-NEXT:    br label [[INIT_END3]]
 // CHECK11:       init.end3:
 // CHECK11-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB2]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, [2 x i32]*, [2 x %struct.S.0]*, %struct.S.0*)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i32* @_ZZ5tmainIiET_vE5t_var, [2 x i32]* @_ZZ5tmainIiET_vE3vec, [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, %struct.S.0* @_ZZ5tmainIiET_vE3var)
 // CHECK11-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB2]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*)* @.omp_outlined..4 to void (i32*, i32*, ...)*), i32* @_ZZ5tmainIiET_vE5t_var)

 //
 // CHECK11-LABEL: define {{[^@]+}}@__cxx_global_array_dtor.2
 // CHECK11-SAME: (i8* [[TMP0:%.*]]) #[[ATTR3]] section ".text.startup" {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK11-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK11-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK11:       arraydestroy.body:
-// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S.0* [ getelementptr inbounds ([[STRUCT_S_0:%.*]], %struct.S.0* getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], %struct.S.0* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S.0* [ getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0:%.*]], %struct.S.0* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK11-NEXT:    call void @_ZN1SIiED1Ev(%struct.S.0* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
 // CHECK11-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S.0* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i32 0, i32 0)
 // CHECK11-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK11:       arraydestroy.done1:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@.omp_outlined..3

 // CHECK11-NEXT:    [[TMP5:%.*]] = icmp ne i64 [[TMP4]], ptrtoint (i32* @_ZZ5tmainIiET_vE5t_var to i64)
 // CHECK11-NEXT:    br i1 [[TMP5]], label [[COPYIN_NOT_MASTER:%.*]], label [[COPYIN_NOT_MASTER_END:%.*]]
 // CHECK11:       copyin.not.master:
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP0]], align 128
 // CHECK11-NEXT:    store i32 [[TMP6]], i32* @_ZZ5tmainIiET_vE5t_var, align 128
 // CHECK11-NEXT:    [[TMP7:%.*]] = bitcast [2 x i32]* [[TMP1]] to i8*
 // CHECK11-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 128 bitcast ([2 x i32]* @_ZZ5tmainIiET_vE3vec to i8*), i8* align 128 [[TMP7]], i64 8, i1 false)
 // CHECK11-NEXT:    [[TMP8:%.*]] = bitcast [2 x %struct.S.0]* [[TMP2]] to %struct.S.0*
-// CHECK11-NEXT:    br i1 icmp eq (%struct.S.0* getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i32 0, i32 0), %struct.S.0* getelementptr ([[STRUCT_S_0:%.*]], %struct.S.0* getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i32 0, i32 0), i64 2)), label [[OMP_ARRAYCPY_DONE1:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
+// CHECK11-NEXT:    br i1 icmp eq (%struct.S.0* getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i32 0, i32 0), %struct.S.0* getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i64 1, i64 0)), label [[OMP_ARRAYCPY_DONE1:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
 // CHECK11:       omp.arraycpy.body:
 // CHECK11-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi %struct.S.0* [ [[TMP8]], [[COPYIN_NOT_MASTER]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
 // CHECK11-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi %struct.S.0* [ getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i32 0, i32 0), [[COPYIN_NOT_MASTER]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
 // CHECK11-NEXT:    [[CALL:%.*]] = call nonnull align 4 dereferenceable(4) %struct.S.0* @_ZN1SIiEaSERKS0_(%struct.S.0* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_DESTELEMENTPAST]], %struct.S.0* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_SRCELEMENTPAST]])
-// CHECK11-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr [[STRUCT_S_0]], %struct.S.0* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
+// CHECK11-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr [[STRUCT_S_0:%.*]], %struct.S.0* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
 // CHECK11-NEXT:    [[OMP_ARRAYCPY_SRC_ELEMENT]] = getelementptr [[STRUCT_S_0]], %struct.S.0* [[OMP_ARRAYCPY_SRCELEMENTPAST]], i32 1
-// CHECK11-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq %struct.S.0* [[OMP_ARRAYCPY_DEST_ELEMENT]], getelementptr ([[STRUCT_S_0]], %struct.S.0* getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i32 0, i32 0), i64 2)
+// CHECK11-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq %struct.S.0* [[OMP_ARRAYCPY_DEST_ELEMENT]], getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i64 1, i64 0)
 // CHECK11-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYCPY_DONE1]], label [[OMP_ARRAYCPY_BODY]]
 // CHECK11:       omp.arraycpy.done1:
 // CHECK11-NEXT:    [[CALL2:%.*]] = call nonnull align 4 dereferenceable(4) %struct.S.0* @_ZN1SIiEaSERKS0_(%struct.S.0* nonnull align 4 dereferenceable(4) @_ZZ5tmainIiET_vE3var, %struct.S.0* nonnull align 4 dereferenceable(4) [[TMP3]])
 // CHECK11-NEXT:    br label [[COPYIN_NOT_MASTER_END]]
 // CHECK11:       copyin.not.master.end:
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK11-NEXT:    [[TMP10:%.*]] = load i32, i32* [[TMP9]], align 4
 // CHECK11-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB1]], i32 [[TMP10]])

 // CHECK12-NEXT:    [[REF_TMP:%.*]] = alloca [[STRUCT_S]], align 4
 // CHECK12-NEXT:    store i32 0, i32* [[RETVAL]], align 4
 // CHECK12-NEXT:    call void @_ZN1SIfEC1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[TEST]])
 // CHECK12-NEXT:    call void @_ZN1SIfEC1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[REF_TMP]])
 // CHECK12-NEXT:    [[CALL:%.*]] = call nonnull align 4 dereferenceable(4) %struct.S* @_ZN1SIfEaSERKS0_(%struct.S* nonnull align 4 dereferenceable(4) [[TEST]], %struct.S* nonnull align 4 dereferenceable(4) [[REF_TMP]])
 // CHECK12-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[REF_TMP]]) #[[ATTR4:[0-9]+]]
 // CHECK12-NEXT:    [[TMP0:%.*]] = load i8, i8* @_ZGVZ4mainE5s_arr, align 1
 // CHECK12-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0
-// CHECK12-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF2:![0-9]+]]
+// CHECK12-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF3:![0-9]+]]
 // CHECK12:       init.check:
 // CHECK12-NEXT:    call void @_ZN1SIfEC1Ef(%struct.S* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i64 0, i64 0), float 1.000000e+00)
 // CHECK12-NEXT:    call void @_ZN1SIfEC1Ef(%struct.S* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i64 0, i64 1), float 2.000000e+00)
 // CHECK12-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_thread_atexit(void (i8*)* @__cxx_global_array_dtor, i8* null, i8* @__dso_handle) #[[ATTR4]]
 // CHECK12-NEXT:    store i8 1, i8* @_ZGVZ4mainE5s_arr, align 1
 // CHECK12-NEXT:    br label [[INIT_END]]
 // CHECK12:       init.end:
 // CHECK12-NEXT:    [[TMP2:%.*]] = load i8, i8* @_ZGVZ4mainE3var, align 1
 // CHECK12-NEXT:    [[GUARD_UNINITIALIZED1:%.*]] = icmp eq i8 [[TMP2]], 0
-// CHECK12-NEXT:    br i1 [[GUARD_UNINITIALIZED1]], label [[INIT_CHECK2:%.*]], label [[INIT_END3:%.*]], !prof [[PROF2]]
+// CHECK12-NEXT:    br i1 [[GUARD_UNINITIALIZED1]], label [[INIT_CHECK2:%.*]], label [[INIT_END3:%.*]], !prof [[PROF3]]
 // CHECK12:       init.check2:
 // CHECK12-NEXT:    call void @_ZN1SIfEC1Ef(%struct.S* nonnull align 4 dereferenceable(4) @_ZZ4mainE3var, float 3.000000e+00)
 // CHECK12-NEXT:    [[TMP3:%.*]] = call i32 @__cxa_thread_atexit(void (i8*)* bitcast (void (%struct.S*)* @_ZN1SIfED1Ev to void (i8*)*), i8* bitcast (%struct.S* @_ZZ4mainE3var to i8*), i8* @__dso_handle) #[[ATTR4]]
 // CHECK12-NEXT:    store i8 1, i8* @_ZGVZ4mainE3var, align 1
 // CHECK12-NEXT:    br label [[INIT_END3]]
 // CHECK12:       init.end3:
 // CHECK12-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, [2 x i32]*, [2 x %struct.S]*, %struct.S*)* @.omp_outlined. to void (i32*, i32*, ...)*), i32* @_ZZ4mainE5t_var, [2 x i32]* @_ZZ4mainE3vec, [2 x %struct.S]* @_ZZ4mainE5s_arr, %struct.S* @_ZZ4mainE3var)
 // CHECK12-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB2]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*)* @.omp_outlined..1 to void (i32*, i32*, ...)*), i32* @_ZZ4mainE5t_var)

 //
 // CHECK12-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK12-SAME: (i8* [[TMP0:%.*]]) #[[ATTR3:[0-9]+]] section ".text.startup" {
 // CHECK12-NEXT:  entry:
 // CHECK12-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK12-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK12-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK12:       arraydestroy.body:
-// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK12-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
 // CHECK12-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i32 0, i32 0)
 // CHECK12-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK12:       arraydestroy.done1:
 // CHECK12-NEXT:    ret void
 //
 //
 // CHECK12-LABEL: define {{[^@]+}}@.omp_outlined.

 // CHECK12-NEXT:    [[TMP5:%.*]] = icmp ne i64 [[TMP4]], ptrtoint (i32* @_ZZ4mainE5t_var to i64)
 // CHECK12-NEXT:    br i1 [[TMP5]], label [[COPYIN_NOT_MASTER:%.*]], label [[COPYIN_NOT_MASTER_END:%.*]]
 // CHECK12:       copyin.not.master:
 // CHECK12-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP0]], align 4
 // CHECK12-NEXT:    store i32 [[TMP6]], i32* @_ZZ4mainE5t_var, align 4
 // CHECK12-NEXT:    [[TMP7:%.*]] = bitcast [2 x i32]* [[TMP1]] to i8*
 // CHECK12-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 bitcast ([2 x i32]* @_ZZ4mainE3vec to i8*), i8* align 4 [[TMP7]], i64 8, i1 false)
 // CHECK12-NEXT:    [[TMP8:%.*]] = bitcast [2 x %struct.S]* [[TMP2]] to %struct.S*
-// CHECK12-NEXT:    br i1 icmp eq (%struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i32 0, i32 0), %struct.S* getelementptr ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i32 0, i32 0), i64 2)), label [[OMP_ARRAYCPY_DONE1:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
+// CHECK12-NEXT:    br i1 icmp eq (%struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i32 0, i32 0), %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i64 1, i64 0)), label [[OMP_ARRAYCPY_DONE1:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
 // CHECK12:       omp.arraycpy.body:
 // CHECK12-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi %struct.S* [ [[TMP8]], [[COPYIN_NOT_MASTER]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
 // CHECK12-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i32 0, i32 0), [[COPYIN_NOT_MASTER]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
 // CHECK12-NEXT:    [[CALL:%.*]] = call nonnull align 4 dereferenceable(4) %struct.S* @_ZN1SIfEaSERKS0_(%struct.S* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_DESTELEMENTPAST]], %struct.S* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_SRCELEMENTPAST]])
-// CHECK12-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr [[STRUCT_S]], %struct.S* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
+// CHECK12-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr [[STRUCT_S:%.*]], %struct.S* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
 // CHECK12-NEXT:    [[OMP_ARRAYCPY_SRC_ELEMENT]] = getelementptr [[STRUCT_S]], %struct.S* [[OMP_ARRAYCPY_SRCELEMENTPAST]], i32 1
-// CHECK12-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq %struct.S* [[OMP_ARRAYCPY_DEST_ELEMENT]], getelementptr ([[STRUCT_S]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i32 0, i32 0), i64 2)
+// CHECK12-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq %struct.S* [[OMP_ARRAYCPY_DEST_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @_ZZ4mainE5s_arr, i64 1, i64 0)
 // CHECK12-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYCPY_DONE1]], label [[OMP_ARRAYCPY_BODY]]
 // CHECK12:       omp.arraycpy.done1:
 // CHECK12-NEXT:    [[CALL2:%.*]] = call nonnull align 4 dereferenceable(4) %struct.S* @_ZN1SIfEaSERKS0_(%struct.S* nonnull align 4 dereferenceable(4) @_ZZ4mainE3var, %struct.S* nonnull align 4 dereferenceable(4) [[TMP3]])
 // CHECK12-NEXT:    br label [[COPYIN_NOT_MASTER_END]]
 // CHECK12:       copyin.not.master.end:
 // CHECK12-NEXT:    [[TMP9:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK12-NEXT:    [[TMP10:%.*]] = load i32, i32* [[TMP9]], align 4
 // CHECK12-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB1:[0-9]+]], i32 [[TMP10]])

 // CHECK12-NEXT:    [[TEST:%.*]] = alloca [[STRUCT_S_0:%.*]], align 4
 // CHECK12-NEXT:    [[REF_TMP:%.*]] = alloca [[STRUCT_S_0]], align 4
 // CHECK12-NEXT:    call void @_ZN1SIiEC1Ev(%struct.S.0* nonnull align 4 dereferenceable(4) [[TEST]])
 // CHECK12-NEXT:    call void @_ZN1SIiEC1Ev(%struct.S.0* nonnull align 4 dereferenceable(4) [[REF_TMP]])
 // CHECK12-NEXT:    [[CALL:%.*]] = call nonnull align 4 dereferenceable(4) %struct.S.0* @_ZN1SIiEaSERKS0_(%struct.S.0* nonnull align 4 dereferenceable(4) [[TEST]], %struct.S.0* nonnull align 4 dereferenceable(4) [[REF_TMP]])
 // CHECK12-NEXT:    call void @_ZN1SIiED1Ev(%struct.S.0* nonnull align 4 dereferenceable(4) [[REF_TMP]]) #[[ATTR4]]
 // CHECK12-NEXT:    [[TMP0:%.*]] = load i8, i8* bitcast (i64* @_ZGVZ5tmainIiET_vE5s_arr to i8*), align 8
 // CHECK12-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0
-// CHECK12-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF2]]
+// CHECK12-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF3]]
 // CHECK12:       init.check:
 // CHECK12-NEXT:    call void @_ZN1SIiEC1Ei(%struct.S.0* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i64 0, i64 0), i32 1)
 // CHECK12-NEXT:    call void @_ZN1SIiEC1Ei(%struct.S.0* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i64 0, i64 1), i32 2)
 // CHECK12-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_thread_atexit(void (i8*)* @__cxx_global_array_dtor.2, i8* null, i8* @__dso_handle) #[[ATTR4]]
 // CHECK12-NEXT:    store i8 1, i8* bitcast (i64* @_ZGVZ5tmainIiET_vE5s_arr to i8*), align 8
 // CHECK12-NEXT:    br label [[INIT_END]]
 // CHECK12:       init.end:
 // CHECK12-NEXT:    [[TMP2:%.*]] = load i8, i8* bitcast (i64* @_ZGVZ5tmainIiET_vE3var to i8*), align 8
 // CHECK12-NEXT:    [[GUARD_UNINITIALIZED1:%.*]] = icmp eq i8 [[TMP2]], 0
-// CHECK12-NEXT:    br i1 [[GUARD_UNINITIALIZED1]], label [[INIT_CHECK2:%.*]], label [[INIT_END3:%.*]], !prof [[PROF2]]
+// CHECK12-NEXT:    br i1 [[GUARD_UNINITIALIZED1]], label [[INIT_CHECK2:%.*]], label [[INIT_END3:%.*]], !prof [[PROF3]]
 // CHECK12:       init.check2:
 // CHECK12-NEXT:    call void @_ZN1SIiEC1Ei(%struct.S.0* nonnull align 4 dereferenceable(4) @_ZZ5tmainIiET_vE3var, i32 3)
 // CHECK12-NEXT:    [[TMP3:%.*]] = call i32 @__cxa_thread_atexit(void (i8*)* bitcast (void (%struct.S.0*)* @_ZN1SIiED1Ev to void (i8*)*), i8* bitcast (%struct.S.0* @_ZZ5tmainIiET_vE3var to i8*), i8* @__dso_handle) #[[ATTR4]]
 // CHECK12-NEXT:    store i8 1, i8* bitcast (i64* @_ZGVZ5tmainIiET_vE3var to i8*), align 8
 // CHECK12-NEXT:    br label [[INIT_END3]]
 // CHECK12:       init.end3:
 // CHECK12-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB2]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, [2 x i32]*, [2 x %struct.S.0]*, %struct.S.0*)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i32* @_ZZ5tmainIiET_vE5t_var, [2 x i32]* @_ZZ5tmainIiET_vE3vec, [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, %struct.S.0* @_ZZ5tmainIiET_vE3var)
 // CHECK12-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB2]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*)* @.omp_outlined..4 to void (i32*, i32*, ...)*), i32* @_ZZ5tmainIiET_vE5t_var)

 //
 // CHECK12-LABEL: define {{[^@]+}}@__cxx_global_array_dtor.2
 // CHECK12-SAME: (i8* [[TMP0:%.*]]) #[[ATTR3]] section ".text.startup" {
 // CHECK12-NEXT:  entry:
 // CHECK12-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK12-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK12-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK12:       arraydestroy.body:
-// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S.0* [ getelementptr inbounds ([[STRUCT_S_0:%.*]], %struct.S.0* getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], %struct.S.0* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S.0* [ getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0:%.*]], %struct.S.0* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK12-NEXT:    call void @_ZN1SIiED1Ev(%struct.S.0* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
 // CHECK12-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S.0* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i32 0, i32 0)
 // CHECK12-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK12:       arraydestroy.done1:
 // CHECK12-NEXT:    ret void
 //
 //
 // CHECK12-LABEL: define {{[^@]+}}@.omp_outlined..3

 // CHECK12-NEXT:    [[TMP5:%.*]] = icmp ne i64 [[TMP4]], ptrtoint (i32* @_ZZ5tmainIiET_vE5t_var to i64)
 // CHECK12-NEXT:    br i1 [[TMP5]], label [[COPYIN_NOT_MASTER:%.*]], label [[COPYIN_NOT_MASTER_END:%.*]]
 // CHECK12:       copyin.not.master:
 // CHECK12-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP0]], align 128
 // CHECK12-NEXT:    store i32 [[TMP6]], i32* @_ZZ5tmainIiET_vE5t_var, align 128
 // CHECK12-NEXT:    [[TMP7:%.*]] = bitcast [2 x i32]* [[TMP1]] to i8*
 // CHECK12-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 128 bitcast ([2 x i32]* @_ZZ5tmainIiET_vE3vec to i8*), i8* align 128 [[TMP7]], i64 8, i1 false)
 // CHECK12-NEXT:    [[TMP8:%.*]] = bitcast [2 x %struct.S.0]* [[TMP2]] to %struct.S.0*
-// CHECK12-NEXT:    br i1 icmp eq (%struct.S.0* getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i32 0, i32 0), %struct.S.0* getelementptr ([[STRUCT_S_0:%.*]], %struct.S.0* getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i32 0, i32 0), i64 2)), label [[OMP_ARRAYCPY_DONE1:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
+// CHECK12-NEXT:    br i1 icmp eq (%struct.S.0* getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i32 0, i32 0), %struct.S.0* getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i64 1, i64 0)), label [[OMP_ARRAYCPY_DONE1:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
 // CHECK12:       omp.arraycpy.body:
 // CHECK12-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi %struct.S.0* [ [[TMP8]], [[COPYIN_NOT_MASTER]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
 // CHECK12-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi %struct.S.0* [ getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i32 0, i32 0), [[COPYIN_NOT_MASTER]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
 // CHECK12-NEXT:    [[CALL:%.*]] = call nonnull align 4 dereferenceable(4) %struct.S.0* @_ZN1SIiEaSERKS0_(%struct.S.0* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_DESTELEMENTPAST]], %struct.S.0* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_SRCELEMENTPAST]])
-// CHECK12-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr [[STRUCT_S_0]], %struct.S.0* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
+// CHECK12-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr [[STRUCT_S_0:%.*]], %struct.S.0* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
 // CHECK12-NEXT:    [[OMP_ARRAYCPY_SRC_ELEMENT]] = getelementptr [[STRUCT_S_0]], %struct.S.0* [[OMP_ARRAYCPY_SRCELEMENTPAST]], i32 1
-// CHECK12-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq %struct.S.0* [[OMP_ARRAYCPY_DEST_ELEMENT]], getelementptr ([[STRUCT_S_0]], %struct.S.0* getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i32 0, i32 0), i64 2)
+// CHECK12-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq %struct.S.0* [[OMP_ARRAYCPY_DEST_ELEMENT]], getelementptr inbounds ([2 x %struct.S.0], [2 x %struct.S.0]* @_ZZ5tmainIiET_vE5s_arr, i64 1, i64 0)
 // CHECK12-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYCPY_DONE1]], label [[OMP_ARRAYCPY_BODY]]
 // CHECK12:       omp.arraycpy.done1:
 // CHECK12-NEXT:    [[CALL2:%.*]] = call nonnull align 4 dereferenceable(4) %struct.S.0* @_ZN1SIiEaSERKS0_(%struct.S.0* nonnull align 4 dereferenceable(4) @_ZZ5tmainIiET_vE3var, %struct.S.0* nonnull align 4 dereferenceable(4) [[TMP3]])
 // CHECK12-NEXT:    br label [[COPYIN_NOT_MASTER_END]]
 // CHECK12:       copyin.not.master.end:
 // CHECK12-NEXT:    [[TMP9:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK12-NEXT:    [[TMP10:%.*]] = load i32, i32* [[TMP9]], align 4
 // CHECK12-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB1]], i32 [[TMP10]])

 // CHECK14-NEXT:    ret i32* @g
 //
 //
 // CHECK15-LABEL: define {{[^@]+}}@_Z10array_funcv
 // CHECK15-SAME: () #[[ATTR0:[0-9]+]] {
 // CHECK15-NEXT:  entry:
 // CHECK15-NEXT:    [[TMP0:%.*]] = load i8, i8* @_ZGVZ10array_funcvE1s, align 1
 // CHECK15-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0
-// CHECK15-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF2:![0-9]+]]
+// CHECK15-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF3:![0-9]+]]
 // CHECK15:       init.check:
 // CHECK15-NEXT:    br label [[ARRAYCTOR_LOOP:%.*]]
 // CHECK15:       arrayctor.loop:
 // CHECK15-NEXT:    [[ARRAYCTOR_CUR:%.*]] = phi %struct.St* [ getelementptr inbounds ([2 x %struct.St], [2 x %struct.St]* @_ZZ10array_funcvE1s, i32 0, i32 0), [[INIT_CHECK]] ], [ [[ARRAYCTOR_NEXT:%.*]], [[ARRAYCTOR_LOOP]] ]
 // CHECK15-NEXT:    call void @_ZN2StC1Ev(%struct.St* nonnull align 4 dereferenceable(8) [[ARRAYCTOR_CUR]])
 // CHECK15-NEXT:    [[ARRAYCTOR_NEXT]] = getelementptr inbounds [[STRUCT_ST:%.*]], %struct.St* [[ARRAYCTOR_CUR]], i64 1
-// CHECK15-NEXT:    [[ARRAYCTOR_DONE:%.*]] = icmp eq %struct.St* [[ARRAYCTOR_NEXT]], getelementptr inbounds ([[STRUCT_ST]], %struct.St* getelementptr inbounds ([2 x %struct.St], [2 x %struct.St]* @_ZZ10array_funcvE1s, i32 0, i32 0), i64 2)
+// CHECK15-NEXT:    [[ARRAYCTOR_DONE:%.*]] = icmp eq %struct.St* [[ARRAYCTOR_NEXT]], getelementptr inbounds ([2 x %struct.St], [2 x %struct.St]* @_ZZ10array_funcvE1s, i64 1, i64 0)
 // CHECK15-NEXT:    br i1 [[ARRAYCTOR_DONE]], label [[ARRAYCTOR_CONT:%.*]], label [[ARRAYCTOR_LOOP]]
 // CHECK15:       arrayctor.cont:
 // CHECK15-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_thread_atexit(void (i8*)* @__cxx_global_array_dtor, i8* null, i8* @__dso_handle) #[[ATTR3:[0-9]+]]
 // CHECK15-NEXT:    store i8 1, i8* @_ZGVZ10array_funcvE1s, align 1
 // CHECK15-NEXT:    br label [[INIT_END]]
 // CHECK15:       init.end:
 // CHECK15-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, [2 x i32]*, [2 x %struct.St]*)* @.omp_outlined. to void (i32*, i32*, ...)*), [2 x i32]* @_ZZ10array_funcvE1a, [2 x %struct.St]* @_ZZ10array_funcvE1s)
 // CHECK15-NEXT:    ret void

 //
 // CHECK15-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK15-SAME: (i8* [[TMP0:%.*]]) #[[ATTR2:[0-9]+]] section ".text.startup" {
 // CHECK15-NEXT:  entry:
 // CHECK15-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK15-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK15-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK15:       arraydestroy.body:
-// CHECK15-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.St* [ getelementptr inbounds ([[STRUCT_ST:%.*]], %struct.St* getelementptr inbounds ([2 x %struct.St], [2 x %struct.St]* @_ZZ10array_funcvE1s, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK15-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_ST]], %struct.St* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK15-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.St* [ getelementptr inbounds ([2 x %struct.St], [2 x %struct.St]* @_ZZ10array_funcvE1s, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK15-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_ST:%.*]], %struct.St* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK15-NEXT:    call void @_ZN2StD1Ev(%struct.St* nonnull align 4 dereferenceable(8) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
 // CHECK15-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.St* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.St], [2 x %struct.St]* @_ZZ10array_funcvE1s, i32 0, i32 0)
 // CHECK15-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK15:       arraydestroy.done1:
 // CHECK15-NEXT:    ret void
 //
 //
 // CHECK15-LABEL: define {{[^@]+}}@_ZN2StD1Ev

 // CHECK15-NEXT:    [[TMP1:%.*]] = load [2 x %struct.St]*, [2 x %struct.St]** [[S_ADDR]], align 8
 // CHECK15-NEXT:    [[TMP2:%.*]] = ptrtoint [2 x i32]* [[TMP0]] to i64
 // CHECK15-NEXT:    [[TMP3:%.*]] = icmp ne i64 [[TMP2]], ptrtoint ([2 x i32]* @_ZZ10array_funcvE1a to i64)
 // CHECK15-NEXT:    br i1 [[TMP3]], label [[COPYIN_NOT_MASTER:%.*]], label [[COPYIN_NOT_MASTER_END:%.*]]
 // CHECK15:       copyin.not.master:
 // CHECK15-NEXT:    [[TMP4:%.*]] = bitcast [2 x i32]* [[TMP0]] to i8*
 // CHECK15-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 bitcast ([2 x i32]* @_ZZ10array_funcvE1a to i8*), i8* align 4 [[TMP4]], i64 8, i1 false)
 // CHECK15-NEXT:    [[TMP5:%.*]] = bitcast [2 x %struct.St]* [[TMP1]] to %struct.St*
-// CHECK15-NEXT:    br i1 icmp eq (%struct.St* getelementptr inbounds ([2 x %struct.St], [2 x %struct.St]* @_ZZ10array_funcvE1s, i32 0, i32 0), %struct.St* getelementptr ([[STRUCT_ST:%.*]], %struct.St* getelementptr inbounds ([2 x %struct.St], [2 x %struct.St]* @_ZZ10array_funcvE1s, i32 0, i32 0), i64 2)), label [[OMP_ARRAYCPY_DONE1:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
+// CHECK15-NEXT:    br i1 icmp eq (%struct.St* getelementptr inbounds ([2 x %struct.St], [2 x %struct.St]* @_ZZ10array_funcvE1s, i32 0, i32 0), %struct.St* getelementptr inbounds ([2 x %struct.St], [2 x %struct.St]* @_ZZ10array_funcvE1s, i64 1, i64 0)), label [[OMP_ARRAYCPY_DONE1:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
 // CHECK15:       omp.arraycpy.body:
 // CHECK15-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi %struct.St* [ [[TMP5]], [[COPYIN_NOT_MASTER]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
 // CHECK15-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi %struct.St* [ getelementptr inbounds ([2 x %struct.St], [2 x %struct.St]* @_ZZ10array_funcvE1s, i32 0, i32 0), [[COPYIN_NOT_MASTER]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
 // CHECK15-NEXT:    [[CALL:%.*]] = call nonnull align 4 dereferenceable(8) %struct.St* @_ZN2StaSERKS_(%struct.St* nonnull align 4 dereferenceable(8) [[OMP_ARRAYCPY_DESTELEMENTPAST]], %struct.St* nonnull align 4 dereferenceable(8) [[OMP_ARRAYCPY_SRCELEMENTPAST]])
-// CHECK15-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr [[STRUCT_ST]], %struct.St* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
+// CHECK15-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr [[STRUCT_ST:%.*]], %struct.St* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
 // CHECK15-NEXT:    [[OMP_ARRAYCPY_SRC_ELEMENT]] = getelementptr [[STRUCT_ST]], %struct.St* [[OMP_ARRAYCPY_SRCELEMENTPAST]], i32 1
-// CHECK15-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq %struct.St* [[OMP_ARRAYCPY_DEST_ELEMENT]], getelementptr ([[STRUCT_ST]], %struct.St* getelementptr inbounds ([2 x %struct.St], [2 x %struct.St]* @_ZZ10array_funcvE1s, i32 0, i32 0), i64 2)
+// CHECK15-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq %struct.St* [[OMP_ARRAYCPY_DEST_ELEMENT]], getelementptr inbounds ([2 x %struct.St], [2 x %struct.St]* @_ZZ10array_funcvE1s, i64 1, i64 0)
 // CHECK15-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYCPY_DONE1]], label [[OMP_ARRAYCPY_BODY]]
 // CHECK15:       omp.arraycpy.done1:
 // CHECK15-NEXT:    br label [[COPYIN_NOT_MASTER_END]]
 // CHECK15:       copyin.not.master.end:
 // CHECK15-NEXT:    [[TMP6:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK15-NEXT:    [[TMP7:%.*]] = load i32, i32* [[TMP6]], align 4
 // CHECK15-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB1:[0-9]+]], i32 [[TMP7]])
 // CHECK15-NEXT:    ret void

 // CHECK16-NEXT:    ret i32* @t
 //
 //
 // CHECK16-LABEL: define {{[^@]+}}@__tls_init
 // CHECK16-SAME: () #[[ATTR0]] {
 // CHECK16-NEXT:  entry:
 // CHECK16-NEXT:    [[TMP0:%.*]] = load i8, i8* @__tls_guard, align 1
 // CHECK16-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0
-// CHECK16-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT:%.*]], label [[EXIT:%.*]], !prof [[PROF4:![0-9]+]]
+// CHECK16-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT:%.*]], label [[EXIT:%.*]], !prof [[PROF5:![0-9]+]]
 // CHECK16:       init:
 // CHECK16-NEXT:    store i8 1, i8* @__tls_guard, align 1
 // CHECK16-NEXT:    call void @__cxx_global_var_init()
 // CHECK16-NEXT:    br label [[EXIT]]
 // CHECK16:       exit:
 // CHECK16-NEXT:    ret void
 //
-//
 

clang/test/OpenMP/sections_firstprivate_codegen.cpp

 //
 // CHECK1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK1:       arraydestroy.body:
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK1-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK1:       arraydestroy.done1:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK2:       arraydestroy.body:
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK2-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK2:       arraydestroy.done1:
 // CHECK2-NEXT:    ret void
 //
 //
 // CHECK2-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK3-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK3-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK3:       arraydestroy.body:
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK3-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK3-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK3-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK3:       arraydestroy.done1:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK4-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
 // CHECK4-NEXT:  entry:
 // CHECK4-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK4-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK4-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK4:       arraydestroy.body:
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK4-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK4-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK4-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK4:       arraydestroy.done1:
 // CHECK4-NEXT:    ret void
 //
 //
 // CHECK4-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 // CHECK4-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_sections_firstprivate_codegen.cpp
 // CHECK4-SAME: () #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
 // CHECK4-NEXT:  entry:
 // CHECK4-NEXT:    call void @__cxx_global_var_init()
 // CHECK4-NEXT:    call void @__cxx_global_var_init.1()
 // CHECK4-NEXT:    call void @__cxx_global_var_init.2()
 // CHECK4-NEXT:    ret void
 //
-//
 

clang/test/OpenMP/single_codegen.cpp

 // CHECK1-NEXT:    [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    br label [[ARRAYCTOR_LOOP:%.*]]
 // CHECK1:       arrayctor.loop:
 // CHECK1-NEXT:    [[ARRAYCTOR_CUR:%.*]] = phi %class.TestClass* [ getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYCTOR_NEXT:%.*]], [[INVOKE_CONT:%.*]] ]
 // CHECK1-NEXT:    invoke void @_ZN9TestClassC1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYCTOR_CUR]])
 // CHECK1-NEXT:    to label [[INVOKE_CONT]] unwind label [[LPAD:%.*]]
 // CHECK1:       invoke.cont:
 // CHECK1-NEXT:    [[ARRAYCTOR_NEXT]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[ARRAYCTOR_CUR]], i64 1
-// CHECK1-NEXT:    [[ARRAYCTOR_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYCTOR_NEXT]], getelementptr inbounds ([[CLASS_TESTCLASS]], %class.TestClass* getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), i64 2)
+// CHECK1-NEXT:    [[ARRAYCTOR_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYCTOR_NEXT]], getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i64 1, i64 0)
 // CHECK1-NEXT:    br i1 [[ARRAYCTOR_DONE]], label [[ARRAYCTOR_CONT:%.*]], label [[ARRAYCTOR_LOOP]]
 // CHECK1:       arrayctor.cont:
 // CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* @__cxx_global_array_dtor, i8* null, i8* @__dso_handle) #[[ATTR3]]
 // CHECK1-NEXT:    ret void
 // CHECK1:       lpad:
 // CHECK1-NEXT:    [[TMP1:%.*]] = landingpad { i8*, i32 }
 // CHECK1-NEXT:    cleanup
 // CHECK1-NEXT:    [[TMP2:%.*]] = extractvalue { i8*, i32 } [[TMP1]], 0

 //
 // CHECK1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK1:       arraydestroy.body:
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ getelementptr inbounds ([[CLASS_TESTCLASS:%.*]], %class.TestClass* getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK1-NEXT:    call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
 // CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0)
 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK1:       arraydestroy.done1:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN9TestClassC2Ev

 // CHECK2-NEXT:    [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    br label [[ARRAYCTOR_LOOP:%.*]]
 // CHECK2:       arrayctor.loop:
 // CHECK2-NEXT:    [[ARRAYCTOR_CUR:%.*]] = phi %class.TestClass* [ getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYCTOR_NEXT:%.*]], [[INVOKE_CONT:%.*]] ]
 // CHECK2-NEXT:    invoke void @_ZN9TestClassC1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYCTOR_CUR]])
 // CHECK2-NEXT:    to label [[INVOKE_CONT]] unwind label [[LPAD:%.*]]
 // CHECK2:       invoke.cont:
 // CHECK2-NEXT:    [[ARRAYCTOR_NEXT]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[ARRAYCTOR_CUR]], i64 1
-// CHECK2-NEXT:    [[ARRAYCTOR_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYCTOR_NEXT]], getelementptr inbounds ([[CLASS_TESTCLASS]], %class.TestClass* getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), i64 2)
+// CHECK2-NEXT:    [[ARRAYCTOR_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYCTOR_NEXT]], getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i64 1, i64 0)
 // CHECK2-NEXT:    br i1 [[ARRAYCTOR_DONE]], label [[ARRAYCTOR_CONT:%.*]], label [[ARRAYCTOR_LOOP]]
 // CHECK2:       arrayctor.cont:
 // CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* @__cxx_global_array_dtor, i8* null, i8* @__dso_handle) #[[ATTR3]]
 // CHECK2-NEXT:    ret void
 // CHECK2:       lpad:
 // CHECK2-NEXT:    [[TMP1:%.*]] = landingpad { i8*, i32 }
 // CHECK2-NEXT:    cleanup
 // CHECK2-NEXT:    [[TMP2:%.*]] = extractvalue { i8*, i32 } [[TMP1]], 0

 //
 // CHECK2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK2:       arraydestroy.body:
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ getelementptr inbounds ([[CLASS_TESTCLASS:%.*]], %class.TestClass* getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK2-NEXT:    call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
 // CHECK2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0)
 // CHECK2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK2:       arraydestroy.done1:
 // CHECK2-NEXT:    ret void
 //
 //
 // CHECK2-LABEL: define {{[^@]+}}@.__kmpc_global_ctor_.

 // CHECK3-NEXT:    [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    br label [[ARRAYCTOR_LOOP:%.*]]
 // CHECK3:       arrayctor.loop:
 // CHECK3-NEXT:    [[ARRAYCTOR_CUR:%.*]] = phi %class.TestClass* [ getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYCTOR_NEXT:%.*]], [[INVOKE_CONT:%.*]] ]
 // CHECK3-NEXT:    invoke void @_ZN9TestClassC1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYCTOR_CUR]])
 // CHECK3-NEXT:    to label [[INVOKE_CONT]] unwind label [[LPAD:%.*]]
 // CHECK3:       invoke.cont:
 // CHECK3-NEXT:    [[ARRAYCTOR_NEXT]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[ARRAYCTOR_CUR]], i64 1
-// CHECK3-NEXT:    [[ARRAYCTOR_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYCTOR_NEXT]], getelementptr inbounds ([[CLASS_TESTCLASS]], %class.TestClass* getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), i64 2)
+// CHECK3-NEXT:    [[ARRAYCTOR_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYCTOR_NEXT]], getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i64 1, i64 0)
 // CHECK3-NEXT:    br i1 [[ARRAYCTOR_DONE]], label [[ARRAYCTOR_CONT:%.*]], label [[ARRAYCTOR_LOOP]]
 // CHECK3:       arrayctor.cont:
 // CHECK3-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* @__cxx_global_array_dtor, i8* null, i8* @__dso_handle) #[[ATTR3]]
 // CHECK3-NEXT:    ret void
 // CHECK3:       lpad:
 // CHECK3-NEXT:    [[TMP1:%.*]] = landingpad { i8*, i32 }
 // CHECK3-NEXT:    cleanup
 // CHECK3-NEXT:    [[TMP2:%.*]] = extractvalue { i8*, i32 } [[TMP1]], 0

 //
 // CHECK3-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK3-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK3:       arraydestroy.body:
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ getelementptr inbounds ([[CLASS_TESTCLASS:%.*]], %class.TestClass* getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK3-NEXT:    call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
 // CHECK3-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0)
 // CHECK3-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK3:       arraydestroy.done1:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@_ZN9TestClassC2Ev

 // CHECK4-NEXT:    call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB1]], i8* bitcast ([2 x %class.TestClass]* @tc2 to i8*), i8* (i8*)* @.__kmpc_global_ctor_..2, i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_..3)
 // CHECK4-NEXT:    br label [[ARRAYCTOR_LOOP:%.*]]
 // CHECK4:       arrayctor.loop:
 // CHECK4-NEXT:    [[ARRAYCTOR_CUR:%.*]] = phi %class.TestClass* [ getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYCTOR_NEXT:%.*]], [[INVOKE_CONT:%.*]] ]
 // CHECK4-NEXT:    invoke void @_ZN9TestClassC1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYCTOR_CUR]])
 // CHECK4-NEXT:    to label [[INVOKE_CONT]] unwind label [[LPAD:%.*]]
 // CHECK4:       invoke.cont:
 // CHECK4-NEXT:    [[ARRAYCTOR_NEXT]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[ARRAYCTOR_CUR]], i64 1
-// CHECK4-NEXT:    [[ARRAYCTOR_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYCTOR_NEXT]], getelementptr inbounds ([[CLASS_TESTCLASS]], %class.TestClass* getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), i64 2)
+// CHECK4-NEXT:    [[ARRAYCTOR_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYCTOR_NEXT]], getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i64 1, i64 0)
 // CHECK4-NEXT:    br i1 [[ARRAYCTOR_DONE]], label [[ARRAYCTOR_CONT:%.*]], label [[ARRAYCTOR_LOOP]]
 // CHECK4:       arrayctor.cont:
 // CHECK4-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_atexit(void (i8*)* @__cxx_global_array_dtor, i8* null, i8* @__dso_handle) #[[ATTR3]]
 // CHECK4-NEXT:    ret void
 // CHECK4:       lpad:
 // CHECK4-NEXT:    [[TMP2:%.*]] = landingpad { i8*, i32 }
 // CHECK4-NEXT:    cleanup
 // CHECK4-NEXT:    [[TMP3:%.*]] = extractvalue { i8*, i32 } [[TMP2]], 0

 //
 // CHECK4-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK4-NEXT:  entry:
 // CHECK4-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK4-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK4-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK4:       arraydestroy.body:
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ getelementptr inbounds ([[CLASS_TESTCLASS:%.*]], %class.TestClass* getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK4-NEXT:    call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
 // CHECK4-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0)
 // CHECK4-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK4:       arraydestroy.done1:
 // CHECK4-NEXT:    ret void
 //
 //
 // CHECK4-LABEL: define {{[^@]+}}@_Z3foov

 // CHECK5-NEXT:    [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
 // CHECK5-NEXT:    br label [[ARRAYCTOR_LOOP:%.*]], !dbg [[DBG37:![0-9]+]]
 // CHECK5:       arrayctor.loop:
 // CHECK5-NEXT:    [[ARRAYCTOR_CUR:%.*]] = phi %class.TestClass* [ getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYCTOR_NEXT:%.*]], [[INVOKE_CONT:%.*]] ], !dbg [[DBG37]]
 // CHECK5-NEXT:    invoke void @_ZN9TestClassC1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYCTOR_CUR]])
 // CHECK5-NEXT:    to label [[INVOKE_CONT]] unwind label [[LPAD:%.*]], !dbg [[DBG37]]
 // CHECK5:       invoke.cont:
 // CHECK5-NEXT:    [[ARRAYCTOR_NEXT]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[ARRAYCTOR_CUR]], i64 1, !dbg [[DBG37]]
-// CHECK5-NEXT:    [[ARRAYCTOR_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYCTOR_NEXT]], getelementptr inbounds ([[CLASS_TESTCLASS]], %class.TestClass* getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), i64 2), !dbg [[DBG37]]
+// CHECK5-NEXT:    [[ARRAYCTOR_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYCTOR_NEXT]], getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i64 1, i64 0), !dbg [[DBG37]]
 // CHECK5-NEXT:    br i1 [[ARRAYCTOR_DONE]], label [[ARRAYCTOR_CONT:%.*]], label [[ARRAYCTOR_LOOP]], !dbg [[DBG37]]
 // CHECK5:       arrayctor.cont:
 // CHECK5-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* @__cxx_global_array_dtor, i8* null, i8* @__dso_handle) #[[ATTR3]], !dbg [[DBG39:![0-9]+]]
 // CHECK5-NEXT:    ret void, !dbg [[DBG39]]
 // CHECK5:       lpad:
 // CHECK5-NEXT:    [[TMP1:%.*]] = landingpad { i8*, i32 }
 // CHECK5-NEXT:    cleanup, !dbg [[DBG40:![0-9]+]]
 // CHECK5-NEXT:    [[TMP2:%.*]] = extractvalue { i8*, i32 } [[TMP1]], 0, !dbg [[DBG40]]

 //
 // CHECK5-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK5-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" !dbg [[DBG41:![0-9]+]] {
 // CHECK5-NEXT:  entry:
 // CHECK5-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK5-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK5-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]], !dbg [[DBG42:![0-9]+]]
 // CHECK5:       arraydestroy.body:
-// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ getelementptr inbounds ([[CLASS_TESTCLASS:%.*]], %class.TestClass* getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG42]]
-// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG42]]
+// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG42]]
+// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG42]]
 // CHECK5-NEXT:    call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]], !dbg [[DBG42]]
 // CHECK5-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), !dbg [[DBG42]]
 // CHECK5-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]], !dbg [[DBG42]]
 // CHECK5:       arraydestroy.done1:
 // CHECK5-NEXT:    ret void, !dbg [[DBG42]]
 //
 //
 // CHECK5-LABEL: define {{[^@]+}}@_ZN9TestClassC2Ev

clang/test/OpenMP/single_firstprivate_codegen.cpp

 //
 // CHECK1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK1:       arraydestroy.body:
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK1-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK1:       arraydestroy.done1:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK2:       arraydestroy.body:
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK2-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK2:       arraydestroy.done1:
 // CHECK2-NEXT:    ret void
 //
 //
 // CHECK2-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK3-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK3-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK3:       arraydestroy.body:
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK3-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK3-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK3-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK3:       arraydestroy.done1:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK4-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
 // CHECK4-NEXT:  entry:
 // CHECK4-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK4-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK4-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK4:       arraydestroy.body:
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK4-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK4-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK4-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK4:       arraydestroy.done1:
 // CHECK4-NEXT:    ret void
 //
 //
 // CHECK4-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 // CHECK4-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_single_firstprivate_codegen.cpp
 // CHECK4-SAME: () #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
 // CHECK4-NEXT:  entry:
 // CHECK4-NEXT:    call void @__cxx_global_var_init()
 // CHECK4-NEXT:    call void @__cxx_global_var_init.1()
 // CHECK4-NEXT:    call void @__cxx_global_var_init.2()
 // CHECK4-NEXT:    ret void
 //
-//
 

clang/test/OpenMP/target_teams_distribute_firstprivate_codegen.cpp

 //
 // CHECK1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK1:       arraydestroy.body:
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK1-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK1:       arraydestroy.done1:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK2:       arraydestroy.body:
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK2-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK2:       arraydestroy.done1:
 // CHECK2-NEXT:    ret void
 //
 //
 // CHECK2-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK3-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK3-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK3:       arraydestroy.body:
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK3-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK3-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK3-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK3:       arraydestroy.done1:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK4-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK4-NEXT:  entry:
 // CHECK4-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK4-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK4-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK4:       arraydestroy.body:
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK4-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK4-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK4-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK4:       arraydestroy.done1:
 // CHECK4-NEXT:    ret void
 //
 //
 // CHECK4-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK9-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK9-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK9-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK9-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK9:       arraydestroy.body:
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK9-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK9-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK9-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK9:       arraydestroy.done1:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK10-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK10-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK10-NEXT:  entry:
 // CHECK10-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK10-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK10-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK10:       arraydestroy.body:
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK10-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK10-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK10-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK10:       arraydestroy.done1:
 // CHECK10-NEXT:    ret void
 //
 //
 // CHECK10-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

clang/test/OpenMP/target_teams_distribute_parallel_for_firstprivate_codegen.cpp

 //
 // CHECK1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK1:       arraydestroy.body:
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK1-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK1:       arraydestroy.done1:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK2:       arraydestroy.body:
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK2-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK2:       arraydestroy.done1:
 // CHECK2-NEXT:    ret void
 //
 //
 // CHECK2-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK3-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK3-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK3:       arraydestroy.body:
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK3-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK3-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK3-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK3:       arraydestroy.done1:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK4-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK4-NEXT:  entry:
 // CHECK4-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK4-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK4-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK4:       arraydestroy.body:
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK4-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK4-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK4-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK4:       arraydestroy.done1:
 // CHECK4-NEXT:    ret void
 //
 //
 // CHECK4-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK5-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK5-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK5-NEXT:  entry:
 // CHECK5-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK5-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK5-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK5:       arraydestroy.body:
-// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK5-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK5-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK5-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK5:       arraydestroy.done1:
 // CHECK5-NEXT:    ret void
 //
 //
 // CHECK5-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK6-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK6-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK6-NEXT:  entry:
 // CHECK6-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK6-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK6-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK6:       arraydestroy.body:
-// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK6-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK6-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK6-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK6:       arraydestroy.done1:
 // CHECK6-NEXT:    ret void
 //
 //
 // CHECK6-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

clang/test/OpenMP/target_teams_distribute_parallel_for_private_codegen.cpp

 //
 // CHECK1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK1:       arraydestroy.body:
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK1-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK1:       arraydestroy.done1:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK2:       arraydestroy.body:
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK2-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK2:       arraydestroy.done1:
 // CHECK2-NEXT:    ret void
 //
 //
 // CHECK2-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK3-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK3-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK3:       arraydestroy.body:
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK3-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK3-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK3-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK3:       arraydestroy.done1:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK4-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK4-NEXT:  entry:
 // CHECK4-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK4-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK4-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK4:       arraydestroy.body:
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK4-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK4-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK4-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK4:       arraydestroy.done1:
 // CHECK4-NEXT:    ret void
 //
 //
 // CHECK4-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK5-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK5-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK5-NEXT:  entry:
 // CHECK5-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK5-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK5-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK5:       arraydestroy.body:
-// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK5-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK5-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK5-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK5:       arraydestroy.done1:
 // CHECK5-NEXT:    ret void
 //
 //
 // CHECK5-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK6-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK6-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK6-NEXT:  entry:
 // CHECK6-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK6-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK6-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK6:       arraydestroy.body:
-// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK6-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK6-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK6-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK6:       arraydestroy.done1:
 // CHECK6-NEXT:    ret void
 //
 //
 // CHECK6-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 // CHECK17-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4
 // CHECK17-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK17:       omp.inner.for.end:
 // CHECK17-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK17:       omp.loop.exit:
 // CHECK17-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
 // CHECK17-NEXT:    ret void
 //
-//
 

clang/test/OpenMP/target_teams_distribute_parallel_for_simd_firstprivate_codegen.cpp

 //
 // CHECK1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK1:       arraydestroy.body:
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK1-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK1:       arraydestroy.done1:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK2:       arraydestroy.body:
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK2-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK2:       arraydestroy.done1:
 // CHECK2-NEXT:    ret void
 //
 //
 // CHECK2-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK3-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK3-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK3:       arraydestroy.body:
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK3-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK3-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK3-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK3:       arraydestroy.done1:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK4-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK4-NEXT:  entry:
 // CHECK4-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK4-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK4-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK4:       arraydestroy.body:
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK4-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK4-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK4-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK4:       arraydestroy.done1:
 // CHECK4-NEXT:    ret void
 //
 //
 // CHECK4-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK5-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK5-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK5-NEXT:  entry:
 // CHECK5-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK5-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK5-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK5:       arraydestroy.body:
-// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK5-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK5-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK5-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK5:       arraydestroy.done1:
 // CHECK5-NEXT:    ret void
 //
 //
 // CHECK5-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK6-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK6-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK6-NEXT:  entry:
 // CHECK6-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK6-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK6-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK6:       arraydestroy.body:
-// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK6-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK6-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK6-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK6:       arraydestroy.done1:
 // CHECK6-NEXT:    ret void
 //
 //
 // CHECK6-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK7-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK7-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK7-NEXT:  entry:
 // CHECK7-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK7-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK7-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK7:       arraydestroy.body:
-// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK7-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK7-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK7-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK7:       arraydestroy.done1:
 // CHECK7-NEXT:    ret void
 //
 //
 // CHECK7-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK8-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK8-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK8-NEXT:  entry:
 // CHECK8-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK8-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK8-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK8:       arraydestroy.body:
-// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK8-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK8-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK8-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK8:       arraydestroy.done1:
 // CHECK8-NEXT:    ret void
 //
 //
 // CHECK8-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK9-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK9-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK9-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK9-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK9:       arraydestroy.body:
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK9-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK9-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK9-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK9:       arraydestroy.done1:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK10-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK10-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK10-NEXT:  entry:
 // CHECK10-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK10-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK10-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK10:       arraydestroy.body:
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK10-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK10-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK10-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK10:       arraydestroy.done1:
 // CHECK10-NEXT:    ret void
 //
 //
 // CHECK10-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK11-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK11-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK11-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK11-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK11:       arraydestroy.body:
-// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK11-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK11-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK11-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK11:       arraydestroy.done1:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK12-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK12-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK12-NEXT:  entry:
 // CHECK12-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK12-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK12-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK12:       arraydestroy.body:
-// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK12-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK12-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK12-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK12:       arraydestroy.done1:
 // CHECK12-NEXT:    ret void
 //
 //
 // CHECK12-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK18-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK18-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK18-NEXT:  entry:
 // CHECK18-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK18-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK18-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK18:       arraydestroy.body:
-// CHECK18-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK18-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK18-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK18-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK18-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK18-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK18-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK18:       arraydestroy.done1:
 // CHECK18-NEXT:    ret void
 //
 //
 // CHECK18-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK19-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK19-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK19-NEXT:  entry:
 // CHECK19-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK19-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK19-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK19:       arraydestroy.body:
-// CHECK19-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK19-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK19-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK19-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK19-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK19-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK19-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK19:       arraydestroy.done1:
 // CHECK19-NEXT:    ret void
 //
 //
 // CHECK19-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK20-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK20-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK20-NEXT:  entry:
 // CHECK20-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK20-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK20-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK20:       arraydestroy.body:
-// CHECK20-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK20-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK20-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK20-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK20-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK20-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK20-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK20:       arraydestroy.done1:
 // CHECK20-NEXT:    ret void
 //
 //
 // CHECK20-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK21-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK21-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK21-NEXT:  entry:
 // CHECK21-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK21-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK21-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK21:       arraydestroy.body:
-// CHECK21-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK21-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK21-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK21-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK21-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK21-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK21-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK21:       arraydestroy.done1:
 // CHECK21-NEXT:    ret void
 //
 //
 // CHECK21-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK22-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK22-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK22-NEXT:  entry:
 // CHECK22-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK22-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK22-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK22:       arraydestroy.body:
-// CHECK22-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK22-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK22-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK22-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK22-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK22-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK22-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK22:       arraydestroy.done1:
 // CHECK22-NEXT:    ret void
 //
 //
 // CHECK22-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

clang/test/OpenMP/target_teams_distribute_parallel_for_simd_private_codegen.cpp

 //
 // CHECK1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK1:       arraydestroy.body:
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK1-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK1:       arraydestroy.done1:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 // CHECK1-NEXT:    br label [[COND_END]]
 // CHECK1:       cond.end:
 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK1-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK1:       omp.inner.for.cond:
-// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !5
+// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !5
 // CHECK1-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK1-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK1:       omp.inner.for.cond.cleanup:
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK1:       omp.inner.for.body:
-// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !5
 // CHECK1-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
-// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !5
 // CHECK1-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
-// CHECK1-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]])
+// CHECK1-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]]), !llvm.access.group !5
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK1:       omp.inner.for.inc:
-// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !5
+// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !5
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
-// CHECK1-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP4:![0-9]+]]
+// CHECK1-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !5
+// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP6:![0-9]+]]
 // CHECK1:       omp.inner.for.end:
 // CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK1:       omp.loop.exit:
 // CHECK1-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP14:%.*]] = load i32, i32* [[TMP13]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP14]])
 // CHECK1-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK1-NEXT:    [[TMP16:%.*]] = icmp ne i32 [[TMP15]], 0

 // CHECK1-NEXT:    br label [[COND_END]]
 // CHECK1:       cond.end:
 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK1-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK1:       omp.inner.for.cond:
-// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
+// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !9
 // CHECK1-NEXT:    [[CMP2:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK1-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK1:       omp.inner.for.cond.cleanup:
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK1:       omp.inner.for.body:
-// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
 // CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK1-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK1-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !9
+// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !9
+// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !9
 // CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP11]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i64 0, i64 [[IDXPROM]]
-// CHECK1-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4
-// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4
+// CHECK1-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !9
+// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !9
 // CHECK1-NEXT:    [[IDXPROM3:%.*]] = sext i32 [[TMP12]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds [2 x %struct.S], [2 x %struct.S]* [[S_ARR]], i64 0, i64 [[IDXPROM3]]
 // CHECK1-NEXT:    [[TMP13:%.*]] = bitcast %struct.S* [[ARRAYIDX4]] to i8*
 // CHECK1-NEXT:    [[TMP14:%.*]] = bitcast %struct.S* [[VAR]] to i8*
-// CHECK1-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i64 4, i1 false)
-// CHECK1-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
-// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, i32* [[SIVAR]], align 4
+// CHECK1-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i64 4, i1 false), !llvm.access.group !9
+// CHECK1-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !9
+// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, i32* [[SIVAR]], align 4, !llvm.access.group !9
 // CHECK1-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP16]], [[TMP15]]
-// CHECK1-NEXT:    store i32 [[ADD5]], i32* [[SIVAR]], align 4
+// CHECK1-NEXT:    store i32 [[ADD5]], i32* [[SIVAR]], align 4, !llvm.access.group !9
 // CHECK1-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK1:       omp.body.continue:
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK1:       omp.inner.for.inc:
-// CHECK1-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
 // CHECK1-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP17]], 1
-// CHECK1-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
-// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP6:![0-9]+]]
+// CHECK1-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
+// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP10:![0-9]+]]
 // CHECK1:       omp.inner.for.end:
 // CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK1:       omp.loop.exit:
 // CHECK1-NEXT:    [[TMP18:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP19:%.*]] = load i32, i32* [[TMP18]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP19]])
 // CHECK1-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK1-NEXT:    [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0

 // CHECK1-NEXT:    br label [[COND_END]]
 // CHECK1:       cond.end:
 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK1-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK1:       omp.inner.for.cond:
-// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !14
+// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !14
 // CHECK1-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK1-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK1:       omp.inner.for.cond.cleanup:
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK1:       omp.inner.for.body:
-// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !14
 // CHECK1-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
-// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !14
 // CHECK1-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
-// CHECK1-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..5 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]])
+// CHECK1-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..5 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]]), !llvm.access.group !14
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK1:       omp.inner.for.inc:
-// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !14
+// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !14
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
-// CHECK1-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP9:![0-9]+]]
+// CHECK1-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !14
+// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP15:![0-9]+]]
 // CHECK1:       omp.inner.for.end:
 // CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK1:       omp.loop.exit:
 // CHECK1-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP14:%.*]] = load i32, i32* [[TMP13]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP14]])
 // CHECK1-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK1-NEXT:    [[TMP16:%.*]] = icmp ne i32 [[TMP15]], 0

 // CHECK1-NEXT:    br label [[COND_END]]
 // CHECK1:       cond.end:
 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK1-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK1:       omp.inner.for.cond:
-// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !17
+// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !17
 // CHECK1-NEXT:    [[CMP4:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK1-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK1:       omp.inner.for.cond.cleanup:
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK1:       omp.inner.for.body:
-// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !17
 // CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK1-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK1-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !17
+// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !17
+// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !17
 // CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP11]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i64 0, i64 [[IDXPROM]]
-// CHECK1-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4
-// CHECK1-NEXT:    [[TMP12:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP3]], align 8
-// CHECK1-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4
+// CHECK1-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !17
+// CHECK1-NEXT:    [[TMP12:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP3]], align 8, !llvm.access.group !17
+// CHECK1-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !17
 // CHECK1-NEXT:    [[IDXPROM5:%.*]] = sext i32 [[TMP13]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR]], i64 0, i64 [[IDXPROM5]]
 // CHECK1-NEXT:    [[TMP14:%.*]] = bitcast %struct.S.0* [[ARRAYIDX6]] to i8*
 // CHECK1-NEXT:    [[TMP15:%.*]] = bitcast %struct.S.0* [[TMP12]] to i8*
-// CHECK1-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP14]], i8* align 4 [[TMP15]], i64 4, i1 false)
+// CHECK1-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP14]], i8* align 4 [[TMP15]], i64 4, i1 false), !llvm.access.group !17
 // CHECK1-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK1:       omp.body.continue:
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK1:       omp.inner.for.inc:
-// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !17
 // CHECK1-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP16]], 1
-// CHECK1-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
-// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP10:![0-9]+]]
+// CHECK1-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !17
+// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP18:![0-9]+]]
 // CHECK1:       omp.inner.for.end:
 // CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK1:       omp.loop.exit:
 // CHECK1-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP18]])
 // CHECK1-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK1-NEXT:    [[TMP20:%.*]] = icmp ne i32 [[TMP19]], 0

 //
 // CHECK2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK2:       arraydestroy.body:
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK2-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK2:       arraydestroy.done1:
 // CHECK2-NEXT:    ret void
 //
 //
 // CHECK2-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 // CHECK2-NEXT:    br label [[COND_END]]
 // CHECK2:       cond.end:
 // CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK2-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK2-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK2-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK2:       omp.inner.for.cond:
-// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !5
+// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !5
 // CHECK2-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK2-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK2:       omp.inner.for.cond.cleanup:
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK2:       omp.inner.for.body:
-// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
+// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !5
 // CHECK2-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
-// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !5
 // CHECK2-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
-// CHECK2-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]])
+// CHECK2-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]]), !llvm.access.group !5
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK2:       omp.inner.for.inc:
-// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !5
+// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !5
 // CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
-// CHECK2-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP4:![0-9]+]]
+// CHECK2-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !5
+// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP6:![0-9]+]]
 // CHECK2:       omp.inner.for.end:
 // CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK2:       omp.loop.exit:
 // CHECK2-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK2-NEXT:    [[TMP14:%.*]] = load i32, i32* [[TMP13]], align 4
 // CHECK2-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP14]])
 // CHECK2-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK2-NEXT:    [[TMP16:%.*]] = icmp ne i32 [[TMP15]], 0

 // CHECK2-NEXT:    br label [[COND_END]]
 // CHECK2:       cond.end:
 // CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK2-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK2-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK2:       omp.inner.for.cond:
-// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
+// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !9
 // CHECK2-NEXT:    [[CMP2:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK2-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK2:       omp.inner.for.cond.cleanup:
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK2:       omp.inner.for.body:
-// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
 // CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK2-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK2-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !9
+// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !9
+// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !9
 // CHECK2-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP11]] to i64
 // CHECK2-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i64 0, i64 [[IDXPROM]]
-// CHECK2-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4
-// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4
+// CHECK2-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !9
+// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !9
 // CHECK2-NEXT:    [[IDXPROM3:%.*]] = sext i32 [[TMP12]] to i64
 // CHECK2-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds [2 x %struct.S], [2 x %struct.S]* [[S_ARR]], i64 0, i64 [[IDXPROM3]]
 // CHECK2-NEXT:    [[TMP13:%.*]] = bitcast %struct.S* [[ARRAYIDX4]] to i8*
 // CHECK2-NEXT:    [[TMP14:%.*]] = bitcast %struct.S* [[VAR]] to i8*
-// CHECK2-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i64 4, i1 false)
-// CHECK2-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
-// CHECK2-NEXT:    [[TMP16:%.*]] = load i32, i32* [[SIVAR]], align 4
+// CHECK2-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i64 4, i1 false), !llvm.access.group !9
+// CHECK2-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !9
+// CHECK2-NEXT:    [[TMP16:%.*]] = load i32, i32* [[SIVAR]], align 4, !llvm.access.group !9
 // CHECK2-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP16]], [[TMP15]]
-// CHECK2-NEXT:    store i32 [[ADD5]], i32* [[SIVAR]], align 4
+// CHECK2-NEXT:    store i32 [[ADD5]], i32* [[SIVAR]], align 4, !llvm.access.group !9
 // CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK2:       omp.body.continue:
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK2:       omp.inner.for.inc:
-// CHECK2-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
 // CHECK2-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP17]], 1
-// CHECK2-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
-// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP6:![0-9]+]]
+// CHECK2-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
+// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP10:![0-9]+]]
 // CHECK2:       omp.inner.for.end:
 // CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK2:       omp.loop.exit:
 // CHECK2-NEXT:    [[TMP18:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK2-NEXT:    [[TMP19:%.*]] = load i32, i32* [[TMP18]], align 4
 // CHECK2-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP19]])
 // CHECK2-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK2-NEXT:    [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0

 // CHECK2-NEXT:    br label [[COND_END]]
 // CHECK2:       cond.end:
 // CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK2-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK2-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK2-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK2:       omp.inner.for.cond:
-// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !14
+// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !14
 // CHECK2-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK2-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK2:       omp.inner.for.cond.cleanup:
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK2:       omp.inner.for.body:
-// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
+// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !14
 // CHECK2-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
-// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !14
 // CHECK2-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
-// CHECK2-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..5 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]])
+// CHECK2-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..5 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]]), !llvm.access.group !14
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK2:       omp.inner.for.inc:
-// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !14
+// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !14
 // CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
-// CHECK2-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP9:![0-9]+]]
+// CHECK2-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !14
+// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP15:![0-9]+]]
 // CHECK2:       omp.inner.for.end:
 // CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK2:       omp.loop.exit:
 // CHECK2-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK2-NEXT:    [[TMP14:%.*]] = load i32, i32* [[TMP13]], align 4
 // CHECK2-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP14]])
 // CHECK2-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK2-NEXT:    [[TMP16:%.*]] = icmp ne i32 [[TMP15]], 0

 // CHECK2-NEXT:    br label [[COND_END]]
 // CHECK2:       cond.end:
 // CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK2-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK2-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK2:       omp.inner.for.cond:
-// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !17
+// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !17
 // CHECK2-NEXT:    [[CMP4:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK2-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK2:       omp.inner.for.cond.cleanup:
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK2:       omp.inner.for.body:
-// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !17
 // CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK2-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK2-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !17
+// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !17
+// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !17
 // CHECK2-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP11]] to i64
 // CHECK2-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i64 0, i64 [[IDXPROM]]
-// CHECK2-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4
-// CHECK2-NEXT:    [[TMP12:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP3]], align 8
-// CHECK2-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4
+// CHECK2-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !17
+// CHECK2-NEXT:    [[TMP12:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP3]], align 8, !llvm.access.group !17
+// CHECK2-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !17
 // CHECK2-NEXT:    [[IDXPROM5:%.*]] = sext i32 [[TMP13]] to i64
 // CHECK2-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR]], i64 0, i64 [[IDXPROM5]]
 // CHECK2-NEXT:    [[TMP14:%.*]] = bitcast %struct.S.0* [[ARRAYIDX6]] to i8*
 // CHECK2-NEXT:    [[TMP15:%.*]] = bitcast %struct.S.0* [[TMP12]] to i8*
-// CHECK2-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP14]], i8* align 4 [[TMP15]], i64 4, i1 false)
+// CHECK2-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP14]], i8* align 4 [[TMP15]], i64 4, i1 false), !llvm.access.group !17
 // CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK2:       omp.body.continue:
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK2:       omp.inner.for.inc:
-// CHECK2-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !17
 // CHECK2-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP16]], 1
-// CHECK2-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
-// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP10:![0-9]+]]
+// CHECK2-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !17
+// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP18:![0-9]+]]
 // CHECK2:       omp.inner.for.end:
 // CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK2:       omp.loop.exit:
 // CHECK2-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK2-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
 // CHECK2-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP18]])
 // CHECK2-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK2-NEXT:    [[TMP20:%.*]] = icmp ne i32 [[TMP19]], 0

 //
 // CHECK3-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK3-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK3:       arraydestroy.body:
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK3-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK3-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK3-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK3:       arraydestroy.done1:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 // CHECK3-NEXT:    br label [[COND_END]]
 // CHECK3:       cond.end:
 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK3-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK3:       omp.inner.for.cond:
-// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
+// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !6
 // CHECK3-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK3-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK3:       omp.inner.for.cond.cleanup:
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK3:       omp.inner.for.body:
-// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
-// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
-// CHECK3-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32, i32)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i32 [[TMP7]], i32 [[TMP8]])
+// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !6
+// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !6
+// CHECK3-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32, i32)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i32 [[TMP7]], i32 [[TMP8]]), !llvm.access.group !6
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK3:       omp.inner.for.inc:
-// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
+// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !6
 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP9]], [[TMP10]]
-// CHECK3-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
+// CHECK3-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
+// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP7:![0-9]+]]
 // CHECK3:       omp.inner.for.end:
 // CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK3:       omp.loop.exit:
 // CHECK3-NEXT:    [[TMP11:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP12]])
 // CHECK3-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK3-NEXT:    [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0

 // CHECK3-NEXT:    br label [[COND_END]]
 // CHECK3:       cond.end:
 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK3-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK3:       omp.inner.for.cond:
-// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !10
+// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !10
 // CHECK3-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK3-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK3:       omp.inner.for.cond.cleanup:
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK3:       omp.inner.for.body:
-// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !10
 // CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK3-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK3-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK3-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !10
+// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !10
+// CHECK3-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !10
 // CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i32 0, i32 [[TMP11]]
-// CHECK3-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4
-// CHECK3-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4
+// CHECK3-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !10
+// CHECK3-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !10
 // CHECK3-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds [2 x %struct.S], [2 x %struct.S]* [[S_ARR]], i32 0, i32 [[TMP12]]
 // CHECK3-NEXT:    [[TMP13:%.*]] = bitcast %struct.S* [[ARRAYIDX2]] to i8*
 // CHECK3-NEXT:    [[TMP14:%.*]] = bitcast %struct.S* [[VAR]] to i8*
-// CHECK3-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i32 4, i1 false)
-// CHECK3-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
-// CHECK3-NEXT:    [[TMP16:%.*]] = load i32, i32* [[SIVAR]], align 4
+// CHECK3-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i32 4, i1 false), !llvm.access.group !10
+// CHECK3-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !10
+// CHECK3-NEXT:    [[TMP16:%.*]] = load i32, i32* [[SIVAR]], align 4, !llvm.access.group !10
 // CHECK3-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP16]], [[TMP15]]
-// CHECK3-NEXT:    store i32 [[ADD3]], i32* [[SIVAR]], align 4
+// CHECK3-NEXT:    store i32 [[ADD3]], i32* [[SIVAR]], align 4, !llvm.access.group !10
 // CHECK3-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK3:       omp.body.continue:
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK3:       omp.inner.for.inc:
-// CHECK3-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK3-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !10
 // CHECK3-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP17]], 1
-// CHECK3-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4
-// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP7:![0-9]+]]
+// CHECK3-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !10
+// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP11:![0-9]+]]
 // CHECK3:       omp.inner.for.end:
 // CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK3:       omp.loop.exit:
 // CHECK3-NEXT:    [[TMP18:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP19:%.*]] = load i32, i32* [[TMP18]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP19]])
 // CHECK3-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK3-NEXT:    [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0

 // CHECK3-NEXT:    br label [[COND_END]]
 // CHECK3:       cond.end:
 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK3-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK3:       omp.inner.for.cond:
-// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !15
+// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !15
 // CHECK3-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK3-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK3:       omp.inner.for.cond.cleanup:
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK3:       omp.inner.for.body:
-// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
-// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
-// CHECK3-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32, i32)* @.omp_outlined..5 to void (i32*, i32*, ...)*), i32 [[TMP7]], i32 [[TMP8]])
+// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !15
+// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !15
+// CHECK3-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32, i32)* @.omp_outlined..5 to void (i32*, i32*, ...)*), i32 [[TMP7]], i32 [[TMP8]]), !llvm.access.group !15
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK3:       omp.inner.for.inc:
-// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !15
+// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !15
 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP9]], [[TMP10]]
-// CHECK3-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP10:![0-9]+]]
+// CHECK3-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !15
+// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP16:![0-9]+]]
 // CHECK3:       omp.inner.for.end:
 // CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK3:       omp.loop.exit:
 // CHECK3-NEXT:    [[TMP11:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP12]])
 // CHECK3-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK3-NEXT:    [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0

 // CHECK3-NEXT:    br label [[COND_END]]
 // CHECK3:       cond.end:
 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK3-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK3:       omp.inner.for.cond:
-// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !18
+// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !18
 // CHECK3-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK3-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK3:       omp.inner.for.cond.cleanup:
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK3:       omp.inner.for.body:
-// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !18
 // CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK3-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK3-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK3-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !18
+// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !18
+// CHECK3-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !18
 // CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i32 0, i32 [[TMP11]]
-// CHECK3-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4
-// CHECK3-NEXT:    [[TMP12:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP2]], align 4
-// CHECK3-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4
+// CHECK3-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !18
+// CHECK3-NEXT:    [[TMP12:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP2]], align 4, !llvm.access.group !18
+// CHECK3-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !18
 // CHECK3-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR]], i32 0, i32 [[TMP13]]
 // CHECK3-NEXT:    [[TMP14:%.*]] = bitcast %struct.S.0* [[ARRAYIDX4]] to i8*
 // CHECK3-NEXT:    [[TMP15:%.*]] = bitcast %struct.S.0* [[TMP12]] to i8*
-// CHECK3-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP14]], i8* align 4 [[TMP15]], i32 4, i1 false)
+// CHECK3-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP14]], i8* align 4 [[TMP15]], i32 4, i1 false), !llvm.access.group !18
 // CHECK3-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK3:       omp.body.continue:
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK3:       omp.inner.for.inc:
-// CHECK3-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK3-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !18
 // CHECK3-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP16]], 1
-// CHECK3-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4
-// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP11:![0-9]+]]
+// CHECK3-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !18
+// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP19:![0-9]+]]
 // CHECK3:       omp.inner.for.end:
 // CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK3:       omp.loop.exit:
 // CHECK3-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP18]])
 // CHECK3-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK3-NEXT:    [[TMP20:%.*]] = icmp ne i32 [[TMP19]], 0

 //
 // CHECK4-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK4-NEXT:  entry:
 // CHECK4-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK4-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK4-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK4:       arraydestroy.body:
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK4-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK4-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK4-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK4:       arraydestroy.done1:
 // CHECK4-NEXT:    ret void
 //
 //
 // CHECK4-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 // CHECK4-NEXT:    br label [[COND_END]]
 // CHECK4:       cond.end:
 // CHECK4-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK4-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK4-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK4:       omp.inner.for.cond:
-// CHECK4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
+// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !6
 // CHECK4-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK4-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK4:       omp.inner.for.cond.cleanup:
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK4:       omp.inner.for.body:
-// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
-// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
-// CHECK4-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32, i32)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i32 [[TMP7]], i32 [[TMP8]])
+// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !6
+// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !6
+// CHECK4-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32, i32)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i32 [[TMP7]], i32 [[TMP8]]), !llvm.access.group !6
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK4:       omp.inner.for.inc:
-// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
+// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !6
 // CHECK4-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP9]], [[TMP10]]
-// CHECK4-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
+// CHECK4-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
+// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP7:![0-9]+]]
 // CHECK4:       omp.inner.for.end:
 // CHECK4-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK4:       omp.loop.exit:
 // CHECK4-NEXT:    [[TMP11:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK4-NEXT:    [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 4
 // CHECK4-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP12]])
 // CHECK4-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK4-NEXT:    [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0

 // CHECK4-NEXT:    br label [[COND_END]]
 // CHECK4:       cond.end:
 // CHECK4-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK4-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK4-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK4:       omp.inner.for.cond:
-// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !10
+// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !10
 // CHECK4-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK4-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK4:       omp.inner.for.cond.cleanup:
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK4:       omp.inner.for.body:
-// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !10
 // CHECK4-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK4-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK4-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK4-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK4-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !10
+// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !10
+// CHECK4-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !10
 // CHECK4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i32 0, i32 [[TMP11]]
-// CHECK4-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4
-// CHECK4-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4
+// CHECK4-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !10
+// CHECK4-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !10
 // CHECK4-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds [2 x %struct.S], [2 x %struct.S]* [[S_ARR]], i32 0, i32 [[TMP12]]
 // CHECK4-NEXT:    [[TMP13:%.*]] = bitcast %struct.S* [[ARRAYIDX2]] to i8*
 // CHECK4-NEXT:    [[TMP14:%.*]] = bitcast %struct.S* [[VAR]] to i8*
-// CHECK4-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i32 4, i1 false)
-// CHECK4-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
-// CHECK4-NEXT:    [[TMP16:%.*]] = load i32, i32* [[SIVAR]], align 4
+// CHECK4-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i32 4, i1 false), !llvm.access.group !10
+// CHECK4-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !10
+// CHECK4-NEXT:    [[TMP16:%.*]] = load i32, i32* [[SIVAR]], align 4, !llvm.access.group !10
 // CHECK4-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP16]], [[TMP15]]
-// CHECK4-NEXT:    store i32 [[ADD3]], i32* [[SIVAR]], align 4
+// CHECK4-NEXT:    store i32 [[ADD3]], i32* [[SIVAR]], align 4, !llvm.access.group !10
 // CHECK4-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK4:       omp.body.continue:
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK4:       omp.inner.for.inc:
-// CHECK4-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK4-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !10
 // CHECK4-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP17]], 1
-// CHECK4-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4
-// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP7:![0-9]+]]
+// CHECK4-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !10
+// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP11:![0-9]+]]
 // CHECK4:       omp.inner.for.end:
 // CHECK4-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK4:       omp.loop.exit:
 // CHECK4-NEXT:    [[TMP18:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK4-NEXT:    [[TMP19:%.*]] = load i32, i32* [[TMP18]], align 4
 // CHECK4-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP19]])
 // CHECK4-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK4-NEXT:    [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0

 // CHECK4-NEXT:    br label [[COND_END]]
 // CHECK4:       cond.end:
 // CHECK4-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK4-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK4-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK4:       omp.inner.for.cond:
-// CHECK4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !15
+// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !15
 // CHECK4-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK4-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK4:       omp.inner.for.cond.cleanup:
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK4:       omp.inner.for.body:
-// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
-// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
-// CHECK4-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32, i32)* @.omp_outlined..5 to void (i32*, i32*, ...)*), i32 [[TMP7]], i32 [[TMP8]])
+// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !15
+// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !15
+// CHECK4-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32, i32)* @.omp_outlined..5 to void (i32*, i32*, ...)*), i32 [[TMP7]], i32 [[TMP8]]), !llvm.access.group !15
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK4:       omp.inner.for.inc:
-// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !15
+// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !15
 // CHECK4-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP9]], [[TMP10]]
-// CHECK4-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP10:![0-9]+]]
+// CHECK4-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !15
+// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP16:![0-9]+]]
 // CHECK4:       omp.inner.for.end:
 // CHECK4-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK4:       omp.loop.exit:
 // CHECK4-NEXT:    [[TMP11:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK4-NEXT:    [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 4
 // CHECK4-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP12]])
 // CHECK4-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK4-NEXT:    [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0

 // CHECK4-NEXT:    br label [[COND_END]]
 // CHECK4:       cond.end:
 // CHECK4-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK4-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK4-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK4:       omp.inner.for.cond:
-// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !18
+// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !18
 // CHECK4-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK4-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK4:       omp.inner.for.cond.cleanup:
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK4:       omp.inner.for.body:
-// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !18
 // CHECK4-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK4-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK4-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK4-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK4-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !18
+// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !18
+// CHECK4-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !18
 // CHECK4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i32 0, i32 [[TMP11]]
-// CHECK4-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4
-// CHECK4-NEXT:    [[TMP12:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP2]], align 4
-// CHECK4-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4
+// CHECK4-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !18
+// CHECK4-NEXT:    [[TMP12:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP2]], align 4, !llvm.access.group !18
+// CHECK4-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !18
 // CHECK4-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR]], i32 0, i32 [[TMP13]]
 // CHECK4-NEXT:    [[TMP14:%.*]] = bitcast %struct.S.0* [[ARRAYIDX4]] to i8*
 // CHECK4-NEXT:    [[TMP15:%.*]] = bitcast %struct.S.0* [[TMP12]] to i8*
-// CHECK4-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP14]], i8* align 4 [[TMP15]], i32 4, i1 false)
+// CHECK4-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP14]], i8* align 4 [[TMP15]], i32 4, i1 false), !llvm.access.group !18
 // CHECK4-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK4:       omp.body.continue:
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK4:       omp.inner.for.inc:
-// CHECK4-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK4-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !18
 // CHECK4-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP16]], 1
-// CHECK4-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4
-// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP11:![0-9]+]]
+// CHECK4-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !18
+// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP19:![0-9]+]]
 // CHECK4:       omp.inner.for.end:
 // CHECK4-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK4:       omp.loop.exit:
 // CHECK4-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK4-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
 // CHECK4-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP18]])
 // CHECK4-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK4-NEXT:    [[TMP20:%.*]] = icmp ne i32 [[TMP19]], 0

 //
 // CHECK5-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK5-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK5-NEXT:  entry:
 // CHECK5-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK5-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK5-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK5:       arraydestroy.body:
-// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK5-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK5-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK5-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK5:       arraydestroy.done1:
 // CHECK5-NEXT:    ret void
 //
 //
 // CHECK5-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 // CHECK5-NEXT:    br label [[COND_END]]
 // CHECK5:       cond.end:
 // CHECK5-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK5-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK5-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK5-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK5:       omp.inner.for.cond:
-// CHECK5-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK5-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK5-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !4
+// CHECK5-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !4
 // CHECK5-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK5-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK5:       omp.inner.for.body:
-// CHECK5-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
+// CHECK5-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !4
 // CHECK5-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
-// CHECK5-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK5-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !4
 // CHECK5-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
-// CHECK5-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]])
+// CHECK5-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]]), !llvm.access.group !4
 // CHECK5-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK5:       omp.inner.for.inc:
-// CHECK5-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK5-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK5-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !4
+// CHECK5-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !4
 // CHECK5-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
-// CHECK5-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP3:![0-9]+]]
+// CHECK5-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !4
+// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
 // CHECK5:       omp.inner.for.end:
 // CHECK5-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK5:       omp.loop.exit:
 // CHECK5-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]])
 // CHECK5-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK5-NEXT:    [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
 // CHECK5-NEXT:    br i1 [[TMP14]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
 // CHECK5:       .omp.final.then:

 // CHECK5-NEXT:    br label [[COND_END]]
 // CHECK5:       cond.end:
 // CHECK5-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK5-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK5-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK5-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK5:       omp.inner.for.cond:
-// CHECK5-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK5-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK5-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !8
+// CHECK5-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !8
 // CHECK5-NEXT:    [[CMP4:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK5-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK5:       omp.inner.for.body:
-// CHECK5-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK5-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !8
 // CHECK5-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK5-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK5-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK5-NEXT:    store i32 1, i32* [[G]], align 4
-// CHECK5-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[_TMP3]], align 8
-// CHECK5-NEXT:    store volatile i32 1, i32* [[TMP10]], align 4
-// CHECK5-NEXT:    store i32 2, i32* [[SIVAR]], align 4
+// CHECK5-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !8
+// CHECK5-NEXT:    store i32 1, i32* [[G]], align 4, !llvm.access.group !8
+// CHECK5-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[_TMP3]], align 8, !llvm.access.group !8
+// CHECK5-NEXT:    store volatile i32 1, i32* [[TMP10]], align 4, !llvm.access.group !8
+// CHECK5-NEXT:    store i32 2, i32* [[SIVAR]], align 4, !llvm.access.group !8
 // CHECK5-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 0
-// CHECK5-NEXT:    store i32* [[G]], i32** [[TMP11]], align 8
+// CHECK5-NEXT:    store i32* [[G]], i32** [[TMP11]], align 8, !llvm.access.group !8
 // CHECK5-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 1
-// CHECK5-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[_TMP3]], align 8
-// CHECK5-NEXT:    store i32* [[TMP13]], i32** [[TMP12]], align 8
+// CHECK5-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[_TMP3]], align 8, !llvm.access.group !8
+// CHECK5-NEXT:    store i32* [[TMP13]], i32** [[TMP12]], align 8, !llvm.access.group !8
 // CHECK5-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 2
-// CHECK5-NEXT:    store i32* [[SIVAR]], i32** [[TMP14]], align 8
-// CHECK5-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE_clEv"(%class.anon.0* nonnull align 8 dereferenceable(24) [[REF_TMP]])
+// CHECK5-NEXT:    store i32* [[SIVAR]], i32** [[TMP14]], align 8, !llvm.access.group !8
+// CHECK5-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE_clEv"(%class.anon.0* nonnull align 8 dereferenceable(24) [[REF_TMP]]), !llvm.access.group !8
 // CHECK5-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK5:       omp.body.continue:
 // CHECK5-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK5:       omp.inner.for.inc:
-// CHECK5-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK5-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !8
 // CHECK5-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP15]], 1
-// CHECK5-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4
-// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
+// CHECK5-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !8
+// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP9:![0-9]+]]
 // CHECK5:       omp.inner.for.end:
 // CHECK5-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK5:       omp.loop.exit:
 // CHECK5-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
 // CHECK5-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK5-NEXT:    [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
 // CHECK5-NEXT:    br i1 [[TMP17]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
 // CHECK5:       .omp.final.then:

 //
 // CHECK6-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK6-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK6-NEXT:  entry:
 // CHECK6-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK6-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK6-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK6:       arraydestroy.body:
-// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK6-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK6-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK6-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK6:       arraydestroy.done1:
 // CHECK6-NEXT:    ret void
 //
 //
 // CHECK6-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 // CHECK6-NEXT:    br label [[COND_END]]
 // CHECK6:       cond.end:
 // CHECK6-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK6-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK6-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK6-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK6:       omp.inner.for.cond:
-// CHECK6-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK6-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK6-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !4
+// CHECK6-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !4
 // CHECK6-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK6-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK6:       omp.inner.for.body:
-// CHECK6-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
+// CHECK6-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !4
 // CHECK6-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
-// CHECK6-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK6-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !4
 // CHECK6-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
-// CHECK6-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]])
+// CHECK6-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]]), !llvm.access.group !4
 // CHECK6-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK6:       omp.inner.for.inc:
-// CHECK6-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK6-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK6-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !4
+// CHECK6-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !4
 // CHECK6-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
-// CHECK6-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP3:![0-9]+]]
+// CHECK6-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !4
+// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
 // CHECK6:       omp.inner.for.end:
 // CHECK6-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK6:       omp.loop.exit:
 // CHECK6-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]])
 // CHECK6-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK6-NEXT:    [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
 // CHECK6-NEXT:    br i1 [[TMP14]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
 // CHECK6:       .omp.final.then:

 // CHECK6-NEXT:    br label [[COND_END]]
 // CHECK6:       cond.end:
 // CHECK6-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK6-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK6-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK6-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK6:       omp.inner.for.cond:
-// CHECK6-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK6-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK6-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !8
+// CHECK6-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !8
 // CHECK6-NEXT:    [[CMP4:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK6-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK6:       omp.inner.for.body:
-// CHECK6-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK6-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !8
 // CHECK6-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK6-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK6-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK6-NEXT:    store i32 1, i32* [[G]], align 4
-// CHECK6-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[_TMP3]], align 8
-// CHECK6-NEXT:    store volatile i32 1, i32* [[TMP10]], align 4
-// CHECK6-NEXT:    store i32 2, i32* [[SIVAR]], align 4
+// CHECK6-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !8
+// CHECK6-NEXT:    store i32 1, i32* [[G]], align 4, !llvm.access.group !8
+// CHECK6-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[_TMP3]], align 8, !llvm.access.group !8
+// CHECK6-NEXT:    store volatile i32 1, i32* [[TMP10]], align 4, !llvm.access.group !8
+// CHECK6-NEXT:    store i32 2, i32* [[SIVAR]], align 4, !llvm.access.group !8
 // CHECK6-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 0
-// CHECK6-NEXT:    store i32* [[G]], i32** [[TMP11]], align 8
+// CHECK6-NEXT:    store i32* [[G]], i32** [[TMP11]], align 8, !llvm.access.group !8
 // CHECK6-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 1
-// CHECK6-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[_TMP3]], align 8
-// CHECK6-NEXT:    store i32* [[TMP13]], i32** [[TMP12]], align 8
+// CHECK6-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[_TMP3]], align 8, !llvm.access.group !8
+// CHECK6-NEXT:    store i32* [[TMP13]], i32** [[TMP12]], align 8, !llvm.access.group !8
 // CHECK6-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 2
-// CHECK6-NEXT:    store i32* [[SIVAR]], i32** [[TMP14]], align 8
-// CHECK6-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE_clEv"(%class.anon.0* nonnull align 8 dereferenceable(24) [[REF_TMP]])
+// CHECK6-NEXT:    store i32* [[SIVAR]], i32** [[TMP14]], align 8, !llvm.access.group !8
+// CHECK6-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE_clEv"(%class.anon.0* nonnull align 8 dereferenceable(24) [[REF_TMP]]), !llvm.access.group !8
 // CHECK6-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK6:       omp.body.continue:
 // CHECK6-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK6:       omp.inner.for.inc:
-// CHECK6-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK6-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !8
 // CHECK6-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP15]], 1
-// CHECK6-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4
-// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
+// CHECK6-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !8
+// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP9:![0-9]+]]
 // CHECK6:       omp.inner.for.end:
 // CHECK6-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK6:       omp.loop.exit:
 // CHECK6-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
 // CHECK6-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK6-NEXT:    [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
 // CHECK6-NEXT:    br i1 [[TMP17]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
 // CHECK6:       .omp.final.then:

 //
 // CHECK7-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK7-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK7-NEXT:  entry:
 // CHECK7-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK7-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK7-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK7:       arraydestroy.body:
-// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK7-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK7-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK7-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK7:       arraydestroy.done1:
 // CHECK7-NEXT:    ret void
 //
 //
 // CHECK7-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK8-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK8-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK8-NEXT:  entry:
 // CHECK8-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK8-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK8-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK8:       arraydestroy.body:
-// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK8-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK8-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK8-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK8:       arraydestroy.done1:
 // CHECK8-NEXT:    ret void
 //
 //
 // CHECK8-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK9-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK9-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK9-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK9-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK9:       arraydestroy.body:
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK9-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK9-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK9-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK9:       arraydestroy.done1:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK10-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK10-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK10-NEXT:  entry:
 // CHECK10-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK10-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK10-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK10:       arraydestroy.body:
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK10-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK10-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK10-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK10:       arraydestroy.done1:
 // CHECK10-NEXT:    ret void
 //
 //
 // CHECK10-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK11-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK11-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK11-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK11-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK11:       arraydestroy.body:
-// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK11-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK11-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK11-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK11:       arraydestroy.done1:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK12-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK12-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK12-NEXT:  entry:
 // CHECK12-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK12-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK12-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK12:       arraydestroy.body:
-// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK12-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK12-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK12-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK12:       arraydestroy.done1:
 // CHECK12-NEXT:    ret void
 //
 //
 // CHECK12-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 // CHECK13-NEXT:    br label [[COND_END]]
 // CHECK13:       cond.end:
 // CHECK13-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK13-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK13-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK13-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK13:       omp.inner.for.cond:
-// CHECK13-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK13-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK13-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
+// CHECK13-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !6
 // CHECK13-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK13-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK13:       omp.inner.for.cond.cleanup:
 // CHECK13-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK13:       omp.inner.for.body:
-// CHECK13-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
+// CHECK13-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !6
 // CHECK13-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
-// CHECK13-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK13-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !6
 // CHECK13-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
-// CHECK13-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..1 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]])
+// CHECK13-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..1 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]]), !llvm.access.group !6
 // CHECK13-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK13:       omp.inner.for.inc:
-// CHECK13-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK13-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK13-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
+// CHECK13-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !6
 // CHECK13-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
-// CHECK13-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP4:![0-9]+]]
+// CHECK13-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
+// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP7:![0-9]+]]
 // CHECK13:       omp.inner.for.end:
 // CHECK13-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK13:       omp.loop.exit:
 // CHECK13-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK13-NEXT:    [[TMP14:%.*]] = load i32, i32* [[TMP13]], align 4
 // CHECK13-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP14]])
 // CHECK13-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK13-NEXT:    [[TMP16:%.*]] = icmp ne i32 [[TMP15]], 0

 // CHECK13-NEXT:    br label [[COND_END]]
 // CHECK13:       cond.end:
 // CHECK13-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK13-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK13-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK13-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK13:       omp.inner.for.cond:
-// CHECK13-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK13-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK13-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !10
+// CHECK13-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !10
 // CHECK13-NEXT:    [[CMP2:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK13-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK13:       omp.inner.for.cond.cleanup:
 // CHECK13-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK13:       omp.inner.for.body:
-// CHECK13-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK13-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !10
 // CHECK13-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK13-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK13-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK13-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK13-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK13-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !10
+// CHECK13-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !10
+// CHECK13-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !10
 // CHECK13-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP11]] to i64
 // CHECK13-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i64 0, i64 [[IDXPROM]]
-// CHECK13-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4
-// CHECK13-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4
+// CHECK13-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !10
+// CHECK13-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !10
 // CHECK13-NEXT:    [[IDXPROM3:%.*]] = sext i32 [[TMP12]] to i64
 // CHECK13-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds [2 x %struct.S], [2 x %struct.S]* [[S_ARR]], i64 0, i64 [[IDXPROM3]]
 // CHECK13-NEXT:    [[TMP13:%.*]] = bitcast %struct.S* [[ARRAYIDX4]] to i8*
 // CHECK13-NEXT:    [[TMP14:%.*]] = bitcast %struct.S* [[VAR]] to i8*
-// CHECK13-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i64 4, i1 false)
-// CHECK13-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
-// CHECK13-NEXT:    [[TMP16:%.*]] = load i32, i32* [[SIVAR]], align 4
+// CHECK13-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i64 4, i1 false), !llvm.access.group !10
+// CHECK13-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !10
+// CHECK13-NEXT:    [[TMP16:%.*]] = load i32, i32* [[SIVAR]], align 4, !llvm.access.group !10
 // CHECK13-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP16]], [[TMP15]]
-// CHECK13-NEXT:    store i32 [[ADD5]], i32* [[SIVAR]], align 4
+// CHECK13-NEXT:    store i32 [[ADD5]], i32* [[SIVAR]], align 4, !llvm.access.group !10
 // CHECK13-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK13:       omp.body.continue:
 // CHECK13-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK13:       omp.inner.for.inc:
-// CHECK13-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK13-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !10
 // CHECK13-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP17]], 1
-// CHECK13-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
-// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP6:![0-9]+]]
+// CHECK13-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !10
+// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP11:![0-9]+]]
 // CHECK13:       omp.inner.for.end:
 // CHECK13-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK13:       omp.loop.exit:
 // CHECK13-NEXT:    [[TMP18:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK13-NEXT:    [[TMP19:%.*]] = load i32, i32* [[TMP18]], align 4
 // CHECK13-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP19]])
 // CHECK13-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK13-NEXT:    [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0

 // CHECK13-NEXT:    br label [[COND_END]]
 // CHECK13:       cond.end:
 // CHECK13-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK13-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK13-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK13-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK13:       omp.inner.for.cond:
-// CHECK13-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK13-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK13-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !15
+// CHECK13-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !15
 // CHECK13-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK13-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK13:       omp.inner.for.cond.cleanup:
 // CHECK13-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK13:       omp.inner.for.body:
-// CHECK13-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
+// CHECK13-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !15
 // CHECK13-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
-// CHECK13-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK13-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !15
 // CHECK13-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
-// CHECK13-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]])
+// CHECK13-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]]), !llvm.access.group !15
 // CHECK13-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK13:       omp.inner.for.inc:
-// CHECK13-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK13-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK13-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !15
+// CHECK13-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !15
 // CHECK13-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
-// CHECK13-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP9:![0-9]+]]
+// CHECK13-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !15
+// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP16:![0-9]+]]
 // CHECK13:       omp.inner.for.end:
 // CHECK13-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK13:       omp.loop.exit:
 // CHECK13-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK13-NEXT:    [[TMP14:%.*]] = load i32, i32* [[TMP13]], align 4
 // CHECK13-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP14]])
 // CHECK13-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK13-NEXT:    [[TMP16:%.*]] = icmp ne i32 [[TMP15]], 0

 // CHECK13-NEXT:    br label [[COND_END]]
 // CHECK13:       cond.end:
 // CHECK13-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK13-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK13-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK13-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK13:       omp.inner.for.cond:
-// CHECK13-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK13-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK13-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !18
+// CHECK13-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !18
 // CHECK13-NEXT:    [[CMP4:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK13-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK13:       omp.inner.for.cond.cleanup:
 // CHECK13-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK13:       omp.inner.for.body:
-// CHECK13-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK13-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !18
 // CHECK13-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK13-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK13-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK13-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK13-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK13-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !18
+// CHECK13-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !18
+// CHECK13-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !18
 // CHECK13-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP11]] to i64
 // CHECK13-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i64 0, i64 [[IDXPROM]]
-// CHECK13-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4
-// CHECK13-NEXT:    [[TMP12:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP3]], align 8
-// CHECK13-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4
+// CHECK13-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !18
+// CHECK13-NEXT:    [[TMP12:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP3]], align 8, !llvm.access.group !18
+// CHECK13-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !18
 // CHECK13-NEXT:    [[IDXPROM5:%.*]] = sext i32 [[TMP13]] to i64
 // CHECK13-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR]], i64 0, i64 [[IDXPROM5]]
 // CHECK13-NEXT:    [[TMP14:%.*]] = bitcast %struct.S.0* [[ARRAYIDX6]] to i8*
 // CHECK13-NEXT:    [[TMP15:%.*]] = bitcast %struct.S.0* [[TMP12]] to i8*
-// CHECK13-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP14]], i8* align 4 [[TMP15]], i64 4, i1 false)
+// CHECK13-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP14]], i8* align 4 [[TMP15]], i64 4, i1 false), !llvm.access.group !18
 // CHECK13-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK13:       omp.body.continue:
 // CHECK13-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK13:       omp.inner.for.inc:
-// CHECK13-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK13-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !18
 // CHECK13-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP16]], 1
-// CHECK13-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
-// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP10:![0-9]+]]
+// CHECK13-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !18
+// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP19:![0-9]+]]
 // CHECK13:       omp.inner.for.end:
 // CHECK13-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK13:       omp.loop.exit:
 // CHECK13-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK13-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
 // CHECK13-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP18]])
 // CHECK13-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK13-NEXT:    [[TMP20:%.*]] = icmp ne i32 [[TMP19]], 0

 // CHECK14-NEXT:    br label [[COND_END]]
 // CHECK14:       cond.end:
 // CHECK14-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK14-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK14-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK14-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK14:       omp.inner.for.cond:
-// CHECK14-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK14-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK14-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
+// CHECK14-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !6
 // CHECK14-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK14-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK14:       omp.inner.for.cond.cleanup:
 // CHECK14-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK14:       omp.inner.for.body:
-// CHECK14-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
+// CHECK14-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !6
 // CHECK14-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
-// CHECK14-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK14-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !6
 // CHECK14-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
-// CHECK14-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..1 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]])
+// CHECK14-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..1 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]]), !llvm.access.group !6
 // CHECK14-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK14:       omp.inner.for.inc:
-// CHECK14-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK14-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK14-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
+// CHECK14-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !6
 // CHECK14-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
-// CHECK14-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP4:![0-9]+]]
+// CHECK14-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
+// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP7:![0-9]+]]
 // CHECK14:       omp.inner.for.end:
 // CHECK14-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK14:       omp.loop.exit:
 // CHECK14-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK14-NEXT:    [[TMP14:%.*]] = load i32, i32* [[TMP13]], align 4
 // CHECK14-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP14]])
 // CHECK14-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK14-NEXT:    [[TMP16:%.*]] = icmp ne i32 [[TMP15]], 0

 // CHECK14-NEXT:    br label [[COND_END]]
 // CHECK14:       cond.end:
 // CHECK14-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK14-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK14-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK14-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK14:       omp.inner.for.cond:
-// CHECK14-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK14-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK14-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !10
+// CHECK14-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !10
 // CHECK14-NEXT:    [[CMP2:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK14-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK14:       omp.inner.for.cond.cleanup:
 // CHECK14-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK14:       omp.inner.for.body:
-// CHECK14-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK14-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !10
 // CHECK14-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK14-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK14-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK14-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK14-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK14-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !10
+// CHECK14-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !10
+// CHECK14-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !10
 // CHECK14-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP11]] to i64
 // CHECK14-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i64 0, i64 [[IDXPROM]]
-// CHECK14-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4
-// CHECK14-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4
+// CHECK14-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !10
+// CHECK14-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !10
 // CHECK14-NEXT:    [[IDXPROM3:%.*]] = sext i32 [[TMP12]] to i64
 // CHECK14-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds [2 x %struct.S], [2 x %struct.S]* [[S_ARR]], i64 0, i64 [[IDXPROM3]]
 // CHECK14-NEXT:    [[TMP13:%.*]] = bitcast %struct.S* [[ARRAYIDX4]] to i8*
 // CHECK14-NEXT:    [[TMP14:%.*]] = bitcast %struct.S* [[VAR]] to i8*
-// CHECK14-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i64 4, i1 false)
-// CHECK14-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
-// CHECK14-NEXT:    [[TMP16:%.*]] = load i32, i32* [[SIVAR]], align 4
+// CHECK14-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i64 4, i1 false), !llvm.access.group !10
+// CHECK14-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !10
+// CHECK14-NEXT:    [[TMP16:%.*]] = load i32, i32* [[SIVAR]], align 4, !llvm.access.group !10
 // CHECK14-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP16]], [[TMP15]]
-// CHECK14-NEXT:    store i32 [[ADD5]], i32* [[SIVAR]], align 4
+// CHECK14-NEXT:    store i32 [[ADD5]], i32* [[SIVAR]], align 4, !llvm.access.group !10
 // CHECK14-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK14:       omp.body.continue:
 // CHECK14-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK14:       omp.inner.for.inc:
-// CHECK14-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK14-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !10
 // CHECK14-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP17]], 1
-// CHECK14-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
-// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP6:![0-9]+]]
+// CHECK14-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !10
+// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP11:![0-9]+]]
 // CHECK14:       omp.inner.for.end:
 // CHECK14-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK14:       omp.loop.exit:
 // CHECK14-NEXT:    [[TMP18:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK14-NEXT:    [[TMP19:%.*]] = load i32, i32* [[TMP18]], align 4
 // CHECK14-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP19]])
 // CHECK14-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK14-NEXT:    [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0

 // CHECK14-NEXT:    br label [[COND_END]]
 // CHECK14:       cond.end:
 // CHECK14-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK14-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK14-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK14-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK14:       omp.inner.for.cond:
-// CHECK14-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK14-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK14-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !15
+// CHECK14-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !15
 // CHECK14-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK14-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK14:       omp.inner.for.cond.cleanup:
 // CHECK14-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK14:       omp.inner.for.body:
-// CHECK14-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
+// CHECK14-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !15
 // CHECK14-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
-// CHECK14-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK14-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !15
 // CHECK14-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
-// CHECK14-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]])
+// CHECK14-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]]), !llvm.access.group !15
 // CHECK14-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK14:       omp.inner.for.inc:
-// CHECK14-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK14-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK14-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !15
+// CHECK14-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !15
 // CHECK14-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
-// CHECK14-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP9:![0-9]+]]
+// CHECK14-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !15
+// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP16:![0-9]+]]
 // CHECK14:       omp.inner.for.end:
 // CHECK14-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK14:       omp.loop.exit:
 // CHECK14-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK14-NEXT:    [[TMP14:%.*]] = load i32, i32* [[TMP13]], align 4
 // CHECK14-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP14]])
 // CHECK14-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK14-NEXT:    [[TMP16:%.*]] = icmp ne i32 [[TMP15]], 0

 // CHECK14-NEXT:    br label [[COND_END]]
 // CHECK14:       cond.end:
 // CHECK14-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK14-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK14-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK14-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK14:       omp.inner.for.cond:
-// CHECK14-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK14-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK14-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !18
+// CHECK14-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !18
 // CHECK14-NEXT:    [[CMP4:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK14-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK14:       omp.inner.for.cond.cleanup:
 // CHECK14-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK14:       omp.inner.for.body:
-// CHECK14-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK14-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !18
 // CHECK14-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK14-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK14-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK14-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK14-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK14-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !18
+// CHECK14-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !18
+// CHECK14-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !18
 // CHECK14-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP11]] to i64
 // CHECK14-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i64 0, i64 [[IDXPROM]]
-// CHECK14-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4
-// CHECK14-NEXT:    [[TMP12:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP3]], align 8
-// CHECK14-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4
+// CHECK14-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !18
+// CHECK14-NEXT:    [[TMP12:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP3]], align 8, !llvm.access.group !18
+// CHECK14-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !18
 // CHECK14-NEXT:    [[IDXPROM5:%.*]] = sext i32 [[TMP13]] to i64
 // CHECK14-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR]], i64 0, i64 [[IDXPROM5]]
 // CHECK14-NEXT:    [[TMP14:%.*]] = bitcast %struct.S.0* [[ARRAYIDX6]] to i8*
 // CHECK14-NEXT:    [[TMP15:%.*]] = bitcast %struct.S.0* [[TMP12]] to i8*
-// CHECK14-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP14]], i8* align 4 [[TMP15]], i64 4, i1 false)
+// CHECK14-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP14]], i8* align 4 [[TMP15]], i64 4, i1 false), !llvm.access.group !18
 // CHECK14-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK14:       omp.body.continue:
 // CHECK14-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK14:       omp.inner.for.inc:
-// CHECK14-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK14-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !18
 // CHECK14-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP16]], 1
-// CHECK14-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
-// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP10:![0-9]+]]
+// CHECK14-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !18
+// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP19:![0-9]+]]
 // CHECK14:       omp.inner.for.end:
 // CHECK14-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK14:       omp.loop.exit:
 // CHECK14-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK14-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
 // CHECK14-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP18]])
 // CHECK14-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK14-NEXT:    [[TMP20:%.*]] = icmp ne i32 [[TMP19]], 0

 // CHECK15-NEXT:    br label [[COND_END]]
 // CHECK15:       cond.end:
 // CHECK15-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK15-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK15-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK15-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK15:       omp.inner.for.cond:
-// CHECK15-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK15-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK15-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !7
+// CHECK15-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !7
 // CHECK15-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK15-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK15:       omp.inner.for.cond.cleanup:
 // CHECK15-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK15:       omp.inner.for.body:
-// CHECK15-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
-// CHECK15-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
-// CHECK15-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32, i32)* @.omp_outlined..1 to void (i32*, i32*, ...)*), i32 [[TMP7]], i32 [[TMP8]])
+// CHECK15-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !7
+// CHECK15-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !7
+// CHECK15-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32, i32)* @.omp_outlined..1 to void (i32*, i32*, ...)*), i32 [[TMP7]], i32 [[TMP8]]), !llvm.access.group !7
 // CHECK15-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK15:       omp.inner.for.inc:
-// CHECK15-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK15-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK15-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !7
+// CHECK15-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !7
 // CHECK15-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP9]], [[TMP10]]
-// CHECK15-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
+// CHECK15-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !7
+// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP8:![0-9]+]]
 // CHECK15:       omp.inner.for.end:
 // CHECK15-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK15:       omp.loop.exit:
 // CHECK15-NEXT:    [[TMP11:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK15-NEXT:    [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 4
 // CHECK15-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP12]])
 // CHECK15-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK15-NEXT:    [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0

 // CHECK15-NEXT:    br label [[COND_END]]
 // CHECK15:       cond.end:
 // CHECK15-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK15-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK15-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK15-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK15:       omp.inner.for.cond:
-// CHECK15-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK15-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK15-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !11
+// CHECK15-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !11
 // CHECK15-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK15-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK15:       omp.inner.for.cond.cleanup:
 // CHECK15-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK15:       omp.inner.for.body:
-// CHECK15-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK15-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !11
 // CHECK15-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK15-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK15-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK15-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK15-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK15-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !11
+// CHECK15-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !11
+// CHECK15-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !11
 // CHECK15-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i32 0, i32 [[TMP11]]
-// CHECK15-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4
-// CHECK15-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4
+// CHECK15-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !11
+// CHECK15-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !11
 // CHECK15-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds [2 x %struct.S], [2 x %struct.S]* [[S_ARR]], i32 0, i32 [[TMP12]]
 // CHECK15-NEXT:    [[TMP13:%.*]] = bitcast %struct.S* [[ARRAYIDX2]] to i8*
 // CHECK15-NEXT:    [[TMP14:%.*]] = bitcast %struct.S* [[VAR]] to i8*
-// CHECK15-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i32 4, i1 false)
-// CHECK15-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
-// CHECK15-NEXT:    [[TMP16:%.*]] = load i32, i32* [[SIVAR]], align 4
+// CHECK15-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i32 4, i1 false), !llvm.access.group !11
+// CHECK15-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !11
+// CHECK15-NEXT:    [[TMP16:%.*]] = load i32, i32* [[SIVAR]], align 4, !llvm.access.group !11
 // CHECK15-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP16]], [[TMP15]]
-// CHECK15-NEXT:    store i32 [[ADD3]], i32* [[SIVAR]], align 4
+// CHECK15-NEXT:    store i32 [[ADD3]], i32* [[SIVAR]], align 4, !llvm.access.group !11
 // CHECK15-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK15:       omp.body.continue:
 // CHECK15-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK15:       omp.inner.for.inc:
-// CHECK15-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK15-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !11
 // CHECK15-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP17]], 1
-// CHECK15-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4
-// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP7:![0-9]+]]
+// CHECK15-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !11
+// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP12:![0-9]+]]
 // CHECK15:       omp.inner.for.end:
 // CHECK15-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK15:       omp.loop.exit:
 // CHECK15-NEXT:    [[TMP18:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK15-NEXT:    [[TMP19:%.*]] = load i32, i32* [[TMP18]], align 4
 // CHECK15-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP19]])
 // CHECK15-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK15-NEXT:    [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0

 // CHECK15-NEXT:    br label [[COND_END]]
 // CHECK15:       cond.end:
 // CHECK15-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK15-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK15-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK15-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK15:       omp.inner.for.cond:
-// CHECK15-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK15-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK15-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !16
+// CHECK15-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !16
 // CHECK15-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK15-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK15:       omp.inner.for.cond.cleanup:
 // CHECK15-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK15:       omp.inner.for.body:
-// CHECK15-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
-// CHECK15-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
-// CHECK15-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32, i32)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i32 [[TMP7]], i32 [[TMP8]])
+// CHECK15-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !16
+// CHECK15-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !16
+// CHECK15-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32, i32)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i32 [[TMP7]], i32 [[TMP8]]), !llvm.access.group !16
 // CHECK15-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK15:       omp.inner.for.inc:
-// CHECK15-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK15-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK15-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !16
+// CHECK15-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !16
 // CHECK15-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP9]], [[TMP10]]
-// CHECK15-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP10:![0-9]+]]
+// CHECK15-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !16
+// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP17:![0-9]+]]
 // CHECK15:       omp.inner.for.end:
 // CHECK15-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK15:       omp.loop.exit:
 // CHECK15-NEXT:    [[TMP11:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK15-NEXT:    [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 4
 // CHECK15-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP12]])
 // CHECK15-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK15-NEXT:    [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0

 // CHECK15-NEXT:    br label [[COND_END]]
 // CHECK15:       cond.end:
 // CHECK15-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK15-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK15-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK15-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK15:       omp.inner.for.cond:
-// CHECK15-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK15-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK15-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !19
+// CHECK15-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !19
 // CHECK15-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK15-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK15:       omp.inner.for.cond.cleanup:
 // CHECK15-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK15:       omp.inner.for.body:
-// CHECK15-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK15-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !19
 // CHECK15-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK15-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK15-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK15-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK15-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK15-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !19
+// CHECK15-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !19
+// CHECK15-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !19
 // CHECK15-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i32 0, i32 [[TMP11]]
-// CHECK15-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4
-// CHECK15-NEXT:    [[TMP12:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP2]], align 4
-// CHECK15-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4
+// CHECK15-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !19
+// CHECK15-NEXT:    [[TMP12:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP2]], align 4, !llvm.access.group !19
+// CHECK15-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !19
 // CHECK15-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR]], i32 0, i32 [[TMP13]]
 // CHECK15-NEXT:    [[TMP14:%.*]] = bitcast %struct.S.0* [[ARRAYIDX4]] to i8*
 // CHECK15-NEXT:    [[TMP15:%.*]] = bitcast %struct.S.0* [[TMP12]] to i8*
-// CHECK15-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP14]], i8* align 4 [[TMP15]], i32 4, i1 false)
+// CHECK15-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP14]], i8* align 4 [[TMP15]], i32 4, i1 false), !llvm.access.group !19
 // CHECK15-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK15:       omp.body.continue:
 // CHECK15-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK15:       omp.inner.for.inc:
-// CHECK15-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK15-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !19
 // CHECK15-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP16]], 1
-// CHECK15-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4
-// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP11:![0-9]+]]
+// CHECK15-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !19
+// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP20:![0-9]+]]
 // CHECK15:       omp.inner.for.end:
 // CHECK15-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK15:       omp.loop.exit:
 // CHECK15-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK15-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
 // CHECK15-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP18]])
 // CHECK15-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK15-NEXT:    [[TMP20:%.*]] = icmp ne i32 [[TMP19]], 0

 // CHECK16-NEXT:    br label [[COND_END]]
 // CHECK16:       cond.end:
 // CHECK16-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK16-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK16-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK16-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK16:       omp.inner.for.cond:
-// CHECK16-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK16-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK16-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !7
+// CHECK16-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !7
 // CHECK16-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK16-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK16:       omp.inner.for.cond.cleanup:
 // CHECK16-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK16:       omp.inner.for.body:
-// CHECK16-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
-// CHECK16-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
-// CHECK16-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32, i32)* @.omp_outlined..1 to void (i32*, i32*, ...)*), i32 [[TMP7]], i32 [[TMP8]])
+// CHECK16-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !7
+// CHECK16-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !7
+// CHECK16-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32, i32)* @.omp_outlined..1 to void (i32*, i32*, ...)*), i32 [[TMP7]], i32 [[TMP8]]), !llvm.access.group !7
 // CHECK16-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK16:       omp.inner.for.inc:
-// CHECK16-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK16-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK16-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !7
+// CHECK16-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !7
 // CHECK16-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP9]], [[TMP10]]
-// CHECK16-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
+// CHECK16-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !7
+// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP8:![0-9]+]]
 // CHECK16:       omp.inner.for.end:
 // CHECK16-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK16:       omp.loop.exit:
 // CHECK16-NEXT:    [[TMP11:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK16-NEXT:    [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 4
 // CHECK16-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP12]])
 // CHECK16-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK16-NEXT:    [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0

 // CHECK16-NEXT:    br label [[COND_END]]
 // CHECK16:       cond.end:
 // CHECK16-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK16-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK16-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK16-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK16:       omp.inner.for.cond:
-// CHECK16-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK16-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK16-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !11
+// CHECK16-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !11
 // CHECK16-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK16-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK16:       omp.inner.for.cond.cleanup:
 // CHECK16-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK16:       omp.inner.for.body:
-// CHECK16-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK16-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !11
 // CHECK16-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK16-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK16-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK16-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK16-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK16-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !11
+// CHECK16-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !11
+// CHECK16-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !11
 // CHECK16-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i32 0, i32 [[TMP11]]
-// CHECK16-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4
-// CHECK16-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4
+// CHECK16-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !11
+// CHECK16-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !11
 // CHECK16-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds [2 x %struct.S], [2 x %struct.S]* [[S_ARR]], i32 0, i32 [[TMP12]]
 // CHECK16-NEXT:    [[TMP13:%.*]] = bitcast %struct.S* [[ARRAYIDX2]] to i8*
 // CHECK16-NEXT:    [[TMP14:%.*]] = bitcast %struct.S* [[VAR]] to i8*
-// CHECK16-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i32 4, i1 false)
-// CHECK16-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
-// CHECK16-NEXT:    [[TMP16:%.*]] = load i32, i32* [[SIVAR]], align 4
+// CHECK16-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i32 4, i1 false), !llvm.access.group !11
+// CHECK16-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !11
+// CHECK16-NEXT:    [[TMP16:%.*]] = load i32, i32* [[SIVAR]], align 4, !llvm.access.group !11
 // CHECK16-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP16]], [[TMP15]]
-// CHECK16-NEXT:    store i32 [[ADD3]], i32* [[SIVAR]], align 4
+// CHECK16-NEXT:    store i32 [[ADD3]], i32* [[SIVAR]], align 4, !llvm.access.group !11
 // CHECK16-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK16:       omp.body.continue:
 // CHECK16-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK16:       omp.inner.for.inc:
-// CHECK16-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK16-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !11
 // CHECK16-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP17]], 1
-// CHECK16-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4
-// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP7:![0-9]+]]
+// CHECK16-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !11
+// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP12:![0-9]+]]
 // CHECK16:       omp.inner.for.end:
 // CHECK16-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK16:       omp.loop.exit:
 // CHECK16-NEXT:    [[TMP18:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK16-NEXT:    [[TMP19:%.*]] = load i32, i32* [[TMP18]], align 4
 // CHECK16-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP19]])
 // CHECK16-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK16-NEXT:    [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0

 // CHECK16-NEXT:    br label [[COND_END]]
 // CHECK16:       cond.end:
 // CHECK16-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK16-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK16-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK16-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK16:       omp.inner.for.cond:
-// CHECK16-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK16-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK16-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !16
+// CHECK16-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !16
 // CHECK16-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK16-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK16:       omp.inner.for.cond.cleanup:
 // CHECK16-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK16:       omp.inner.for.body:
-// CHECK16-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
-// CHECK16-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
-// CHECK16-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32, i32)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i32 [[TMP7]], i32 [[TMP8]])
+// CHECK16-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !16
+// CHECK16-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !16
+// CHECK16-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32, i32)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i32 [[TMP7]], i32 [[TMP8]]), !llvm.access.group !16
 // CHECK16-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK16:       omp.inner.for.inc:
-// CHECK16-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK16-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK16-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !16
+// CHECK16-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !16
 // CHECK16-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP9]], [[TMP10]]
-// CHECK16-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP10:![0-9]+]]
+// CHECK16-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !16
+// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP17:![0-9]+]]
 // CHECK16:       omp.inner.for.end:
 // CHECK16-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK16:       omp.loop.exit:
 // CHECK16-NEXT:    [[TMP11:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK16-NEXT:    [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 4
 // CHECK16-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP12]])
 // CHECK16-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK16-NEXT:    [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0

 // CHECK16-NEXT:    br label [[COND_END]]
 // CHECK16:       cond.end:
 // CHECK16-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK16-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK16-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK16-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK16:       omp.inner.for.cond:
-// CHECK16-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK16-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK16-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !19
+// CHECK16-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !19
 // CHECK16-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK16-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK16:       omp.inner.for.cond.cleanup:
 // CHECK16-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK16:       omp.inner.for.body:
-// CHECK16-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK16-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !19
 // CHECK16-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK16-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK16-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK16-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK16-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK16-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !19
+// CHECK16-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !19
+// CHECK16-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !19
 // CHECK16-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i32 0, i32 [[TMP11]]
-// CHECK16-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4
-// CHECK16-NEXT:    [[TMP12:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP2]], align 4
-// CHECK16-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4
+// CHECK16-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !19
+// CHECK16-NEXT:    [[TMP12:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP2]], align 4, !llvm.access.group !19
+// CHECK16-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !19
 // CHECK16-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR]], i32 0, i32 [[TMP13]]
 // CHECK16-NEXT:    [[TMP14:%.*]] = bitcast %struct.S.0* [[ARRAYIDX4]] to i8*
 // CHECK16-NEXT:    [[TMP15:%.*]] = bitcast %struct.S.0* [[TMP12]] to i8*
-// CHECK16-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP14]], i8* align 4 [[TMP15]], i32 4, i1 false)
+// CHECK16-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP14]], i8* align 4 [[TMP15]], i32 4, i1 false), !llvm.access.group !19
 // CHECK16-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK16:       omp.body.continue:
 // CHECK16-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK16:       omp.inner.for.inc:
-// CHECK16-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK16-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !19
 // CHECK16-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP16]], 1
-// CHECK16-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4
-// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP11:![0-9]+]]
+// CHECK16-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !19
+// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP20:![0-9]+]]
 // CHECK16:       omp.inner.for.end:
 // CHECK16-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK16:       omp.loop.exit:
 // CHECK16-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK16-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
 // CHECK16-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP18]])
 // CHECK16-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK16-NEXT:    [[TMP20:%.*]] = icmp ne i32 [[TMP19]], 0

 // CHECK17-NEXT:    br label [[COND_END]]
 // CHECK17:       cond.end:
 // CHECK17-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK17-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK17-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK17-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK17-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK17:       omp.inner.for.cond:
-// CHECK17-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK17-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK17-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !5
+// CHECK17-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !5
 // CHECK17-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK17-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK17:       omp.inner.for.body:
-// CHECK17-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
+// CHECK17-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !5
 // CHECK17-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
-// CHECK17-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK17-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !5
 // CHECK17-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
-// CHECK17-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..1 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]])
+// CHECK17-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..1 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]]), !llvm.access.group !5
 // CHECK17-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK17:       omp.inner.for.inc:
-// CHECK17-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK17-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK17-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !5
+// CHECK17-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !5
 // CHECK17-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
-// CHECK17-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK17-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP3:![0-9]+]]
+// CHECK17-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !5
+// CHECK17-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP6:![0-9]+]]
 // CHECK17:       omp.inner.for.end:
 // CHECK17-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK17:       omp.loop.exit:
 // CHECK17-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]])
 // CHECK17-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK17-NEXT:    [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
 // CHECK17-NEXT:    br i1 [[TMP14]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
 // CHECK17:       .omp.final.then:

 // CHECK17-NEXT:    br label [[COND_END]]
 // CHECK17:       cond.end:
 // CHECK17-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK17-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK17-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK17-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK17-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK17:       omp.inner.for.cond:
-// CHECK17-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK17-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK17-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
+// CHECK17-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !9
 // CHECK17-NEXT:    [[CMP4:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK17-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK17:       omp.inner.for.body:
-// CHECK17-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK17-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
 // CHECK17-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK17-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK17-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK17-NEXT:    store i32 1, i32* [[G]], align 4
-// CHECK17-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[_TMP3]], align 8
-// CHECK17-NEXT:    store volatile i32 1, i32* [[TMP10]], align 4
-// CHECK17-NEXT:    store i32 2, i32* [[SIVAR]], align 4
+// CHECK17-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !9
+// CHECK17-NEXT:    store i32 1, i32* [[G]], align 4, !llvm.access.group !9
+// CHECK17-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[_TMP3]], align 8, !llvm.access.group !9
+// CHECK17-NEXT:    store volatile i32 1, i32* [[TMP10]], align 4, !llvm.access.group !9
+// CHECK17-NEXT:    store i32 2, i32* [[SIVAR]], align 4, !llvm.access.group !9
 // CHECK17-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[REF_TMP]], i32 0, i32 0
-// CHECK17-NEXT:    store i32* [[G]], i32** [[TMP11]], align 8
+// CHECK17-NEXT:    store i32* [[G]], i32** [[TMP11]], align 8, !llvm.access.group !9
 // CHECK17-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[REF_TMP]], i32 0, i32 1
-// CHECK17-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[_TMP3]], align 8
-// CHECK17-NEXT:    store i32* [[TMP13]], i32** [[TMP12]], align 8
+// CHECK17-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[_TMP3]], align 8, !llvm.access.group !9
+// CHECK17-NEXT:    store i32* [[TMP13]], i32** [[TMP12]], align 8, !llvm.access.group !9
 // CHECK17-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[REF_TMP]], i32 0, i32 2
-// CHECK17-NEXT:    store i32* [[SIVAR]], i32** [[TMP14]], align 8
-// CHECK17-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE_clEv"(%class.anon* nonnull align 8 dereferenceable(24) [[REF_TMP]]) #[[ATTR3:[0-9]+]]
+// CHECK17-NEXT:    store i32* [[SIVAR]], i32** [[TMP14]], align 8, !llvm.access.group !9
+// CHECK17-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE_clEv"(%class.anon* nonnull align 8 dereferenceable(24) [[REF_TMP]]) #[[ATTR3:[0-9]+]], !llvm.access.group !9
 // CHECK17-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK17:       omp.body.continue:
 // CHECK17-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK17:       omp.inner.for.inc:
-// CHECK17-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK17-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
 // CHECK17-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP15]], 1
-// CHECK17-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4
-// CHECK17-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
+// CHECK17-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
+// CHECK17-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP10:![0-9]+]]
 // CHECK17:       omp.inner.for.end:
 // CHECK17-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK17:       omp.loop.exit:
 // CHECK17-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
 // CHECK17-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK17-NEXT:    [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
 // CHECK17-NEXT:    br i1 [[TMP17]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
 // CHECK17:       .omp.final.then:

 //
 // CHECK18-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK18-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK18-NEXT:  entry:
 // CHECK18-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK18-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK18-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK18:       arraydestroy.body:
-// CHECK18-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK18-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK18-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK18-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK18-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK18-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK18-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK18:       arraydestroy.done1:
 // CHECK18-NEXT:    ret void
 //
 //
 // CHECK18-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK19-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK19-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK19-NEXT:  entry:
 // CHECK19-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK19-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK19-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK19:       arraydestroy.body:
-// CHECK19-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK19-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK19-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK19-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK19-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK19-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK19-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK19:       arraydestroy.done1:
 // CHECK19-NEXT:    ret void
 //
 //
 // CHECK19-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK20-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK20-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK20-NEXT:  entry:
 // CHECK20-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK20-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK20-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK20:       arraydestroy.body:
-// CHECK20-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK20-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK20-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK20-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK20-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK20-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK20-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK20:       arraydestroy.done1:
 // CHECK20-NEXT:    ret void
 //
 //
 // CHECK20-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK21-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK21-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK21-NEXT:  entry:
 // CHECK21-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK21-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK21-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK21:       arraydestroy.body:
-// CHECK21-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK21-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK21-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK21-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK21-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK21-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK21-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK21:       arraydestroy.done1:
 // CHECK21-NEXT:    ret void
 //
 //
 // CHECK21-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK22-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK22-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK22-NEXT:  entry:
 // CHECK22-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK22-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK22-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK22:       arraydestroy.body:
-// CHECK22-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK22-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK22-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK22-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK22-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK22-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK22-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK22:       arraydestroy.done1:
 // CHECK22-NEXT:    ret void
 //
 //
 // CHECK22-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

clang/test/OpenMP/target_teams_distribute_private_codegen.cpp

 //
 // CHECK1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK1:       arraydestroy.body:
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK1-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK1:       arraydestroy.done1:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK2:       arraydestroy.body:
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK2-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK2:       arraydestroy.done1:
 // CHECK2-NEXT:    ret void
 //
 //
 // CHECK2-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK3-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK3-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK3:       arraydestroy.body:
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK3-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK3-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK3-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK3:       arraydestroy.done1:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK4-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK4-NEXT:  entry:
 // CHECK4-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK4-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK4-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK4:       arraydestroy.body:
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK4-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK4-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK4-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK4:       arraydestroy.done1:
 // CHECK4-NEXT:    ret void
 //
 //
 // CHECK4-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK9-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK9-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK9-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK9-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK9:       arraydestroy.body:
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK9-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK9-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK9-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK9:       arraydestroy.done1:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK10-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK10-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK10-NEXT:  entry:
 // CHECK10-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK10-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK10-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK10:       arraydestroy.body:
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK10-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK10-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK10-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK10:       arraydestroy.done1:
 // CHECK10-NEXT:    ret void
 //
 //
 // CHECK10-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 //
 // CHECK10-LABEL: define {{[^@]+}}@.omp_offloading.requires_reg
 // CHECK10-SAME: () #[[ATTR0]] {
 // CHECK10-NEXT:  entry:
 // CHECK10-NEXT:    call void @__tgt_register_requires(i64 1)
 // CHECK10-NEXT:    ret void
 //
-//
 

clang/test/OpenMP/target_teams_distribute_simd_firstprivate_codegen.cpp

 //
 // CHECK1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK1:       arraydestroy.body:
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK1-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK1:       arraydestroy.done1:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK2:       arraydestroy.body:
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK2-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK2:       arraydestroy.done1:
 // CHECK2-NEXT:    ret void
 //
 //
 // CHECK2-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK3-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK3-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK3:       arraydestroy.body:
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK3-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK3-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK3-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK3:       arraydestroy.done1:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK4-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK4-NEXT:  entry:
 // CHECK4-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK4-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK4-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK4:       arraydestroy.body:
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK4-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK4-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK4-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK4:       arraydestroy.done1:
 // CHECK4-NEXT:    ret void
 //
 //
 // CHECK4-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK5-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK5-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK5-NEXT:  entry:
 // CHECK5-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK5-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK5-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK5:       arraydestroy.body:
-// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK5-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK5-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK5-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK5:       arraydestroy.done1:
 // CHECK5-NEXT:    ret void
 //
 //
 // CHECK5-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK6-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK6-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK6-NEXT:  entry:
 // CHECK6-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK6-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK6-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK6:       arraydestroy.body:
-// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK6-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK6-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK6-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK6:       arraydestroy.done1:
 // CHECK6-NEXT:    ret void
 //
 //
 // CHECK6-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK7-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK7-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK7-NEXT:  entry:
 // CHECK7-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK7-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK7-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK7:       arraydestroy.body:
-// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK7-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK7-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK7-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK7:       arraydestroy.done1:
 // CHECK7-NEXT:    ret void
 //
 //
 // CHECK7-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK8-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK8-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK8-NEXT:  entry:
 // CHECK8-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK8-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK8-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK8:       arraydestroy.body:
-// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK8-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK8-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK8-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK8:       arraydestroy.done1:
 // CHECK8-NEXT:    ret void
 //
 //
 // CHECK8-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK9-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK9-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK9-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK9-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK9:       arraydestroy.body:
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK9-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK9-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK9-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK9:       arraydestroy.done1:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK10-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK10-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK10-NEXT:  entry:
 // CHECK10-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK10-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK10-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK10:       arraydestroy.body:
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK10-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK10-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK10-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK10:       arraydestroy.done1:
 // CHECK10-NEXT:    ret void
 //
 //
 // CHECK10-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK11-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK11-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK11-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK11-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK11:       arraydestroy.body:
-// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK11-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK11-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK11-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK11:       arraydestroy.done1:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK12-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK12-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK12-NEXT:  entry:
 // CHECK12-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK12-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK12-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK12:       arraydestroy.body:
-// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK12-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK12-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK12-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK12:       arraydestroy.done1:
 // CHECK12-NEXT:    ret void
 //
 //
 // CHECK12-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

clang/test/OpenMP/target_teams_distribute_simd_private_codegen.cpp

 //
 // CHECK1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK1:       arraydestroy.body:
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK1-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK1:       arraydestroy.done1:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK2:       arraydestroy.body:
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK2-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK2:       arraydestroy.done1:
 // CHECK2-NEXT:    ret void
 //
 //
 // CHECK2-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK3-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK3-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK3:       arraydestroy.body:
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK3-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK3-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK3-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK3:       arraydestroy.done1:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK4-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK4-NEXT:  entry:
 // CHECK4-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK4-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK4-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK4:       arraydestroy.body:
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK4-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK4-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK4-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK4:       arraydestroy.done1:
 // CHECK4-NEXT:    ret void
 //
 //
 // CHECK4-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK5-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK5-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK5-NEXT:  entry:
 // CHECK5-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK5-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK5-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK5:       arraydestroy.body:
-// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK5-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK5-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK5-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK5:       arraydestroy.done1:
 // CHECK5-NEXT:    ret void
 //
 //
 // CHECK5-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK6-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK6-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK6-NEXT:  entry:
 // CHECK6-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK6-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK6-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK6:       arraydestroy.body:
-// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK6-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK6-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK6-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK6:       arraydestroy.done1:
 // CHECK6-NEXT:    ret void
 //
 //
 // CHECK6-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK7-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK7-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK7-NEXT:  entry:
 // CHECK7-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK7-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK7-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK7:       arraydestroy.body:
-// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK7-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK7-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK7-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK7:       arraydestroy.done1:
 // CHECK7-NEXT:    ret void
 //
 //
 // CHECK7-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK8-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK8-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK8-NEXT:  entry:
 // CHECK8-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK8-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK8-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK8:       arraydestroy.body:
-// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK8-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK8-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK8-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK8:       arraydestroy.done1:
 // CHECK8-NEXT:    ret void
 //
 //
 // CHECK8-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK9-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK9-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK9-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK9-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK9:       arraydestroy.body:
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK9-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK9-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK9-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK9:       arraydestroy.done1:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK10-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK10-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK10-NEXT:  entry:
 // CHECK10-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK10-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK10-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK10:       arraydestroy.body:
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK10-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK10-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK10-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK10:       arraydestroy.done1:
 // CHECK10-NEXT:    ret void
 //
 //
 // CHECK10-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK11-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK11-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK11-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK11-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK11:       arraydestroy.body:
-// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK11-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK11-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK11-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK11:       arraydestroy.done1:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK12-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK12-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK12-NEXT:  entry:
 // CHECK12-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK12-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK12-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK12:       arraydestroy.body:
-// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK12-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK12-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK12-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK12:       arraydestroy.done1:
 // CHECK12-NEXT:    ret void
 //
 //
 // CHECK12-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

clang/test/OpenMP/teams_distribute_firstprivate_codegen.cpp

 //
 // CHECK1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK1:       arraydestroy.body:
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK1-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK1:       arraydestroy.done1:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK2:       arraydestroy.body:
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK2-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK2:       arraydestroy.done1:
 // CHECK2-NEXT:    ret void
 //
 //
 // CHECK2-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK3-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK3-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK3:       arraydestroy.body:
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK3-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK3-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK3-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK3:       arraydestroy.done1:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK4-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK4-NEXT:  entry:
 // CHECK4-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK4-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK4-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK4:       arraydestroy.body:
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK4-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK4-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK4-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK4:       arraydestroy.done1:
 // CHECK4-NEXT:    ret void
 //
 //
 // CHECK4-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK9-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK9-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK9-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK9-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK9:       arraydestroy.body:
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK9-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK9-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK9-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK9:       arraydestroy.done1:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK10-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK10-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK10-NEXT:  entry:
 // CHECK10-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK10-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK10-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK10:       arraydestroy.body:
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK10-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK10-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK10-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK10:       arraydestroy.done1:
 // CHECK10-NEXT:    ret void
 //
 //
 // CHECK10-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

clang/test/OpenMP/teams_distribute_parallel_for_firstprivate_codegen.cpp

 //
 // CHECK1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK1:       arraydestroy.body:
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK1-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK1:       arraydestroy.done1:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK2:       arraydestroy.body:
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK2-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK2:       arraydestroy.done1:
 // CHECK2-NEXT:    ret void
 //
 //
 // CHECK2-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK3-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK3-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK3:       arraydestroy.body:
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK3-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK3-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK3-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK3:       arraydestroy.done1:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK4-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK4-NEXT:  entry:
 // CHECK4-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK4-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK4-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK4:       arraydestroy.body:
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK4-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK4-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK4-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK4:       arraydestroy.done1:
 // CHECK4-NEXT:    ret void
 //
 //
 // CHECK4-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK9-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK9-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK9-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK9-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK9:       arraydestroy.body:
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK9-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK9-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK9-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK9:       arraydestroy.done1:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK10-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK10-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK10-NEXT:  entry:
 // CHECK10-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK10-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK10-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK10:       arraydestroy.body:
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK10-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK10-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK10-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK10:       arraydestroy.done1:
 // CHECK10-NEXT:    ret void
 //
 //
 // CHECK10-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

clang/test/OpenMP/teams_distribute_parallel_for_private_codegen.cpp

 //
 // CHECK1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK1:       arraydestroy.body:
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK1-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK1:       arraydestroy.done1:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK2:       arraydestroy.body:
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK2-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK2:       arraydestroy.done1:
 // CHECK2-NEXT:    ret void
 //
 //
 // CHECK2-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK3-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK3-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK3:       arraydestroy.body:
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK3-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK3-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK3-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK3:       arraydestroy.done1:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK4-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK4-NEXT:  entry:
 // CHECK4-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK4-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK4-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK4:       arraydestroy.body:
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK4-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK4-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK4-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK4:       arraydestroy.done1:
 // CHECK4-NEXT:    ret void
 //
 //
 // CHECK4-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK9-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK9-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK9-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK9-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK9:       arraydestroy.body:
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK9-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK9-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK9-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK9:       arraydestroy.done1:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK10-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK10-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK10-NEXT:  entry:
 // CHECK10-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK10-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK10-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK10:       arraydestroy.body:
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK10-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK10-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK10-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK10:       arraydestroy.done1:
 // CHECK10-NEXT:    ret void
 //
 //
 // CHECK10-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 //
 // CHECK10-LABEL: define {{[^@]+}}@.omp_offloading.requires_reg
 // CHECK10-SAME: () #[[ATTR0]] {
 // CHECK10-NEXT:  entry:
 // CHECK10-NEXT:    call void @__tgt_register_requires(i64 1)
 // CHECK10-NEXT:    ret void
 //
-//
 

clang/test/OpenMP/teams_distribute_parallel_for_simd_firstprivate_codegen.cpp

 //
 // CHECK1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK1:       arraydestroy.body:
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK1-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK1:       arraydestroy.done1:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK2:       arraydestroy.body:
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK2-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK2:       arraydestroy.done1:
 // CHECK2-NEXT:    ret void
 //
 //
 // CHECK2-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK3-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK3-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK3:       arraydestroy.body:
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK3-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK3-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK3-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK3:       arraydestroy.done1:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK4-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK4-NEXT:  entry:
 // CHECK4-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK4-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK4-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK4:       arraydestroy.body:
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK4-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK4-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK4-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK4:       arraydestroy.done1:
 // CHECK4-NEXT:    ret void
 //
 //
 // CHECK4-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK5-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK5-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK5-NEXT:  entry:
 // CHECK5-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK5-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK5-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK5:       arraydestroy.body:
-// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK5-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK5-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK5-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK5:       arraydestroy.done1:
 // CHECK5-NEXT:    ret void
 //
 //
 // CHECK5-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK6-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK6-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK6-NEXT:  entry:
 // CHECK6-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK6-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK6-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK6:       arraydestroy.body:
-// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK6-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK6-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK6-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK6:       arraydestroy.done1:
 // CHECK6-NEXT:    ret void
 //
 //
 // CHECK6-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK7-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK7-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK7-NEXT:  entry:
 // CHECK7-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK7-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK7-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK7:       arraydestroy.body:
-// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK7-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK7-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK7-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK7:       arraydestroy.done1:
 // CHECK7-NEXT:    ret void
 //
 //
 // CHECK7-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK8-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK8-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK8-NEXT:  entry:
 // CHECK8-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK8-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK8-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK8:       arraydestroy.body:
-// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK8-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK8-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK8-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK8:       arraydestroy.done1:
 // CHECK8-NEXT:    ret void
 //
 //
 // CHECK8-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK9-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK9-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK9-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK9-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK9:       arraydestroy.body:
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK9-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK9-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK9-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK9:       arraydestroy.done1:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK10-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK10-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK10-NEXT:  entry:
 // CHECK10-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK10-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK10-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK10:       arraydestroy.body:
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK10-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK10-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK10-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK10:       arraydestroy.done1:
 // CHECK10-NEXT:    ret void
 //
 //
 // CHECK10-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK11-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK11-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK11-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK11-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK11:       arraydestroy.body:
-// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK11-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK11-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK11-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK11:       arraydestroy.done1:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK12-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK12-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK12-NEXT:  entry:
 // CHECK12-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK12-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK12-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK12:       arraydestroy.body:
-// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK12-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK12-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK12-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK12:       arraydestroy.done1:
 // CHECK12-NEXT:    ret void
 //
 //
 // CHECK12-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

clang/test/OpenMP/teams_distribute_parallel_for_simd_private_codegen.cpp

 //
 // CHECK1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK1:       arraydestroy.body:
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK1-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK1:       arraydestroy.done1:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 // CHECK1-NEXT:    br label [[COND_END]]
 // CHECK1:       cond.end:
 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK1-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK1:       omp.inner.for.cond:
-// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !5
+// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !5
 // CHECK1-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK1-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK1:       omp.inner.for.cond.cleanup:
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK1:       omp.inner.for.body:
-// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !5
 // CHECK1-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
-// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !5
 // CHECK1-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
-// CHECK1-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]])
+// CHECK1-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]]), !llvm.access.group !5
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK1:       omp.inner.for.inc:
-// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !5
+// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !5
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
-// CHECK1-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP4:![0-9]+]]
+// CHECK1-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !5
+// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP6:![0-9]+]]
 // CHECK1:       omp.inner.for.end:
 // CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK1:       omp.loop.exit:
 // CHECK1-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP14:%.*]] = load i32, i32* [[TMP13]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP14]])
 // CHECK1-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK1-NEXT:    [[TMP16:%.*]] = icmp ne i32 [[TMP15]], 0

 // CHECK1-NEXT:    br label [[COND_END]]
 // CHECK1:       cond.end:
 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK1-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK1:       omp.inner.for.cond:
-// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
+// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !9
 // CHECK1-NEXT:    [[CMP2:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK1-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK1:       omp.inner.for.cond.cleanup:
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK1:       omp.inner.for.body:
-// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
 // CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK1-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK1-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !9
+// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !9
+// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !9
 // CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP11]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i64 0, i64 [[IDXPROM]]
-// CHECK1-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4
-// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4
+// CHECK1-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !9
+// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !9
 // CHECK1-NEXT:    [[IDXPROM3:%.*]] = sext i32 [[TMP12]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds [2 x %struct.S], [2 x %struct.S]* [[S_ARR]], i64 0, i64 [[IDXPROM3]]
 // CHECK1-NEXT:    [[TMP13:%.*]] = bitcast %struct.S* [[ARRAYIDX4]] to i8*
 // CHECK1-NEXT:    [[TMP14:%.*]] = bitcast %struct.S* [[VAR]] to i8*
-// CHECK1-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i64 4, i1 false)
-// CHECK1-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
-// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, i32* [[SIVAR]], align 4
+// CHECK1-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i64 4, i1 false), !llvm.access.group !9
+// CHECK1-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !9
+// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, i32* [[SIVAR]], align 4, !llvm.access.group !9
 // CHECK1-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP16]], [[TMP15]]
-// CHECK1-NEXT:    store i32 [[ADD5]], i32* [[SIVAR]], align 4
+// CHECK1-NEXT:    store i32 [[ADD5]], i32* [[SIVAR]], align 4, !llvm.access.group !9
 // CHECK1-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK1:       omp.body.continue:
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK1:       omp.inner.for.inc:
-// CHECK1-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
 // CHECK1-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP17]], 1
-// CHECK1-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
-// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP6:![0-9]+]]
+// CHECK1-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
+// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP10:![0-9]+]]
 // CHECK1:       omp.inner.for.end:
 // CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK1:       omp.loop.exit:
 // CHECK1-NEXT:    [[TMP18:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP19:%.*]] = load i32, i32* [[TMP18]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP19]])
 // CHECK1-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK1-NEXT:    [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0

 // CHECK1-NEXT:    br label [[COND_END]]
 // CHECK1:       cond.end:
 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK1-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK1:       omp.inner.for.cond:
-// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !14
+// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !14
 // CHECK1-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK1-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK1:       omp.inner.for.cond.cleanup:
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK1:       omp.inner.for.body:
-// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !14
 // CHECK1-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
-// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !14
 // CHECK1-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
-// CHECK1-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..5 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]])
+// CHECK1-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..5 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]]), !llvm.access.group !14
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK1:       omp.inner.for.inc:
-// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !14
+// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !14
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
-// CHECK1-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP9:![0-9]+]]
+// CHECK1-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !14
+// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP15:![0-9]+]]
 // CHECK1:       omp.inner.for.end:
 // CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK1:       omp.loop.exit:
 // CHECK1-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP14:%.*]] = load i32, i32* [[TMP13]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP14]])
 // CHECK1-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK1-NEXT:    [[TMP16:%.*]] = icmp ne i32 [[TMP15]], 0

 // CHECK1-NEXT:    br label [[COND_END]]
 // CHECK1:       cond.end:
 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK1-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK1:       omp.inner.for.cond:
-// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !17
+// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !17
 // CHECK1-NEXT:    [[CMP4:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK1-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK1:       omp.inner.for.cond.cleanup:
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK1:       omp.inner.for.body:
-// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !17
 // CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK1-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK1-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !17
+// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !17
+// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !17
 // CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP11]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i64 0, i64 [[IDXPROM]]
-// CHECK1-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4
-// CHECK1-NEXT:    [[TMP12:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP3]], align 8
-// CHECK1-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4
+// CHECK1-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !17
+// CHECK1-NEXT:    [[TMP12:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP3]], align 8, !llvm.access.group !17
+// CHECK1-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !17
 // CHECK1-NEXT:    [[IDXPROM5:%.*]] = sext i32 [[TMP13]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR]], i64 0, i64 [[IDXPROM5]]
 // CHECK1-NEXT:    [[TMP14:%.*]] = bitcast %struct.S.0* [[ARRAYIDX6]] to i8*
 // CHECK1-NEXT:    [[TMP15:%.*]] = bitcast %struct.S.0* [[TMP12]] to i8*
-// CHECK1-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP14]], i8* align 4 [[TMP15]], i64 4, i1 false)
+// CHECK1-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP14]], i8* align 4 [[TMP15]], i64 4, i1 false), !llvm.access.group !17
 // CHECK1-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK1:       omp.body.continue:
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK1:       omp.inner.for.inc:
-// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !17
 // CHECK1-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP16]], 1
-// CHECK1-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
-// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP10:![0-9]+]]
+// CHECK1-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !17
+// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP18:![0-9]+]]
 // CHECK1:       omp.inner.for.end:
 // CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK1:       omp.loop.exit:
 // CHECK1-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP18]])
 // CHECK1-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK1-NEXT:    [[TMP20:%.*]] = icmp ne i32 [[TMP19]], 0

 //
 // CHECK2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK2:       arraydestroy.body:
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK2-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK2:       arraydestroy.done1:
 // CHECK2-NEXT:    ret void
 //
 //
 // CHECK2-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 // CHECK2-NEXT:    br label [[COND_END]]
 // CHECK2:       cond.end:
 // CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK2-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK2-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK2-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK2:       omp.inner.for.cond:
-// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !5
+// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !5
 // CHECK2-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK2-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK2:       omp.inner.for.cond.cleanup:
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK2:       omp.inner.for.body:
-// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
+// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !5
 // CHECK2-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
-// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !5
 // CHECK2-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
-// CHECK2-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]])
+// CHECK2-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]]), !llvm.access.group !5
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK2:       omp.inner.for.inc:
-// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !5
+// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !5
 // CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
-// CHECK2-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP4:![0-9]+]]
+// CHECK2-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !5
+// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP6:![0-9]+]]
 // CHECK2:       omp.inner.for.end:
 // CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK2:       omp.loop.exit:
 // CHECK2-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK2-NEXT:    [[TMP14:%.*]] = load i32, i32* [[TMP13]], align 4
 // CHECK2-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP14]])
 // CHECK2-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK2-NEXT:    [[TMP16:%.*]] = icmp ne i32 [[TMP15]], 0

 // CHECK2-NEXT:    br label [[COND_END]]
 // CHECK2:       cond.end:
 // CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK2-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK2-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK2:       omp.inner.for.cond:
-// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
+// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !9
 // CHECK2-NEXT:    [[CMP2:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK2-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK2:       omp.inner.for.cond.cleanup:
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK2:       omp.inner.for.body:
-// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
 // CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK2-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK2-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !9
+// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !9
+// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !9
 // CHECK2-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP11]] to i64
 // CHECK2-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i64 0, i64 [[IDXPROM]]
-// CHECK2-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4
-// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4
+// CHECK2-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !9
+// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !9
 // CHECK2-NEXT:    [[IDXPROM3:%.*]] = sext i32 [[TMP12]] to i64
 // CHECK2-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds [2 x %struct.S], [2 x %struct.S]* [[S_ARR]], i64 0, i64 [[IDXPROM3]]
 // CHECK2-NEXT:    [[TMP13:%.*]] = bitcast %struct.S* [[ARRAYIDX4]] to i8*
 // CHECK2-NEXT:    [[TMP14:%.*]] = bitcast %struct.S* [[VAR]] to i8*
-// CHECK2-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i64 4, i1 false)
-// CHECK2-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
-// CHECK2-NEXT:    [[TMP16:%.*]] = load i32, i32* [[SIVAR]], align 4
+// CHECK2-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i64 4, i1 false), !llvm.access.group !9
+// CHECK2-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !9
+// CHECK2-NEXT:    [[TMP16:%.*]] = load i32, i32* [[SIVAR]], align 4, !llvm.access.group !9
 // CHECK2-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP16]], [[TMP15]]
-// CHECK2-NEXT:    store i32 [[ADD5]], i32* [[SIVAR]], align 4
+// CHECK2-NEXT:    store i32 [[ADD5]], i32* [[SIVAR]], align 4, !llvm.access.group !9
 // CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK2:       omp.body.continue:
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK2:       omp.inner.for.inc:
-// CHECK2-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
 // CHECK2-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP17]], 1
-// CHECK2-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
-// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP6:![0-9]+]]
+// CHECK2-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
+// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP10:![0-9]+]]
 // CHECK2:       omp.inner.for.end:
 // CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK2:       omp.loop.exit:
 // CHECK2-NEXT:    [[TMP18:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK2-NEXT:    [[TMP19:%.*]] = load i32, i32* [[TMP18]], align 4
 // CHECK2-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP19]])
 // CHECK2-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK2-NEXT:    [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0

 // CHECK2-NEXT:    br label [[COND_END]]
 // CHECK2:       cond.end:
 // CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK2-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK2-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK2-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK2:       omp.inner.for.cond:
-// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !14
+// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !14
 // CHECK2-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK2-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK2:       omp.inner.for.cond.cleanup:
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK2:       omp.inner.for.body:
-// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
+// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !14
 // CHECK2-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
-// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !14
 // CHECK2-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
-// CHECK2-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..5 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]])
+// CHECK2-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..5 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]]), !llvm.access.group !14
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK2:       omp.inner.for.inc:
-// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !14
+// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !14
 // CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
-// CHECK2-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP9:![0-9]+]]
+// CHECK2-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !14
+// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP15:![0-9]+]]
 // CHECK2:       omp.inner.for.end:
 // CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK2:       omp.loop.exit:
 // CHECK2-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK2-NEXT:    [[TMP14:%.*]] = load i32, i32* [[TMP13]], align 4
 // CHECK2-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP14]])
 // CHECK2-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK2-NEXT:    [[TMP16:%.*]] = icmp ne i32 [[TMP15]], 0

 // CHECK2-NEXT:    br label [[COND_END]]
 // CHECK2:       cond.end:
 // CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK2-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK2-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK2:       omp.inner.for.cond:
-// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !17
+// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !17
 // CHECK2-NEXT:    [[CMP4:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK2-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK2:       omp.inner.for.cond.cleanup:
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK2:       omp.inner.for.body:
-// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !17
 // CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK2-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK2-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !17
+// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !17
+// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !17
 // CHECK2-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP11]] to i64
 // CHECK2-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i64 0, i64 [[IDXPROM]]
-// CHECK2-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4
-// CHECK2-NEXT:    [[TMP12:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP3]], align 8
-// CHECK2-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4
+// CHECK2-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !17
+// CHECK2-NEXT:    [[TMP12:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP3]], align 8, !llvm.access.group !17
+// CHECK2-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !17
 // CHECK2-NEXT:    [[IDXPROM5:%.*]] = sext i32 [[TMP13]] to i64
 // CHECK2-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR]], i64 0, i64 [[IDXPROM5]]
 // CHECK2-NEXT:    [[TMP14:%.*]] = bitcast %struct.S.0* [[ARRAYIDX6]] to i8*
 // CHECK2-NEXT:    [[TMP15:%.*]] = bitcast %struct.S.0* [[TMP12]] to i8*
-// CHECK2-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP14]], i8* align 4 [[TMP15]], i64 4, i1 false)
+// CHECK2-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP14]], i8* align 4 [[TMP15]], i64 4, i1 false), !llvm.access.group !17
 // CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK2:       omp.body.continue:
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK2:       omp.inner.for.inc:
-// CHECK2-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !17
 // CHECK2-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP16]], 1
-// CHECK2-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
-// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP10:![0-9]+]]
+// CHECK2-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !17
+// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP18:![0-9]+]]
 // CHECK2:       omp.inner.for.end:
 // CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK2:       omp.loop.exit:
 // CHECK2-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK2-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
 // CHECK2-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP18]])
 // CHECK2-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK2-NEXT:    [[TMP20:%.*]] = icmp ne i32 [[TMP19]], 0

 //
 // CHECK3-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK3-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK3:       arraydestroy.body:
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK3-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK3-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK3-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK3:       arraydestroy.done1:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 // CHECK3-NEXT:    br label [[COND_END]]
 // CHECK3:       cond.end:
 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK3-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK3:       omp.inner.for.cond:
-// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
+// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !6
 // CHECK3-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK3-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK3:       omp.inner.for.cond.cleanup:
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK3:       omp.inner.for.body:
-// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
-// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
-// CHECK3-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32, i32)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i32 [[TMP7]], i32 [[TMP8]])
+// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !6
+// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !6
+// CHECK3-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32, i32)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i32 [[TMP7]], i32 [[TMP8]]), !llvm.access.group !6
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK3:       omp.inner.for.inc:
-// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
+// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !6
 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP9]], [[TMP10]]
-// CHECK3-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
+// CHECK3-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
+// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP7:![0-9]+]]
 // CHECK3:       omp.inner.for.end:
 // CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK3:       omp.loop.exit:
 // CHECK3-NEXT:    [[TMP11:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP12]])
 // CHECK3-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK3-NEXT:    [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0

 // CHECK3-NEXT:    br label [[COND_END]]
 // CHECK3:       cond.end:
 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK3-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK3:       omp.inner.for.cond:
-// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !10
+// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !10
 // CHECK3-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK3-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK3:       omp.inner.for.cond.cleanup:
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK3:       omp.inner.for.body:
-// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !10
 // CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK3-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK3-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK3-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !10
+// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !10
+// CHECK3-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !10
 // CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i32 0, i32 [[TMP11]]
-// CHECK3-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4
-// CHECK3-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4
+// CHECK3-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !10
+// CHECK3-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !10
 // CHECK3-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds [2 x %struct.S], [2 x %struct.S]* [[S_ARR]], i32 0, i32 [[TMP12]]
 // CHECK3-NEXT:    [[TMP13:%.*]] = bitcast %struct.S* [[ARRAYIDX2]] to i8*
 // CHECK3-NEXT:    [[TMP14:%.*]] = bitcast %struct.S* [[VAR]] to i8*
-// CHECK3-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i32 4, i1 false)
-// CHECK3-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
-// CHECK3-NEXT:    [[TMP16:%.*]] = load i32, i32* [[SIVAR]], align 4
+// CHECK3-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i32 4, i1 false), !llvm.access.group !10
+// CHECK3-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !10
+// CHECK3-NEXT:    [[TMP16:%.*]] = load i32, i32* [[SIVAR]], align 4, !llvm.access.group !10
 // CHECK3-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP16]], [[TMP15]]
-// CHECK3-NEXT:    store i32 [[ADD3]], i32* [[SIVAR]], align 4
+// CHECK3-NEXT:    store i32 [[ADD3]], i32* [[SIVAR]], align 4, !llvm.access.group !10
 // CHECK3-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK3:       omp.body.continue:
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK3:       omp.inner.for.inc:
-// CHECK3-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK3-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !10
 // CHECK3-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP17]], 1
-// CHECK3-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4
-// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP7:![0-9]+]]
+// CHECK3-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !10
+// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP11:![0-9]+]]
 // CHECK3:       omp.inner.for.end:
 // CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK3:       omp.loop.exit:
 // CHECK3-NEXT:    [[TMP18:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP19:%.*]] = load i32, i32* [[TMP18]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP19]])
 // CHECK3-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK3-NEXT:    [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0

 // CHECK3-NEXT:    br label [[COND_END]]
 // CHECK3:       cond.end:
 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK3-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK3:       omp.inner.for.cond:
-// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !15
+// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !15
 // CHECK3-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK3-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK3:       omp.inner.for.cond.cleanup:
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK3:       omp.inner.for.body:
-// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
-// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
-// CHECK3-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32, i32)* @.omp_outlined..5 to void (i32*, i32*, ...)*), i32 [[TMP7]], i32 [[TMP8]])
+// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !15
+// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !15
+// CHECK3-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32, i32)* @.omp_outlined..5 to void (i32*, i32*, ...)*), i32 [[TMP7]], i32 [[TMP8]]), !llvm.access.group !15
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK3:       omp.inner.for.inc:
-// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !15
+// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !15
 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP9]], [[TMP10]]
-// CHECK3-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP10:![0-9]+]]
+// CHECK3-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !15
+// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP16:![0-9]+]]
 // CHECK3:       omp.inner.for.end:
 // CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK3:       omp.loop.exit:
 // CHECK3-NEXT:    [[TMP11:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP12]])
 // CHECK3-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK3-NEXT:    [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0

 // CHECK3-NEXT:    br label [[COND_END]]
 // CHECK3:       cond.end:
 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK3-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK3:       omp.inner.for.cond:
-// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !18
+// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !18
 // CHECK3-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK3-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK3:       omp.inner.for.cond.cleanup:
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK3:       omp.inner.for.body:
-// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !18
 // CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK3-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK3-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK3-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !18
+// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !18
+// CHECK3-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !18
 // CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i32 0, i32 [[TMP11]]
-// CHECK3-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4
-// CHECK3-NEXT:    [[TMP12:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP2]], align 4
-// CHECK3-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4
+// CHECK3-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !18
+// CHECK3-NEXT:    [[TMP12:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP2]], align 4, !llvm.access.group !18
+// CHECK3-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !18
 // CHECK3-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR]], i32 0, i32 [[TMP13]]
 // CHECK3-NEXT:    [[TMP14:%.*]] = bitcast %struct.S.0* [[ARRAYIDX4]] to i8*
 // CHECK3-NEXT:    [[TMP15:%.*]] = bitcast %struct.S.0* [[TMP12]] to i8*
-// CHECK3-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP14]], i8* align 4 [[TMP15]], i32 4, i1 false)
+// CHECK3-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP14]], i8* align 4 [[TMP15]], i32 4, i1 false), !llvm.access.group !18
 // CHECK3-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK3:       omp.body.continue:
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK3:       omp.inner.for.inc:
-// CHECK3-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK3-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !18
 // CHECK3-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP16]], 1
-// CHECK3-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4
-// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP11:![0-9]+]]
+// CHECK3-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !18
+// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP19:![0-9]+]]
 // CHECK3:       omp.inner.for.end:
 // CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK3:       omp.loop.exit:
 // CHECK3-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP18]])
 // CHECK3-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK3-NEXT:    [[TMP20:%.*]] = icmp ne i32 [[TMP19]], 0

 //
 // CHECK4-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK4-NEXT:  entry:
 // CHECK4-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK4-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK4-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK4:       arraydestroy.body:
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK4-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK4-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK4-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK4:       arraydestroy.done1:
 // CHECK4-NEXT:    ret void
 //
 //
 // CHECK4-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 // CHECK4-NEXT:    br label [[COND_END]]
 // CHECK4:       cond.end:
 // CHECK4-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK4-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK4-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK4:       omp.inner.for.cond:
-// CHECK4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
+// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !6
 // CHECK4-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK4-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK4:       omp.inner.for.cond.cleanup:
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK4:       omp.inner.for.body:
-// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
-// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
-// CHECK4-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32, i32)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i32 [[TMP7]], i32 [[TMP8]])
+// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !6
+// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !6
+// CHECK4-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32, i32)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i32 [[TMP7]], i32 [[TMP8]]), !llvm.access.group !6
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK4:       omp.inner.for.inc:
-// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
+// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !6
 // CHECK4-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP9]], [[TMP10]]
-// CHECK4-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
+// CHECK4-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
+// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP7:![0-9]+]]
 // CHECK4:       omp.inner.for.end:
 // CHECK4-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK4:       omp.loop.exit:
 // CHECK4-NEXT:    [[TMP11:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK4-NEXT:    [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 4
 // CHECK4-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP12]])
 // CHECK4-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK4-NEXT:    [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0

 // CHECK4-NEXT:    br label [[COND_END]]
 // CHECK4:       cond.end:
 // CHECK4-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK4-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK4-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK4:       omp.inner.for.cond:
-// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !10
+// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !10
 // CHECK4-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK4-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK4:       omp.inner.for.cond.cleanup:
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK4:       omp.inner.for.body:
-// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !10
 // CHECK4-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK4-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK4-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK4-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK4-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !10
+// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !10
+// CHECK4-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !10
 // CHECK4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i32 0, i32 [[TMP11]]
-// CHECK4-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4
-// CHECK4-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4
+// CHECK4-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !10
+// CHECK4-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !10
 // CHECK4-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds [2 x %struct.S], [2 x %struct.S]* [[S_ARR]], i32 0, i32 [[TMP12]]
 // CHECK4-NEXT:    [[TMP13:%.*]] = bitcast %struct.S* [[ARRAYIDX2]] to i8*
 // CHECK4-NEXT:    [[TMP14:%.*]] = bitcast %struct.S* [[VAR]] to i8*
-// CHECK4-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i32 4, i1 false)
-// CHECK4-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
-// CHECK4-NEXT:    [[TMP16:%.*]] = load i32, i32* [[SIVAR]], align 4
+// CHECK4-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i32 4, i1 false), !llvm.access.group !10
+// CHECK4-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !10
+// CHECK4-NEXT:    [[TMP16:%.*]] = load i32, i32* [[SIVAR]], align 4, !llvm.access.group !10
 // CHECK4-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP16]], [[TMP15]]
-// CHECK4-NEXT:    store i32 [[ADD3]], i32* [[SIVAR]], align 4
+// CHECK4-NEXT:    store i32 [[ADD3]], i32* [[SIVAR]], align 4, !llvm.access.group !10
 // CHECK4-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK4:       omp.body.continue:
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK4:       omp.inner.for.inc:
-// CHECK4-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK4-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !10
 // CHECK4-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP17]], 1
-// CHECK4-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4
-// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP7:![0-9]+]]
+// CHECK4-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !10
+// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP11:![0-9]+]]
 // CHECK4:       omp.inner.for.end:
 // CHECK4-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK4:       omp.loop.exit:
 // CHECK4-NEXT:    [[TMP18:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK4-NEXT:    [[TMP19:%.*]] = load i32, i32* [[TMP18]], align 4
 // CHECK4-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP19]])
 // CHECK4-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK4-NEXT:    [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0

 // CHECK4-NEXT:    br label [[COND_END]]
 // CHECK4:       cond.end:
 // CHECK4-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK4-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK4-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK4:       omp.inner.for.cond:
-// CHECK4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !15
+// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !15
 // CHECK4-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK4-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK4:       omp.inner.for.cond.cleanup:
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK4:       omp.inner.for.body:
-// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
-// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
-// CHECK4-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32, i32)* @.omp_outlined..5 to void (i32*, i32*, ...)*), i32 [[TMP7]], i32 [[TMP8]])
+// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !15
+// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !15
+// CHECK4-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32, i32)* @.omp_outlined..5 to void (i32*, i32*, ...)*), i32 [[TMP7]], i32 [[TMP8]]), !llvm.access.group !15
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK4:       omp.inner.for.inc:
-// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !15
+// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !15
 // CHECK4-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP9]], [[TMP10]]
-// CHECK4-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP10:![0-9]+]]
+// CHECK4-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !15
+// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP16:![0-9]+]]
 // CHECK4:       omp.inner.for.end:
 // CHECK4-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK4:       omp.loop.exit:
 // CHECK4-NEXT:    [[TMP11:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK4-NEXT:    [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 4
 // CHECK4-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP12]])
 // CHECK4-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK4-NEXT:    [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0

 // CHECK4-NEXT:    br label [[COND_END]]
 // CHECK4:       cond.end:
 // CHECK4-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK4-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK4-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK4:       omp.inner.for.cond:
-// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !18
+// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !18
 // CHECK4-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK4-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK4:       omp.inner.for.cond.cleanup:
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK4:       omp.inner.for.body:
-// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !18
 // CHECK4-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK4-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK4-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK4-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK4-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !18
+// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !18
+// CHECK4-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !18
 // CHECK4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i32 0, i32 [[TMP11]]
-// CHECK4-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4
-// CHECK4-NEXT:    [[TMP12:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP2]], align 4
-// CHECK4-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4
+// CHECK4-NEXT:    store i32 [[TMP10]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !18
+// CHECK4-NEXT:    [[TMP12:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP2]], align 4, !llvm.access.group !18
+// CHECK4-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !18
 // CHECK4-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR]], i32 0, i32 [[TMP13]]
 // CHECK4-NEXT:    [[TMP14:%.*]] = bitcast %struct.S.0* [[ARRAYIDX4]] to i8*
 // CHECK4-NEXT:    [[TMP15:%.*]] = bitcast %struct.S.0* [[TMP12]] to i8*
-// CHECK4-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP14]], i8* align 4 [[TMP15]], i32 4, i1 false)
+// CHECK4-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP14]], i8* align 4 [[TMP15]], i32 4, i1 false), !llvm.access.group !18
 // CHECK4-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK4:       omp.body.continue:
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK4:       omp.inner.for.inc:
-// CHECK4-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK4-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !18
 // CHECK4-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP16]], 1
-// CHECK4-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4
-// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP11:![0-9]+]]
+// CHECK4-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !18
+// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP19:![0-9]+]]
 // CHECK4:       omp.inner.for.end:
 // CHECK4-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK4:       omp.loop.exit:
 // CHECK4-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK4-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
 // CHECK4-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP18]])
 // CHECK4-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK4-NEXT:    [[TMP20:%.*]] = icmp ne i32 [[TMP19]], 0

 //
 // CHECK5-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK5-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK5-NEXT:  entry:
 // CHECK5-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK5-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK5-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK5:       arraydestroy.body:
-// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK5-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK5-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK5-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK5:       arraydestroy.done1:
 // CHECK5-NEXT:    ret void
 //
 //
 // CHECK5-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK6-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK6-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK6-NEXT:  entry:
 // CHECK6-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK6-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK6-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK6:       arraydestroy.body:
-// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK6-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK6-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK6-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK6:       arraydestroy.done1:
 // CHECK6-NEXT:    ret void
 //
 //
 // CHECK6-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK7-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK7-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK7-NEXT:  entry:
 // CHECK7-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK7-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK7-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK7:       arraydestroy.body:
-// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK7-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK7-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK7-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK7:       arraydestroy.done1:
 // CHECK7-NEXT:    ret void
 //
 //
 // CHECK7-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK8-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK8-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK8-NEXT:  entry:
 // CHECK8-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK8-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK8-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK8:       arraydestroy.body:
-// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK8-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK8-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK8-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK8:       arraydestroy.done1:
 // CHECK8-NEXT:    ret void
 //
 //
 // CHECK8-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK9-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK9-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK9-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK9-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK9:       arraydestroy.body:
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK9-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK9-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK9-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK9:       arraydestroy.done1:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
-// CHECK9-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK9-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK9-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !4
+// CHECK9-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !4
 // CHECK9-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK9-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
-// CHECK9-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
+// CHECK9-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !4
 // CHECK9-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
-// CHECK9-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK9-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !4
 // CHECK9-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
-// CHECK9-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]])
+// CHECK9-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]]), !llvm.access.group !4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK9:       omp.inner.for.inc:
-// CHECK9-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK9-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK9-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !4
+// CHECK9-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !4
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
-// CHECK9-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP3:![0-9]+]]
+// CHECK9-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !4
+// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK9:       omp.loop.exit:
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]])
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
 // CHECK9-NEXT:    br i1 [[TMP14]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
 // CHECK9:       .omp.final.then:

 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
-// CHECK9-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK9-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK9-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !8
+// CHECK9-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !8
 // CHECK9-NEXT:    [[CMP4:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK9-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
-// CHECK9-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK9-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !8
 // CHECK9-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK9-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK9-NEXT:    store i32 1, i32* [[G]], align 4
-// CHECK9-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[_TMP3]], align 8
-// CHECK9-NEXT:    store volatile i32 1, i32* [[TMP10]], align 4
-// CHECK9-NEXT:    store i32 2, i32* [[SIVAR]], align 4
+// CHECK9-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !8
+// CHECK9-NEXT:    store i32 1, i32* [[G]], align 4, !llvm.access.group !8
+// CHECK9-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[_TMP3]], align 8, !llvm.access.group !8
+// CHECK9-NEXT:    store volatile i32 1, i32* [[TMP10]], align 4, !llvm.access.group !8
+// CHECK9-NEXT:    store i32 2, i32* [[SIVAR]], align 4, !llvm.access.group !8
 // CHECK9-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 0
-// CHECK9-NEXT:    store i32* [[G]], i32** [[TMP11]], align 8
+// CHECK9-NEXT:    store i32* [[G]], i32** [[TMP11]], align 8, !llvm.access.group !8
 // CHECK9-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 1
-// CHECK9-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[_TMP3]], align 8
-// CHECK9-NEXT:    store i32* [[TMP13]], i32** [[TMP12]], align 8
+// CHECK9-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[_TMP3]], align 8, !llvm.access.group !8
+// CHECK9-NEXT:    store i32* [[TMP13]], i32** [[TMP12]], align 8, !llvm.access.group !8
 // CHECK9-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 2
-// CHECK9-NEXT:    store i32* [[SIVAR]], i32** [[TMP14]], align 8
-// CHECK9-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE_clEv"(%class.anon.0* nonnull align 8 dereferenceable(24) [[REF_TMP]])
+// CHECK9-NEXT:    store i32* [[SIVAR]], i32** [[TMP14]], align 8, !llvm.access.group !8
+// CHECK9-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE_clEv"(%class.anon.0* nonnull align 8 dereferenceable(24) [[REF_TMP]]), !llvm.access.group !8
 // CHECK9-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK9:       omp.body.continue:
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK9:       omp.inner.for.inc:
-// CHECK9-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK9-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !8
 // CHECK9-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP15]], 1
-// CHECK9-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4
-// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
+// CHECK9-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !8
+// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP9:![0-9]+]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK9:       omp.loop.exit:
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
 // CHECK9-NEXT:    br i1 [[TMP17]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
 // CHECK9:       .omp.final.then:

 //
 // CHECK10-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK10-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK10-NEXT:  entry:
 // CHECK10-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK10-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK10-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK10:       arraydestroy.body:
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK10-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK10-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK10-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK10:       arraydestroy.done1:
 // CHECK10-NEXT:    ret void
 //
 //
 // CHECK10-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 // CHECK10-NEXT:    br label [[COND_END]]
 // CHECK10:       cond.end:
 // CHECK10-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK10-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
 // CHECK10-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
 // CHECK10-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK10:       omp.inner.for.cond:
-// CHECK10-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK10-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK10-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !4
+// CHECK10-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !4
 // CHECK10-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK10-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK10:       omp.inner.for.body:
-// CHECK10-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
+// CHECK10-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4, !llvm.access.group !4
 // CHECK10-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
-// CHECK10-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
+// CHECK10-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4, !llvm.access.group !4
 // CHECK10-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
-// CHECK10-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]])
+// CHECK10-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i64)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i64 [[TMP8]], i64 [[TMP10]]), !llvm.access.group !4
 // CHECK10-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK10:       omp.inner.for.inc:
-// CHECK10-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK10-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
+// CHECK10-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !4
+// CHECK10-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !llvm.access.group !4
 // CHECK10-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
-// CHECK10-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
-// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP3:![0-9]+]]
+// CHECK10-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !4
+// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
 // CHECK10:       omp.inner.for.end:
 // CHECK10-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK10:       omp.loop.exit:
 // CHECK10-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]])
 // CHECK10-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK10-NEXT:    [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
 // CHECK10-NEXT:    br i1 [[TMP14]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
 // CHECK10:       .omp.final.then:

 // CHECK10-NEXT:    br label [[COND_END]]
 // CHECK10:       cond.end:
 // CHECK10-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]
 // CHECK10-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK10-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK10-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_IV]], align 4
 // CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK10:       omp.inner.for.cond:
-// CHECK10-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK10-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK10-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !8
+// CHECK10-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !8
 // CHECK10-NEXT:    [[CMP4:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
 // CHECK10-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK10:       omp.inner.for.body:
-// CHECK10-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK10-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !8
 // CHECK10-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
 // CHECK10-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK10-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK10-NEXT:    store i32 1, i32* [[G]], align 4
-// CHECK10-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[_TMP3]], align 8
-// CHECK10-NEXT:    store volatile i32 1, i32* [[TMP10]], align 4
-// CHECK10-NEXT:    store i32 2, i32* [[SIVAR]], align 4
+// CHECK10-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !8
+// CHECK10-NEXT:    store i32 1, i32* [[G]], align 4, !llvm.access.group !8
+// CHECK10-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[_TMP3]], align 8, !llvm.access.group !8
+// CHECK10-NEXT:    store volatile i32 1, i32* [[TMP10]], align 4, !llvm.access.group !8
+// CHECK10-NEXT:    store i32 2, i32* [[SIVAR]], align 4, !llvm.access.group !8
 // CHECK10-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 0
-// CHECK10-NEXT:    store i32* [[G]], i32** [[TMP11]], align 8
+// CHECK10-NEXT:    store i32* [[G]], i32** [[TMP11]], align 8, !llvm.access.group !8
 // CHECK10-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 1
-// CHECK10-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[_TMP3]], align 8
-// CHECK10-NEXT:    store i32* [[TMP13]], i32** [[TMP12]], align 8
+// CHECK10-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[_TMP3]], align 8, !llvm.access.group !8
+// CHECK10-NEXT:    store i32* [[TMP13]], i32** [[TMP12]], align 8, !llvm.access.group !8
 // CHECK10-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 2
-// CHECK10-NEXT:    store i32* [[SIVAR]], i32** [[TMP14]], align 8
-// CHECK10-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE_clEv"(%class.anon.0* nonnull align 8 dereferenceable(24) [[REF_TMP]])
+// CHECK10-NEXT:    store i32* [[SIVAR]], i32** [[TMP14]], align 8, !llvm.access.group !8
+// CHECK10-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE_clEv"(%class.anon.0* nonnull align 8 dereferenceable(24) [[REF_TMP]]), !llvm.access.group !8
 // CHECK10-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK10:       omp.body.continue:
 // CHECK10-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK10:       omp.inner.for.inc:
-// CHECK10-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK10-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !8
 // CHECK10-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP15]], 1
-// CHECK10-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4
-// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
+// CHECK10-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !8
+// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP9:![0-9]+]]
 // CHECK10:       omp.inner.for.end:
 // CHECK10-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK10:       omp.loop.exit:
 // CHECK10-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
 // CHECK10-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK10-NEXT:    [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
 // CHECK10-NEXT:    br i1 [[TMP17]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
 // CHECK10:       .omp.final.then:

 //
 // CHECK11-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK11-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK11-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK11-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK11:       arraydestroy.body:
-// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK11-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK11-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK11-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK11:       arraydestroy.done1:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK12-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK12-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK12-NEXT:  entry:
 // CHECK12-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK12-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK12-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK12:       arraydestroy.body:
-// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK12-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK12-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK12-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK12:       arraydestroy.done1:
 // CHECK12-NEXT:    ret void
 //
 //
 // CHECK12-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

clang/test/OpenMP/teams_distribute_private_codegen.cpp

 //
 // CHECK1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK1:       arraydestroy.body:
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK1-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK1:       arraydestroy.done1:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK2:       arraydestroy.body:
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK2-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK2:       arraydestroy.done1:
 // CHECK2-NEXT:    ret void
 //
 //
 // CHECK2-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK3-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK3-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK3:       arraydestroy.body:
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK3-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK3-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK3-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK3:       arraydestroy.done1:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK4-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK4-NEXT:  entry:
 // CHECK4-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK4-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK4-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK4:       arraydestroy.body:
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK4-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK4-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK4-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK4:       arraydestroy.done1:
 // CHECK4-NEXT:    ret void
 //
 //
 // CHECK4-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK9-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK9-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK9-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK9-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK9:       arraydestroy.body:
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK9-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK9-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK9-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK9:       arraydestroy.done1:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK10-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK10-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK10-NEXT:  entry:
 // CHECK10-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK10-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK10-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK10:       arraydestroy.body:
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK10-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK10-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK10-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK10:       arraydestroy.done1:
 // CHECK10-NEXT:    ret void
 //
 //
 // CHECK10-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 //
 // CHECK10-LABEL: define {{[^@]+}}@.omp_offloading.requires_reg
 // CHECK10-SAME: () #[[ATTR0]] {
 // CHECK10-NEXT:  entry:
 // CHECK10-NEXT:    call void @__tgt_register_requires(i64 1)
 // CHECK10-NEXT:    ret void
 //
-//
 

clang/test/OpenMP/teams_distribute_simd_firstprivate_codegen.cpp

 //
 // CHECK1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK1:       arraydestroy.body:
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK1-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK1:       arraydestroy.done1:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK2:       arraydestroy.body:
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK2-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK2:       arraydestroy.done1:
 // CHECK2-NEXT:    ret void
 //
 //
 // CHECK2-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK3-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK3-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK3:       arraydestroy.body:
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK3-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK3-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK3-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK3:       arraydestroy.done1:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK4-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK4-NEXT:  entry:
 // CHECK4-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK4-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK4-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK4:       arraydestroy.body:
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK4-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK4-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK4-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK4:       arraydestroy.done1:
 // CHECK4-NEXT:    ret void
 //
 //
 // CHECK4-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK5-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK5-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK5-NEXT:  entry:
 // CHECK5-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK5-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK5-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK5:       arraydestroy.body:
-// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK5-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK5-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK5-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK5:       arraydestroy.done1:
 // CHECK5-NEXT:    ret void
 //
 //
 // CHECK5-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK6-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK6-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK6-NEXT:  entry:
 // CHECK6-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK6-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK6-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK6:       arraydestroy.body:
-// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK6-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK6-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK6-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK6:       arraydestroy.done1:
 // CHECK6-NEXT:    ret void
 //
 //
 // CHECK6-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK7-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK7-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK7-NEXT:  entry:
 // CHECK7-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK7-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK7-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK7:       arraydestroy.body:
-// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK7-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK7-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK7-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK7:       arraydestroy.done1:
 // CHECK7-NEXT:    ret void
 //
 //
 // CHECK7-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK8-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK8-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK8-NEXT:  entry:
 // CHECK8-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK8-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK8-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK8:       arraydestroy.body:
-// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK8-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK8-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK8-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK8:       arraydestroy.done1:
 // CHECK8-NEXT:    ret void
 //
 //
 // CHECK8-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK9-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK9-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK9-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK9-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK9:       arraydestroy.body:
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK9-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK9-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK9-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK9:       arraydestroy.done1:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK10-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK10-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK10-NEXT:  entry:
 // CHECK10-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK10-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK10-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK10:       arraydestroy.body:
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK10-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK10-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK10-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK10:       arraydestroy.done1:
 // CHECK10-NEXT:    ret void
 //
 //
 // CHECK10-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 //
 // CHECK11-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK11-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK11-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK11-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK11:       arraydestroy.body:
-// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK11-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK11-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK11-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK11:       arraydestroy.done1:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK12-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK12-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK12-NEXT:  entry:
 // CHECK12-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK12-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK12-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK12:       arraydestroy.body:
-// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK12-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK12-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK12-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK12:       arraydestroy.done1:
 // CHECK12-NEXT:    ret void
 //
 //
 // CHECK12-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

clang/test/OpenMP/teams_distribute_simd_private_codegen.cpp

 //
 // CHECK1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK1:       arraydestroy.body:
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK1-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK1:       arraydestroy.done1:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 // CHECK1-NEXT:    br label [[COND_END]]
 // CHECK1:       cond.end:
 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK1-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK1:       omp.inner.for.cond:
-// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !5
+// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !5
 // CHECK1-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK1-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK1:       omp.inner.for.cond.cleanup:
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK1:       omp.inner.for.body:
-// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !5
 // CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK1-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4
+// CHECK1-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !5
+// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !5
+// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !5
 // CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP9]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i64 0, i64 [[IDXPROM]]
-// CHECK1-NEXT:    store i32 [[TMP8]], i32* [[ARRAYIDX]], align 4
-// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, i32* [[I]], align 4
+// CHECK1-NEXT:    store i32 [[TMP8]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !5
+// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !5
 // CHECK1-NEXT:    [[IDXPROM2:%.*]] = sext i32 [[TMP10]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX3:%.*]] = getelementptr inbounds [2 x %struct.S], [2 x %struct.S]* [[S_ARR]], i64 0, i64 [[IDXPROM2]]
 // CHECK1-NEXT:    [[TMP11:%.*]] = bitcast %struct.S* [[ARRAYIDX3]] to i8*
 // CHECK1-NEXT:    [[TMP12:%.*]] = bitcast %struct.S* [[VAR]] to i8*
-// CHECK1-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP11]], i8* align 4 [[TMP12]], i64 4, i1 false)
-// CHECK1-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4
-// CHECK1-NEXT:    [[TMP14:%.*]] = load i32, i32* [[SIVAR]], align 4
+// CHECK1-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP11]], i8* align 4 [[TMP12]], i64 4, i1 false), !llvm.access.group !5
+// CHECK1-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !5
+// CHECK1-NEXT:    [[TMP14:%.*]] = load i32, i32* [[SIVAR]], align 4, !llvm.access.group !5
 // CHECK1-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP14]], [[TMP13]]
-// CHECK1-NEXT:    store i32 [[ADD4]], i32* [[SIVAR]], align 4
+// CHECK1-NEXT:    store i32 [[ADD4]], i32* [[SIVAR]], align 4, !llvm.access.group !5
 // CHECK1-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK1:       omp.body.continue:
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK1:       omp.inner.for.inc:
-// CHECK1-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !5
 // CHECK1-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP15]], 1
-// CHECK1-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4
-// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP4:![0-9]+]]
+// CHECK1-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !5
+// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP6:![0-9]+]]
 // CHECK1:       omp.inner.for.end:
 // CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK1:       omp.loop.exit:
 // CHECK1-NEXT:    [[TMP16:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP17:%.*]] = load i32, i32* [[TMP16]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP17]])
 // CHECK1-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK1-NEXT:    [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0

 // CHECK1-NEXT:    br label [[COND_END]]
 // CHECK1:       cond.end:
 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK1-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK1:       omp.inner.for.cond:
-// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !11
+// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !11
 // CHECK1-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK1-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK1:       omp.inner.for.cond.cleanup:
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK1:       omp.inner.for.body:
-// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !11
 // CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK1-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4
+// CHECK1-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !11
+// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !11
+// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !11
 // CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP9]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i64 0, i64 [[IDXPROM]]
-// CHECK1-NEXT:    store i32 [[TMP8]], i32* [[ARRAYIDX]], align 4
-// CHECK1-NEXT:    [[TMP10:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP2]], align 8
-// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK1-NEXT:    store i32 [[TMP8]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !11
+// CHECK1-NEXT:    [[TMP10:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP2]], align 8, !llvm.access.group !11
+// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !11
 // CHECK1-NEXT:    [[IDXPROM4:%.*]] = sext i32 [[TMP11]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR]], i64 0, i64 [[IDXPROM4]]
 // CHECK1-NEXT:    [[TMP12:%.*]] = bitcast %struct.S.0* [[ARRAYIDX5]] to i8*
 // CHECK1-NEXT:    [[TMP13:%.*]] = bitcast %struct.S.0* [[TMP10]] to i8*
-// CHECK1-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP12]], i8* align 4 [[TMP13]], i64 4, i1 false)
+// CHECK1-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP12]], i8* align 4 [[TMP13]], i64 4, i1 false), !llvm.access.group !11
 // CHECK1-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK1:       omp.body.continue:
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK1:       omp.inner.for.inc:
-// CHECK1-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK1-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !11
 // CHECK1-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP14]], 1
-// CHECK1-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
-// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP8:![0-9]+]]
+// CHECK1-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !11
+// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP12:![0-9]+]]
 // CHECK1:       omp.inner.for.end:
 // CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK1:       omp.loop.exit:
 // CHECK1-NEXT:    [[TMP15:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP16:%.*]] = load i32, i32* [[TMP15]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP16]])
 // CHECK1-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK1-NEXT:    [[TMP18:%.*]] = icmp ne i32 [[TMP17]], 0

 //
 // CHECK2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK2:       arraydestroy.body:
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK2-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK2:       arraydestroy.done1:
 // CHECK2-NEXT:    ret void
 //
 //
 // CHECK2-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 // CHECK2-NEXT:    br label [[COND_END]]
 // CHECK2:       cond.end:
 // CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK2-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK2-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK2-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK2:       omp.inner.for.cond:
-// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !5
+// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !5
 // CHECK2-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK2-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK2:       omp.inner.for.cond.cleanup:
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK2:       omp.inner.for.body:
-// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !5
 // CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
 // CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK2-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4
+// CHECK2-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !5
+// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !5
+// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !5
 // CHECK2-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP9]] to i64
 // CHECK2-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i64 0, i64 [[IDXPROM]]
-// CHECK2-NEXT:    store i32 [[TMP8]], i32* [[ARRAYIDX]], align 4
-// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[I]], align 4
+// CHECK2-NEXT:    store i32 [[TMP8]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !5
+// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !5
 // CHECK2-NEXT:    [[IDXPROM2:%.*]] = sext i32 [[TMP10]] to i64
 // CHECK2-NEXT:    [[ARRAYIDX3:%.*]] = getelementptr inbounds [2 x %struct.S], [2 x %struct.S]* [[S_ARR]], i64 0, i64 [[IDXPROM2]]
 // CHECK2-NEXT:    [[TMP11:%.*]] = bitcast %struct.S* [[ARRAYIDX3]] to i8*
 // CHECK2-NEXT:    [[TMP12:%.*]] = bitcast %struct.S* [[VAR]] to i8*
-// CHECK2-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP11]], i8* align 4 [[TMP12]], i64 4, i1 false)
-// CHECK2-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4
-// CHECK2-NEXT:    [[TMP14:%.*]] = load i32, i32* [[SIVAR]], align 4
+// CHECK2-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP11]], i8* align 4 [[TMP12]], i64 4, i1 false), !llvm.access.group !5
+// CHECK2-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !5
+// CHECK2-NEXT:    [[TMP14:%.*]] = load i32, i32* [[SIVAR]], align 4, !llvm.access.group !5
 // CHECK2-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP14]], [[TMP13]]
-// CHECK2-NEXT:    store i32 [[ADD4]], i32* [[SIVAR]], align 4
+// CHECK2-NEXT:    store i32 [[ADD4]], i32* [[SIVAR]], align 4, !llvm.access.group !5
 // CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK2:       omp.body.continue:
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK2:       omp.inner.for.inc:
-// CHECK2-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !5
 // CHECK2-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP15]], 1
-// CHECK2-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4
-// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP4:![0-9]+]]
+// CHECK2-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !5
+// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP6:![0-9]+]]
 // CHECK2:       omp.inner.for.end:
 // CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK2:       omp.loop.exit:
 // CHECK2-NEXT:    [[TMP16:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK2-NEXT:    [[TMP17:%.*]] = load i32, i32* [[TMP16]], align 4
 // CHECK2-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP17]])
 // CHECK2-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK2-NEXT:    [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0

 // CHECK2-NEXT:    br label [[COND_END]]
 // CHECK2:       cond.end:
 // CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK2-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK2-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK2-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK2:       omp.inner.for.cond:
-// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !11
+// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !11
 // CHECK2-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK2-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK2:       omp.inner.for.cond.cleanup:
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK2:       omp.inner.for.body:
-// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !11
 // CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
 // CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK2-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4
+// CHECK2-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !11
+// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !11
+// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !11
 // CHECK2-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP9]] to i64
 // CHECK2-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i64 0, i64 [[IDXPROM]]
-// CHECK2-NEXT:    store i32 [[TMP8]], i32* [[ARRAYIDX]], align 4
-// CHECK2-NEXT:    [[TMP10:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP2]], align 8
-// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK2-NEXT:    store i32 [[TMP8]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !11
+// CHECK2-NEXT:    [[TMP10:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP2]], align 8, !llvm.access.group !11
+// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !11
 // CHECK2-NEXT:    [[IDXPROM4:%.*]] = sext i32 [[TMP11]] to i64
 // CHECK2-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR]], i64 0, i64 [[IDXPROM4]]
 // CHECK2-NEXT:    [[TMP12:%.*]] = bitcast %struct.S.0* [[ARRAYIDX5]] to i8*
 // CHECK2-NEXT:    [[TMP13:%.*]] = bitcast %struct.S.0* [[TMP10]] to i8*
-// CHECK2-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP12]], i8* align 4 [[TMP13]], i64 4, i1 false)
+// CHECK2-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP12]], i8* align 4 [[TMP13]], i64 4, i1 false), !llvm.access.group !11
 // CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK2:       omp.body.continue:
 // CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK2:       omp.inner.for.inc:
-// CHECK2-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK2-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !11
 // CHECK2-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP14]], 1
-// CHECK2-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
-// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP8:![0-9]+]]
+// CHECK2-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !11
+// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP12:![0-9]+]]
 // CHECK2:       omp.inner.for.end:
 // CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK2:       omp.loop.exit:
 // CHECK2-NEXT:    [[TMP15:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK2-NEXT:    [[TMP16:%.*]] = load i32, i32* [[TMP15]], align 4
 // CHECK2-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP16]])
 // CHECK2-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK2-NEXT:    [[TMP18:%.*]] = icmp ne i32 [[TMP17]], 0

 //
 // CHECK3-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK3-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK3:       arraydestroy.body:
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK3-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK3-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK3-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK3:       arraydestroy.done1:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 // CHECK3-NEXT:    br label [[COND_END]]
 // CHECK3:       cond.end:
 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK3-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK3:       omp.inner.for.cond:
-// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
+// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !6
 // CHECK3-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK3-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK3:       omp.inner.for.cond.cleanup:
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK3:       omp.inner.for.body:
-// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
 // CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK3-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4
+// CHECK3-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !6
+// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !6
+// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !6
 // CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i32 0, i32 [[TMP9]]
-// CHECK3-NEXT:    store i32 [[TMP8]], i32* [[ARRAYIDX]], align 4
-// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[I]], align 4
+// CHECK3-NEXT:    store i32 [[TMP8]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !6
+// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !6
 // CHECK3-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds [2 x %struct.S], [2 x %struct.S]* [[S_ARR]], i32 0, i32 [[TMP10]]
 // CHECK3-NEXT:    [[TMP11:%.*]] = bitcast %struct.S* [[ARRAYIDX2]] to i8*
 // CHECK3-NEXT:    [[TMP12:%.*]] = bitcast %struct.S* [[VAR]] to i8*
-// CHECK3-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP11]], i8* align 4 [[TMP12]], i32 4, i1 false)
-// CHECK3-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4
-// CHECK3-NEXT:    [[TMP14:%.*]] = load i32, i32* [[SIVAR]], align 4
+// CHECK3-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP11]], i8* align 4 [[TMP12]], i32 4, i1 false), !llvm.access.group !6
+// CHECK3-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !6
+// CHECK3-NEXT:    [[TMP14:%.*]] = load i32, i32* [[SIVAR]], align 4, !llvm.access.group !6
 // CHECK3-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP14]], [[TMP13]]
-// CHECK3-NEXT:    store i32 [[ADD3]], i32* [[SIVAR]], align 4
+// CHECK3-NEXT:    store i32 [[ADD3]], i32* [[SIVAR]], align 4, !llvm.access.group !6
 // CHECK3-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK3:       omp.body.continue:
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK3:       omp.inner.for.inc:
-// CHECK3-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK3-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
 // CHECK3-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP15]], 1
-// CHECK3-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4
-// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
+// CHECK3-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
+// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP7:![0-9]+]]
 // CHECK3:       omp.inner.for.end:
 // CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK3:       omp.loop.exit:
 // CHECK3-NEXT:    [[TMP16:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP17:%.*]] = load i32, i32* [[TMP16]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP17]])
 // CHECK3-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK3-NEXT:    [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0

 // CHECK3-NEXT:    br label [[COND_END]]
 // CHECK3:       cond.end:
 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK3-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK3:       omp.inner.for.cond:
-// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !12
+// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !12
 // CHECK3-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK3-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK3:       omp.inner.for.cond.cleanup:
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK3:       omp.inner.for.body:
-// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !12
 // CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK3-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4
+// CHECK3-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !12
+// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !12
+// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !12
 // CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i32 0, i32 [[TMP9]]
-// CHECK3-NEXT:    store i32 [[TMP8]], i32* [[ARRAYIDX]], align 4
-// CHECK3-NEXT:    [[TMP10:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP2]], align 4
-// CHECK3-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK3-NEXT:    store i32 [[TMP8]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !12
+// CHECK3-NEXT:    [[TMP10:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP2]], align 4, !llvm.access.group !12
+// CHECK3-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !12
 // CHECK3-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR]], i32 0, i32 [[TMP11]]
 // CHECK3-NEXT:    [[TMP12:%.*]] = bitcast %struct.S.0* [[ARRAYIDX4]] to i8*
 // CHECK3-NEXT:    [[TMP13:%.*]] = bitcast %struct.S.0* [[TMP10]] to i8*
-// CHECK3-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP12]], i8* align 4 [[TMP13]], i32 4, i1 false)
+// CHECK3-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP12]], i8* align 4 [[TMP13]], i32 4, i1 false), !llvm.access.group !12
 // CHECK3-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK3:       omp.body.continue:
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK3:       omp.inner.for.inc:
-// CHECK3-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK3-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !12
 // CHECK3-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP14]], 1
-// CHECK3-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4
-// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP9:![0-9]+]]
+// CHECK3-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !12
+// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP13:![0-9]+]]
 // CHECK3:       omp.inner.for.end:
 // CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK3:       omp.loop.exit:
 // CHECK3-NEXT:    [[TMP15:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP16:%.*]] = load i32, i32* [[TMP15]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP16]])
 // CHECK3-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK3-NEXT:    [[TMP18:%.*]] = icmp ne i32 [[TMP17]], 0

 //
 // CHECK4-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK4-NEXT:  entry:
 // CHECK4-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK4-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK4-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK4:       arraydestroy.body:
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK4-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK4-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK4-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK4:       arraydestroy.done1:
 // CHECK4-NEXT:    ret void
 //
 //
 // CHECK4-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 // CHECK4-NEXT:    br label [[COND_END]]
 // CHECK4:       cond.end:
 // CHECK4-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK4-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK4-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK4:       omp.inner.for.cond:
-// CHECK4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
+// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !6
 // CHECK4-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK4-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK4:       omp.inner.for.cond.cleanup:
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK4:       omp.inner.for.body:
-// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
 // CHECK4-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
 // CHECK4-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK4-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4
+// CHECK4-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !6
+// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !6
+// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !6
 // CHECK4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i32 0, i32 [[TMP9]]
-// CHECK4-NEXT:    store i32 [[TMP8]], i32* [[ARRAYIDX]], align 4
-// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[I]], align 4
+// CHECK4-NEXT:    store i32 [[TMP8]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !6
+// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !6
 // CHECK4-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds [2 x %struct.S], [2 x %struct.S]* [[S_ARR]], i32 0, i32 [[TMP10]]
 // CHECK4-NEXT:    [[TMP11:%.*]] = bitcast %struct.S* [[ARRAYIDX2]] to i8*
 // CHECK4-NEXT:    [[TMP12:%.*]] = bitcast %struct.S* [[VAR]] to i8*
-// CHECK4-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP11]], i8* align 4 [[TMP12]], i32 4, i1 false)
-// CHECK4-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4
-// CHECK4-NEXT:    [[TMP14:%.*]] = load i32, i32* [[SIVAR]], align 4
+// CHECK4-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP11]], i8* align 4 [[TMP12]], i32 4, i1 false), !llvm.access.group !6
+// CHECK4-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !6
+// CHECK4-NEXT:    [[TMP14:%.*]] = load i32, i32* [[SIVAR]], align 4, !llvm.access.group !6
 // CHECK4-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP14]], [[TMP13]]
-// CHECK4-NEXT:    store i32 [[ADD3]], i32* [[SIVAR]], align 4
+// CHECK4-NEXT:    store i32 [[ADD3]], i32* [[SIVAR]], align 4, !llvm.access.group !6
 // CHECK4-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK4:       omp.body.continue:
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK4:       omp.inner.for.inc:
-// CHECK4-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK4-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
 // CHECK4-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP15]], 1
-// CHECK4-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4
-// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
+// CHECK4-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !6
+// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP7:![0-9]+]]
 // CHECK4:       omp.inner.for.end:
 // CHECK4-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK4:       omp.loop.exit:
 // CHECK4-NEXT:    [[TMP16:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK4-NEXT:    [[TMP17:%.*]] = load i32, i32* [[TMP16]], align 4
 // CHECK4-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP17]])
 // CHECK4-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK4-NEXT:    [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0

 // CHECK4-NEXT:    br label [[COND_END]]
 // CHECK4:       cond.end:
 // CHECK4-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK4-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK4-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK4:       omp.inner.for.cond:
-// CHECK4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !12
+// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !12
 // CHECK4-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK4-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_COND_CLEANUP:%.*]]
 // CHECK4:       omp.inner.for.cond.cleanup:
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_END:%.*]]
 // CHECK4:       omp.inner.for.body:
-// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !12
 // CHECK4-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
 // CHECK4-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK4-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[T_VAR]], align 4
-// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4
+// CHECK4-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !12
+// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[T_VAR]], align 4, !llvm.access.group !12
+// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !12
 // CHECK4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i32 0, i32 [[TMP9]]
-// CHECK4-NEXT:    store i32 [[TMP8]], i32* [[ARRAYIDX]], align 4
-// CHECK4-NEXT:    [[TMP10:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP2]], align 4
-// CHECK4-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
+// CHECK4-NEXT:    store i32 [[TMP8]], i32* [[ARRAYIDX]], align 4, !llvm.access.group !12
+// CHECK4-NEXT:    [[TMP10:%.*]] = load %struct.S.0*, %struct.S.0** [[_TMP2]], align 4, !llvm.access.group !12
+// CHECK4-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4, !llvm.access.group !12
 // CHECK4-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR]], i32 0, i32 [[TMP11]]
 // CHECK4-NEXT:    [[TMP12:%.*]] = bitcast %struct.S.0* [[ARRAYIDX4]] to i8*
 // CHECK4-NEXT:    [[TMP13:%.*]] = bitcast %struct.S.0* [[TMP10]] to i8*
-// CHECK4-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP12]], i8* align 4 [[TMP13]], i32 4, i1 false)
+// CHECK4-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 [[TMP12]], i8* align 4 [[TMP13]], i32 4, i1 false), !llvm.access.group !12
 // CHECK4-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK4:       omp.body.continue:
 // CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK4:       omp.inner.for.inc:
-// CHECK4-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK4-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !12
 // CHECK4-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP14]], 1
-// CHECK4-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4
-// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP9:![0-9]+]]
+// CHECK4-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !12
+// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP13:![0-9]+]]
 // CHECK4:       omp.inner.for.end:
 // CHECK4-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK4:       omp.loop.exit:
 // CHECK4-NEXT:    [[TMP15:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK4-NEXT:    [[TMP16:%.*]] = load i32, i32* [[TMP15]], align 4
 // CHECK4-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP16]])
 // CHECK4-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK4-NEXT:    [[TMP18:%.*]] = icmp ne i32 [[TMP17]], 0

 //
 // CHECK5-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK5-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK5-NEXT:  entry:
 // CHECK5-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK5-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK5-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK5:       arraydestroy.body:
-// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK5-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK5-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK5-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK5-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK5:       arraydestroy.done1:
 // CHECK5-NEXT:    ret void
 //
 //
 // CHECK5-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK6-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK6-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK6-NEXT:  entry:
 // CHECK6-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK6-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK6-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK6:       arraydestroy.body:
-// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK6-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK6-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK6-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK6-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK6:       arraydestroy.done1:
 // CHECK6-NEXT:    ret void
 //
 //
 // CHECK6-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK7-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK7-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK7-NEXT:  entry:
 // CHECK7-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK7-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK7-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK7:       arraydestroy.body:
-// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK7-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK7-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK7-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK7-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK7:       arraydestroy.done1:
 // CHECK7-NEXT:    ret void
 //
 //
 // CHECK7-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK8-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK8-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK8-NEXT:  entry:
 // CHECK8-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 4
 // CHECK8-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 4
 // CHECK8-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK8:       arraydestroy.body:
-// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i32 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
+// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK8-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i32 -1
 // CHECK8-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK8-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK8-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK8:       arraydestroy.done1:
 // CHECK8-NEXT:    ret void
 //
 //
 // CHECK8-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK9-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK9-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK9-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK9-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK9:       arraydestroy.body:
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK9-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK9-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK9-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK9-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK9:       arraydestroy.done1:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
-// CHECK9-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK9-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK9-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !4
+// CHECK9-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !4
 // CHECK9-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK9-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
-// CHECK9-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK9-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !4
 // CHECK9-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK9-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK9-NEXT:    store i32 1, i32* [[G]], align 4
-// CHECK9-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[_TMP2]], align 8
-// CHECK9-NEXT:    store volatile i32 1, i32* [[TMP8]], align 4
-// CHECK9-NEXT:    store i32 2, i32* [[SIVAR]], align 4
+// CHECK9-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !4
+// CHECK9-NEXT:    store i32 1, i32* [[G]], align 4, !llvm.access.group !4
+// CHECK9-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[_TMP2]], align 8, !llvm.access.group !4
+// CHECK9-NEXT:    store volatile i32 1, i32* [[TMP8]], align 4, !llvm.access.group !4
+// CHECK9-NEXT:    store i32 2, i32* [[SIVAR]], align 4, !llvm.access.group !4
 // CHECK9-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 0
-// CHECK9-NEXT:    store i32* [[G]], i32** [[TMP9]], align 8
+// CHECK9-NEXT:    store i32* [[G]], i32** [[TMP9]], align 8, !llvm.access.group !4
 // CHECK9-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 1
-// CHECK9-NEXT:    [[TMP11:%.*]] = load i32*, i32** [[_TMP2]], align 8
-// CHECK9-NEXT:    store i32* [[TMP11]], i32** [[TMP10]], align 8
+// CHECK9-NEXT:    [[TMP11:%.*]] = load i32*, i32** [[_TMP2]], align 8, !llvm.access.group !4
+// CHECK9-NEXT:    store i32* [[TMP11]], i32** [[TMP10]], align 8, !llvm.access.group !4
 // CHECK9-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 2
-// CHECK9-NEXT:    store i32* [[SIVAR]], i32** [[TMP12]], align 8
-// CHECK9-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE_clEv"(%class.anon.0* nonnull align 8 dereferenceable(24) [[REF_TMP]])
+// CHECK9-NEXT:    store i32* [[SIVAR]], i32** [[TMP12]], align 8, !llvm.access.group !4
+// CHECK9-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE_clEv"(%class.anon.0* nonnull align 8 dereferenceable(24) [[REF_TMP]]), !llvm.access.group !4
 // CHECK9-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK9:       omp.body.continue:
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK9:       omp.inner.for.inc:
-// CHECK9-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK9-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !4
 // CHECK9-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP13]], 1
-// CHECK9-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4
-// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP3:![0-9]+]]
+// CHECK9-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !4
+// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK9:       omp.loop.exit:
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]])
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
 // CHECK9-NEXT:    br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
 // CHECK9:       .omp.final.then:

 //
 // CHECK10-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK10-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK10-NEXT:  entry:
 // CHECK10-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK10-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK10-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK10:       arraydestroy.body:
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK10-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK10-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK10-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK10-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK10:       arraydestroy.done1:
 // CHECK10-NEXT:    ret void
 //
 //
 // CHECK10-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef

 // CHECK10-NEXT:    br label [[COND_END]]
 // CHECK10:       cond.end:
 // CHECK10-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
 // CHECK10-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
 // CHECK10-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
 // CHECK10-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_IV]], align 4
 // CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK10:       omp.inner.for.cond:
-// CHECK10-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
-// CHECK10-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
+// CHECK10-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !4
+// CHECK10-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !4
 // CHECK10-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
 // CHECK10-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK10:       omp.inner.for.body:
-// CHECK10-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK10-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !4
 // CHECK10-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
 // CHECK10-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
-// CHECK10-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
-// CHECK10-NEXT:    store i32 1, i32* [[G]], align 4
-// CHECK10-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[_TMP2]], align 8
-// CHECK10-NEXT:    store volatile i32 1, i32* [[TMP8]], align 4
-// CHECK10-NEXT:    store i32 2, i32* [[SIVAR]], align 4
+// CHECK10-NEXT:    store i32 [[ADD]], i32* [[I]], align 4, !llvm.access.group !4
+// CHECK10-NEXT:    store i32 1, i32* [[G]], align 4, !llvm.access.group !4
+// CHECK10-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[_TMP2]], align 8, !llvm.access.group !4
+// CHECK10-NEXT:    store volatile i32 1, i32* [[TMP8]], align 4, !llvm.access.group !4
+// CHECK10-NEXT:    store i32 2, i32* [[SIVAR]], align 4, !llvm.access.group !4
 // CHECK10-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 0
-// CHECK10-NEXT:    store i32* [[G]], i32** [[TMP9]], align 8
+// CHECK10-NEXT:    store i32* [[G]], i32** [[TMP9]], align 8, !llvm.access.group !4
 // CHECK10-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 1
-// CHECK10-NEXT:    [[TMP11:%.*]] = load i32*, i32** [[_TMP2]], align 8
-// CHECK10-NEXT:    store i32* [[TMP11]], i32** [[TMP10]], align 8
+// CHECK10-NEXT:    [[TMP11:%.*]] = load i32*, i32** [[_TMP2]], align 8, !llvm.access.group !4
+// CHECK10-NEXT:    store i32* [[TMP11]], i32** [[TMP10]], align 8, !llvm.access.group !4
 // CHECK10-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 2
-// CHECK10-NEXT:    store i32* [[SIVAR]], i32** [[TMP12]], align 8
-// CHECK10-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE_clEv"(%class.anon.0* nonnull align 8 dereferenceable(24) [[REF_TMP]])
+// CHECK10-NEXT:    store i32* [[SIVAR]], i32** [[TMP12]], align 8, !llvm.access.group !4
+// CHECK10-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE_clEv"(%class.anon.0* nonnull align 8 dereferenceable(24) [[REF_TMP]]), !llvm.access.group !4
 // CHECK10-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 // CHECK10:       omp.body.continue:
 // CHECK10-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK10:       omp.inner.for.inc:
-// CHECK10-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
+// CHECK10-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !4
 // CHECK10-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP13]], 1
-// CHECK10-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4
-// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP3:![0-9]+]]
+// CHECK10-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !4
+// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
 // CHECK10:       omp.inner.for.end:
 // CHECK10-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK10:       omp.loop.exit:
 // CHECK10-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]])
 // CHECK10-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
 // CHECK10-NEXT:    [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
 // CHECK10-NEXT:    br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
 // CHECK10:       .omp.final.then:

 //
 // CHECK11-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK11-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK11-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK11-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK11:       arraydestroy.body:
-// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK11-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK11-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK11-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK11-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK11:       arraydestroy.done1:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

 //
 // CHECK12-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK12-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK12-NEXT:  entry:
 // CHECK12-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK12-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK12-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK12:       arraydestroy.body:
-// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([[STRUCT_S:%.*]], %struct.S* getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK12-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK12-NEXT:    call void @_ZN1SIfED1Ev(%struct.S* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
 // CHECK12-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
 // CHECK12-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK12:       arraydestroy.done1:
 // CHECK12-NEXT:    ret void
 //
 //
 // CHECK12-LABEL: define {{[^@]+}}@__cxx_global_var_init.2

clang/test/OpenMP/threadprivate_codegen.cpp

 //
 // CHECK1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK1:       arraydestroy.body:
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ getelementptr inbounds ([[STRUCT_S1:%.*]], %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0), i64 6), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 1, i64 0, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK1-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
 // CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0)
 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK1:       arraydestroy.done1:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@main
 // CHECK1-SAME: () #[[ATTR4:[0-9]+]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[RES:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[EXN_SLOT:%.*]] = alloca i8*, align 8
 // CHECK1-NEXT:    [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK1-NEXT:    store i32 0, i32* [[RETVAL]], align 4
 // CHECK1-NEXT:    [[TMP1:%.*]] = load atomic i8, i8* bitcast (i64* @_ZGVZ4mainE2sm to i8*) acquire, align 8
 // CHECK1-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP1]], 0
-// CHECK1-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF2:![0-9]+]]
+// CHECK1-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF3:![0-9]+]]
 // CHECK1:       init.check:
 // CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__cxa_guard_acquire(i64* @_ZGVZ4mainE2sm) #[[ATTR3]]
 // CHECK1-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP2]], 0
 // CHECK1-NEXT:    br i1 [[TOBOOL]], label [[INIT:%.*]], label [[INIT_END]]
 // CHECK1:       init:
 // CHECK1-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK1-NEXT:    call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB1]], i8* bitcast (%struct.Smain* @_ZZ4mainE2sm to i8*), i8* (i8*)* @.__kmpc_global_ctor_..6, i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_..7)
 // CHECK1-NEXT:    [[TMP4:%.*]] = call i8* @__kmpc_threadprivate_cached(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i8* bitcast (%struct.S1* @_ZL3gs1 to i8*), i64 4, i8*** @_ZL3gs1.cache.)

 //
 // CHECK2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK2:       arraydestroy.body:
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ getelementptr inbounds ([[STRUCT_S1:%.*]], %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0), i64 6), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 1, i64 0, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK2-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
 // CHECK2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0)
 // CHECK2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK2:       arraydestroy.done1:
 // CHECK2-NEXT:    ret void
 //
 //
 // CHECK2-LABEL: define {{[^@]+}}@.__kmpc_global_ctor_..3

 // CHECK2-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    [[RES:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    [[EXN_SLOT:%.*]] = alloca i8*, align 8
 // CHECK2-NEXT:    [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK2-NEXT:    store i32 0, i32* [[RETVAL]], align 4
 // CHECK2-NEXT:    [[TMP1:%.*]] = load atomic i8, i8* bitcast (i64* @_ZGVZ4mainE2sm to i8*) acquire, align 8
 // CHECK2-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP1]], 0
-// CHECK2-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF2:![0-9]+]]
+// CHECK2-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF3:![0-9]+]]
 // CHECK2:       init.check:
 // CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__cxa_guard_acquire(i64* @_ZGVZ4mainE2sm) #[[ATTR3]]
 // CHECK2-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP2]], 0
 // CHECK2-NEXT:    br i1 [[TOBOOL]], label [[INIT:%.*]], label [[INIT_END]]
 // CHECK2:       init:
 // CHECK2-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK2-NEXT:    call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB1]], i8* bitcast (%struct.Smain* @_ZZ4mainE2sm to i8*), i8* (i8*)* @.__kmpc_global_ctor_..6, i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_..7)
 // CHECK2-NEXT:    [[TMP4:%.*]] = call i8* @__kmpc_threadprivate_cached(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i8* bitcast (%struct.S1* @_ZL3gs1 to i8*), i64 4, i8*** @_ZL3gs1.cache.)

 //
 // SIMD1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // SIMD1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // SIMD1-NEXT:  entry:
 // SIMD1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // SIMD1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // SIMD1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // SIMD1:       arraydestroy.body:
-// SIMD1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ getelementptr inbounds ([[STRUCT_S1:%.*]], %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0), i64 6), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// SIMD1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// SIMD1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 1, i64 0, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// SIMD1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // SIMD1-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
 // SIMD1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0)
 // SIMD1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // SIMD1:       arraydestroy.done1:
 // SIMD1-NEXT:    ret void
 //
 //
 // SIMD1-LABEL: define {{[^@]+}}@main

 // SIMD1-NEXT:    call void @__cxx_global_var_init.1()
 // SIMD1-NEXT:    call void @__cxx_global_var_init.2()
 // SIMD1-NEXT:    ret void
 //
 //
 // SIMD2-LABEL: define {{[^@]+}}@__cxx_global_var_init
 // SIMD2-SAME: () #[[ATTR0:[0-9]+]] !dbg [[DBG115:![0-9]+]] {
 // SIMD2-NEXT:  entry:
-// SIMD2-NEXT:    call void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) @_ZL3gs1, i32 5), !dbg [[DBG119:![0-9]+]]
-// SIMD2-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.S1*)* @_ZN2S1D1Ev to void (i8*)*), i8* bitcast (%struct.S1* @_ZL3gs1 to i8*), i8* @__dso_handle) #[[ATTR3:[0-9]+]], !dbg [[DBG121:![0-9]+]]
-// SIMD2-NEXT:    ret void, !dbg [[DBG122:![0-9]+]]
+// SIMD2-NEXT:    call void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) @_ZL3gs1, i32 5), !dbg [[DBG118:![0-9]+]]
+// SIMD2-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.S1*)* @_ZN2S1D1Ev to void (i8*)*), i8* bitcast (%struct.S1* @_ZL3gs1 to i8*), i8* @__dso_handle) #[[ATTR3:[0-9]+]], !dbg [[DBG120:![0-9]+]]
+// SIMD2-NEXT:    ret void, !dbg [[DBG121:![0-9]+]]
 //
 //
 // SIMD2-LABEL: define {{[^@]+}}@_ZN2S1C1Ei
-// SIMD2-SAME: (%struct.S1* nonnull align 4 dereferenceable(4) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR1:[0-9]+]] comdat align 2 !dbg [[DBG123:![0-9]+]] {
+// SIMD2-SAME: (%struct.S1* nonnull align 4 dereferenceable(4) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR1:[0-9]+]] comdat align 2 !dbg [[DBG122:![0-9]+]] {
 // SIMD2-NEXT:  entry:
 // SIMD2-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S1*, align 8
 // SIMD2-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
 // SIMD2-NEXT:    store %struct.S1* [[THIS]], %struct.S1** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.S1** [[THIS_ADDR]], metadata [[META124:![0-9]+]], metadata !DIExpression()), !dbg [[DBG126:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.S1** [[THIS_ADDR]], metadata [[META123:![0-9]+]], metadata !DIExpression()), !dbg [[DBG125:![0-9]+]]
 // SIMD2-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META127:![0-9]+]], metadata !DIExpression()), !dbg [[DBG128:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META126:![0-9]+]], metadata !DIExpression()), !dbg [[DBG127:![0-9]+]]
 // SIMD2-NEXT:    [[THIS1:%.*]] = load %struct.S1*, %struct.S1** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG129:![0-9]+]]
-// SIMD2-NEXT:    call void @_ZN2S1C2Ei(%struct.S1* nonnull align 4 dereferenceable(4) [[THIS1]], i32 [[TMP0]]), !dbg [[DBG129]]
-// SIMD2-NEXT:    ret void, !dbg [[DBG130:![0-9]+]]
+// SIMD2-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG128:![0-9]+]]
+// SIMD2-NEXT:    call void @_ZN2S1C2Ei(%struct.S1* nonnull align 4 dereferenceable(4) [[THIS1]], i32 [[TMP0]]), !dbg [[DBG128]]
+// SIMD2-NEXT:    ret void, !dbg [[DBG129:![0-9]+]]
 //
 //
 // SIMD2-LABEL: define {{[^@]+}}@_ZN2S1D1Ev
-// SIMD2-SAME: (%struct.S1* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR2:[0-9]+]] comdat align 2 !dbg [[DBG131:![0-9]+]] {
+// SIMD2-SAME: (%struct.S1* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR2:[0-9]+]] comdat align 2 !dbg [[DBG130:![0-9]+]] {
 // SIMD2-NEXT:  entry:
 // SIMD2-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S1*, align 8
 // SIMD2-NEXT:    store %struct.S1* [[THIS]], %struct.S1** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.S1** [[THIS_ADDR]], metadata [[META132:![0-9]+]], metadata !DIExpression()), !dbg [[DBG133:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.S1** [[THIS_ADDR]], metadata [[META131:![0-9]+]], metadata !DIExpression()), !dbg [[DBG132:![0-9]+]]
 // SIMD2-NEXT:    [[THIS1:%.*]] = load %struct.S1*, %struct.S1** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    call void @_ZN2S1D2Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[THIS1]]) #[[ATTR3]], !dbg [[DBG134:![0-9]+]]
-// SIMD2-NEXT:    ret void, !dbg [[DBG135:![0-9]+]]
+// SIMD2-NEXT:    call void @_ZN2S1D2Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[THIS1]]) #[[ATTR3]], !dbg [[DBG133:![0-9]+]]
+// SIMD2-NEXT:    ret void, !dbg [[DBG134:![0-9]+]]
 //
 //
 // SIMD2-LABEL: define {{[^@]+}}@__cxx_global_var_init.1
-// SIMD2-SAME: () #[[ATTR0]] !dbg [[DBG136:![0-9]+]] {
+// SIMD2-SAME: () #[[ATTR0]] !dbg [[DBG135:![0-9]+]] {
 // SIMD2-NEXT:  entry:
-// SIMD2-NEXT:    call void @_ZN2S2C1Ei(%struct.S2* nonnull align 8 dereferenceable(16) @_ZL3gs2, i32 27), !dbg [[DBG137:![0-9]+]]
-// SIMD2-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.S2*)* @_ZN2S2D1Ev to void (i8*)*), i8* bitcast (%struct.S2* @_ZL3gs2 to i8*), i8* @__dso_handle) #[[ATTR3]], !dbg [[DBG139:![0-9]+]]
-// SIMD2-NEXT:    ret void, !dbg [[DBG140:![0-9]+]]
+// SIMD2-NEXT:    call void @_ZN2S2C1Ei(%struct.S2* nonnull align 8 dereferenceable(16) @_ZL3gs2, i32 27), !dbg [[DBG136:![0-9]+]]
+// SIMD2-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.S2*)* @_ZN2S2D1Ev to void (i8*)*), i8* bitcast (%struct.S2* @_ZL3gs2 to i8*), i8* @__dso_handle) #[[ATTR3]], !dbg [[DBG138:![0-9]+]]
+// SIMD2-NEXT:    ret void, !dbg [[DBG139:![0-9]+]]
 //
 //
 // SIMD2-LABEL: define {{[^@]+}}@_ZN2S2C1Ei
-// SIMD2-SAME: (%struct.S2* nonnull align 8 dereferenceable(16) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 !dbg [[DBG141:![0-9]+]] {
+// SIMD2-SAME: (%struct.S2* nonnull align 8 dereferenceable(16) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 !dbg [[DBG140:![0-9]+]] {
 // SIMD2-NEXT:  entry:
 // SIMD2-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S2*, align 8
 // SIMD2-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
 // SIMD2-NEXT:    store %struct.S2* [[THIS]], %struct.S2** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.S2** [[THIS_ADDR]], metadata [[META142:![0-9]+]], metadata !DIExpression()), !dbg [[DBG144:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.S2** [[THIS_ADDR]], metadata [[META141:![0-9]+]], metadata !DIExpression()), !dbg [[DBG143:![0-9]+]]
 // SIMD2-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META145:![0-9]+]], metadata !DIExpression()), !dbg [[DBG146:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META144:![0-9]+]], metadata !DIExpression()), !dbg [[DBG145:![0-9]+]]
 // SIMD2-NEXT:    [[THIS1:%.*]] = load %struct.S2*, %struct.S2** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG147:![0-9]+]]
-// SIMD2-NEXT:    call void @_ZN2S2C2Ei(%struct.S2* nonnull align 8 dereferenceable(16) [[THIS1]], i32 [[TMP0]]), !dbg [[DBG147]]
-// SIMD2-NEXT:    ret void, !dbg [[DBG148:![0-9]+]]
+// SIMD2-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG146:![0-9]+]]
+// SIMD2-NEXT:    call void @_ZN2S2C2Ei(%struct.S2* nonnull align 8 dereferenceable(16) [[THIS1]], i32 [[TMP0]]), !dbg [[DBG146]]
+// SIMD2-NEXT:    ret void, !dbg [[DBG147:![0-9]+]]
 //
 //
 // SIMD2-LABEL: define {{[^@]+}}@_ZN2S2D1Ev
-// SIMD2-SAME: (%struct.S2* nonnull align 8 dereferenceable(16) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG149:![0-9]+]] {
+// SIMD2-SAME: (%struct.S2* nonnull align 8 dereferenceable(16) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG148:![0-9]+]] {
 // SIMD2-NEXT:  entry:
 // SIMD2-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S2*, align 8
 // SIMD2-NEXT:    store %struct.S2* [[THIS]], %struct.S2** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.S2** [[THIS_ADDR]], metadata [[META150:![0-9]+]], metadata !DIExpression()), !dbg [[DBG151:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.S2** [[THIS_ADDR]], metadata [[META149:![0-9]+]], metadata !DIExpression()), !dbg [[DBG150:![0-9]+]]
 // SIMD2-NEXT:    [[THIS1:%.*]] = load %struct.S2*, %struct.S2** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    call void @_ZN2S2D2Ev(%struct.S2* nonnull align 8 dereferenceable(16) [[THIS1]]) #[[ATTR3]], !dbg [[DBG152:![0-9]+]]
-// SIMD2-NEXT:    ret void, !dbg [[DBG153:![0-9]+]]
+// SIMD2-NEXT:    call void @_ZN2S2D2Ev(%struct.S2* nonnull align 8 dereferenceable(16) [[THIS1]]) #[[ATTR3]], !dbg [[DBG151:![0-9]+]]
+// SIMD2-NEXT:    ret void, !dbg [[DBG152:![0-9]+]]
 //
 //
 // SIMD2-LABEL: define {{[^@]+}}@__cxx_global_var_init.2
-// SIMD2-SAME: () #[[ATTR0]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) !dbg [[DBG154:![0-9]+]] {
+// SIMD2-SAME: () #[[ATTR0]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) !dbg [[DBG153:![0-9]+]] {
 // SIMD2-NEXT:  entry:
 // SIMD2-NEXT:    [[ARRAYINIT_ENDOFINIT:%.*]] = alloca [3 x %struct.S1]*, align 8
 // SIMD2-NEXT:    [[ARRAYINIT_ENDOFINIT1:%.*]] = alloca %struct.S1*, align 8
 // SIMD2-NEXT:    [[EXN_SLOT:%.*]] = alloca i8*, align 8
 // SIMD2-NEXT:    [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
 // SIMD2-NEXT:    [[ARRAYINIT_ENDOFINIT5:%.*]] = alloca %struct.S1*, align 8
-// SIMD2-NEXT:    store [3 x %struct.S1]* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0), [3 x %struct.S1]** [[ARRAYINIT_ENDOFINIT]], align 8, !dbg [[DBG155:![0-9]+]]
-// SIMD2-NEXT:    store %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 0), %struct.S1** [[ARRAYINIT_ENDOFINIT1]], align 8, !dbg [[DBG157:![0-9]+]]
+// SIMD2-NEXT:    store [3 x %struct.S1]* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0), [3 x %struct.S1]** [[ARRAYINIT_ENDOFINIT]], align 8, !dbg [[DBG154:![0-9]+]]
+// SIMD2-NEXT:    store %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 0), %struct.S1** [[ARRAYINIT_ENDOFINIT1]], align 8, !dbg [[DBG156:![0-9]+]]
 // SIMD2-NEXT:    invoke void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 0), i32 1)
-// SIMD2-NEXT:    to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]], !dbg [[DBG158:![0-9]+]]
+// SIMD2-NEXT:    to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]], !dbg [[DBG157:![0-9]+]]
 // SIMD2:       invoke.cont:
-// SIMD2-NEXT:    store %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 1), %struct.S1** [[ARRAYINIT_ENDOFINIT1]], align 8, !dbg [[DBG157]]
+// SIMD2-NEXT:    store %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 1), %struct.S1** [[ARRAYINIT_ENDOFINIT1]], align 8, !dbg [[DBG156]]
 // SIMD2-NEXT:    invoke void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 1), i32 2)
-// SIMD2-NEXT:    to label [[INVOKE_CONT2:%.*]] unwind label [[LPAD]], !dbg [[DBG159:![0-9]+]]
+// SIMD2-NEXT:    to label [[INVOKE_CONT2:%.*]] unwind label [[LPAD]], !dbg [[DBG158:![0-9]+]]
 // SIMD2:       invoke.cont2:
-// SIMD2-NEXT:    store %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 2), %struct.S1** [[ARRAYINIT_ENDOFINIT1]], align 8, !dbg [[DBG157]]
+// SIMD2-NEXT:    store %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 2), %struct.S1** [[ARRAYINIT_ENDOFINIT1]], align 8, !dbg [[DBG156]]
 // SIMD2-NEXT:    invoke void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 2), i32 3)
-// SIMD2-NEXT:    to label [[INVOKE_CONT3:%.*]] unwind label [[LPAD]], !dbg [[DBG160:![0-9]+]]
+// SIMD2-NEXT:    to label [[INVOKE_CONT3:%.*]] unwind label [[LPAD]], !dbg [[DBG159:![0-9]+]]
 // SIMD2:       invoke.cont3:
-// SIMD2-NEXT:    store [3 x %struct.S1]* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1), [3 x %struct.S1]** [[ARRAYINIT_ENDOFINIT]], align 8, !dbg [[DBG155]]
-// SIMD2-NEXT:    store %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 0), %struct.S1** [[ARRAYINIT_ENDOFINIT5]], align 8, !dbg [[DBG161:![0-9]+]]
+// SIMD2-NEXT:    store [3 x %struct.S1]* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1), [3 x %struct.S1]** [[ARRAYINIT_ENDOFINIT]], align 8, !dbg [[DBG154]]
+// SIMD2-NEXT:    store %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 0), %struct.S1** [[ARRAYINIT_ENDOFINIT5]], align 8, !dbg [[DBG160:![0-9]+]]
 // SIMD2-NEXT:    invoke void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 0), i32 4)
-// SIMD2-NEXT:    to label [[INVOKE_CONT7:%.*]] unwind label [[LPAD6:%.*]], !dbg [[DBG162:![0-9]+]]
+// SIMD2-NEXT:    to label [[INVOKE_CONT7:%.*]] unwind label [[LPAD6:%.*]], !dbg [[DBG161:![0-9]+]]
 // SIMD2:       invoke.cont7:
-// SIMD2-NEXT:    store %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 1), %struct.S1** [[ARRAYINIT_ENDOFINIT5]], align 8, !dbg [[DBG161]]
+// SIMD2-NEXT:    store %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 1), %struct.S1** [[ARRAYINIT_ENDOFINIT5]], align 8, !dbg [[DBG160]]
 // SIMD2-NEXT:    invoke void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 1), i32 5)
-// SIMD2-NEXT:    to label [[INVOKE_CONT8:%.*]] unwind label [[LPAD6]], !dbg [[DBG163:![0-9]+]]
+// SIMD2-NEXT:    to label [[INVOKE_CONT8:%.*]] unwind label [[LPAD6]], !dbg [[DBG162:![0-9]+]]
 // SIMD2:       invoke.cont8:
-// SIMD2-NEXT:    store %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 2), %struct.S1** [[ARRAYINIT_ENDOFINIT5]], align 8, !dbg [[DBG161]]
+// SIMD2-NEXT:    store %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 2), %struct.S1** [[ARRAYINIT_ENDOFINIT5]], align 8, !dbg [[DBG160]]
 // SIMD2-NEXT:    invoke void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 2), i32 6)
-// SIMD2-NEXT:    to label [[INVOKE_CONT9:%.*]] unwind label [[LPAD6]], !dbg [[DBG164:![0-9]+]]
+// SIMD2-NEXT:    to label [[INVOKE_CONT9:%.*]] unwind label [[LPAD6]], !dbg [[DBG163:![0-9]+]]
 // SIMD2:       invoke.cont9:
-// SIMD2-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* @__cxx_global_array_dtor, i8* null, i8* @__dso_handle) #[[ATTR3]], !dbg [[DBG165:![0-9]+]]
-// SIMD2-NEXT:    ret void, !dbg [[DBG165]]
+// SIMD2-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* @__cxx_global_array_dtor, i8* null, i8* @__dso_handle) #[[ATTR3]], !dbg [[DBG164:![0-9]+]]
+// SIMD2-NEXT:    ret void, !dbg [[DBG164]]
 // SIMD2:       lpad:
 // SIMD2-NEXT:    [[TMP1:%.*]] = landingpad { i8*, i32 }
-// SIMD2-NEXT:    cleanup, !dbg [[DBG166:![0-9]+]]
-// SIMD2-NEXT:    [[TMP2:%.*]] = extractvalue { i8*, i32 } [[TMP1]], 0, !dbg [[DBG166]]
-// SIMD2-NEXT:    store i8* [[TMP2]], i8** [[EXN_SLOT]], align 8, !dbg [[DBG166]]
-// SIMD2-NEXT:    [[TMP3:%.*]] = extractvalue { i8*, i32 } [[TMP1]], 1, !dbg [[DBG166]]
-// SIMD2-NEXT:    store i32 [[TMP3]], i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG166]]
-// SIMD2-NEXT:    [[TMP4:%.*]] = load %struct.S1*, %struct.S1** [[ARRAYINIT_ENDOFINIT1]], align 8, !dbg [[DBG157]]
-// SIMD2-NEXT:    [[ARRAYDESTROY_ISEMPTY:%.*]] = icmp eq %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 0), [[TMP4]], !dbg [[DBG157]]
-// SIMD2-NEXT:    br i1 [[ARRAYDESTROY_ISEMPTY]], label [[ARRAYDESTROY_DONE4:%.*]], label [[ARRAYDESTROY_BODY:%.*]], !dbg [[DBG157]]
+// SIMD2-NEXT:    cleanup, !dbg [[DBG165:![0-9]+]]
+// SIMD2-NEXT:    [[TMP2:%.*]] = extractvalue { i8*, i32 } [[TMP1]], 0, !dbg [[DBG165]]
+// SIMD2-NEXT:    store i8* [[TMP2]], i8** [[EXN_SLOT]], align 8, !dbg [[DBG165]]
+// SIMD2-NEXT:    [[TMP3:%.*]] = extractvalue { i8*, i32 } [[TMP1]], 1, !dbg [[DBG165]]
+// SIMD2-NEXT:    store i32 [[TMP3]], i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG165]]
+// SIMD2-NEXT:    [[TMP4:%.*]] = load %struct.S1*, %struct.S1** [[ARRAYINIT_ENDOFINIT1]], align 8, !dbg [[DBG156]]
+// SIMD2-NEXT:    [[ARRAYDESTROY_ISEMPTY:%.*]] = icmp eq %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 0), [[TMP4]], !dbg [[DBG156]]
+// SIMD2-NEXT:    br i1 [[ARRAYDESTROY_ISEMPTY]], label [[ARRAYDESTROY_DONE4:%.*]], label [[ARRAYDESTROY_BODY:%.*]], !dbg [[DBG156]]
 // SIMD2:       arraydestroy.body:
-// SIMD2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ [[TMP4]], [[LPAD]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG157]]
-// SIMD2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG157]]
-// SIMD2-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]], !dbg [[DBG157]]
-// SIMD2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 0), !dbg [[DBG157]]
-// SIMD2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE4]], label [[ARRAYDESTROY_BODY]], !dbg [[DBG157]]
+// SIMD2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ [[TMP4]], [[LPAD]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG156]]
+// SIMD2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG156]]
+// SIMD2-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]], !dbg [[DBG156]]
+// SIMD2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 0), !dbg [[DBG156]]
+// SIMD2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE4]], label [[ARRAYDESTROY_BODY]], !dbg [[DBG156]]
 // SIMD2:       arraydestroy.done4:
-// SIMD2-NEXT:    br label [[EHCLEANUP:%.*]], !dbg [[DBG157]]
+// SIMD2-NEXT:    br label [[EHCLEANUP:%.*]], !dbg [[DBG156]]
 // SIMD2:       lpad6:
 // SIMD2-NEXT:    [[TMP5:%.*]] = landingpad { i8*, i32 }
-// SIMD2-NEXT:    cleanup, !dbg [[DBG166]]
-// SIMD2-NEXT:    [[TMP6:%.*]] = extractvalue { i8*, i32 } [[TMP5]], 0, !dbg [[DBG166]]
-// SIMD2-NEXT:    store i8* [[TMP6]], i8** [[EXN_SLOT]], align 8, !dbg [[DBG166]]
-// SIMD2-NEXT:    [[TMP7:%.*]] = extractvalue { i8*, i32 } [[TMP5]], 1, !dbg [[DBG166]]
-// SIMD2-NEXT:    store i32 [[TMP7]], i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG166]]
-// SIMD2-NEXT:    [[TMP8:%.*]] = load %struct.S1*, %struct.S1** [[ARRAYINIT_ENDOFINIT5]], align 8, !dbg [[DBG161]]
-// SIMD2-NEXT:    [[ARRAYDESTROY_ISEMPTY10:%.*]] = icmp eq %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 0), [[TMP8]], !dbg [[DBG161]]
-// SIMD2-NEXT:    br i1 [[ARRAYDESTROY_ISEMPTY10]], label [[ARRAYDESTROY_DONE15:%.*]], label [[ARRAYDESTROY_BODY11:%.*]], !dbg [[DBG161]]
+// SIMD2-NEXT:    cleanup, !dbg [[DBG165]]
+// SIMD2-NEXT:    [[TMP6:%.*]] = extractvalue { i8*, i32 } [[TMP5]], 0, !dbg [[DBG165]]
+// SIMD2-NEXT:    store i8* [[TMP6]], i8** [[EXN_SLOT]], align 8, !dbg [[DBG165]]
+// SIMD2-NEXT:    [[TMP7:%.*]] = extractvalue { i8*, i32 } [[TMP5]], 1, !dbg [[DBG165]]
+// SIMD2-NEXT:    store i32 [[TMP7]], i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG165]]
+// SIMD2-NEXT:    [[TMP8:%.*]] = load %struct.S1*, %struct.S1** [[ARRAYINIT_ENDOFINIT5]], align 8, !dbg [[DBG160]]
+// SIMD2-NEXT:    [[ARRAYDESTROY_ISEMPTY10:%.*]] = icmp eq %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 0), [[TMP8]], !dbg [[DBG160]]
+// SIMD2-NEXT:    br i1 [[ARRAYDESTROY_ISEMPTY10]], label [[ARRAYDESTROY_DONE15:%.*]], label [[ARRAYDESTROY_BODY11:%.*]], !dbg [[DBG160]]
 // SIMD2:       arraydestroy.body11:
-// SIMD2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST12:%.*]] = phi %struct.S1* [ [[TMP8]], [[LPAD6]] ], [ [[ARRAYDESTROY_ELEMENT13:%.*]], [[ARRAYDESTROY_BODY11]] ], !dbg [[DBG161]]
-// SIMD2-NEXT:    [[ARRAYDESTROY_ELEMENT13]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST12]], i64 -1, !dbg [[DBG161]]
-// SIMD2-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT13]]) #[[ATTR3]], !dbg [[DBG161]]
-// SIMD2-NEXT:    [[ARRAYDESTROY_DONE14:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT13]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 0), !dbg [[DBG161]]
-// SIMD2-NEXT:    br i1 [[ARRAYDESTROY_DONE14]], label [[ARRAYDESTROY_DONE15]], label [[ARRAYDESTROY_BODY11]], !dbg [[DBG161]]
+// SIMD2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST12:%.*]] = phi %struct.S1* [ [[TMP8]], [[LPAD6]] ], [ [[ARRAYDESTROY_ELEMENT13:%.*]], [[ARRAYDESTROY_BODY11]] ], !dbg [[DBG160]]
+// SIMD2-NEXT:    [[ARRAYDESTROY_ELEMENT13]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST12]], i64 -1, !dbg [[DBG160]]
+// SIMD2-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT13]]) #[[ATTR3]], !dbg [[DBG160]]
+// SIMD2-NEXT:    [[ARRAYDESTROY_DONE14:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT13]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 0), !dbg [[DBG160]]
+// SIMD2-NEXT:    br i1 [[ARRAYDESTROY_DONE14]], label [[ARRAYDESTROY_DONE15]], label [[ARRAYDESTROY_BODY11]], !dbg [[DBG160]]
 // SIMD2:       arraydestroy.done15:
-// SIMD2-NEXT:    br label [[EHCLEANUP]], !dbg [[DBG161]]
+// SIMD2-NEXT:    br label [[EHCLEANUP]], !dbg [[DBG160]]
 // SIMD2:       ehcleanup:
-// SIMD2-NEXT:    [[TMP9:%.*]] = load [3 x %struct.S1]*, [3 x %struct.S1]** [[ARRAYINIT_ENDOFINIT]], align 8, !dbg [[DBG155]]
-// SIMD2-NEXT:    [[PAD_ARRAYEND:%.*]] = getelementptr inbounds [3 x %struct.S1], [3 x %struct.S1]* [[TMP9]], i64 0, i64 0, !dbg [[DBG155]]
-// SIMD2-NEXT:    [[ARRAYDESTROY_ISEMPTY16:%.*]] = icmp eq %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 0), [[PAD_ARRAYEND]], !dbg [[DBG155]]
-// SIMD2-NEXT:    br i1 [[ARRAYDESTROY_ISEMPTY16]], label [[ARRAYDESTROY_DONE21:%.*]], label [[ARRAYDESTROY_BODY17:%.*]], !dbg [[DBG155]]
+// SIMD2-NEXT:    [[TMP9:%.*]] = load [3 x %struct.S1]*, [3 x %struct.S1]** [[ARRAYINIT_ENDOFINIT]], align 8, !dbg [[DBG154]]
+// SIMD2-NEXT:    [[PAD_ARRAYEND:%.*]] = getelementptr inbounds [3 x %struct.S1], [3 x %struct.S1]* [[TMP9]], i64 0, i64 0, !dbg [[DBG154]]
+// SIMD2-NEXT:    [[ARRAYDESTROY_ISEMPTY16:%.*]] = icmp eq %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 0), [[PAD_ARRAYEND]], !dbg [[DBG154]]
+// SIMD2-NEXT:    br i1 [[ARRAYDESTROY_ISEMPTY16]], label [[ARRAYDESTROY_DONE21:%.*]], label [[ARRAYDESTROY_BODY17:%.*]], !dbg [[DBG154]]
 // SIMD2:       arraydestroy.body17:
-// SIMD2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST18:%.*]] = phi %struct.S1* [ [[PAD_ARRAYEND]], [[EHCLEANUP]] ], [ [[ARRAYDESTROY_ELEMENT19:%.*]], [[ARRAYDESTROY_BODY17]] ], !dbg [[DBG155]]
-// SIMD2-NEXT:    [[ARRAYDESTROY_ELEMENT19]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST18]], i64 -1, !dbg [[DBG155]]
-// SIMD2-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT19]]) #[[ATTR3]], !dbg [[DBG155]]
-// SIMD2-NEXT:    [[ARRAYDESTROY_DONE20:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT19]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 0), !dbg [[DBG155]]
-// SIMD2-NEXT:    br i1 [[ARRAYDESTROY_DONE20]], label [[ARRAYDESTROY_DONE21]], label [[ARRAYDESTROY_BODY17]], !dbg [[DBG155]]
+// SIMD2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST18:%.*]] = phi %struct.S1* [ [[PAD_ARRAYEND]], [[EHCLEANUP]] ], [ [[ARRAYDESTROY_ELEMENT19:%.*]], [[ARRAYDESTROY_BODY17]] ], !dbg [[DBG154]]
+// SIMD2-NEXT:    [[ARRAYDESTROY_ELEMENT19]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST18]], i64 -1, !dbg [[DBG154]]
+// SIMD2-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT19]]) #[[ATTR3]], !dbg [[DBG154]]
+// SIMD2-NEXT:    [[ARRAYDESTROY_DONE20:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT19]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 0), !dbg [[DBG154]]
+// SIMD2-NEXT:    br i1 [[ARRAYDESTROY_DONE20]], label [[ARRAYDESTROY_DONE21]], label [[ARRAYDESTROY_BODY17]], !dbg [[DBG154]]
 // SIMD2:       arraydestroy.done21:
-// SIMD2-NEXT:    br label [[EH_RESUME:%.*]], !dbg [[DBG155]]
+// SIMD2-NEXT:    br label [[EH_RESUME:%.*]], !dbg [[DBG154]]
 // SIMD2:       eh.resume:
-// SIMD2-NEXT:    [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8, !dbg [[DBG155]]
-// SIMD2-NEXT:    [[SEL:%.*]] = load i32, i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG155]]
-// SIMD2-NEXT:    [[LPAD_VAL:%.*]] = insertvalue { i8*, i32 } undef, i8* [[EXN]], 0, !dbg [[DBG155]]
-// SIMD2-NEXT:    [[LPAD_VAL22:%.*]] = insertvalue { i8*, i32 } [[LPAD_VAL]], i32 [[SEL]], 1, !dbg [[DBG155]]
-// SIMD2-NEXT:    resume { i8*, i32 } [[LPAD_VAL22]], !dbg [[DBG155]]
+// SIMD2-NEXT:    [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8, !dbg [[DBG154]]
+// SIMD2-NEXT:    [[SEL:%.*]] = load i32, i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG154]]
+// SIMD2-NEXT:    [[LPAD_VAL:%.*]] = insertvalue { i8*, i32 } undef, i8* [[EXN]], 0, !dbg [[DBG154]]
+// SIMD2-NEXT:    [[LPAD_VAL22:%.*]] = insertvalue { i8*, i32 } [[LPAD_VAL]], i32 [[SEL]], 1, !dbg [[DBG154]]
+// SIMD2-NEXT:    resume { i8*, i32 } [[LPAD_VAL22]], !dbg [[DBG154]]
 //
 //
 // SIMD2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
-// SIMD2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] !dbg [[DBG167:![0-9]+]] {
+// SIMD2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] !dbg [[DBG166:![0-9]+]] {
 // SIMD2-NEXT:  entry:
 // SIMD2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // SIMD2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata i8** [[DOTADDR]], metadata [[META171:![0-9]+]], metadata !DIExpression()), !dbg [[DBG172:![0-9]+]]
-// SIMD2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]], !dbg [[DBG172]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata i8** [[DOTADDR]], metadata [[META170:![0-9]+]], metadata !DIExpression()), !dbg [[DBG171:![0-9]+]]
+// SIMD2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]], !dbg [[DBG171]]
 // SIMD2:       arraydestroy.body:
-// SIMD2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ getelementptr inbounds ([[STRUCT_S1:%.*]], %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0), i64 6), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG172]]
-// SIMD2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG172]]
-// SIMD2-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]], !dbg [[DBG172]]
-// SIMD2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0), !dbg [[DBG172]]
-// SIMD2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]], !dbg [[DBG172]]
+// SIMD2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 1, i64 0, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG171]]
+// SIMD2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG171]]
+// SIMD2-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]], !dbg [[DBG171]]
+// SIMD2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0), !dbg [[DBG171]]
+// SIMD2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]], !dbg [[DBG171]]
 // SIMD2:       arraydestroy.done1:
-// SIMD2-NEXT:    ret void, !dbg [[DBG172]]
+// SIMD2-NEXT:    ret void, !dbg [[DBG171]]
 //
 //
 // SIMD2-LABEL: define {{[^@]+}}@main
-// SIMD2-SAME: () #[[ATTR5:[0-9]+]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) !dbg [[DBG53:![0-9]+]] {
+// SIMD2-SAME: () #[[ATTR5:[0-9]+]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) !dbg [[DBG52:![0-9]+]] {
 // SIMD2-NEXT:  entry:
 // SIMD2-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
 // SIMD2-NEXT:    [[RES:%.*]] = alloca i32, align 4
 // SIMD2-NEXT:    [[EXN_SLOT:%.*]] = alloca i8*, align 8
 // SIMD2-NEXT:    [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
 // SIMD2-NEXT:    store i32 0, i32* [[RETVAL]], align 4
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata i32* [[RES]], metadata [[META173:![0-9]+]], metadata !DIExpression()), !dbg [[DBG174:![0-9]+]]
-// SIMD2-NEXT:    [[TMP0:%.*]] = load atomic i8, i8* bitcast (i64* @_ZGVZ4mainE2sm to i8*) acquire, align 8, !dbg [[DBG175:![0-9]+]]
-// SIMD2-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0, !dbg [[DBG175]]
-// SIMD2-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !dbg [[DBG175]], !prof [[PROF176:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata i32* [[RES]], metadata [[META172:![0-9]+]], metadata !DIExpression()), !dbg [[DBG173:![0-9]+]]
+// SIMD2-NEXT:    [[TMP0:%.*]] = load atomic i8, i8* bitcast (i64* @_ZGVZ4mainE2sm to i8*) acquire, align 8, !dbg [[DBG174:![0-9]+]]
+// SIMD2-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0, !dbg [[DBG174]]
+// SIMD2-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !dbg [[DBG174]], !prof [[PROF175:![0-9]+]]
 // SIMD2:       init.check:
-// SIMD2-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_guard_acquire(i64* @_ZGVZ4mainE2sm) #[[ATTR3]], !dbg [[DBG175]]
-// SIMD2-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP1]], 0, !dbg [[DBG175]]
-// SIMD2-NEXT:    br i1 [[TOBOOL]], label [[INIT:%.*]], label [[INIT_END]], !dbg [[DBG175]]
+// SIMD2-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_guard_acquire(i64* @_ZGVZ4mainE2sm) #[[ATTR3]], !dbg [[DBG174]]
+// SIMD2-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP1]], 0, !dbg [[DBG174]]
+// SIMD2-NEXT:    br i1 [[TOBOOL]], label [[INIT:%.*]], label [[INIT_END]], !dbg [[DBG174]]
 // SIMD2:       init:
-// SIMD2-NEXT:    [[TMP2:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S1:%.*]], %struct.S1* @_ZL3gs1, i32 0, i32 0), align 4, !dbg [[DBG177:![0-9]+]]
+// SIMD2-NEXT:    [[TMP2:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S1:%.*]], %struct.S1* @_ZL3gs1, i32 0, i32 0), align 4, !dbg [[DBG176:![0-9]+]]
 // SIMD2-NEXT:    invoke void @_ZZ4mainEN5SmainC1Ei(%struct.Smain* nonnull align 8 dereferenceable(24) @_ZZ4mainE2sm, i32 [[TMP2]])
-// SIMD2-NEXT:    to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]], !dbg [[DBG178:![0-9]+]]
+// SIMD2-NEXT:    to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]], !dbg [[DBG177:![0-9]+]]
 // SIMD2:       invoke.cont:
-// SIMD2-NEXT:    [[TMP3:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.Smain*)* @_ZZ4mainEN5SmainD1Ev to void (i8*)*), i8* bitcast (%struct.Smain* @_ZZ4mainE2sm to i8*), i8* @__dso_handle) #[[ATTR3]], !dbg [[DBG175]]
-// SIMD2-NEXT:    call void @__cxa_guard_release(i64* @_ZGVZ4mainE2sm) #[[ATTR3]], !dbg [[DBG175]]
-// SIMD2-NEXT:    br label [[INIT_END]], !dbg [[DBG175]]
+// SIMD2-NEXT:    [[TMP3:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.Smain*)* @_ZZ4mainEN5SmainD1Ev to void (i8*)*), i8* bitcast (%struct.Smain* @_ZZ4mainE2sm to i8*), i8* @__dso_handle) #[[ATTR3]], !dbg [[DBG174]]
+// SIMD2-NEXT:    call void @__cxa_guard_release(i64* @_ZGVZ4mainE2sm) #[[ATTR3]], !dbg [[DBG174]]
+// SIMD2-NEXT:    br label [[INIT_END]], !dbg [[DBG174]]
 // SIMD2:       init.end:
-// SIMD2-NEXT:    [[TMP4:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S3:%.*]], %struct.S3* @_ZN6Static1sE, i32 0, i32 0), align 4, !dbg [[DBG179:![0-9]+]]
-// SIMD2-NEXT:    store i32 [[TMP4]], i32* [[RES]], align 4, !dbg [[DBG180:![0-9]+]]
-// SIMD2-NEXT:    [[TMP5:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_SMAIN:%.*]], %struct.Smain* @_ZZ4mainE2sm, i32 0, i32 0), align 8, !dbg [[DBG181:![0-9]+]]
-// SIMD2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG182:![0-9]+]]
-// SIMD2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP6]], [[TMP5]], !dbg [[DBG182]]
-// SIMD2-NEXT:    store i32 [[ADD]], i32* [[RES]], align 4, !dbg [[DBG182]]
-// SIMD2-NEXT:    [[TMP7:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S1]], %struct.S1* @_ZL3gs1, i32 0, i32 0), align 4, !dbg [[DBG183:![0-9]+]]
-// SIMD2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG184:![0-9]+]]
-// SIMD2-NEXT:    [[ADD1:%.*]] = add nsw i32 [[TMP8]], [[TMP7]], !dbg [[DBG184]]
-// SIMD2-NEXT:    store i32 [[ADD1]], i32* [[RES]], align 4, !dbg [[DBG184]]
-// SIMD2-NEXT:    [[TMP9:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S2:%.*]], %struct.S2* @_ZL3gs2, i32 0, i32 0), align 8, !dbg [[DBG185:![0-9]+]]
-// SIMD2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG186:![0-9]+]]
-// SIMD2-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP10]], [[TMP9]], !dbg [[DBG186]]
-// SIMD2-NEXT:    store i32 [[ADD2]], i32* [[RES]], align 4, !dbg [[DBG186]]
-// SIMD2-NEXT:    [[TMP11:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S5:%.*]], %struct.S5* @gs3, i32 0, i32 0), align 4, !dbg [[DBG187:![0-9]+]]
-// SIMD2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG188:![0-9]+]]
-// SIMD2-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP12]], [[TMP11]], !dbg [[DBG188]]
-// SIMD2-NEXT:    store i32 [[ADD3]], i32* [[RES]], align 4, !dbg [[DBG188]]
-// SIMD2-NEXT:    [[TMP13:%.*]] = load i32, i32* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 1, i32 0), align 4, !dbg [[DBG189:![0-9]+]]
-// SIMD2-NEXT:    [[TMP14:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG190:![0-9]+]]
-// SIMD2-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP14]], [[TMP13]], !dbg [[DBG190]]
-// SIMD2-NEXT:    store i32 [[ADD4]], i32* [[RES]], align 4, !dbg [[DBG190]]
-// SIMD2-NEXT:    [[TMP15:%.*]] = load i32, i32* @_ZN2STIiE2stE, align 4, !dbg [[DBG191:![0-9]+]]
-// SIMD2-NEXT:    [[TMP16:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG192:![0-9]+]]
-// SIMD2-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP16]], [[TMP15]], !dbg [[DBG192]]
-// SIMD2-NEXT:    store i32 [[ADD5]], i32* [[RES]], align 4, !dbg [[DBG192]]
-// SIMD2-NEXT:    [[TMP17:%.*]] = load float, float* @_ZN2STIfE2stE, align 4, !dbg [[DBG193:![0-9]+]]
-// SIMD2-NEXT:    [[CONV:%.*]] = fptosi float [[TMP17]] to i32, !dbg [[DBG193]]
-// SIMD2-NEXT:    [[TMP18:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG194:![0-9]+]]
-// SIMD2-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP18]], [[CONV]], !dbg [[DBG194]]
-// SIMD2-NEXT:    store i32 [[ADD6]], i32* [[RES]], align 4, !dbg [[DBG194]]
-// SIMD2-NEXT:    [[TMP19:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S4:%.*]], %struct.S4* @_ZN2STI2S4E2stE, i32 0, i32 0), align 4, !dbg [[DBG195:![0-9]+]]
-// SIMD2-NEXT:    [[TMP20:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG196:![0-9]+]]
-// SIMD2-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP20]], [[TMP19]], !dbg [[DBG196]]
-// SIMD2-NEXT:    store i32 [[ADD7]], i32* [[RES]], align 4, !dbg [[DBG196]]
-// SIMD2-NEXT:    [[TMP21:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG197:![0-9]+]]
-// SIMD2-NEXT:    ret i32 [[TMP21]], !dbg [[DBG198:![0-9]+]]
+// SIMD2-NEXT:    [[TMP4:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S3:%.*]], %struct.S3* @_ZN6Static1sE, i32 0, i32 0), align 4, !dbg [[DBG178:![0-9]+]]
+// SIMD2-NEXT:    store i32 [[TMP4]], i32* [[RES]], align 4, !dbg [[DBG179:![0-9]+]]
+// SIMD2-NEXT:    [[TMP5:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_SMAIN:%.*]], %struct.Smain* @_ZZ4mainE2sm, i32 0, i32 0), align 8, !dbg [[DBG180:![0-9]+]]
+// SIMD2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG181:![0-9]+]]
+// SIMD2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP6]], [[TMP5]], !dbg [[DBG181]]
+// SIMD2-NEXT:    store i32 [[ADD]], i32* [[RES]], align 4, !dbg [[DBG181]]
+// SIMD2-NEXT:    [[TMP7:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S1]], %struct.S1* @_ZL3gs1, i32 0, i32 0), align 4, !dbg [[DBG182:![0-9]+]]
+// SIMD2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG183:![0-9]+]]
+// SIMD2-NEXT:    [[ADD1:%.*]] = add nsw i32 [[TMP8]], [[TMP7]], !dbg [[DBG183]]
+// SIMD2-NEXT:    store i32 [[ADD1]], i32* [[RES]], align 4, !dbg [[DBG183]]
+// SIMD2-NEXT:    [[TMP9:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S2:%.*]], %struct.S2* @_ZL3gs2, i32 0, i32 0), align 8, !dbg [[DBG184:![0-9]+]]
+// SIMD2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG185:![0-9]+]]
+// SIMD2-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP10]], [[TMP9]], !dbg [[DBG185]]
+// SIMD2-NEXT:    store i32 [[ADD2]], i32* [[RES]], align 4, !dbg [[DBG185]]
+// SIMD2-NEXT:    [[TMP11:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S5:%.*]], %struct.S5* @gs3, i32 0, i32 0), align 4, !dbg [[DBG186:![0-9]+]]
+// SIMD2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG187:![0-9]+]]
+// SIMD2-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP12]], [[TMP11]], !dbg [[DBG187]]
+// SIMD2-NEXT:    store i32 [[ADD3]], i32* [[RES]], align 4, !dbg [[DBG187]]
+// SIMD2-NEXT:    [[TMP13:%.*]] = load i32, i32* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 1, i32 0), align 4, !dbg [[DBG188:![0-9]+]]
+// SIMD2-NEXT:    [[TMP14:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG189:![0-9]+]]
+// SIMD2-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP14]], [[TMP13]], !dbg [[DBG189]]
+// SIMD2-NEXT:    store i32 [[ADD4]], i32* [[RES]], align 4, !dbg [[DBG189]]
+// SIMD2-NEXT:    [[TMP15:%.*]] = load i32, i32* @_ZN2STIiE2stE, align 4, !dbg [[DBG190:![0-9]+]]
+// SIMD2-NEXT:    [[TMP16:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG191:![0-9]+]]
+// SIMD2-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP16]], [[TMP15]], !dbg [[DBG191]]
+// SIMD2-NEXT:    store i32 [[ADD5]], i32* [[RES]], align 4, !dbg [[DBG191]]
+// SIMD2-NEXT:    [[TMP17:%.*]] = load float, float* @_ZN2STIfE2stE, align 4, !dbg [[DBG192:![0-9]+]]
+// SIMD2-NEXT:    [[CONV:%.*]] = fptosi float [[TMP17]] to i32, !dbg [[DBG192]]
+// SIMD2-NEXT:    [[TMP18:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG193:![0-9]+]]
+// SIMD2-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP18]], [[CONV]], !dbg [[DBG193]]
+// SIMD2-NEXT:    store i32 [[ADD6]], i32* [[RES]], align 4, !dbg [[DBG193]]
+// SIMD2-NEXT:    [[TMP19:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S4:%.*]], %struct.S4* @_ZN2STI2S4E2stE, i32 0, i32 0), align 4, !dbg [[DBG194:![0-9]+]]
+// SIMD2-NEXT:    [[TMP20:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG195:![0-9]+]]
+// SIMD2-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP20]], [[TMP19]], !dbg [[DBG195]]
+// SIMD2-NEXT:    store i32 [[ADD7]], i32* [[RES]], align 4, !dbg [[DBG195]]
+// SIMD2-NEXT:    [[TMP21:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG196:![0-9]+]]
+// SIMD2-NEXT:    ret i32 [[TMP21]], !dbg [[DBG197:![0-9]+]]
 // SIMD2:       lpad:
 // SIMD2-NEXT:    [[TMP22:%.*]] = landingpad { i8*, i32 }
-// SIMD2-NEXT:    cleanup, !dbg [[DBG199:![0-9]+]]
-// SIMD2-NEXT:    [[TMP23:%.*]] = extractvalue { i8*, i32 } [[TMP22]], 0, !dbg [[DBG199]]
-// SIMD2-NEXT:    store i8* [[TMP23]], i8** [[EXN_SLOT]], align 8, !dbg [[DBG199]]
-// SIMD2-NEXT:    [[TMP24:%.*]] = extractvalue { i8*, i32 } [[TMP22]], 1, !dbg [[DBG199]]
-// SIMD2-NEXT:    store i32 [[TMP24]], i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG199]]
-// SIMD2-NEXT:    call void @__cxa_guard_abort(i64* @_ZGVZ4mainE2sm) #[[ATTR3]], !dbg [[DBG175]]
-// SIMD2-NEXT:    br label [[EH_RESUME:%.*]], !dbg [[DBG175]]
+// SIMD2-NEXT:    cleanup, !dbg [[DBG198:![0-9]+]]
+// SIMD2-NEXT:    [[TMP23:%.*]] = extractvalue { i8*, i32 } [[TMP22]], 0, !dbg [[DBG198]]
+// SIMD2-NEXT:    store i8* [[TMP23]], i8** [[EXN_SLOT]], align 8, !dbg [[DBG198]]
+// SIMD2-NEXT:    [[TMP24:%.*]] = extractvalue { i8*, i32 } [[TMP22]], 1, !dbg [[DBG198]]
+// SIMD2-NEXT:    store i32 [[TMP24]], i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG198]]
+// SIMD2-NEXT:    call void @__cxa_guard_abort(i64* @_ZGVZ4mainE2sm) #[[ATTR3]], !dbg [[DBG174]]
+// SIMD2-NEXT:    br label [[EH_RESUME:%.*]], !dbg [[DBG174]]
 // SIMD2:       eh.resume:
-// SIMD2-NEXT:    [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8, !dbg [[DBG175]]
-// SIMD2-NEXT:    [[SEL:%.*]] = load i32, i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG175]]
-// SIMD2-NEXT:    [[LPAD_VAL:%.*]] = insertvalue { i8*, i32 } undef, i8* [[EXN]], 0, !dbg [[DBG175]]
-// SIMD2-NEXT:    [[LPAD_VAL8:%.*]] = insertvalue { i8*, i32 } [[LPAD_VAL]], i32 [[SEL]], 1, !dbg [[DBG175]]
-// SIMD2-NEXT:    resume { i8*, i32 } [[LPAD_VAL8]], !dbg [[DBG175]]
+// SIMD2-NEXT:    [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8, !dbg [[DBG174]]
+// SIMD2-NEXT:    [[SEL:%.*]] = load i32, i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG174]]
+// SIMD2-NEXT:    [[LPAD_VAL:%.*]] = insertvalue { i8*, i32 } undef, i8* [[EXN]], 0, !dbg [[DBG174]]
+// SIMD2-NEXT:    [[LPAD_VAL8:%.*]] = insertvalue { i8*, i32 } [[LPAD_VAL]], i32 [[SEL]], 1, !dbg [[DBG174]]
+// SIMD2-NEXT:    resume { i8*, i32 } [[LPAD_VAL8]], !dbg [[DBG174]]
 //
 //
 // SIMD2-LABEL: define {{[^@]+}}@_ZZ4mainEN5SmainC1Ei
-// SIMD2-SAME: (%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR1]] align 2 !dbg [[DBG200:![0-9]+]] {
+// SIMD2-SAME: (%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR1]] align 2 !dbg [[DBG199:![0-9]+]] {
 // SIMD2-NEXT:  entry:
 // SIMD2-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.Smain*, align 8
 // SIMD2-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
 // SIMD2-NEXT:    store %struct.Smain* [[THIS]], %struct.Smain** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.Smain** [[THIS_ADDR]], metadata [[META201:![0-9]+]], metadata !DIExpression()), !dbg [[DBG203:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.Smain** [[THIS_ADDR]], metadata [[META200:![0-9]+]], metadata !DIExpression()), !dbg [[DBG202:![0-9]+]]
 // SIMD2-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META204:![0-9]+]], metadata !DIExpression()), !dbg [[DBG205:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META203:![0-9]+]], metadata !DIExpression()), !dbg [[DBG204:![0-9]+]]
 // SIMD2-NEXT:    [[THIS1:%.*]] = load %struct.Smain*, %struct.Smain** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG206:![0-9]+]]
-// SIMD2-NEXT:    call void @_ZZ4mainEN5SmainC2Ei(%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS1]], i32 [[TMP0]]), !dbg [[DBG206]]
-// SIMD2-NEXT:    ret void, !dbg [[DBG207:![0-9]+]]
+// SIMD2-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG205:![0-9]+]]
+// SIMD2-NEXT:    call void @_ZZ4mainEN5SmainC2Ei(%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS1]], i32 [[TMP0]]), !dbg [[DBG205]]
+// SIMD2-NEXT:    ret void, !dbg [[DBG206:![0-9]+]]
 //
 //
 // SIMD2-LABEL: define {{[^@]+}}@_ZZ4mainEN5SmainD1Ev
-// SIMD2-SAME: (%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] align 2 !dbg [[DBG208:![0-9]+]] {
+// SIMD2-SAME: (%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] align 2 !dbg [[DBG207:![0-9]+]] {
 // SIMD2-NEXT:  entry:
 // SIMD2-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.Smain*, align 8
 // SIMD2-NEXT:    store %struct.Smain* [[THIS]], %struct.Smain** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.Smain** [[THIS_ADDR]], metadata [[META209:![0-9]+]], metadata !DIExpression()), !dbg [[DBG210:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.Smain** [[THIS_ADDR]], metadata [[META208:![0-9]+]], metadata !DIExpression()), !dbg [[DBG209:![0-9]+]]
 // SIMD2-NEXT:    [[THIS1:%.*]] = load %struct.Smain*, %struct.Smain** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    call void @_ZZ4mainEN5SmainD2Ev(%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS1]]) #[[ATTR3]], !dbg [[DBG211:![0-9]+]]
-// SIMD2-NEXT:    ret void, !dbg [[DBG212:![0-9]+]]
+// SIMD2-NEXT:    call void @_ZZ4mainEN5SmainD2Ev(%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS1]]) #[[ATTR3]], !dbg [[DBG210:![0-9]+]]
+// SIMD2-NEXT:    ret void, !dbg [[DBG211:![0-9]+]]
 //
 //
 // SIMD2-LABEL: define {{[^@]+}}@_Z6foobarv
-// SIMD2-SAME: () #[[ATTR6:[0-9]+]] !dbg [[DBG213:![0-9]+]] {
+// SIMD2-SAME: () #[[ATTR6:[0-9]+]] !dbg [[DBG212:![0-9]+]] {
 // SIMD2-NEXT:  entry:
 // SIMD2-NEXT:    [[RES:%.*]] = alloca i32, align 4
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata i32* [[RES]], metadata [[META214:![0-9]+]], metadata !DIExpression()), !dbg [[DBG215:![0-9]+]]
-// SIMD2-NEXT:    [[TMP0:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S3:%.*]], %struct.S3* @_ZN6Static1sE, i32 0, i32 0), align 4, !dbg [[DBG216:![0-9]+]]
-// SIMD2-NEXT:    store i32 [[TMP0]], i32* [[RES]], align 4, !dbg [[DBG217:![0-9]+]]
-// SIMD2-NEXT:    [[TMP1:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S1:%.*]], %struct.S1* @_ZL3gs1, i32 0, i32 0), align 4, !dbg [[DBG218:![0-9]+]]
-// SIMD2-NEXT:    [[TMP2:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG219:![0-9]+]]
-// SIMD2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], [[TMP1]], !dbg [[DBG219]]
-// SIMD2-NEXT:    store i32 [[ADD]], i32* [[RES]], align 4, !dbg [[DBG219]]
-// SIMD2-NEXT:    [[TMP3:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S2:%.*]], %struct.S2* @_ZL3gs2, i32 0, i32 0), align 8, !dbg [[DBG220:![0-9]+]]
-// SIMD2-NEXT:    [[TMP4:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG221:![0-9]+]]
-// SIMD2-NEXT:    [[ADD1:%.*]] = add nsw i32 [[TMP4]], [[TMP3]], !dbg [[DBG221]]
-// SIMD2-NEXT:    store i32 [[ADD1]], i32* [[RES]], align 4, !dbg [[DBG221]]
-// SIMD2-NEXT:    [[TMP5:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S5:%.*]], %struct.S5* @gs3, i32 0, i32 0), align 4, !dbg [[DBG222:![0-9]+]]
-// SIMD2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG223:![0-9]+]]
-// SIMD2-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP6]], [[TMP5]], !dbg [[DBG223]]
-// SIMD2-NEXT:    store i32 [[ADD2]], i32* [[RES]], align 4, !dbg [[DBG223]]
-// SIMD2-NEXT:    [[TMP7:%.*]] = load i32, i32* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 1, i32 0), align 4, !dbg [[DBG224:![0-9]+]]
-// SIMD2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG225:![0-9]+]]
-// SIMD2-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP8]], [[TMP7]], !dbg [[DBG225]]
-// SIMD2-NEXT:    store i32 [[ADD3]], i32* [[RES]], align 4, !dbg [[DBG225]]
-// SIMD2-NEXT:    [[TMP9:%.*]] = load i32, i32* @_ZN2STIiE2stE, align 4, !dbg [[DBG226:![0-9]+]]
-// SIMD2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG227:![0-9]+]]
-// SIMD2-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP10]], [[TMP9]], !dbg [[DBG227]]
-// SIMD2-NEXT:    store i32 [[ADD4]], i32* [[RES]], align 4, !dbg [[DBG227]]
-// SIMD2-NEXT:    [[TMP11:%.*]] = load float, float* @_ZN2STIfE2stE, align 4, !dbg [[DBG228:![0-9]+]]
-// SIMD2-NEXT:    [[CONV:%.*]] = fptosi float [[TMP11]] to i32, !dbg [[DBG228]]
-// SIMD2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG229:![0-9]+]]
-// SIMD2-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP12]], [[CONV]], !dbg [[DBG229]]
-// SIMD2-NEXT:    store i32 [[ADD5]], i32* [[RES]], align 4, !dbg [[DBG229]]
-// SIMD2-NEXT:    [[TMP13:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S4:%.*]], %struct.S4* @_ZN2STI2S4E2stE, i32 0, i32 0), align 4, !dbg [[DBG230:![0-9]+]]
-// SIMD2-NEXT:    [[TMP14:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG231:![0-9]+]]
-// SIMD2-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP14]], [[TMP13]], !dbg [[DBG231]]
-// SIMD2-NEXT:    store i32 [[ADD6]], i32* [[RES]], align 4, !dbg [[DBG231]]
-// SIMD2-NEXT:    [[TMP15:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG232:![0-9]+]]
-// SIMD2-NEXT:    ret i32 [[TMP15]], !dbg [[DBG233:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata i32* [[RES]], metadata [[META213:![0-9]+]], metadata !DIExpression()), !dbg [[DBG214:![0-9]+]]
+// SIMD2-NEXT:    [[TMP0:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S3:%.*]], %struct.S3* @_ZN6Static1sE, i32 0, i32 0), align 4, !dbg [[DBG215:![0-9]+]]
+// SIMD2-NEXT:    store i32 [[TMP0]], i32* [[RES]], align 4, !dbg [[DBG216:![0-9]+]]
+// SIMD2-NEXT:    [[TMP1:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S1:%.*]], %struct.S1* @_ZL3gs1, i32 0, i32 0), align 4, !dbg [[DBG217:![0-9]+]]
+// SIMD2-NEXT:    [[TMP2:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG218:![0-9]+]]
+// SIMD2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], [[TMP1]], !dbg [[DBG218]]
+// SIMD2-NEXT:    store i32 [[ADD]], i32* [[RES]], align 4, !dbg [[DBG218]]
+// SIMD2-NEXT:    [[TMP3:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S2:%.*]], %struct.S2* @_ZL3gs2, i32 0, i32 0), align 8, !dbg [[DBG219:![0-9]+]]
+// SIMD2-NEXT:    [[TMP4:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG220:![0-9]+]]
+// SIMD2-NEXT:    [[ADD1:%.*]] = add nsw i32 [[TMP4]], [[TMP3]], !dbg [[DBG220]]
+// SIMD2-NEXT:    store i32 [[ADD1]], i32* [[RES]], align 4, !dbg [[DBG220]]
+// SIMD2-NEXT:    [[TMP5:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S5:%.*]], %struct.S5* @gs3, i32 0, i32 0), align 4, !dbg [[DBG221:![0-9]+]]
+// SIMD2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG222:![0-9]+]]
+// SIMD2-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP6]], [[TMP5]], !dbg [[DBG222]]
+// SIMD2-NEXT:    store i32 [[ADD2]], i32* [[RES]], align 4, !dbg [[DBG222]]
+// SIMD2-NEXT:    [[TMP7:%.*]] = load i32, i32* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 1, i32 0), align 4, !dbg [[DBG223:![0-9]+]]
+// SIMD2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG224:![0-9]+]]
+// SIMD2-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP8]], [[TMP7]], !dbg [[DBG224]]
+// SIMD2-NEXT:    store i32 [[ADD3]], i32* [[RES]], align 4, !dbg [[DBG224]]
+// SIMD2-NEXT:    [[TMP9:%.*]] = load i32, i32* @_ZN2STIiE2stE, align 4, !dbg [[DBG225:![0-9]+]]
+// SIMD2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG226:![0-9]+]]
+// SIMD2-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP10]], [[TMP9]], !dbg [[DBG226]]
+// SIMD2-NEXT:    store i32 [[ADD4]], i32* [[RES]], align 4, !dbg [[DBG226]]
+// SIMD2-NEXT:    [[TMP11:%.*]] = load float, float* @_ZN2STIfE2stE, align 4, !dbg [[DBG227:![0-9]+]]
+// SIMD2-NEXT:    [[CONV:%.*]] = fptosi float [[TMP11]] to i32, !dbg [[DBG227]]
+// SIMD2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG228:![0-9]+]]
+// SIMD2-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP12]], [[CONV]], !dbg [[DBG228]]
+// SIMD2-NEXT:    store i32 [[ADD5]], i32* [[RES]], align 4, !dbg [[DBG228]]
+// SIMD2-NEXT:    [[TMP13:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S4:%.*]], %struct.S4* @_ZN2STI2S4E2stE, i32 0, i32 0), align 4, !dbg [[DBG229:![0-9]+]]
+// SIMD2-NEXT:    [[TMP14:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG230:![0-9]+]]
+// SIMD2-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP14]], [[TMP13]], !dbg [[DBG230]]
+// SIMD2-NEXT:    store i32 [[ADD6]], i32* [[RES]], align 4, !dbg [[DBG230]]
+// SIMD2-NEXT:    [[TMP15:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG231:![0-9]+]]
+// SIMD2-NEXT:    ret i32 [[TMP15]], !dbg [[DBG232:![0-9]+]]
 //
 //
 // SIMD2-LABEL: define {{[^@]+}}@__cxx_global_var_init.3
-// SIMD2-SAME: () #[[ATTR0]] comdat($_ZN2STI2S4E2stE) !dbg [[DBG234:![0-9]+]] {
+// SIMD2-SAME: () #[[ATTR0]] comdat($_ZN2STI2S4E2stE) !dbg [[DBG233:![0-9]+]] {
 // SIMD2-NEXT:  entry:
-// SIMD2-NEXT:    [[TMP0:%.*]] = load i8, i8* bitcast (i64* @_ZGVN2STI2S4E2stE to i8*), align 8, !dbg [[DBG235:![0-9]+]]
-// SIMD2-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0, !dbg [[DBG235]]
-// SIMD2-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !dbg [[DBG235]]
+// SIMD2-NEXT:    [[TMP0:%.*]] = load i8, i8* bitcast (i64* @_ZGVN2STI2S4E2stE to i8*), align 8, !dbg [[DBG234:![0-9]+]]
+// SIMD2-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0, !dbg [[DBG234]]
+// SIMD2-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !dbg [[DBG234]]
 // SIMD2:       init.check:
-// SIMD2-NEXT:    call void @_ZN2S4C1Ei(%struct.S4* nonnull align 4 dereferenceable(8) @_ZN2STI2S4E2stE, i32 23), !dbg [[DBG236:![0-9]+]]
-// SIMD2-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.S4*)* @_ZN2S4D1Ev to void (i8*)*), i8* bitcast (%struct.S4* @_ZN2STI2S4E2stE to i8*), i8* @__dso_handle) #[[ATTR3]], !dbg [[DBG235]]
-// SIMD2-NEXT:    store i8 1, i8* bitcast (i64* @_ZGVN2STI2S4E2stE to i8*), align 8, !dbg [[DBG235]]
-// SIMD2-NEXT:    br label [[INIT_END]], !dbg [[DBG235]]
+// SIMD2-NEXT:    call void @_ZN2S4C1Ei(%struct.S4* nonnull align 4 dereferenceable(8) @_ZN2STI2S4E2stE, i32 23), !dbg [[DBG235:![0-9]+]]
+// SIMD2-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.S4*)* @_ZN2S4D1Ev to void (i8*)*), i8* bitcast (%struct.S4* @_ZN2STI2S4E2stE to i8*), i8* @__dso_handle) #[[ATTR3]], !dbg [[DBG234]]
+// SIMD2-NEXT:    store i8 1, i8* bitcast (i64* @_ZGVN2STI2S4E2stE to i8*), align 8, !dbg [[DBG234]]
+// SIMD2-NEXT:    br label [[INIT_END]], !dbg [[DBG234]]
 // SIMD2:       init.end:
-// SIMD2-NEXT:    ret void, !dbg [[DBG238:![0-9]+]]
+// SIMD2-NEXT:    ret void, !dbg [[DBG237:![0-9]+]]
 //
 //
 // SIMD2-LABEL: define {{[^@]+}}@_ZN2S4C1Ei
-// SIMD2-SAME: (%struct.S4* nonnull align 4 dereferenceable(8) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 !dbg [[DBG239:![0-9]+]] {
+// SIMD2-SAME: (%struct.S4* nonnull align 4 dereferenceable(8) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 !dbg [[DBG238:![0-9]+]] {
 // SIMD2-NEXT:  entry:
 // SIMD2-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S4*, align 8
 // SIMD2-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
 // SIMD2-NEXT:    store %struct.S4* [[THIS]], %struct.S4** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.S4** [[THIS_ADDR]], metadata [[META240:![0-9]+]], metadata !DIExpression()), !dbg [[DBG242:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.S4** [[THIS_ADDR]], metadata [[META239:![0-9]+]], metadata !DIExpression()), !dbg [[DBG241:![0-9]+]]
 // SIMD2-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META243:![0-9]+]], metadata !DIExpression()), !dbg [[DBG244:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META242:![0-9]+]], metadata !DIExpression()), !dbg [[DBG243:![0-9]+]]
 // SIMD2-NEXT:    [[THIS1:%.*]] = load %struct.S4*, %struct.S4** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG245:![0-9]+]]
-// SIMD2-NEXT:    call void @_ZN2S4C2Ei(%struct.S4* nonnull align 4 dereferenceable(8) [[THIS1]], i32 [[TMP0]]), !dbg [[DBG245]]
-// SIMD2-NEXT:    ret void, !dbg [[DBG246:![0-9]+]]
+// SIMD2-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG244:![0-9]+]]
+// SIMD2-NEXT:    call void @_ZN2S4C2Ei(%struct.S4* nonnull align 4 dereferenceable(8) [[THIS1]], i32 [[TMP0]]), !dbg [[DBG244]]
+// SIMD2-NEXT:    ret void, !dbg [[DBG245:![0-9]+]]
 //
 //
 // SIMD2-LABEL: define {{[^@]+}}@_ZN2S4D1Ev
-// SIMD2-SAME: (%struct.S4* nonnull align 4 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG247:![0-9]+]] {
+// SIMD2-SAME: (%struct.S4* nonnull align 4 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG246:![0-9]+]] {
 // SIMD2-NEXT:  entry:
 // SIMD2-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S4*, align 8
 // SIMD2-NEXT:    store %struct.S4* [[THIS]], %struct.S4** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.S4** [[THIS_ADDR]], metadata [[META248:![0-9]+]], metadata !DIExpression()), !dbg [[DBG249:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.S4** [[THIS_ADDR]], metadata [[META247:![0-9]+]], metadata !DIExpression()), !dbg [[DBG248:![0-9]+]]
 // SIMD2-NEXT:    [[THIS1:%.*]] = load %struct.S4*, %struct.S4** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    call void @_ZN2S4D2Ev(%struct.S4* nonnull align 4 dereferenceable(8) [[THIS1]]) #[[ATTR3]], !dbg [[DBG250:![0-9]+]]
-// SIMD2-NEXT:    ret void, !dbg [[DBG251:![0-9]+]]
+// SIMD2-NEXT:    call void @_ZN2S4D2Ev(%struct.S4* nonnull align 4 dereferenceable(8) [[THIS1]]) #[[ATTR3]], !dbg [[DBG249:![0-9]+]]
+// SIMD2-NEXT:    ret void, !dbg [[DBG250:![0-9]+]]
 //
 //
 // SIMD2-LABEL: define {{[^@]+}}@_ZN2S1C2Ei
-// SIMD2-SAME: (%struct.S1* nonnull align 4 dereferenceable(4) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG252:![0-9]+]] {
+// SIMD2-SAME: (%struct.S1* nonnull align 4 dereferenceable(4) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG251:![0-9]+]] {
 // SIMD2-NEXT:  entry:
 // SIMD2-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S1*, align 8
 // SIMD2-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
 // SIMD2-NEXT:    store %struct.S1* [[THIS]], %struct.S1** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.S1** [[THIS_ADDR]], metadata [[META253:![0-9]+]], metadata !DIExpression()), !dbg [[DBG254:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.S1** [[THIS_ADDR]], metadata [[META252:![0-9]+]], metadata !DIExpression()), !dbg [[DBG253:![0-9]+]]
 // SIMD2-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META255:![0-9]+]], metadata !DIExpression()), !dbg [[DBG256:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META254:![0-9]+]], metadata !DIExpression()), !dbg [[DBG255:![0-9]+]]
 // SIMD2-NEXT:    [[THIS1:%.*]] = load %struct.S1*, %struct.S1** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    [[A2:%.*]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[THIS1]], i32 0, i32 0, !dbg [[DBG257:![0-9]+]]
-// SIMD2-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG258:![0-9]+]]
-// SIMD2-NEXT:    store i32 [[TMP0]], i32* [[A2]], align 4, !dbg [[DBG257]]
-// SIMD2-NEXT:    ret void, !dbg [[DBG259:![0-9]+]]
+// SIMD2-NEXT:    [[A2:%.*]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[THIS1]], i32 0, i32 0, !dbg [[DBG256:![0-9]+]]
+// SIMD2-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG257:![0-9]+]]
+// SIMD2-NEXT:    store i32 [[TMP0]], i32* [[A2]], align 4, !dbg [[DBG256]]
+// SIMD2-NEXT:    ret void, !dbg [[DBG258:![0-9]+]]
 //
 //
 // SIMD2-LABEL: define {{[^@]+}}@_ZN2S1D2Ev
-// SIMD2-SAME: (%struct.S1* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG260:![0-9]+]] {
+// SIMD2-SAME: (%struct.S1* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG259:![0-9]+]] {
 // SIMD2-NEXT:  entry:
 // SIMD2-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S1*, align 8
 // SIMD2-NEXT:    store %struct.S1* [[THIS]], %struct.S1** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.S1** [[THIS_ADDR]], metadata [[META261:![0-9]+]], metadata !DIExpression()), !dbg [[DBG262:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.S1** [[THIS_ADDR]], metadata [[META260:![0-9]+]], metadata !DIExpression()), !dbg [[DBG261:![0-9]+]]
 // SIMD2-NEXT:    [[THIS1:%.*]] = load %struct.S1*, %struct.S1** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[THIS1]], i32 0, i32 0, !dbg [[DBG263:![0-9]+]]
-// SIMD2-NEXT:    store i32 0, i32* [[A]], align 4, !dbg [[DBG265:![0-9]+]]
-// SIMD2-NEXT:    ret void, !dbg [[DBG266:![0-9]+]]
+// SIMD2-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[THIS1]], i32 0, i32 0, !dbg [[DBG262:![0-9]+]]
+// SIMD2-NEXT:    store i32 0, i32* [[A]], align 4, !dbg [[DBG264:![0-9]+]]
+// SIMD2-NEXT:    ret void, !dbg [[DBG265:![0-9]+]]
 //
 //
 // SIMD2-LABEL: define {{[^@]+}}@_ZN2S2C2Ei
-// SIMD2-SAME: (%struct.S2* nonnull align 8 dereferenceable(16) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG267:![0-9]+]] {
+// SIMD2-SAME: (%struct.S2* nonnull align 8 dereferenceable(16) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG266:![0-9]+]] {
 // SIMD2-NEXT:  entry:
 // SIMD2-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S2*, align 8
 // SIMD2-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
 // SIMD2-NEXT:    store %struct.S2* [[THIS]], %struct.S2** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.S2** [[THIS_ADDR]], metadata [[META268:![0-9]+]], metadata !DIExpression()), !dbg [[DBG269:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.S2** [[THIS_ADDR]], metadata [[META267:![0-9]+]], metadata !DIExpression()), !dbg [[DBG268:![0-9]+]]
 // SIMD2-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META270:![0-9]+]], metadata !DIExpression()), !dbg [[DBG271:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META269:![0-9]+]], metadata !DIExpression()), !dbg [[DBG270:![0-9]+]]
 // SIMD2-NEXT:    [[THIS1:%.*]] = load %struct.S2*, %struct.S2** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    [[A2:%.*]] = getelementptr inbounds [[STRUCT_S2:%.*]], %struct.S2* [[THIS1]], i32 0, i32 0, !dbg [[DBG272:![0-9]+]]
-// SIMD2-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG273:![0-9]+]]
-// SIMD2-NEXT:    store i32 [[TMP0]], i32* [[A2]], align 8, !dbg [[DBG272]]
-// SIMD2-NEXT:    ret void, !dbg [[DBG274:![0-9]+]]
+// SIMD2-NEXT:    [[A2:%.*]] = getelementptr inbounds [[STRUCT_S2:%.*]], %struct.S2* [[THIS1]], i32 0, i32 0, !dbg [[DBG271:![0-9]+]]
+// SIMD2-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG272:![0-9]+]]
+// SIMD2-NEXT:    store i32 [[TMP0]], i32* [[A2]], align 8, !dbg [[DBG271]]
+// SIMD2-NEXT:    ret void, !dbg [[DBG273:![0-9]+]]
 //
 //
 // SIMD2-LABEL: define {{[^@]+}}@_ZN2S2D2Ev
-// SIMD2-SAME: (%struct.S2* nonnull align 8 dereferenceable(16) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG275:![0-9]+]] {
+// SIMD2-SAME: (%struct.S2* nonnull align 8 dereferenceable(16) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG274:![0-9]+]] {
 // SIMD2-NEXT:  entry:
 // SIMD2-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S2*, align 8
 // SIMD2-NEXT:    store %struct.S2* [[THIS]], %struct.S2** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.S2** [[THIS_ADDR]], metadata [[META276:![0-9]+]], metadata !DIExpression()), !dbg [[DBG277:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.S2** [[THIS_ADDR]], metadata [[META275:![0-9]+]], metadata !DIExpression()), !dbg [[DBG276:![0-9]+]]
 // SIMD2-NEXT:    [[THIS1:%.*]] = load %struct.S2*, %struct.S2** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_S2:%.*]], %struct.S2* [[THIS1]], i32 0, i32 0, !dbg [[DBG278:![0-9]+]]
-// SIMD2-NEXT:    store i32 0, i32* [[A]], align 8, !dbg [[DBG280:![0-9]+]]
-// SIMD2-NEXT:    ret void, !dbg [[DBG281:![0-9]+]]
+// SIMD2-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_S2:%.*]], %struct.S2* [[THIS1]], i32 0, i32 0, !dbg [[DBG277:![0-9]+]]
+// SIMD2-NEXT:    store i32 0, i32* [[A]], align 8, !dbg [[DBG279:![0-9]+]]
+// SIMD2-NEXT:    ret void, !dbg [[DBG280:![0-9]+]]
 //
 //
 // SIMD2-LABEL: define {{[^@]+}}@_ZZ4mainEN5SmainC2Ei
-// SIMD2-SAME: (%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR2]] align 2 !dbg [[DBG282:![0-9]+]] {
+// SIMD2-SAME: (%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR2]] align 2 !dbg [[DBG281:![0-9]+]] {
 // SIMD2-NEXT:  entry:
 // SIMD2-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.Smain*, align 8
 // SIMD2-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
 // SIMD2-NEXT:    store %struct.Smain* [[THIS]], %struct.Smain** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.Smain** [[THIS_ADDR]], metadata [[META283:![0-9]+]], metadata !DIExpression()), !dbg [[DBG284:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.Smain** [[THIS_ADDR]], metadata [[META282:![0-9]+]], metadata !DIExpression()), !dbg [[DBG283:![0-9]+]]
 // SIMD2-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META285:![0-9]+]], metadata !DIExpression()), !dbg [[DBG286:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META284:![0-9]+]], metadata !DIExpression()), !dbg [[DBG285:![0-9]+]]
 // SIMD2-NEXT:    [[THIS1:%.*]] = load %struct.Smain*, %struct.Smain** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    [[A2:%.*]] = getelementptr inbounds [[STRUCT_SMAIN:%.*]], %struct.Smain* [[THIS1]], i32 0, i32 0, !dbg [[DBG287:![0-9]+]]
-// SIMD2-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG288:![0-9]+]]
-// SIMD2-NEXT:    store i32 [[TMP0]], i32* [[A2]], align 8, !dbg [[DBG287]]
-// SIMD2-NEXT:    ret void, !dbg [[DBG289:![0-9]+]]
+// SIMD2-NEXT:    [[A2:%.*]] = getelementptr inbounds [[STRUCT_SMAIN:%.*]], %struct.Smain* [[THIS1]], i32 0, i32 0, !dbg [[DBG286:![0-9]+]]
+// SIMD2-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG287:![0-9]+]]
+// SIMD2-NEXT:    store i32 [[TMP0]], i32* [[A2]], align 8, !dbg [[DBG286]]
+// SIMD2-NEXT:    ret void, !dbg [[DBG288:![0-9]+]]
 //
 //
 // SIMD2-LABEL: define {{[^@]+}}@_ZZ4mainEN5SmainD2Ev
-// SIMD2-SAME: (%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] align 2 !dbg [[DBG290:![0-9]+]] {
+// SIMD2-SAME: (%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] align 2 !dbg [[DBG289:![0-9]+]] {
 // SIMD2-NEXT:  entry:
 // SIMD2-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.Smain*, align 8
 // SIMD2-NEXT:    store %struct.Smain* [[THIS]], %struct.Smain** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.Smain** [[THIS_ADDR]], metadata [[META291:![0-9]+]], metadata !DIExpression()), !dbg [[DBG292:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.Smain** [[THIS_ADDR]], metadata [[META290:![0-9]+]], metadata !DIExpression()), !dbg [[DBG291:![0-9]+]]
 // SIMD2-NEXT:    [[THIS1:%.*]] = load %struct.Smain*, %struct.Smain** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_SMAIN:%.*]], %struct.Smain* [[THIS1]], i32 0, i32 0, !dbg [[DBG293:![0-9]+]]
-// SIMD2-NEXT:    store i32 0, i32* [[A]], align 8, !dbg [[DBG295:![0-9]+]]
-// SIMD2-NEXT:    ret void, !dbg [[DBG296:![0-9]+]]
+// SIMD2-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_SMAIN:%.*]], %struct.Smain* [[THIS1]], i32 0, i32 0, !dbg [[DBG292:![0-9]+]]
+// SIMD2-NEXT:    store i32 0, i32* [[A]], align 8, !dbg [[DBG294:![0-9]+]]
+// SIMD2-NEXT:    ret void, !dbg [[DBG295:![0-9]+]]
 //
 //
 // SIMD2-LABEL: define {{[^@]+}}@_ZN2S4C2Ei
-// SIMD2-SAME: (%struct.S4* nonnull align 4 dereferenceable(8) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG297:![0-9]+]] {
+// SIMD2-SAME: (%struct.S4* nonnull align 4 dereferenceable(8) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG296:![0-9]+]] {
 // SIMD2-NEXT:  entry:
 // SIMD2-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S4*, align 8
 // SIMD2-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
 // SIMD2-NEXT:    store %struct.S4* [[THIS]], %struct.S4** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.S4** [[THIS_ADDR]], metadata [[META298:![0-9]+]], metadata !DIExpression()), !dbg [[DBG299:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.S4** [[THIS_ADDR]], metadata [[META297:![0-9]+]], metadata !DIExpression()), !dbg [[DBG298:![0-9]+]]
 // SIMD2-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META300:![0-9]+]], metadata !DIExpression()), !dbg [[DBG301:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META299:![0-9]+]], metadata !DIExpression()), !dbg [[DBG300:![0-9]+]]
 // SIMD2-NEXT:    [[THIS1:%.*]] = load %struct.S4*, %struct.S4** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    [[A2:%.*]] = getelementptr inbounds [[STRUCT_S4:%.*]], %struct.S4* [[THIS1]], i32 0, i32 0, !dbg [[DBG302:![0-9]+]]
-// SIMD2-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG303:![0-9]+]]
-// SIMD2-NEXT:    store i32 [[TMP0]], i32* [[A2]], align 4, !dbg [[DBG302]]
-// SIMD2-NEXT:    ret void, !dbg [[DBG304:![0-9]+]]
+// SIMD2-NEXT:    [[A2:%.*]] = getelementptr inbounds [[STRUCT_S4:%.*]], %struct.S4* [[THIS1]], i32 0, i32 0, !dbg [[DBG301:![0-9]+]]
+// SIMD2-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG302:![0-9]+]]
+// SIMD2-NEXT:    store i32 [[TMP0]], i32* [[A2]], align 4, !dbg [[DBG301]]
+// SIMD2-NEXT:    ret void, !dbg [[DBG303:![0-9]+]]
 //
 //
 // SIMD2-LABEL: define {{[^@]+}}@_ZN2S4D2Ev
-// SIMD2-SAME: (%struct.S4* nonnull align 4 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG305:![0-9]+]] {
+// SIMD2-SAME: (%struct.S4* nonnull align 4 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG304:![0-9]+]] {
 // SIMD2-NEXT:  entry:
 // SIMD2-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S4*, align 8
 // SIMD2-NEXT:    store %struct.S4* [[THIS]], %struct.S4** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.S4** [[THIS_ADDR]], metadata [[META306:![0-9]+]], metadata !DIExpression()), !dbg [[DBG307:![0-9]+]]
+// SIMD2-NEXT:    call void @llvm.dbg.declare(metadata %struct.S4** [[THIS_ADDR]], metadata [[META305:![0-9]+]], metadata !DIExpression()), !dbg [[DBG306:![0-9]+]]
 // SIMD2-NEXT:    [[THIS1:%.*]] = load %struct.S4*, %struct.S4** [[THIS_ADDR]], align 8
-// SIMD2-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_S4:%.*]], %struct.S4* [[THIS1]], i32 0, i32 0, !dbg [[DBG308:![0-9]+]]
-// SIMD2-NEXT:    store i32 0, i32* [[A]], align 4, !dbg [[DBG310:![0-9]+]]
-// SIMD2-NEXT:    ret void, !dbg [[DBG311:![0-9]+]]
+// SIMD2-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_S4:%.*]], %struct.S4* [[THIS1]], i32 0, i32 0, !dbg [[DBG307:![0-9]+]]
+// SIMD2-NEXT:    store i32 0, i32* [[A]], align 4, !dbg [[DBG309:![0-9]+]]
+// SIMD2-NEXT:    ret void, !dbg [[DBG310:![0-9]+]]
 //
 //
 // SIMD2-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_threadprivate_codegen.cpp
-// SIMD2-SAME: () #[[ATTR0]] !dbg [[DBG312:![0-9]+]] {
+// SIMD2-SAME: () #[[ATTR0]] !dbg [[DBG311:![0-9]+]] {
 // SIMD2-NEXT:  entry:
-// SIMD2-NEXT:    call void @__cxx_global_var_init(), !dbg [[DBG314:![0-9]+]]
-// SIMD2-NEXT:    call void @__cxx_global_var_init.1(), !dbg [[DBG314]]
-// SIMD2-NEXT:    call void @__cxx_global_var_init.2(), !dbg [[DBG314]]
+// SIMD2-NEXT:    call void @__cxx_global_var_init(), !dbg [[DBG313:![0-9]+]]
+// SIMD2-NEXT:    call void @__cxx_global_var_init.1(), !dbg [[DBG313]]
+// SIMD2-NEXT:    call void @__cxx_global_var_init.2(), !dbg [[DBG313]]
 // SIMD2-NEXT:    ret void
 //
 //
 // CHECK-TLS1-LABEL: define {{[^@]+}}@__cxx_global_var_init
 // CHECK-TLS1-SAME: () #[[ATTR0:[0-9]+]] {
 // CHECK-TLS1-NEXT:  entry:
 // CHECK-TLS1-NEXT:    call void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) @_ZL3gs1, i32 5)
 // CHECK-TLS1-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_thread_atexit(void (i8*)* bitcast (void (%struct.S1*)* @_ZN2S1D1Ev to void (i8*)*), i8* bitcast (%struct.S1* @_ZL3gs1 to i8*), i8* @__dso_handle) #[[ATTR3:[0-9]+]]

 //
 // CHECK-TLS1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK-TLS1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // CHECK-TLS1-NEXT:  entry:
 // CHECK-TLS1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK-TLS1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK-TLS1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK-TLS1:       arraydestroy.body:
-// CHECK-TLS1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ getelementptr inbounds ([[STRUCT_S1:%.*]], %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0), i64 6), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK-TLS1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK-TLS1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 1, i64 0, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK-TLS1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK-TLS1-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
 // CHECK-TLS1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0)
 // CHECK-TLS1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK-TLS1:       arraydestroy.done1:
 // CHECK-TLS1-NEXT:    ret void
 //
 //
 // CHECK-TLS1-LABEL: define {{[^@]+}}@main
 // CHECK-TLS1-SAME: () #[[ATTR4:[0-9]+]] {
 // CHECK-TLS1-NEXT:  entry:
 // CHECK-TLS1-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
 // CHECK-TLS1-NEXT:    [[RES:%.*]] = alloca i32, align 4
 // CHECK-TLS1-NEXT:    store i32 0, i32* [[RETVAL]], align 4
 // CHECK-TLS1-NEXT:    [[TMP0:%.*]] = load i8, i8* @_ZGVZ4mainE2sm, align 1
 // CHECK-TLS1-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0
-// CHECK-TLS1-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF2:![0-9]+]]
+// CHECK-TLS1-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF3:![0-9]+]]
 // CHECK-TLS1:       init.check:
 // CHECK-TLS1-NEXT:    [[TMP1:%.*]] = call %struct.S1* @_ZTWL3gs1()
 // CHECK-TLS1-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[TMP1]], i32 0, i32 0
 // CHECK-TLS1-NEXT:    [[TMP2:%.*]] = load i32, i32* [[A]], align 4
 // CHECK-TLS1-NEXT:    call void @_ZZ4mainEN5SmainC1Ei(%struct.Smain* nonnull align 8 dereferenceable(24) @_ZZ4mainE2sm, i32 [[TMP2]])
 // CHECK-TLS1-NEXT:    [[TMP3:%.*]] = call i32 @__cxa_thread_atexit(void (i8*)* bitcast (void (%struct.Smain*)* @_ZZ4mainEN5SmainD1Ev to void (i8*)*), i8* bitcast (%struct.Smain* @_ZZ4mainE2sm to i8*), i8* @__dso_handle) #[[ATTR3]]
 // CHECK-TLS1-NEXT:    store i8 1, i8* @_ZGVZ4mainE2sm, align 1
 // CHECK-TLS1-NEXT:    br label [[INIT_END]]

 // CHECK-TLS1-NEXT:    ret void
 //
 //
 // CHECK-TLS1-LABEL: define {{[^@]+}}@__tls_init
 // CHECK-TLS1-SAME: () #[[ATTR0]] {
 // CHECK-TLS1-NEXT:  entry:
 // CHECK-TLS1-NEXT:    [[TMP0:%.*]] = load i8, i8* @__tls_guard, align 1
 // CHECK-TLS1-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0
-// CHECK-TLS1-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT:%.*]], label [[EXIT:%.*]], !prof [[PROF2]]
+// CHECK-TLS1-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT:%.*]], label [[EXIT:%.*]], !prof [[PROF3]]
 // CHECK-TLS1:       init:
 // CHECK-TLS1-NEXT:    store i8 1, i8* @__tls_guard, align 1
 // CHECK-TLS1-NEXT:    call void @__cxx_global_var_init()
 // CHECK-TLS1-NEXT:    call void @__cxx_global_var_init.2()
 // CHECK-TLS1-NEXT:    br label [[EXIT]]
 // CHECK-TLS1:       exit:
 // CHECK-TLS1-NEXT:    ret void
 //
 //
 // CHECK-TLS2-LABEL: define {{[^@]+}}@main
 // CHECK-TLS2-SAME: () #[[ATTR0:[0-9]+]] {
 // CHECK-TLS2-NEXT:  entry:
 // CHECK-TLS2-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
 // CHECK-TLS2-NEXT:    [[RES:%.*]] = alloca i32, align 4
 // CHECK-TLS2-NEXT:    store i32 0, i32* [[RETVAL]], align 4
 // CHECK-TLS2-NEXT:    [[TMP0:%.*]] = load i8, i8* @_ZGVZ4mainE2sm, align 1
 // CHECK-TLS2-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0
-// CHECK-TLS2-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF2:![0-9]+]]
+// CHECK-TLS2-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !prof [[PROF3:![0-9]+]]
 // CHECK-TLS2:       init.check:
 // CHECK-TLS2-NEXT:    [[TMP1:%.*]] = call %struct.S1* @_ZTWL3gs1()
 // CHECK-TLS2-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[TMP1]], i32 0, i32 0
 // CHECK-TLS2-NEXT:    [[TMP2:%.*]] = load i32, i32* [[A]], align 4
 // CHECK-TLS2-NEXT:    call void @_ZZ4mainEN5SmainC1Ei(%struct.Smain* nonnull align 8 dereferenceable(24) @_ZZ4mainE2sm, i32 [[TMP2]])
 // CHECK-TLS2-NEXT:    [[TMP3:%.*]] = call i32 @__cxa_thread_atexit(void (i8*)* bitcast (void (%struct.Smain*)* @_ZZ4mainEN5SmainD1Ev to void (i8*)*), i8* bitcast (%struct.Smain* @_ZZ4mainE2sm to i8*), i8* @__dso_handle) #[[ATTR4:[0-9]+]]
 // CHECK-TLS2-NEXT:    store i8 1, i8* @_ZGVZ4mainE2sm, align 1
 // CHECK-TLS2-NEXT:    br label [[INIT_END]]

 //
 // CHECK-TLS2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK-TLS2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR6]] {
 // CHECK-TLS2-NEXT:  entry:
 // CHECK-TLS2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK-TLS2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK-TLS2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // CHECK-TLS2:       arraydestroy.body:
-// CHECK-TLS2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ getelementptr inbounds ([[STRUCT_S1:%.*]], %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0), i64 6), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// CHECK-TLS2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// CHECK-TLS2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 1, i64 0, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// CHECK-TLS2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // CHECK-TLS2-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]]
 // CHECK-TLS2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0)
 // CHECK-TLS2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // CHECK-TLS2:       arraydestroy.done1:
 // CHECK-TLS2-NEXT:    ret void
 //
 //
 // CHECK-TLS2-LABEL: define {{[^@]+}}@_ZZ4mainEN5SmainC2Ei

 // CHECK-TLS2-NEXT:    ret void
 //
 //
 // CHECK-TLS2-LABEL: define {{[^@]+}}@__tls_init
 // CHECK-TLS2-SAME: () #[[ATTR6]] {
 // CHECK-TLS2-NEXT:  entry:
 // CHECK-TLS2-NEXT:    [[TMP0:%.*]] = load i8, i8* @__tls_guard, align 1
 // CHECK-TLS2-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0
-// CHECK-TLS2-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT:%.*]], label [[EXIT:%.*]], !prof [[PROF2]]
+// CHECK-TLS2-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT:%.*]], label [[EXIT:%.*]], !prof [[PROF3]]
 // CHECK-TLS2:       init:
 // CHECK-TLS2-NEXT:    store i8 1, i8* @__tls_guard, align 1
 // CHECK-TLS2-NEXT:    call void @__cxx_global_var_init()
 // CHECK-TLS2-NEXT:    call void @__cxx_global_var_init.2()
 // CHECK-TLS2-NEXT:    br label [[EXIT]]
 // CHECK-TLS2:       exit:
 // CHECK-TLS2-NEXT:    ret void
 //
 //
 // CHECK-TLS3-LABEL: define {{[^@]+}}@__cxx_global_var_init
-// CHECK-TLS3-SAME: () #[[ATTR0:[0-9]+]] !dbg [[DBG115:![0-9]+]] {
+// CHECK-TLS3-SAME: () #[[ATTR0:[0-9]+]] !dbg [[DBG116:![0-9]+]] {
 // CHECK-TLS3-NEXT:  entry:
 // CHECK-TLS3-NEXT:    call void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) @_ZL3gs1, i32 5), !dbg [[DBG119:![0-9]+]]
 // CHECK-TLS3-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_thread_atexit(void (i8*)* bitcast (void (%struct.S1*)* @_ZN2S1D1Ev to void (i8*)*), i8* bitcast (%struct.S1* @_ZL3gs1 to i8*), i8* @__dso_handle) #[[ATTR3:[0-9]+]], !dbg [[DBG121:![0-9]+]]
 // CHECK-TLS3-NEXT:    ret void, !dbg [[DBG122:![0-9]+]]
 //
 //
 // CHECK-TLS3-LABEL: define {{[^@]+}}@_ZN2S1C1Ei
 // CHECK-TLS3-SAME: (%struct.S1* nonnull align 4 dereferenceable(4) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR1:[0-9]+]] comdat align 2 !dbg [[DBG123:![0-9]+]] {

 // CHECK-TLS3-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK-TLS3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] !dbg [[DBG197:![0-9]+]] {
 // CHECK-TLS3-NEXT:  entry:
 // CHECK-TLS3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK-TLS3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK-TLS3-NEXT:    call void @llvm.dbg.declare(metadata i8** [[DOTADDR]], metadata [[META201:![0-9]+]], metadata !DIExpression()), !dbg [[DBG202:![0-9]+]]
 // CHECK-TLS3-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]], !dbg [[DBG202]]
 // CHECK-TLS3:       arraydestroy.body:
-// CHECK-TLS3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ getelementptr inbounds ([[STRUCT_S1:%.*]], %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0), i64 6), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG202]]
-// CHECK-TLS3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG202]]
+// CHECK-TLS3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 1, i64 0, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG202]]
+// CHECK-TLS3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG202]]
 // CHECK-TLS3-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]], !dbg [[DBG202]]
 // CHECK-TLS3-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0), !dbg [[DBG202]]
 // CHECK-TLS3-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]], !dbg [[DBG202]]
 // CHECK-TLS3:       arraydestroy.done1:
 // CHECK-TLS3-NEXT:    ret void, !dbg [[DBG202]]
 //
 //
 // CHECK-TLS3-LABEL: define {{[^@]+}}@main
-// CHECK-TLS3-SAME: () #[[ATTR5:[0-9]+]] !dbg [[DBG53:![0-9]+]] {
+// CHECK-TLS3-SAME: () #[[ATTR5:[0-9]+]] !dbg [[DBG52:![0-9]+]] {
 // CHECK-TLS3-NEXT:  entry:
 // CHECK-TLS3-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
 // CHECK-TLS3-NEXT:    [[RES:%.*]] = alloca i32, align 4
 // CHECK-TLS3-NEXT:    store i32 0, i32* [[RETVAL]], align 4
 // CHECK-TLS3-NEXT:    call void @llvm.dbg.declare(metadata i32* [[RES]], metadata [[META203:![0-9]+]], metadata !DIExpression()), !dbg [[DBG204:![0-9]+]]
 // CHECK-TLS3-NEXT:    [[TMP0:%.*]] = load i8, i8* @_ZGVZ4mainE2sm, align 1, !dbg [[DBG205:![0-9]+]]
 // CHECK-TLS3-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0, !dbg [[DBG205]]
 // CHECK-TLS3-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !dbg [[DBG205]], !prof [[PROF206:![0-9]+]]

 // CHECK-TLS3-NEXT:    call void @__cxx_global_var_init(), !dbg [[DBG326]]
 // CHECK-TLS3-NEXT:    call void @__cxx_global_var_init.2(), !dbg [[DBG326]]
 // CHECK-TLS3-NEXT:    br label [[EXIT]], !dbg [[DBG326]]
 // CHECK-TLS3:       exit:
 // CHECK-TLS3-NEXT:    ret void
 //
 //
 // CHECK-TLS4-LABEL: define {{[^@]+}}@main
-// CHECK-TLS4-SAME: () #[[ATTR0:[0-9]+]] !dbg [[DBG10:![0-9]+]] {
+// CHECK-TLS4-SAME: () #[[ATTR0:[0-9]+]] !dbg [[DBG9:![0-9]+]] {
 // CHECK-TLS4-NEXT:  entry:
 // CHECK-TLS4-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
 // CHECK-TLS4-NEXT:    [[RES:%.*]] = alloca i32, align 4
 // CHECK-TLS4-NEXT:    store i32 0, i32* [[RETVAL]], align 4
-// CHECK-TLS4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[RES]], metadata [[META115:![0-9]+]], metadata !DIExpression()), !dbg [[DBG116:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[TMP0:%.*]] = load i8, i8* @_ZGVZ4mainE2sm, align 1, !dbg [[DBG117:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0, !dbg [[DBG117]]
-// CHECK-TLS4-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !dbg [[DBG117]], !prof [[PROF118:![0-9]+]]
+// CHECK-TLS4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[RES]], metadata [[META116:![0-9]+]], metadata !DIExpression()), !dbg [[DBG117:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[TMP0:%.*]] = load i8, i8* @_ZGVZ4mainE2sm, align 1, !dbg [[DBG118:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0, !dbg [[DBG118]]
+// CHECK-TLS4-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !dbg [[DBG118]], !prof [[PROF119:![0-9]+]]
 // CHECK-TLS4:       init.check:
-// CHECK-TLS4-NEXT:    [[TMP1:%.*]] = call %struct.S1* @_ZTWL3gs1(), !dbg [[DBG119:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[TMP1]], i32 0, i32 0, !dbg [[DBG120:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[TMP2:%.*]] = load i32, i32* [[A]], align 4, !dbg [[DBG120]]
-// CHECK-TLS4-NEXT:    call void @_ZZ4mainEN5SmainC1Ei(%struct.Smain* nonnull align 8 dereferenceable(24) @_ZZ4mainE2sm, i32 [[TMP2]]), !dbg [[DBG121:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[TMP3:%.*]] = call i32 @__cxa_thread_atexit(void (i8*)* bitcast (void (%struct.Smain*)* @_ZZ4mainEN5SmainD1Ev to void (i8*)*), i8* bitcast (%struct.Smain* @_ZZ4mainE2sm to i8*), i8* @__dso_handle) #[[ATTR5:[0-9]+]], !dbg [[DBG117]]
-// CHECK-TLS4-NEXT:    store i8 1, i8* @_ZGVZ4mainE2sm, align 1, !dbg [[DBG117]]
-// CHECK-TLS4-NEXT:    br label [[INIT_END]], !dbg [[DBG117]]
+// CHECK-TLS4-NEXT:    [[TMP1:%.*]] = call %struct.S1* @_ZTWL3gs1(), !dbg [[DBG120:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[TMP1]], i32 0, i32 0, !dbg [[DBG121:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[TMP2:%.*]] = load i32, i32* [[A]], align 4, !dbg [[DBG121]]
+// CHECK-TLS4-NEXT:    call void @_ZZ4mainEN5SmainC1Ei(%struct.Smain* nonnull align 8 dereferenceable(24) @_ZZ4mainE2sm, i32 [[TMP2]]), !dbg [[DBG122:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[TMP3:%.*]] = call i32 @__cxa_thread_atexit(void (i8*)* bitcast (void (%struct.Smain*)* @_ZZ4mainEN5SmainD1Ev to void (i8*)*), i8* bitcast (%struct.Smain* @_ZZ4mainE2sm to i8*), i8* @__dso_handle) #[[ATTR5:[0-9]+]], !dbg [[DBG118]]
+// CHECK-TLS4-NEXT:    store i8 1, i8* @_ZGVZ4mainE2sm, align 1, !dbg [[DBG118]]
+// CHECK-TLS4-NEXT:    br label [[INIT_END]], !dbg [[DBG118]]
 // CHECK-TLS4:       init.end:
-// CHECK-TLS4-NEXT:    [[TMP4:%.*]] = call %struct.S3* @_ZTWN6Static1sE(), !dbg [[DBG122:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[A1:%.*]] = getelementptr inbounds [[STRUCT_S3:%.*]], %struct.S3* [[TMP4]], i32 0, i32 0, !dbg [[DBG123:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[A1]], align 4, !dbg [[DBG123]]
-// CHECK-TLS4-NEXT:    store i32 [[TMP5]], i32* [[RES]], align 4, !dbg [[DBG124:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[TMP6:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_SMAIN:%.*]], %struct.Smain* @_ZZ4mainE2sm, i32 0, i32 0), align 8, !dbg [[DBG125:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG126:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP7]], [[TMP6]], !dbg [[DBG126]]
-// CHECK-TLS4-NEXT:    store i32 [[ADD]], i32* [[RES]], align 4, !dbg [[DBG126]]
-// CHECK-TLS4-NEXT:    [[TMP8:%.*]] = call %struct.S1* @_ZTWL3gs1(), !dbg [[DBG127:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[A2:%.*]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[TMP8]], i32 0, i32 0, !dbg [[DBG128:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[A2]], align 4, !dbg [[DBG128]]
-// CHECK-TLS4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG129:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP10]], [[TMP9]], !dbg [[DBG129]]
-// CHECK-TLS4-NEXT:    store i32 [[ADD3]], i32* [[RES]], align 4, !dbg [[DBG129]]
-// CHECK-TLS4-NEXT:    [[TMP11:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S2:%.*]], %struct.S2* @_ZL3gs2, i32 0, i32 0), align 8, !dbg [[DBG130:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[TMP12:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG131:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP12]], [[TMP11]], !dbg [[DBG131]]
-// CHECK-TLS4-NEXT:    store i32 [[ADD4]], i32* [[RES]], align 4, !dbg [[DBG131]]
-// CHECK-TLS4-NEXT:    [[TMP13:%.*]] = call %struct.S5* @_ZTW3gs3(), !dbg [[DBG132:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[A5:%.*]] = getelementptr inbounds [[STRUCT_S5:%.*]], %struct.S5* [[TMP13]], i32 0, i32 0, !dbg [[DBG133:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[TMP14:%.*]] = load i32, i32* [[A5]], align 4, !dbg [[DBG133]]
-// CHECK-TLS4-NEXT:    [[TMP15:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG134:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP15]], [[TMP14]], !dbg [[DBG134]]
-// CHECK-TLS4-NEXT:    store i32 [[ADD6]], i32* [[RES]], align 4, !dbg [[DBG134]]
-// CHECK-TLS4-NEXT:    [[TMP16:%.*]] = call [2 x [3 x %struct.S1]]* @_ZTW5arr_x(), !dbg [[DBG135:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* [[TMP16]], i64 0, i64 1, !dbg [[DBG135]]
-// CHECK-TLS4-NEXT:    [[ARRAYIDX7:%.*]] = getelementptr inbounds [3 x %struct.S1], [3 x %struct.S1]* [[ARRAYIDX]], i64 0, i64 1, !dbg [[DBG135]]
-// CHECK-TLS4-NEXT:    [[A8:%.*]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYIDX7]], i32 0, i32 0, !dbg [[DBG136:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[TMP17:%.*]] = load i32, i32* [[A8]], align 4, !dbg [[DBG136]]
-// CHECK-TLS4-NEXT:    [[TMP18:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG137:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP18]], [[TMP17]], !dbg [[DBG137]]
-// CHECK-TLS4-NEXT:    store i32 [[ADD9]], i32* [[RES]], align 4, !dbg [[DBG137]]
-// CHECK-TLS4-NEXT:    [[TMP19:%.*]] = call i32* @_ZTWN2STIiE2stE(), !dbg [[DBG138:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[TMP20:%.*]] = load i32, i32* [[TMP19]], align 4, !dbg [[DBG138]]
-// CHECK-TLS4-NEXT:    [[TMP21:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG139:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP21]], [[TMP20]], !dbg [[DBG139]]
-// CHECK-TLS4-NEXT:    store i32 [[ADD10]], i32* [[RES]], align 4, !dbg [[DBG139]]
-// CHECK-TLS4-NEXT:    [[TMP22:%.*]] = call float* @_ZTWN2STIfE2stE(), !dbg [[DBG140:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[TMP23:%.*]] = load float, float* [[TMP22]], align 4, !dbg [[DBG140]]
-// CHECK-TLS4-NEXT:    [[CONV:%.*]] = fptosi float [[TMP23]] to i32, !dbg [[DBG140]]
-// CHECK-TLS4-NEXT:    [[TMP24:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG141:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP24]], [[CONV]], !dbg [[DBG141]]
-// CHECK-TLS4-NEXT:    store i32 [[ADD11]], i32* [[RES]], align 4, !dbg [[DBG141]]
-// CHECK-TLS4-NEXT:    [[TMP25:%.*]] = call %struct.S4* @_ZTWN2STI2S4E2stE(), !dbg [[DBG142:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[A12:%.*]] = getelementptr inbounds [[STRUCT_S4:%.*]], %struct.S4* [[TMP25]], i32 0, i32 0, !dbg [[DBG143:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[TMP26:%.*]] = load i32, i32* [[A12]], align 4, !dbg [[DBG143]]
-// CHECK-TLS4-NEXT:    [[TMP27:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG144:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[ADD13:%.*]] = add nsw i32 [[TMP27]], [[TMP26]], !dbg [[DBG144]]
-// CHECK-TLS4-NEXT:    store i32 [[ADD13]], i32* [[RES]], align 4, !dbg [[DBG144]]
-// CHECK-TLS4-NEXT:    [[TMP28:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG145:![0-9]+]]
-// CHECK-TLS4-NEXT:    ret i32 [[TMP28]], !dbg [[DBG146:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[TMP4:%.*]] = call %struct.S3* @_ZTWN6Static1sE(), !dbg [[DBG123:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[A1:%.*]] = getelementptr inbounds [[STRUCT_S3:%.*]], %struct.S3* [[TMP4]], i32 0, i32 0, !dbg [[DBG124:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[A1]], align 4, !dbg [[DBG124]]
+// CHECK-TLS4-NEXT:    store i32 [[TMP5]], i32* [[RES]], align 4, !dbg [[DBG125:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[TMP6:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_SMAIN:%.*]], %struct.Smain* @_ZZ4mainE2sm, i32 0, i32 0), align 8, !dbg [[DBG126:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG127:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP7]], [[TMP6]], !dbg [[DBG127]]
+// CHECK-TLS4-NEXT:    store i32 [[ADD]], i32* [[RES]], align 4, !dbg [[DBG127]]
+// CHECK-TLS4-NEXT:    [[TMP8:%.*]] = call %struct.S1* @_ZTWL3gs1(), !dbg [[DBG128:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[A2:%.*]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[TMP8]], i32 0, i32 0, !dbg [[DBG129:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[A2]], align 4, !dbg [[DBG129]]
+// CHECK-TLS4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG130:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP10]], [[TMP9]], !dbg [[DBG130]]
+// CHECK-TLS4-NEXT:    store i32 [[ADD3]], i32* [[RES]], align 4, !dbg [[DBG130]]
+// CHECK-TLS4-NEXT:    [[TMP11:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S2:%.*]], %struct.S2* @_ZL3gs2, i32 0, i32 0), align 8, !dbg [[DBG131:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[TMP12:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG132:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP12]], [[TMP11]], !dbg [[DBG132]]
+// CHECK-TLS4-NEXT:    store i32 [[ADD4]], i32* [[RES]], align 4, !dbg [[DBG132]]
+// CHECK-TLS4-NEXT:    [[TMP13:%.*]] = call %struct.S5* @_ZTW3gs3(), !dbg [[DBG133:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[A5:%.*]] = getelementptr inbounds [[STRUCT_S5:%.*]], %struct.S5* [[TMP13]], i32 0, i32 0, !dbg [[DBG134:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[TMP14:%.*]] = load i32, i32* [[A5]], align 4, !dbg [[DBG134]]
+// CHECK-TLS4-NEXT:    [[TMP15:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG135:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP15]], [[TMP14]], !dbg [[DBG135]]
+// CHECK-TLS4-NEXT:    store i32 [[ADD6]], i32* [[RES]], align 4, !dbg [[DBG135]]
+// CHECK-TLS4-NEXT:    [[TMP16:%.*]] = call [2 x [3 x %struct.S1]]* @_ZTW5arr_x(), !dbg [[DBG136:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* [[TMP16]], i64 0, i64 1, !dbg [[DBG136]]
+// CHECK-TLS4-NEXT:    [[ARRAYIDX7:%.*]] = getelementptr inbounds [3 x %struct.S1], [3 x %struct.S1]* [[ARRAYIDX]], i64 0, i64 1, !dbg [[DBG136]]
+// CHECK-TLS4-NEXT:    [[A8:%.*]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYIDX7]], i32 0, i32 0, !dbg [[DBG137:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[TMP17:%.*]] = load i32, i32* [[A8]], align 4, !dbg [[DBG137]]
+// CHECK-TLS4-NEXT:    [[TMP18:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG138:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP18]], [[TMP17]], !dbg [[DBG138]]
+// CHECK-TLS4-NEXT:    store i32 [[ADD9]], i32* [[RES]], align 4, !dbg [[DBG138]]
+// CHECK-TLS4-NEXT:    [[TMP19:%.*]] = call i32* @_ZTWN2STIiE2stE(), !dbg [[DBG139:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[TMP20:%.*]] = load i32, i32* [[TMP19]], align 4, !dbg [[DBG139]]
+// CHECK-TLS4-NEXT:    [[TMP21:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG140:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP21]], [[TMP20]], !dbg [[DBG140]]
+// CHECK-TLS4-NEXT:    store i32 [[ADD10]], i32* [[RES]], align 4, !dbg [[DBG140]]
+// CHECK-TLS4-NEXT:    [[TMP22:%.*]] = call float* @_ZTWN2STIfE2stE(), !dbg [[DBG141:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[TMP23:%.*]] = load float, float* [[TMP22]], align 4, !dbg [[DBG141]]
+// CHECK-TLS4-NEXT:    [[CONV:%.*]] = fptosi float [[TMP23]] to i32, !dbg [[DBG141]]
+// CHECK-TLS4-NEXT:    [[TMP24:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG142:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP24]], [[CONV]], !dbg [[DBG142]]
+// CHECK-TLS4-NEXT:    store i32 [[ADD11]], i32* [[RES]], align 4, !dbg [[DBG142]]
+// CHECK-TLS4-NEXT:    [[TMP25:%.*]] = call %struct.S4* @_ZTWN2STI2S4E2stE(), !dbg [[DBG143:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[A12:%.*]] = getelementptr inbounds [[STRUCT_S4:%.*]], %struct.S4* [[TMP25]], i32 0, i32 0, !dbg [[DBG144:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[TMP26:%.*]] = load i32, i32* [[A12]], align 4, !dbg [[DBG144]]
+// CHECK-TLS4-NEXT:    [[TMP27:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG145:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[ADD13:%.*]] = add nsw i32 [[TMP27]], [[TMP26]], !dbg [[DBG145]]
+// CHECK-TLS4-NEXT:    store i32 [[ADD13]], i32* [[RES]], align 4, !dbg [[DBG145]]
+// CHECK-TLS4-NEXT:    [[TMP28:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG146:![0-9]+]]
+// CHECK-TLS4-NEXT:    ret i32 [[TMP28]], !dbg [[DBG147:![0-9]+]]
 //
 //
 // CHECK-TLS4-LABEL: define {{[^@]+}}@_ZTWL3gs1
 // CHECK-TLS4-SAME: () #[[ATTR2:[0-9]+]] {
 // CHECK-TLS4-NEXT:    call void @_ZTHL3gs1()
 // CHECK-TLS4-NEXT:    ret %struct.S1* @_ZL3gs1
 //
 //
 // CHECK-TLS4-LABEL: define {{[^@]+}}@_ZZ4mainEN5SmainC1Ei
-// CHECK-TLS4-SAME: (%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR3:[0-9]+]] align 2 !dbg [[DBG147:![0-9]+]] {
+// CHECK-TLS4-SAME: (%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR3:[0-9]+]] align 2 !dbg [[DBG148:![0-9]+]] {
 // CHECK-TLS4-NEXT:  entry:
 // CHECK-TLS4-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.Smain*, align 8
 // CHECK-TLS4-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
 // CHECK-TLS4-NEXT:    store %struct.Smain* [[THIS]], %struct.Smain** [[THIS_ADDR]], align 8
-// CHECK-TLS4-NEXT:    call void @llvm.dbg.declare(metadata %struct.Smain** [[THIS_ADDR]], metadata [[META148:![0-9]+]], metadata !DIExpression()), !dbg [[DBG150:![0-9]+]]
+// CHECK-TLS4-NEXT:    call void @llvm.dbg.declare(metadata %struct.Smain** [[THIS_ADDR]], metadata [[META149:![0-9]+]], metadata !DIExpression()), !dbg [[DBG151:![0-9]+]]
 // CHECK-TLS4-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
-// CHECK-TLS4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META151:![0-9]+]], metadata !DIExpression()), !dbg [[DBG152:![0-9]+]]
+// CHECK-TLS4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META152:![0-9]+]], metadata !DIExpression()), !dbg [[DBG153:![0-9]+]]
 // CHECK-TLS4-NEXT:    [[THIS1:%.*]] = load %struct.Smain*, %struct.Smain** [[THIS_ADDR]], align 8
-// CHECK-TLS4-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG153:![0-9]+]]
-// CHECK-TLS4-NEXT:    call void @_ZZ4mainEN5SmainC2Ei(%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS1]], i32 [[TMP0]]), !dbg [[DBG153]]
-// CHECK-TLS4-NEXT:    ret void, !dbg [[DBG154:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG154:![0-9]+]]
+// CHECK-TLS4-NEXT:    call void @_ZZ4mainEN5SmainC2Ei(%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS1]], i32 [[TMP0]]), !dbg [[DBG154]]
+// CHECK-TLS4-NEXT:    ret void, !dbg [[DBG155:![0-9]+]]
 //
 //
 // CHECK-TLS4-LABEL: define {{[^@]+}}@_ZZ4mainEN5SmainD1Ev
-// CHECK-TLS4-SAME: (%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS:%.*]]) unnamed_addr #[[ATTR4:[0-9]+]] align 2 !dbg [[DBG155:![0-9]+]] {
+// CHECK-TLS4-SAME: (%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS:%.*]]) unnamed_addr #[[ATTR4:[0-9]+]] align 2 !dbg [[DBG156:![0-9]+]] {
 // CHECK-TLS4-NEXT:  entry:
 // CHECK-TLS4-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.Smain*, align 8
 // CHECK-TLS4-NEXT:    store %struct.Smain* [[THIS]], %struct.Smain** [[THIS_ADDR]], align 8
-// CHECK-TLS4-NEXT:    call void @llvm.dbg.declare(metadata %struct.Smain** [[THIS_ADDR]], metadata [[META156:![0-9]+]], metadata !DIExpression()), !dbg [[DBG157:![0-9]+]]
+// CHECK-TLS4-NEXT:    call void @llvm.dbg.declare(metadata %struct.Smain** [[THIS_ADDR]], metadata [[META157:![0-9]+]], metadata !DIExpression()), !dbg [[DBG158:![0-9]+]]
 // CHECK-TLS4-NEXT:    [[THIS1:%.*]] = load %struct.Smain*, %struct.Smain** [[THIS_ADDR]], align 8
-// CHECK-TLS4-NEXT:    call void @_ZZ4mainEN5SmainD2Ev(%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS1]]) #[[ATTR5]], !dbg [[DBG158:![0-9]+]]
-// CHECK-TLS4-NEXT:    ret void, !dbg [[DBG159:![0-9]+]]
+// CHECK-TLS4-NEXT:    call void @_ZZ4mainEN5SmainD2Ev(%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS1]]) #[[ATTR5]], !dbg [[DBG159:![0-9]+]]
+// CHECK-TLS4-NEXT:    ret void, !dbg [[DBG160:![0-9]+]]
 //
 //
 // CHECK-TLS4-LABEL: define {{[^@]+}}@_ZTWN6Static1sE
 // CHECK-TLS4-SAME: () #[[ATTR2]] comdat {
 // CHECK-TLS4-NEXT:    br i1 icmp ne (void ()* @_ZTHN6Static1sE, void ()* null), label [[TMP1:%.*]], label [[TMP2:%.*]]
 // CHECK-TLS4:       1:
 // CHECK-TLS4-NEXT:    call void @_ZTHN6Static1sE()
 // CHECK-TLS4-NEXT:    br label [[TMP2]]

 //
 // CHECK-TLS4-LABEL: define {{[^@]+}}@_ZTWN2STI2S4E2stE
 // CHECK-TLS4-SAME: () #[[ATTR2]] comdat {
 // CHECK-TLS4-NEXT:    call void @_ZTHN2STI2S4E2stE()
 // CHECK-TLS4-NEXT:    ret %struct.S4* @_ZN2STI2S4E2stE
 //
 //
 // CHECK-TLS4-LABEL: define {{[^@]+}}@_Z6foobarv
-// CHECK-TLS4-SAME: () #[[ATTR6:[0-9]+]] !dbg [[DBG160:![0-9]+]] {
+// CHECK-TLS4-SAME: () #[[ATTR6:[0-9]+]] !dbg [[DBG161:![0-9]+]] {
 // CHECK-TLS4-NEXT:  entry:
 // CHECK-TLS4-NEXT:    [[RES:%.*]] = alloca i32, align 4
-// CHECK-TLS4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[RES]], metadata [[META161:![0-9]+]], metadata !DIExpression()), !dbg [[DBG162:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[TMP0:%.*]] = call %struct.S3* @_ZTWN6Static1sE(), !dbg [[DBG163:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_S3:%.*]], %struct.S3* [[TMP0]], i32 0, i32 0, !dbg [[DBG164:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[TMP1:%.*]] = load i32, i32* [[A]], align 4, !dbg [[DBG164]]
-// CHECK-TLS4-NEXT:    store i32 [[TMP1]], i32* [[RES]], align 4, !dbg [[DBG165:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[TMP2:%.*]] = call %struct.S1* @_ZTWL3gs1(), !dbg [[DBG166:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[A1:%.*]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[TMP2]], i32 0, i32 0, !dbg [[DBG167:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[TMP3:%.*]] = load i32, i32* [[A1]], align 4, !dbg [[DBG167]]
-// CHECK-TLS4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG168:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP4]], [[TMP3]], !dbg [[DBG168]]
-// CHECK-TLS4-NEXT:    store i32 [[ADD]], i32* [[RES]], align 4, !dbg [[DBG168]]
-// CHECK-TLS4-NEXT:    [[TMP5:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S2:%.*]], %struct.S2* @_ZL3gs2, i32 0, i32 0), align 8, !dbg [[DBG169:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG170:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP6]], [[TMP5]], !dbg [[DBG170]]
-// CHECK-TLS4-NEXT:    store i32 [[ADD2]], i32* [[RES]], align 4, !dbg [[DBG170]]
-// CHECK-TLS4-NEXT:    [[TMP7:%.*]] = call %struct.S5* @_ZTW3gs3(), !dbg [[DBG171:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[A3:%.*]] = getelementptr inbounds [[STRUCT_S5:%.*]], %struct.S5* [[TMP7]], i32 0, i32 0, !dbg [[DBG172:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[A3]], align 4, !dbg [[DBG172]]
-// CHECK-TLS4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG173:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP9]], [[TMP8]], !dbg [[DBG173]]
-// CHECK-TLS4-NEXT:    store i32 [[ADD4]], i32* [[RES]], align 4, !dbg [[DBG173]]
-// CHECK-TLS4-NEXT:    [[TMP10:%.*]] = call [2 x [3 x %struct.S1]]* @_ZTW5arr_x(), !dbg [[DBG174:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* [[TMP10]], i64 0, i64 1, !dbg [[DBG174]]
-// CHECK-TLS4-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds [3 x %struct.S1], [3 x %struct.S1]* [[ARRAYIDX]], i64 0, i64 1, !dbg [[DBG174]]
-// CHECK-TLS4-NEXT:    [[A6:%.*]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYIDX5]], i32 0, i32 0, !dbg [[DBG175:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[TMP11:%.*]] = load i32, i32* [[A6]], align 4, !dbg [[DBG175]]
-// CHECK-TLS4-NEXT:    [[TMP12:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG176:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP12]], [[TMP11]], !dbg [[DBG176]]
-// CHECK-TLS4-NEXT:    store i32 [[ADD7]], i32* [[RES]], align 4, !dbg [[DBG176]]
-// CHECK-TLS4-NEXT:    [[TMP13:%.*]] = call i32* @_ZTWN2STIiE2stE(), !dbg [[DBG177:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[TMP14:%.*]] = load i32, i32* [[TMP13]], align 4, !dbg [[DBG177]]
-// CHECK-TLS4-NEXT:    [[TMP15:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG178:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP15]], [[TMP14]], !dbg [[DBG178]]
-// CHECK-TLS4-NEXT:    store i32 [[ADD8]], i32* [[RES]], align 4, !dbg [[DBG178]]
-// CHECK-TLS4-NEXT:    [[TMP16:%.*]] = call float* @_ZTWN2STIfE2stE(), !dbg [[DBG179:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[TMP17:%.*]] = load float, float* [[TMP16]], align 4, !dbg [[DBG179]]
-// CHECK-TLS4-NEXT:    [[CONV:%.*]] = fptosi float [[TMP17]] to i32, !dbg [[DBG179]]
-// CHECK-TLS4-NEXT:    [[TMP18:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG180:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP18]], [[CONV]], !dbg [[DBG180]]
-// CHECK-TLS4-NEXT:    store i32 [[ADD9]], i32* [[RES]], align 4, !dbg [[DBG180]]
-// CHECK-TLS4-NEXT:    [[TMP19:%.*]] = call %struct.S4* @_ZTWN2STI2S4E2stE(), !dbg [[DBG181:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[A10:%.*]] = getelementptr inbounds [[STRUCT_S4:%.*]], %struct.S4* [[TMP19]], i32 0, i32 0, !dbg [[DBG182:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[TMP20:%.*]] = load i32, i32* [[A10]], align 4, !dbg [[DBG182]]
-// CHECK-TLS4-NEXT:    [[TMP21:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG183:![0-9]+]]
-// CHECK-TLS4-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP21]], [[TMP20]], !dbg [[DBG183]]
-// CHECK-TLS4-NEXT:    store i32 [[ADD11]], i32* [[RES]], align 4, !dbg [[DBG183]]
-// CHECK-TLS4-NEXT:    [[TMP22:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG184:![0-9]+]]
-// CHECK-TLS4-NEXT:    ret i32 [[TMP22]], !dbg [[DBG185:![0-9]+]]
+// CHECK-TLS4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[RES]], metadata [[META162:![0-9]+]], metadata !DIExpression()), !dbg [[DBG163:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[TMP0:%.*]] = call %struct.S3* @_ZTWN6Static1sE(), !dbg [[DBG164:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_S3:%.*]], %struct.S3* [[TMP0]], i32 0, i32 0, !dbg [[DBG165:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[TMP1:%.*]] = load i32, i32* [[A]], align 4, !dbg [[DBG165]]
+// CHECK-TLS4-NEXT:    store i32 [[TMP1]], i32* [[RES]], align 4, !dbg [[DBG166:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[TMP2:%.*]] = call %struct.S1* @_ZTWL3gs1(), !dbg [[DBG167:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[A1:%.*]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[TMP2]], i32 0, i32 0, !dbg [[DBG168:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[TMP3:%.*]] = load i32, i32* [[A1]], align 4, !dbg [[DBG168]]
+// CHECK-TLS4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG169:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP4]], [[TMP3]], !dbg [[DBG169]]
+// CHECK-TLS4-NEXT:    store i32 [[ADD]], i32* [[RES]], align 4, !dbg [[DBG169]]
+// CHECK-TLS4-NEXT:    [[TMP5:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S2:%.*]], %struct.S2* @_ZL3gs2, i32 0, i32 0), align 8, !dbg [[DBG170:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG171:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP6]], [[TMP5]], !dbg [[DBG171]]
+// CHECK-TLS4-NEXT:    store i32 [[ADD2]], i32* [[RES]], align 4, !dbg [[DBG171]]
+// CHECK-TLS4-NEXT:    [[TMP7:%.*]] = call %struct.S5* @_ZTW3gs3(), !dbg [[DBG172:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[A3:%.*]] = getelementptr inbounds [[STRUCT_S5:%.*]], %struct.S5* [[TMP7]], i32 0, i32 0, !dbg [[DBG173:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[A3]], align 4, !dbg [[DBG173]]
+// CHECK-TLS4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG174:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP9]], [[TMP8]], !dbg [[DBG174]]
+// CHECK-TLS4-NEXT:    store i32 [[ADD4]], i32* [[RES]], align 4, !dbg [[DBG174]]
+// CHECK-TLS4-NEXT:    [[TMP10:%.*]] = call [2 x [3 x %struct.S1]]* @_ZTW5arr_x(), !dbg [[DBG175:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* [[TMP10]], i64 0, i64 1, !dbg [[DBG175]]
+// CHECK-TLS4-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds [3 x %struct.S1], [3 x %struct.S1]* [[ARRAYIDX]], i64 0, i64 1, !dbg [[DBG175]]
+// CHECK-TLS4-NEXT:    [[A6:%.*]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYIDX5]], i32 0, i32 0, !dbg [[DBG176:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[TMP11:%.*]] = load i32, i32* [[A6]], align 4, !dbg [[DBG176]]
+// CHECK-TLS4-NEXT:    [[TMP12:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG177:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP12]], [[TMP11]], !dbg [[DBG177]]
+// CHECK-TLS4-NEXT:    store i32 [[ADD7]], i32* [[RES]], align 4, !dbg [[DBG177]]
+// CHECK-TLS4-NEXT:    [[TMP13:%.*]] = call i32* @_ZTWN2STIiE2stE(), !dbg [[DBG178:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[TMP14:%.*]] = load i32, i32* [[TMP13]], align 4, !dbg [[DBG178]]
+// CHECK-TLS4-NEXT:    [[TMP15:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG179:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP15]], [[TMP14]], !dbg [[DBG179]]
+// CHECK-TLS4-NEXT:    store i32 [[ADD8]], i32* [[RES]], align 4, !dbg [[DBG179]]
+// CHECK-TLS4-NEXT:    [[TMP16:%.*]] = call float* @_ZTWN2STIfE2stE(), !dbg [[DBG180:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[TMP17:%.*]] = load float, float* [[TMP16]], align 4, !dbg [[DBG180]]
+// CHECK-TLS4-NEXT:    [[CONV:%.*]] = fptosi float [[TMP17]] to i32, !dbg [[DBG180]]
+// CHECK-TLS4-NEXT:    [[TMP18:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG181:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP18]], [[CONV]], !dbg [[DBG181]]
+// CHECK-TLS4-NEXT:    store i32 [[ADD9]], i32* [[RES]], align 4, !dbg [[DBG181]]
+// CHECK-TLS4-NEXT:    [[TMP19:%.*]] = call %struct.S4* @_ZTWN2STI2S4E2stE(), !dbg [[DBG182:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[A10:%.*]] = getelementptr inbounds [[STRUCT_S4:%.*]], %struct.S4* [[TMP19]], i32 0, i32 0, !dbg [[DBG183:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[TMP20:%.*]] = load i32, i32* [[A10]], align 4, !dbg [[DBG183]]
+// CHECK-TLS4-NEXT:    [[TMP21:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG184:![0-9]+]]
+// CHECK-TLS4-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP21]], [[TMP20]], !dbg [[DBG184]]
+// CHECK-TLS4-NEXT:    store i32 [[ADD11]], i32* [[RES]], align 4, !dbg [[DBG184]]
+// CHECK-TLS4-NEXT:    [[TMP22:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG185:![0-9]+]]
+// CHECK-TLS4-NEXT:    ret i32 [[TMP22]], !dbg [[DBG186:![0-9]+]]
 //
 //
 // CHECK-TLS4-LABEL: define {{[^@]+}}@__cxx_global_var_init
-// CHECK-TLS4-SAME: () #[[ATTR7:[0-9]+]] !dbg [[DBG186:![0-9]+]] {
+// CHECK-TLS4-SAME: () #[[ATTR7:[0-9]+]] !dbg [[DBG187:![0-9]+]] {
 // CHECK-TLS4-NEXT:  entry:
 // CHECK-TLS4-NEXT:    call void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) @_ZL3gs1, i32 5), !dbg [[DBG190:![0-9]+]]
 // CHECK-TLS4-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_thread_atexit(void (i8*)* bitcast (void (%struct.S1*)* @_ZN2S1D1Ev to void (i8*)*), i8* bitcast (%struct.S1* @_ZL3gs1 to i8*), i8* @__dso_handle) #[[ATTR5]], !dbg [[DBG192:![0-9]+]]
 // CHECK-TLS4-NEXT:    ret void, !dbg [[DBG193:![0-9]+]]
 //
 //
 // CHECK-TLS4-LABEL: define {{[^@]+}}@_ZN2S1C1Ei
 // CHECK-TLS4-SAME: (%struct.S1* nonnull align 4 dereferenceable(4) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR3]] comdat align 2 !dbg [[DBG194:![0-9]+]] {

 // CHECK-TLS4-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // CHECK-TLS4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR7]] !dbg [[DBG268:![0-9]+]] {
 // CHECK-TLS4-NEXT:  entry:
 // CHECK-TLS4-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // CHECK-TLS4-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // CHECK-TLS4-NEXT:    call void @llvm.dbg.declare(metadata i8** [[DOTADDR]], metadata [[META272:![0-9]+]], metadata !DIExpression()), !dbg [[DBG273:![0-9]+]]
 // CHECK-TLS4-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]], !dbg [[DBG273]]
 // CHECK-TLS4:       arraydestroy.body:
-// CHECK-TLS4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ getelementptr inbounds ([[STRUCT_S1:%.*]], %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0), i64 6), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG273]]
-// CHECK-TLS4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG273]]
+// CHECK-TLS4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 1, i64 0, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG273]]
+// CHECK-TLS4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG273]]
 // CHECK-TLS4-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR5]], !dbg [[DBG273]]
 // CHECK-TLS4-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0), !dbg [[DBG273]]
 // CHECK-TLS4-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]], !dbg [[DBG273]]
 // CHECK-TLS4:       arraydestroy.done1:
 // CHECK-TLS4-NEXT:    ret void, !dbg [[DBG273]]
 //
 //
 // CHECK-TLS4-LABEL: define {{[^@]+}}@_ZZ4mainEN5SmainC2Ei

 // CHECK-TLS4-NEXT:    ret void
 //
 //
 // CHECK-TLS4-LABEL: define {{[^@]+}}@__tls_init
 // CHECK-TLS4-SAME: () #[[ATTR7]] !dbg [[DBG325:![0-9]+]] {
 // CHECK-TLS4-NEXT:  entry:
 // CHECK-TLS4-NEXT:    [[TMP0:%.*]] = load i8, i8* @__tls_guard, align 1, !dbg [[DBG326:![0-9]+]]
 // CHECK-TLS4-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0, !dbg [[DBG326]]
-// CHECK-TLS4-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT:%.*]], label [[EXIT:%.*]], !dbg [[DBG326]], !prof [[PROF118]]
+// CHECK-TLS4-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT:%.*]], label [[EXIT:%.*]], !dbg [[DBG326]], !prof [[PROF119]]
 // CHECK-TLS4:       init:
 // CHECK-TLS4-NEXT:    store i8 1, i8* @__tls_guard, align 1, !dbg [[DBG326]]
 // CHECK-TLS4-NEXT:    call void @__cxx_global_var_init(), !dbg [[DBG326]]
 // CHECK-TLS4-NEXT:    call void @__cxx_global_var_init.2(), !dbg [[DBG326]]
 // CHECK-TLS4-NEXT:    br label [[EXIT]], !dbg [[DBG326]]
 // CHECK-TLS4:       exit:
 // CHECK-TLS4-NEXT:    ret void
 //

 //
 // SIMD3-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // SIMD3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
 // SIMD3-NEXT:  entry:
 // SIMD3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // SIMD3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // SIMD3-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 // SIMD3:       arraydestroy.body:
-// SIMD3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ getelementptr inbounds ([[STRUCT_S1:%.*]], %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0), i64 6), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
-// SIMD3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
+// SIMD3-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 1, i64 0, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
+// SIMD3-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 // SIMD3-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
 // SIMD3-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0)
 // SIMD3-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 // SIMD3:       arraydestroy.done1:
 // SIMD3-NEXT:    ret void
 //
 //
 // SIMD3-LABEL: define {{[^@]+}}@main

 // SIMD3-NEXT:    call void @__cxx_global_var_init.1()
 // SIMD3-NEXT:    call void @__cxx_global_var_init.2()
 // SIMD3-NEXT:    ret void
 //
 //
 // SIMD4-LABEL: define {{[^@]+}}@__cxx_global_var_init
 // SIMD4-SAME: () #[[ATTR0:[0-9]+]] !dbg [[DBG115:![0-9]+]] {
 // SIMD4-NEXT:  entry:
-// SIMD4-NEXT:    call void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) @_ZL3gs1, i32 5), !dbg [[DBG119:![0-9]+]]
-// SIMD4-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.S1*)* @_ZN2S1D1Ev to void (i8*)*), i8* bitcast (%struct.S1* @_ZL3gs1 to i8*), i8* @__dso_handle) #[[ATTR3:[0-9]+]], !dbg [[DBG121:![0-9]+]]
-// SIMD4-NEXT:    ret void, !dbg [[DBG122:![0-9]+]]
+// SIMD4-NEXT:    call void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) @_ZL3gs1, i32 5), !dbg [[DBG118:![0-9]+]]
+// SIMD4-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.S1*)* @_ZN2S1D1Ev to void (i8*)*), i8* bitcast (%struct.S1* @_ZL3gs1 to i8*), i8* @__dso_handle) #[[ATTR3:[0-9]+]], !dbg [[DBG120:![0-9]+]]
+// SIMD4-NEXT:    ret void, !dbg [[DBG121:![0-9]+]]
 //
 //
 // SIMD4-LABEL: define {{[^@]+}}@_ZN2S1C1Ei
-// SIMD4-SAME: (%struct.S1* nonnull align 4 dereferenceable(4) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR1:[0-9]+]] comdat align 2 !dbg [[DBG123:![0-9]+]] {
+// SIMD4-SAME: (%struct.S1* nonnull align 4 dereferenceable(4) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR1:[0-9]+]] comdat align 2 !dbg [[DBG122:![0-9]+]] {
 // SIMD4-NEXT:  entry:
 // SIMD4-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S1*, align 8
 // SIMD4-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
 // SIMD4-NEXT:    store %struct.S1* [[THIS]], %struct.S1** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.S1** [[THIS_ADDR]], metadata [[META124:![0-9]+]], metadata !DIExpression()), !dbg [[DBG126:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.S1** [[THIS_ADDR]], metadata [[META123:![0-9]+]], metadata !DIExpression()), !dbg [[DBG125:![0-9]+]]
 // SIMD4-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META127:![0-9]+]], metadata !DIExpression()), !dbg [[DBG128:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META126:![0-9]+]], metadata !DIExpression()), !dbg [[DBG127:![0-9]+]]
 // SIMD4-NEXT:    [[THIS1:%.*]] = load %struct.S1*, %struct.S1** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG129:![0-9]+]]
-// SIMD4-NEXT:    call void @_ZN2S1C2Ei(%struct.S1* nonnull align 4 dereferenceable(4) [[THIS1]], i32 [[TMP0]]), !dbg [[DBG129]]
-// SIMD4-NEXT:    ret void, !dbg [[DBG130:![0-9]+]]
+// SIMD4-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG128:![0-9]+]]
+// SIMD4-NEXT:    call void @_ZN2S1C2Ei(%struct.S1* nonnull align 4 dereferenceable(4) [[THIS1]], i32 [[TMP0]]), !dbg [[DBG128]]
+// SIMD4-NEXT:    ret void, !dbg [[DBG129:![0-9]+]]
 //
 //
 // SIMD4-LABEL: define {{[^@]+}}@_ZN2S1D1Ev
-// SIMD4-SAME: (%struct.S1* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR2:[0-9]+]] comdat align 2 !dbg [[DBG131:![0-9]+]] {
+// SIMD4-SAME: (%struct.S1* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR2:[0-9]+]] comdat align 2 !dbg [[DBG130:![0-9]+]] {
 // SIMD4-NEXT:  entry:
 // SIMD4-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S1*, align 8
 // SIMD4-NEXT:    store %struct.S1* [[THIS]], %struct.S1** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.S1** [[THIS_ADDR]], metadata [[META132:![0-9]+]], metadata !DIExpression()), !dbg [[DBG133:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.S1** [[THIS_ADDR]], metadata [[META131:![0-9]+]], metadata !DIExpression()), !dbg [[DBG132:![0-9]+]]
 // SIMD4-NEXT:    [[THIS1:%.*]] = load %struct.S1*, %struct.S1** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    call void @_ZN2S1D2Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[THIS1]]) #[[ATTR3]], !dbg [[DBG134:![0-9]+]]
-// SIMD4-NEXT:    ret void, !dbg [[DBG135:![0-9]+]]
+// SIMD4-NEXT:    call void @_ZN2S1D2Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[THIS1]]) #[[ATTR3]], !dbg [[DBG133:![0-9]+]]
+// SIMD4-NEXT:    ret void, !dbg [[DBG134:![0-9]+]]
 //
 //
 // SIMD4-LABEL: define {{[^@]+}}@__cxx_global_var_init.1
-// SIMD4-SAME: () #[[ATTR0]] !dbg [[DBG136:![0-9]+]] {
+// SIMD4-SAME: () #[[ATTR0]] !dbg [[DBG135:![0-9]+]] {
 // SIMD4-NEXT:  entry:
-// SIMD4-NEXT:    call void @_ZN2S2C1Ei(%struct.S2* nonnull align 8 dereferenceable(16) @_ZL3gs2, i32 27), !dbg [[DBG137:![0-9]+]]
-// SIMD4-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.S2*)* @_ZN2S2D1Ev to void (i8*)*), i8* bitcast (%struct.S2* @_ZL3gs2 to i8*), i8* @__dso_handle) #[[ATTR3]], !dbg [[DBG139:![0-9]+]]
-// SIMD4-NEXT:    ret void, !dbg [[DBG140:![0-9]+]]
+// SIMD4-NEXT:    call void @_ZN2S2C1Ei(%struct.S2* nonnull align 8 dereferenceable(16) @_ZL3gs2, i32 27), !dbg [[DBG136:![0-9]+]]
+// SIMD4-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.S2*)* @_ZN2S2D1Ev to void (i8*)*), i8* bitcast (%struct.S2* @_ZL3gs2 to i8*), i8* @__dso_handle) #[[ATTR3]], !dbg [[DBG138:![0-9]+]]
+// SIMD4-NEXT:    ret void, !dbg [[DBG139:![0-9]+]]
 //
 //
 // SIMD4-LABEL: define {{[^@]+}}@_ZN2S2C1Ei
-// SIMD4-SAME: (%struct.S2* nonnull align 8 dereferenceable(16) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 !dbg [[DBG141:![0-9]+]] {
+// SIMD4-SAME: (%struct.S2* nonnull align 8 dereferenceable(16) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 !dbg [[DBG140:![0-9]+]] {
 // SIMD4-NEXT:  entry:
 // SIMD4-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S2*, align 8
 // SIMD4-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
 // SIMD4-NEXT:    store %struct.S2* [[THIS]], %struct.S2** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.S2** [[THIS_ADDR]], metadata [[META142:![0-9]+]], metadata !DIExpression()), !dbg [[DBG144:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.S2** [[THIS_ADDR]], metadata [[META141:![0-9]+]], metadata !DIExpression()), !dbg [[DBG143:![0-9]+]]
 // SIMD4-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META145:![0-9]+]], metadata !DIExpression()), !dbg [[DBG146:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META144:![0-9]+]], metadata !DIExpression()), !dbg [[DBG145:![0-9]+]]
 // SIMD4-NEXT:    [[THIS1:%.*]] = load %struct.S2*, %struct.S2** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG147:![0-9]+]]
-// SIMD4-NEXT:    call void @_ZN2S2C2Ei(%struct.S2* nonnull align 8 dereferenceable(16) [[THIS1]], i32 [[TMP0]]), !dbg [[DBG147]]
-// SIMD4-NEXT:    ret void, !dbg [[DBG148:![0-9]+]]
+// SIMD4-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG146:![0-9]+]]
+// SIMD4-NEXT:    call void @_ZN2S2C2Ei(%struct.S2* nonnull align 8 dereferenceable(16) [[THIS1]], i32 [[TMP0]]), !dbg [[DBG146]]
+// SIMD4-NEXT:    ret void, !dbg [[DBG147:![0-9]+]]
 //
 //
 // SIMD4-LABEL: define {{[^@]+}}@_ZN2S2D1Ev
-// SIMD4-SAME: (%struct.S2* nonnull align 8 dereferenceable(16) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG149:![0-9]+]] {
+// SIMD4-SAME: (%struct.S2* nonnull align 8 dereferenceable(16) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG148:![0-9]+]] {
 // SIMD4-NEXT:  entry:
 // SIMD4-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S2*, align 8
 // SIMD4-NEXT:    store %struct.S2* [[THIS]], %struct.S2** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.S2** [[THIS_ADDR]], metadata [[META150:![0-9]+]], metadata !DIExpression()), !dbg [[DBG151:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.S2** [[THIS_ADDR]], metadata [[META149:![0-9]+]], metadata !DIExpression()), !dbg [[DBG150:![0-9]+]]
 // SIMD4-NEXT:    [[THIS1:%.*]] = load %struct.S2*, %struct.S2** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    call void @_ZN2S2D2Ev(%struct.S2* nonnull align 8 dereferenceable(16) [[THIS1]]) #[[ATTR3]], !dbg [[DBG152:![0-9]+]]
-// SIMD4-NEXT:    ret void, !dbg [[DBG153:![0-9]+]]
+// SIMD4-NEXT:    call void @_ZN2S2D2Ev(%struct.S2* nonnull align 8 dereferenceable(16) [[THIS1]]) #[[ATTR3]], !dbg [[DBG151:![0-9]+]]
+// SIMD4-NEXT:    ret void, !dbg [[DBG152:![0-9]+]]
 //
 //
 // SIMD4-LABEL: define {{[^@]+}}@__cxx_global_var_init.2
-// SIMD4-SAME: () #[[ATTR0]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) !dbg [[DBG154:![0-9]+]] {
+// SIMD4-SAME: () #[[ATTR0]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) !dbg [[DBG153:![0-9]+]] {
 // SIMD4-NEXT:  entry:
 // SIMD4-NEXT:    [[ARRAYINIT_ENDOFINIT:%.*]] = alloca [3 x %struct.S1]*, align 8
 // SIMD4-NEXT:    [[ARRAYINIT_ENDOFINIT1:%.*]] = alloca %struct.S1*, align 8
 // SIMD4-NEXT:    [[EXN_SLOT:%.*]] = alloca i8*, align 8
 // SIMD4-NEXT:    [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
 // SIMD4-NEXT:    [[ARRAYINIT_ENDOFINIT5:%.*]] = alloca %struct.S1*, align 8
-// SIMD4-NEXT:    store [3 x %struct.S1]* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0), [3 x %struct.S1]** [[ARRAYINIT_ENDOFINIT]], align 8, !dbg [[DBG155:![0-9]+]]
-// SIMD4-NEXT:    store %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 0), %struct.S1** [[ARRAYINIT_ENDOFINIT1]], align 8, !dbg [[DBG157:![0-9]+]]
+// SIMD4-NEXT:    store [3 x %struct.S1]* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0), [3 x %struct.S1]** [[ARRAYINIT_ENDOFINIT]], align 8, !dbg [[DBG154:![0-9]+]]
+// SIMD4-NEXT:    store %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 0), %struct.S1** [[ARRAYINIT_ENDOFINIT1]], align 8, !dbg [[DBG156:![0-9]+]]
 // SIMD4-NEXT:    invoke void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 0), i32 1)
-// SIMD4-NEXT:    to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]], !dbg [[DBG158:![0-9]+]]
+// SIMD4-NEXT:    to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]], !dbg [[DBG157:![0-9]+]]
 // SIMD4:       invoke.cont:
-// SIMD4-NEXT:    store %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 1), %struct.S1** [[ARRAYINIT_ENDOFINIT1]], align 8, !dbg [[DBG157]]
+// SIMD4-NEXT:    store %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 1), %struct.S1** [[ARRAYINIT_ENDOFINIT1]], align 8, !dbg [[DBG156]]
 // SIMD4-NEXT:    invoke void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 1), i32 2)
-// SIMD4-NEXT:    to label [[INVOKE_CONT2:%.*]] unwind label [[LPAD]], !dbg [[DBG159:![0-9]+]]
+// SIMD4-NEXT:    to label [[INVOKE_CONT2:%.*]] unwind label [[LPAD]], !dbg [[DBG158:![0-9]+]]
 // SIMD4:       invoke.cont2:
-// SIMD4-NEXT:    store %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 2), %struct.S1** [[ARRAYINIT_ENDOFINIT1]], align 8, !dbg [[DBG157]]
+// SIMD4-NEXT:    store %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 2), %struct.S1** [[ARRAYINIT_ENDOFINIT1]], align 8, !dbg [[DBG156]]
 // SIMD4-NEXT:    invoke void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 2), i32 3)
-// SIMD4-NEXT:    to label [[INVOKE_CONT3:%.*]] unwind label [[LPAD]], !dbg [[DBG160:![0-9]+]]
+// SIMD4-NEXT:    to label [[INVOKE_CONT3:%.*]] unwind label [[LPAD]], !dbg [[DBG159:![0-9]+]]
 // SIMD4:       invoke.cont3:
-// SIMD4-NEXT:    store [3 x %struct.S1]* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1), [3 x %struct.S1]** [[ARRAYINIT_ENDOFINIT]], align 8, !dbg [[DBG155]]
-// SIMD4-NEXT:    store %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 0), %struct.S1** [[ARRAYINIT_ENDOFINIT5]], align 8, !dbg [[DBG161:![0-9]+]]
+// SIMD4-NEXT:    store [3 x %struct.S1]* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1), [3 x %struct.S1]** [[ARRAYINIT_ENDOFINIT]], align 8, !dbg [[DBG154]]
+// SIMD4-NEXT:    store %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 0), %struct.S1** [[ARRAYINIT_ENDOFINIT5]], align 8, !dbg [[DBG160:![0-9]+]]
 // SIMD4-NEXT:    invoke void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 0), i32 4)
-// SIMD4-NEXT:    to label [[INVOKE_CONT7:%.*]] unwind label [[LPAD6:%.*]], !dbg [[DBG162:![0-9]+]]
+// SIMD4-NEXT:    to label [[INVOKE_CONT7:%.*]] unwind label [[LPAD6:%.*]], !dbg [[DBG161:![0-9]+]]
 // SIMD4:       invoke.cont7:
-// SIMD4-NEXT:    store %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 1), %struct.S1** [[ARRAYINIT_ENDOFINIT5]], align 8, !dbg [[DBG161]]
+// SIMD4-NEXT:    store %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 1), %struct.S1** [[ARRAYINIT_ENDOFINIT5]], align 8, !dbg [[DBG160]]
 // SIMD4-NEXT:    invoke void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 1), i32 5)
-// SIMD4-NEXT:    to label [[INVOKE_CONT8:%.*]] unwind label [[LPAD6]], !dbg [[DBG163:![0-9]+]]
+// SIMD4-NEXT:    to label [[INVOKE_CONT8:%.*]] unwind label [[LPAD6]], !dbg [[DBG162:![0-9]+]]
 // SIMD4:       invoke.cont8:
-// SIMD4-NEXT:    store %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 2), %struct.S1** [[ARRAYINIT_ENDOFINIT5]], align 8, !dbg [[DBG161]]
+// SIMD4-NEXT:    store %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 2), %struct.S1** [[ARRAYINIT_ENDOFINIT5]], align 8, !dbg [[DBG160]]
 // SIMD4-NEXT:    invoke void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 2), i32 6)
-// SIMD4-NEXT:    to label [[INVOKE_CONT9:%.*]] unwind label [[LPAD6]], !dbg [[DBG164:![0-9]+]]
+// SIMD4-NEXT:    to label [[INVOKE_CONT9:%.*]] unwind label [[LPAD6]], !dbg [[DBG163:![0-9]+]]
 // SIMD4:       invoke.cont9:
-// SIMD4-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* @__cxx_global_array_dtor, i8* null, i8* @__dso_handle) #[[ATTR3]], !dbg [[DBG165:![0-9]+]]
-// SIMD4-NEXT:    ret void, !dbg [[DBG165]]
+// SIMD4-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* @__cxx_global_array_dtor, i8* null, i8* @__dso_handle) #[[ATTR3]], !dbg [[DBG164:![0-9]+]]
+// SIMD4-NEXT:    ret void, !dbg [[DBG164]]
 // SIMD4:       lpad:
 // SIMD4-NEXT:    [[TMP1:%.*]] = landingpad { i8*, i32 }
-// SIMD4-NEXT:    cleanup, !dbg [[DBG166:![0-9]+]]
-// SIMD4-NEXT:    [[TMP2:%.*]] = extractvalue { i8*, i32 } [[TMP1]], 0, !dbg [[DBG166]]
-// SIMD4-NEXT:    store i8* [[TMP2]], i8** [[EXN_SLOT]], align 8, !dbg [[DBG166]]
-// SIMD4-NEXT:    [[TMP3:%.*]] = extractvalue { i8*, i32 } [[TMP1]], 1, !dbg [[DBG166]]
-// SIMD4-NEXT:    store i32 [[TMP3]], i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG166]]
-// SIMD4-NEXT:    [[TMP4:%.*]] = load %struct.S1*, %struct.S1** [[ARRAYINIT_ENDOFINIT1]], align 8, !dbg [[DBG157]]
-// SIMD4-NEXT:    [[ARRAYDESTROY_ISEMPTY:%.*]] = icmp eq %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 0), [[TMP4]], !dbg [[DBG157]]
-// SIMD4-NEXT:    br i1 [[ARRAYDESTROY_ISEMPTY]], label [[ARRAYDESTROY_DONE4:%.*]], label [[ARRAYDESTROY_BODY:%.*]], !dbg [[DBG157]]
+// SIMD4-NEXT:    cleanup, !dbg [[DBG165:![0-9]+]]
+// SIMD4-NEXT:    [[TMP2:%.*]] = extractvalue { i8*, i32 } [[TMP1]], 0, !dbg [[DBG165]]
+// SIMD4-NEXT:    store i8* [[TMP2]], i8** [[EXN_SLOT]], align 8, !dbg [[DBG165]]
+// SIMD4-NEXT:    [[TMP3:%.*]] = extractvalue { i8*, i32 } [[TMP1]], 1, !dbg [[DBG165]]
+// SIMD4-NEXT:    store i32 [[TMP3]], i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG165]]
+// SIMD4-NEXT:    [[TMP4:%.*]] = load %struct.S1*, %struct.S1** [[ARRAYINIT_ENDOFINIT1]], align 8, !dbg [[DBG156]]
+// SIMD4-NEXT:    [[ARRAYDESTROY_ISEMPTY:%.*]] = icmp eq %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 0), [[TMP4]], !dbg [[DBG156]]
+// SIMD4-NEXT:    br i1 [[ARRAYDESTROY_ISEMPTY]], label [[ARRAYDESTROY_DONE4:%.*]], label [[ARRAYDESTROY_BODY:%.*]], !dbg [[DBG156]]
 // SIMD4:       arraydestroy.body:
-// SIMD4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ [[TMP4]], [[LPAD]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG157]]
-// SIMD4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG157]]
-// SIMD4-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]], !dbg [[DBG157]]
-// SIMD4-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 0), !dbg [[DBG157]]
-// SIMD4-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE4]], label [[ARRAYDESTROY_BODY]], !dbg [[DBG157]]
+// SIMD4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ [[TMP4]], [[LPAD]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG156]]
+// SIMD4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG156]]
+// SIMD4-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]], !dbg [[DBG156]]
+// SIMD4-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 0), !dbg [[DBG156]]
+// SIMD4-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE4]], label [[ARRAYDESTROY_BODY]], !dbg [[DBG156]]
 // SIMD4:       arraydestroy.done4:
-// SIMD4-NEXT:    br label [[EHCLEANUP:%.*]], !dbg [[DBG157]]
+// SIMD4-NEXT:    br label [[EHCLEANUP:%.*]], !dbg [[DBG156]]
 // SIMD4:       lpad6:
 // SIMD4-NEXT:    [[TMP5:%.*]] = landingpad { i8*, i32 }
-// SIMD4-NEXT:    cleanup, !dbg [[DBG166]]
-// SIMD4-NEXT:    [[TMP6:%.*]] = extractvalue { i8*, i32 } [[TMP5]], 0, !dbg [[DBG166]]
-// SIMD4-NEXT:    store i8* [[TMP6]], i8** [[EXN_SLOT]], align 8, !dbg [[DBG166]]
-// SIMD4-NEXT:    [[TMP7:%.*]] = extractvalue { i8*, i32 } [[TMP5]], 1, !dbg [[DBG166]]
-// SIMD4-NEXT:    store i32 [[TMP7]], i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG166]]
-// SIMD4-NEXT:    [[TMP8:%.*]] = load %struct.S1*, %struct.S1** [[ARRAYINIT_ENDOFINIT5]], align 8, !dbg [[DBG161]]
-// SIMD4-NEXT:    [[ARRAYDESTROY_ISEMPTY10:%.*]] = icmp eq %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 0), [[TMP8]], !dbg [[DBG161]]
-// SIMD4-NEXT:    br i1 [[ARRAYDESTROY_ISEMPTY10]], label [[ARRAYDESTROY_DONE15:%.*]], label [[ARRAYDESTROY_BODY11:%.*]], !dbg [[DBG161]]
+// SIMD4-NEXT:    cleanup, !dbg [[DBG165]]
+// SIMD4-NEXT:    [[TMP6:%.*]] = extractvalue { i8*, i32 } [[TMP5]], 0, !dbg [[DBG165]]
+// SIMD4-NEXT:    store i8* [[TMP6]], i8** [[EXN_SLOT]], align 8, !dbg [[DBG165]]
+// SIMD4-NEXT:    [[TMP7:%.*]] = extractvalue { i8*, i32 } [[TMP5]], 1, !dbg [[DBG165]]
+// SIMD4-NEXT:    store i32 [[TMP7]], i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG165]]
+// SIMD4-NEXT:    [[TMP8:%.*]] = load %struct.S1*, %struct.S1** [[ARRAYINIT_ENDOFINIT5]], align 8, !dbg [[DBG160]]
+// SIMD4-NEXT:    [[ARRAYDESTROY_ISEMPTY10:%.*]] = icmp eq %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 0), [[TMP8]], !dbg [[DBG160]]
+// SIMD4-NEXT:    br i1 [[ARRAYDESTROY_ISEMPTY10]], label [[ARRAYDESTROY_DONE15:%.*]], label [[ARRAYDESTROY_BODY11:%.*]], !dbg [[DBG160]]
 // SIMD4:       arraydestroy.body11:
-// SIMD4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST12:%.*]] = phi %struct.S1* [ [[TMP8]], [[LPAD6]] ], [ [[ARRAYDESTROY_ELEMENT13:%.*]], [[ARRAYDESTROY_BODY11]] ], !dbg [[DBG161]]
-// SIMD4-NEXT:    [[ARRAYDESTROY_ELEMENT13]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST12]], i64 -1, !dbg [[DBG161]]
-// SIMD4-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT13]]) #[[ATTR3]], !dbg [[DBG161]]
-// SIMD4-NEXT:    [[ARRAYDESTROY_DONE14:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT13]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 0), !dbg [[DBG161]]
-// SIMD4-NEXT:    br i1 [[ARRAYDESTROY_DONE14]], label [[ARRAYDESTROY_DONE15]], label [[ARRAYDESTROY_BODY11]], !dbg [[DBG161]]
+// SIMD4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST12:%.*]] = phi %struct.S1* [ [[TMP8]], [[LPAD6]] ], [ [[ARRAYDESTROY_ELEMENT13:%.*]], [[ARRAYDESTROY_BODY11]] ], !dbg [[DBG160]]
+// SIMD4-NEXT:    [[ARRAYDESTROY_ELEMENT13]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST12]], i64 -1, !dbg [[DBG160]]
+// SIMD4-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT13]]) #[[ATTR3]], !dbg [[DBG160]]
+// SIMD4-NEXT:    [[ARRAYDESTROY_DONE14:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT13]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 0), !dbg [[DBG160]]
+// SIMD4-NEXT:    br i1 [[ARRAYDESTROY_DONE14]], label [[ARRAYDESTROY_DONE15]], label [[ARRAYDESTROY_BODY11]], !dbg [[DBG160]]
 // SIMD4:       arraydestroy.done15:
-// SIMD4-NEXT:    br label [[EHCLEANUP]], !dbg [[DBG161]]
+// SIMD4-NEXT:    br label [[EHCLEANUP]], !dbg [[DBG160]]
 // SIMD4:       ehcleanup:
-// SIMD4-NEXT:    [[TMP9:%.*]] = load [3 x %struct.S1]*, [3 x %struct.S1]** [[ARRAYINIT_ENDOFINIT]], align 8, !dbg [[DBG155]]
-// SIMD4-NEXT:    [[PAD_ARRAYEND:%.*]] = getelementptr inbounds [3 x %struct.S1], [3 x %struct.S1]* [[TMP9]], i64 0, i64 0, !dbg [[DBG155]]
-// SIMD4-NEXT:    [[ARRAYDESTROY_ISEMPTY16:%.*]] = icmp eq %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 0), [[PAD_ARRAYEND]], !dbg [[DBG155]]
-// SIMD4-NEXT:    br i1 [[ARRAYDESTROY_ISEMPTY16]], label [[ARRAYDESTROY_DONE21:%.*]], label [[ARRAYDESTROY_BODY17:%.*]], !dbg [[DBG155]]
+// SIMD4-NEXT:    [[TMP9:%.*]] = load [3 x %struct.S1]*, [3 x %struct.S1]** [[ARRAYINIT_ENDOFINIT]], align 8, !dbg [[DBG154]]
+// SIMD4-NEXT:    [[PAD_ARRAYEND:%.*]] = getelementptr inbounds [3 x %struct.S1], [3 x %struct.S1]* [[TMP9]], i64 0, i64 0, !dbg [[DBG154]]
+// SIMD4-NEXT:    [[ARRAYDESTROY_ISEMPTY16:%.*]] = icmp eq %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 0), [[PAD_ARRAYEND]], !dbg [[DBG154]]
+// SIMD4-NEXT:    br i1 [[ARRAYDESTROY_ISEMPTY16]], label [[ARRAYDESTROY_DONE21:%.*]], label [[ARRAYDESTROY_BODY17:%.*]], !dbg [[DBG154]]
 // SIMD4:       arraydestroy.body17:
-// SIMD4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST18:%.*]] = phi %struct.S1* [ [[PAD_ARRAYEND]], [[EHCLEANUP]] ], [ [[ARRAYDESTROY_ELEMENT19:%.*]], [[ARRAYDESTROY_BODY17]] ], !dbg [[DBG155]]
-// SIMD4-NEXT:    [[ARRAYDESTROY_ELEMENT19]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST18]], i64 -1, !dbg [[DBG155]]
-// SIMD4-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT19]]) #[[ATTR3]], !dbg [[DBG155]]
-// SIMD4-NEXT:    [[ARRAYDESTROY_DONE20:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT19]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 0), !dbg [[DBG155]]
-// SIMD4-NEXT:    br i1 [[ARRAYDESTROY_DONE20]], label [[ARRAYDESTROY_DONE21]], label [[ARRAYDESTROY_BODY17]], !dbg [[DBG155]]
+// SIMD4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST18:%.*]] = phi %struct.S1* [ [[PAD_ARRAYEND]], [[EHCLEANUP]] ], [ [[ARRAYDESTROY_ELEMENT19:%.*]], [[ARRAYDESTROY_BODY17]] ], !dbg [[DBG154]]
+// SIMD4-NEXT:    [[ARRAYDESTROY_ELEMENT19]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST18]], i64 -1, !dbg [[DBG154]]
+// SIMD4-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT19]]) #[[ATTR3]], !dbg [[DBG154]]
+// SIMD4-NEXT:    [[ARRAYDESTROY_DONE20:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT19]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 0, i64 0), !dbg [[DBG154]]
+// SIMD4-NEXT:    br i1 [[ARRAYDESTROY_DONE20]], label [[ARRAYDESTROY_DONE21]], label [[ARRAYDESTROY_BODY17]], !dbg [[DBG154]]
 // SIMD4:       arraydestroy.done21:
-// SIMD4-NEXT:    br label [[EH_RESUME:%.*]], !dbg [[DBG155]]
+// SIMD4-NEXT:    br label [[EH_RESUME:%.*]], !dbg [[DBG154]]
 // SIMD4:       eh.resume:
-// SIMD4-NEXT:    [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8, !dbg [[DBG155]]
-// SIMD4-NEXT:    [[SEL:%.*]] = load i32, i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG155]]
-// SIMD4-NEXT:    [[LPAD_VAL:%.*]] = insertvalue { i8*, i32 } undef, i8* [[EXN]], 0, !dbg [[DBG155]]
-// SIMD4-NEXT:    [[LPAD_VAL22:%.*]] = insertvalue { i8*, i32 } [[LPAD_VAL]], i32 [[SEL]], 1, !dbg [[DBG155]]
-// SIMD4-NEXT:    resume { i8*, i32 } [[LPAD_VAL22]], !dbg [[DBG155]]
+// SIMD4-NEXT:    [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8, !dbg [[DBG154]]
+// SIMD4-NEXT:    [[SEL:%.*]] = load i32, i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG154]]
+// SIMD4-NEXT:    [[LPAD_VAL:%.*]] = insertvalue { i8*, i32 } undef, i8* [[EXN]], 0, !dbg [[DBG154]]
+// SIMD4-NEXT:    [[LPAD_VAL22:%.*]] = insertvalue { i8*, i32 } [[LPAD_VAL]], i32 [[SEL]], 1, !dbg [[DBG154]]
+// SIMD4-NEXT:    resume { i8*, i32 } [[LPAD_VAL22]], !dbg [[DBG154]]
 //
 //
 // SIMD4-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
-// SIMD4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] !dbg [[DBG167:![0-9]+]] {
+// SIMD4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] !dbg [[DBG166:![0-9]+]] {
 // SIMD4-NEXT:  entry:
 // SIMD4-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // SIMD4-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata i8** [[DOTADDR]], metadata [[META171:![0-9]+]], metadata !DIExpression()), !dbg [[DBG172:![0-9]+]]
-// SIMD4-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]], !dbg [[DBG172]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata i8** [[DOTADDR]], metadata [[META170:![0-9]+]], metadata !DIExpression()), !dbg [[DBG171:![0-9]+]]
+// SIMD4-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]], !dbg [[DBG171]]
 // SIMD4:       arraydestroy.body:
-// SIMD4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ getelementptr inbounds ([[STRUCT_S1:%.*]], %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0), i64 6), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG172]]
-// SIMD4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG172]]
-// SIMD4-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]], !dbg [[DBG172]]
-// SIMD4-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0), !dbg [[DBG172]]
-// SIMD4-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]], !dbg [[DBG172]]
+// SIMD4-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 1, i64 0, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG171]]
+// SIMD4-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG171]]
+// SIMD4-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]], !dbg [[DBG171]]
+// SIMD4-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0), !dbg [[DBG171]]
+// SIMD4-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]], !dbg [[DBG171]]
 // SIMD4:       arraydestroy.done1:
-// SIMD4-NEXT:    ret void, !dbg [[DBG172]]
+// SIMD4-NEXT:    ret void, !dbg [[DBG171]]
 //
 //
 // SIMD4-LABEL: define {{[^@]+}}@main
-// SIMD4-SAME: () #[[ATTR5:[0-9]+]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) !dbg [[DBG53:![0-9]+]] {
+// SIMD4-SAME: () #[[ATTR5:[0-9]+]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) !dbg [[DBG52:![0-9]+]] {
 // SIMD4-NEXT:  entry:
 // SIMD4-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
 // SIMD4-NEXT:    [[RES:%.*]] = alloca i32, align 4
 // SIMD4-NEXT:    [[EXN_SLOT:%.*]] = alloca i8*, align 8
 // SIMD4-NEXT:    [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
 // SIMD4-NEXT:    store i32 0, i32* [[RETVAL]], align 4
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[RES]], metadata [[META173:![0-9]+]], metadata !DIExpression()), !dbg [[DBG174:![0-9]+]]
-// SIMD4-NEXT:    [[TMP0:%.*]] = load atomic i8, i8* bitcast (i64* @_ZGVZ4mainE2sm to i8*) acquire, align 8, !dbg [[DBG175:![0-9]+]]
-// SIMD4-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0, !dbg [[DBG175]]
-// SIMD4-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !dbg [[DBG175]], !prof [[PROF176:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[RES]], metadata [[META172:![0-9]+]], metadata !DIExpression()), !dbg [[DBG173:![0-9]+]]
+// SIMD4-NEXT:    [[TMP0:%.*]] = load atomic i8, i8* bitcast (i64* @_ZGVZ4mainE2sm to i8*) acquire, align 8, !dbg [[DBG174:![0-9]+]]
+// SIMD4-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0, !dbg [[DBG174]]
+// SIMD4-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !dbg [[DBG174]], !prof [[PROF175:![0-9]+]]
 // SIMD4:       init.check:
-// SIMD4-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_guard_acquire(i64* @_ZGVZ4mainE2sm) #[[ATTR3]], !dbg [[DBG175]]
-// SIMD4-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP1]], 0, !dbg [[DBG175]]
-// SIMD4-NEXT:    br i1 [[TOBOOL]], label [[INIT:%.*]], label [[INIT_END]], !dbg [[DBG175]]
+// SIMD4-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_guard_acquire(i64* @_ZGVZ4mainE2sm) #[[ATTR3]], !dbg [[DBG174]]
+// SIMD4-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP1]], 0, !dbg [[DBG174]]
+// SIMD4-NEXT:    br i1 [[TOBOOL]], label [[INIT:%.*]], label [[INIT_END]], !dbg [[DBG174]]
 // SIMD4:       init:
-// SIMD4-NEXT:    [[TMP2:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S1:%.*]], %struct.S1* @_ZL3gs1, i32 0, i32 0), align 4, !dbg [[DBG177:![0-9]+]]
+// SIMD4-NEXT:    [[TMP2:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S1:%.*]], %struct.S1* @_ZL3gs1, i32 0, i32 0), align 4, !dbg [[DBG176:![0-9]+]]
 // SIMD4-NEXT:    invoke void @_ZZ4mainEN5SmainC1Ei(%struct.Smain* nonnull align 8 dereferenceable(24) @_ZZ4mainE2sm, i32 [[TMP2]])
-// SIMD4-NEXT:    to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]], !dbg [[DBG178:![0-9]+]]
+// SIMD4-NEXT:    to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]], !dbg [[DBG177:![0-9]+]]
 // SIMD4:       invoke.cont:
-// SIMD4-NEXT:    [[TMP3:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.Smain*)* @_ZZ4mainEN5SmainD1Ev to void (i8*)*), i8* bitcast (%struct.Smain* @_ZZ4mainE2sm to i8*), i8* @__dso_handle) #[[ATTR3]], !dbg [[DBG175]]
-// SIMD4-NEXT:    call void @__cxa_guard_release(i64* @_ZGVZ4mainE2sm) #[[ATTR3]], !dbg [[DBG175]]
-// SIMD4-NEXT:    br label [[INIT_END]], !dbg [[DBG175]]
+// SIMD4-NEXT:    [[TMP3:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.Smain*)* @_ZZ4mainEN5SmainD1Ev to void (i8*)*), i8* bitcast (%struct.Smain* @_ZZ4mainE2sm to i8*), i8* @__dso_handle) #[[ATTR3]], !dbg [[DBG174]]
+// SIMD4-NEXT:    call void @__cxa_guard_release(i64* @_ZGVZ4mainE2sm) #[[ATTR3]], !dbg [[DBG174]]
+// SIMD4-NEXT:    br label [[INIT_END]], !dbg [[DBG174]]
 // SIMD4:       init.end:
-// SIMD4-NEXT:    [[TMP4:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S3:%.*]], %struct.S3* @_ZN6Static1sE, i32 0, i32 0), align 4, !dbg [[DBG179:![0-9]+]]
-// SIMD4-NEXT:    store i32 [[TMP4]], i32* [[RES]], align 4, !dbg [[DBG180:![0-9]+]]
-// SIMD4-NEXT:    [[TMP5:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_SMAIN:%.*]], %struct.Smain* @_ZZ4mainE2sm, i32 0, i32 0), align 8, !dbg [[DBG181:![0-9]+]]
-// SIMD4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG182:![0-9]+]]
-// SIMD4-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP6]], [[TMP5]], !dbg [[DBG182]]
-// SIMD4-NEXT:    store i32 [[ADD]], i32* [[RES]], align 4, !dbg [[DBG182]]
-// SIMD4-NEXT:    [[TMP7:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S1]], %struct.S1* @_ZL3gs1, i32 0, i32 0), align 4, !dbg [[DBG183:![0-9]+]]
-// SIMD4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG184:![0-9]+]]
-// SIMD4-NEXT:    [[ADD1:%.*]] = add nsw i32 [[TMP8]], [[TMP7]], !dbg [[DBG184]]
-// SIMD4-NEXT:    store i32 [[ADD1]], i32* [[RES]], align 4, !dbg [[DBG184]]
-// SIMD4-NEXT:    [[TMP9:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S2:%.*]], %struct.S2* @_ZL3gs2, i32 0, i32 0), align 8, !dbg [[DBG185:![0-9]+]]
-// SIMD4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG186:![0-9]+]]
-// SIMD4-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP10]], [[TMP9]], !dbg [[DBG186]]
-// SIMD4-NEXT:    store i32 [[ADD2]], i32* [[RES]], align 4, !dbg [[DBG186]]
-// SIMD4-NEXT:    [[TMP11:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S5:%.*]], %struct.S5* @gs3, i32 0, i32 0), align 4, !dbg [[DBG187:![0-9]+]]
-// SIMD4-NEXT:    [[TMP12:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG188:![0-9]+]]
-// SIMD4-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP12]], [[TMP11]], !dbg [[DBG188]]
-// SIMD4-NEXT:    store i32 [[ADD3]], i32* [[RES]], align 4, !dbg [[DBG188]]
-// SIMD4-NEXT:    [[TMP13:%.*]] = load i32, i32* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 1, i32 0), align 4, !dbg [[DBG189:![0-9]+]]
-// SIMD4-NEXT:    [[TMP14:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG190:![0-9]+]]
-// SIMD4-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP14]], [[TMP13]], !dbg [[DBG190]]
-// SIMD4-NEXT:    store i32 [[ADD4]], i32* [[RES]], align 4, !dbg [[DBG190]]
-// SIMD4-NEXT:    [[TMP15:%.*]] = load i32, i32* @_ZN2STIiE2stE, align 4, !dbg [[DBG191:![0-9]+]]
-// SIMD4-NEXT:    [[TMP16:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG192:![0-9]+]]
-// SIMD4-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP16]], [[TMP15]], !dbg [[DBG192]]
-// SIMD4-NEXT:    store i32 [[ADD5]], i32* [[RES]], align 4, !dbg [[DBG192]]
-// SIMD4-NEXT:    [[TMP17:%.*]] = load float, float* @_ZN2STIfE2stE, align 4, !dbg [[DBG193:![0-9]+]]
-// SIMD4-NEXT:    [[CONV:%.*]] = fptosi float [[TMP17]] to i32, !dbg [[DBG193]]
-// SIMD4-NEXT:    [[TMP18:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG194:![0-9]+]]
-// SIMD4-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP18]], [[CONV]], !dbg [[DBG194]]
-// SIMD4-NEXT:    store i32 [[ADD6]], i32* [[RES]], align 4, !dbg [[DBG194]]
-// SIMD4-NEXT:    [[TMP19:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S4:%.*]], %struct.S4* @_ZN2STI2S4E2stE, i32 0, i32 0), align 4, !dbg [[DBG195:![0-9]+]]
-// SIMD4-NEXT:    [[TMP20:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG196:![0-9]+]]
-// SIMD4-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP20]], [[TMP19]], !dbg [[DBG196]]
-// SIMD4-NEXT:    store i32 [[ADD7]], i32* [[RES]], align 4, !dbg [[DBG196]]
-// SIMD4-NEXT:    [[TMP21:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG197:![0-9]+]]
-// SIMD4-NEXT:    ret i32 [[TMP21]], !dbg [[DBG198:![0-9]+]]
+// SIMD4-NEXT:    [[TMP4:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S3:%.*]], %struct.S3* @_ZN6Static1sE, i32 0, i32 0), align 4, !dbg [[DBG178:![0-9]+]]
+// SIMD4-NEXT:    store i32 [[TMP4]], i32* [[RES]], align 4, !dbg [[DBG179:![0-9]+]]
+// SIMD4-NEXT:    [[TMP5:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_SMAIN:%.*]], %struct.Smain* @_ZZ4mainE2sm, i32 0, i32 0), align 8, !dbg [[DBG180:![0-9]+]]
+// SIMD4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG181:![0-9]+]]
+// SIMD4-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP6]], [[TMP5]], !dbg [[DBG181]]
+// SIMD4-NEXT:    store i32 [[ADD]], i32* [[RES]], align 4, !dbg [[DBG181]]
+// SIMD4-NEXT:    [[TMP7:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S1]], %struct.S1* @_ZL3gs1, i32 0, i32 0), align 4, !dbg [[DBG182:![0-9]+]]
+// SIMD4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG183:![0-9]+]]
+// SIMD4-NEXT:    [[ADD1:%.*]] = add nsw i32 [[TMP8]], [[TMP7]], !dbg [[DBG183]]
+// SIMD4-NEXT:    store i32 [[ADD1]], i32* [[RES]], align 4, !dbg [[DBG183]]
+// SIMD4-NEXT:    [[TMP9:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S2:%.*]], %struct.S2* @_ZL3gs2, i32 0, i32 0), align 8, !dbg [[DBG184:![0-9]+]]
+// SIMD4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG185:![0-9]+]]
+// SIMD4-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP10]], [[TMP9]], !dbg [[DBG185]]
+// SIMD4-NEXT:    store i32 [[ADD2]], i32* [[RES]], align 4, !dbg [[DBG185]]
+// SIMD4-NEXT:    [[TMP11:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S5:%.*]], %struct.S5* @gs3, i32 0, i32 0), align 4, !dbg [[DBG186:![0-9]+]]
+// SIMD4-NEXT:    [[TMP12:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG187:![0-9]+]]
+// SIMD4-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP12]], [[TMP11]], !dbg [[DBG187]]
+// SIMD4-NEXT:    store i32 [[ADD3]], i32* [[RES]], align 4, !dbg [[DBG187]]
+// SIMD4-NEXT:    [[TMP13:%.*]] = load i32, i32* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 1, i32 0), align 4, !dbg [[DBG188:![0-9]+]]
+// SIMD4-NEXT:    [[TMP14:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG189:![0-9]+]]
+// SIMD4-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP14]], [[TMP13]], !dbg [[DBG189]]
+// SIMD4-NEXT:    store i32 [[ADD4]], i32* [[RES]], align 4, !dbg [[DBG189]]
+// SIMD4-NEXT:    [[TMP15:%.*]] = load i32, i32* @_ZN2STIiE2stE, align 4, !dbg [[DBG190:![0-9]+]]
+// SIMD4-NEXT:    [[TMP16:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG191:![0-9]+]]
+// SIMD4-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP16]], [[TMP15]], !dbg [[DBG191]]
+// SIMD4-NEXT:    store i32 [[ADD5]], i32* [[RES]], align 4, !dbg [[DBG191]]
+// SIMD4-NEXT:    [[TMP17:%.*]] = load float, float* @_ZN2STIfE2stE, align 4, !dbg [[DBG192:![0-9]+]]
+// SIMD4-NEXT:    [[CONV:%.*]] = fptosi float [[TMP17]] to i32, !dbg [[DBG192]]
+// SIMD4-NEXT:    [[TMP18:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG193:![0-9]+]]
+// SIMD4-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP18]], [[CONV]], !dbg [[DBG193]]
+// SIMD4-NEXT:    store i32 [[ADD6]], i32* [[RES]], align 4, !dbg [[DBG193]]
+// SIMD4-NEXT:    [[TMP19:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S4:%.*]], %struct.S4* @_ZN2STI2S4E2stE, i32 0, i32 0), align 4, !dbg [[DBG194:![0-9]+]]
+// SIMD4-NEXT:    [[TMP20:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG195:![0-9]+]]
+// SIMD4-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP20]], [[TMP19]], !dbg [[DBG195]]
+// SIMD4-NEXT:    store i32 [[ADD7]], i32* [[RES]], align 4, !dbg [[DBG195]]
+// SIMD4-NEXT:    [[TMP21:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG196:![0-9]+]]
+// SIMD4-NEXT:    ret i32 [[TMP21]], !dbg [[DBG197:![0-9]+]]
 // SIMD4:       lpad:
 // SIMD4-NEXT:    [[TMP22:%.*]] = landingpad { i8*, i32 }
-// SIMD4-NEXT:    cleanup, !dbg [[DBG199:![0-9]+]]
-// SIMD4-NEXT:    [[TMP23:%.*]] = extractvalue { i8*, i32 } [[TMP22]], 0, !dbg [[DBG199]]
-// SIMD4-NEXT:    store i8* [[TMP23]], i8** [[EXN_SLOT]], align 8, !dbg [[DBG199]]
-// SIMD4-NEXT:    [[TMP24:%.*]] = extractvalue { i8*, i32 } [[TMP22]], 1, !dbg [[DBG199]]
-// SIMD4-NEXT:    store i32 [[TMP24]], i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG199]]
-// SIMD4-NEXT:    call void @__cxa_guard_abort(i64* @_ZGVZ4mainE2sm) #[[ATTR3]], !dbg [[DBG175]]
-// SIMD4-NEXT:    br label [[EH_RESUME:%.*]], !dbg [[DBG175]]
+// SIMD4-NEXT:    cleanup, !dbg [[DBG198:![0-9]+]]
+// SIMD4-NEXT:    [[TMP23:%.*]] = extractvalue { i8*, i32 } [[TMP22]], 0, !dbg [[DBG198]]
+// SIMD4-NEXT:    store i8* [[TMP23]], i8** [[EXN_SLOT]], align 8, !dbg [[DBG198]]
+// SIMD4-NEXT:    [[TMP24:%.*]] = extractvalue { i8*, i32 } [[TMP22]], 1, !dbg [[DBG198]]
+// SIMD4-NEXT:    store i32 [[TMP24]], i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG198]]
+// SIMD4-NEXT:    call void @__cxa_guard_abort(i64* @_ZGVZ4mainE2sm) #[[ATTR3]], !dbg [[DBG174]]
+// SIMD4-NEXT:    br label [[EH_RESUME:%.*]], !dbg [[DBG174]]
 // SIMD4:       eh.resume:
-// SIMD4-NEXT:    [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8, !dbg [[DBG175]]
-// SIMD4-NEXT:    [[SEL:%.*]] = load i32, i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG175]]
-// SIMD4-NEXT:    [[LPAD_VAL:%.*]] = insertvalue { i8*, i32 } undef, i8* [[EXN]], 0, !dbg [[DBG175]]
-// SIMD4-NEXT:    [[LPAD_VAL8:%.*]] = insertvalue { i8*, i32 } [[LPAD_VAL]], i32 [[SEL]], 1, !dbg [[DBG175]]
-// SIMD4-NEXT:    resume { i8*, i32 } [[LPAD_VAL8]], !dbg [[DBG175]]
+// SIMD4-NEXT:    [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8, !dbg [[DBG174]]
+// SIMD4-NEXT:    [[SEL:%.*]] = load i32, i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG174]]
+// SIMD4-NEXT:    [[LPAD_VAL:%.*]] = insertvalue { i8*, i32 } undef, i8* [[EXN]], 0, !dbg [[DBG174]]
+// SIMD4-NEXT:    [[LPAD_VAL8:%.*]] = insertvalue { i8*, i32 } [[LPAD_VAL]], i32 [[SEL]], 1, !dbg [[DBG174]]
+// SIMD4-NEXT:    resume { i8*, i32 } [[LPAD_VAL8]], !dbg [[DBG174]]
 //
 //
 // SIMD4-LABEL: define {{[^@]+}}@_ZZ4mainEN5SmainC1Ei
-// SIMD4-SAME: (%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR1]] align 2 !dbg [[DBG200:![0-9]+]] {
+// SIMD4-SAME: (%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR1]] align 2 !dbg [[DBG199:![0-9]+]] {
 // SIMD4-NEXT:  entry:
 // SIMD4-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.Smain*, align 8
 // SIMD4-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
 // SIMD4-NEXT:    store %struct.Smain* [[THIS]], %struct.Smain** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.Smain** [[THIS_ADDR]], metadata [[META201:![0-9]+]], metadata !DIExpression()), !dbg [[DBG203:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.Smain** [[THIS_ADDR]], metadata [[META200:![0-9]+]], metadata !DIExpression()), !dbg [[DBG202:![0-9]+]]
 // SIMD4-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META204:![0-9]+]], metadata !DIExpression()), !dbg [[DBG205:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META203:![0-9]+]], metadata !DIExpression()), !dbg [[DBG204:![0-9]+]]
 // SIMD4-NEXT:    [[THIS1:%.*]] = load %struct.Smain*, %struct.Smain** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG206:![0-9]+]]
-// SIMD4-NEXT:    call void @_ZZ4mainEN5SmainC2Ei(%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS1]], i32 [[TMP0]]), !dbg [[DBG206]]
-// SIMD4-NEXT:    ret void, !dbg [[DBG207:![0-9]+]]
+// SIMD4-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG205:![0-9]+]]
+// SIMD4-NEXT:    call void @_ZZ4mainEN5SmainC2Ei(%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS1]], i32 [[TMP0]]), !dbg [[DBG205]]
+// SIMD4-NEXT:    ret void, !dbg [[DBG206:![0-9]+]]
 //
 //
 // SIMD4-LABEL: define {{[^@]+}}@_ZZ4mainEN5SmainD1Ev
-// SIMD4-SAME: (%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] align 2 !dbg [[DBG208:![0-9]+]] {
+// SIMD4-SAME: (%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] align 2 !dbg [[DBG207:![0-9]+]] {
 // SIMD4-NEXT:  entry:
 // SIMD4-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.Smain*, align 8
 // SIMD4-NEXT:    store %struct.Smain* [[THIS]], %struct.Smain** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.Smain** [[THIS_ADDR]], metadata [[META209:![0-9]+]], metadata !DIExpression()), !dbg [[DBG210:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.Smain** [[THIS_ADDR]], metadata [[META208:![0-9]+]], metadata !DIExpression()), !dbg [[DBG209:![0-9]+]]
 // SIMD4-NEXT:    [[THIS1:%.*]] = load %struct.Smain*, %struct.Smain** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    call void @_ZZ4mainEN5SmainD2Ev(%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS1]]) #[[ATTR3]], !dbg [[DBG211:![0-9]+]]
-// SIMD4-NEXT:    ret void, !dbg [[DBG212:![0-9]+]]
+// SIMD4-NEXT:    call void @_ZZ4mainEN5SmainD2Ev(%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS1]]) #[[ATTR3]], !dbg [[DBG210:![0-9]+]]
+// SIMD4-NEXT:    ret void, !dbg [[DBG211:![0-9]+]]
 //
 //
 // SIMD4-LABEL: define {{[^@]+}}@_Z6foobarv
-// SIMD4-SAME: () #[[ATTR6:[0-9]+]] !dbg [[DBG213:![0-9]+]] {
+// SIMD4-SAME: () #[[ATTR6:[0-9]+]] !dbg [[DBG212:![0-9]+]] {
 // SIMD4-NEXT:  entry:
 // SIMD4-NEXT:    [[RES:%.*]] = alloca i32, align 4
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[RES]], metadata [[META214:![0-9]+]], metadata !DIExpression()), !dbg [[DBG215:![0-9]+]]
-// SIMD4-NEXT:    [[TMP0:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S3:%.*]], %struct.S3* @_ZN6Static1sE, i32 0, i32 0), align 4, !dbg [[DBG216:![0-9]+]]
-// SIMD4-NEXT:    store i32 [[TMP0]], i32* [[RES]], align 4, !dbg [[DBG217:![0-9]+]]
-// SIMD4-NEXT:    [[TMP1:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S1:%.*]], %struct.S1* @_ZL3gs1, i32 0, i32 0), align 4, !dbg [[DBG218:![0-9]+]]
-// SIMD4-NEXT:    [[TMP2:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG219:![0-9]+]]
-// SIMD4-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], [[TMP1]], !dbg [[DBG219]]
-// SIMD4-NEXT:    store i32 [[ADD]], i32* [[RES]], align 4, !dbg [[DBG219]]
-// SIMD4-NEXT:    [[TMP3:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S2:%.*]], %struct.S2* @_ZL3gs2, i32 0, i32 0), align 8, !dbg [[DBG220:![0-9]+]]
-// SIMD4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG221:![0-9]+]]
-// SIMD4-NEXT:    [[ADD1:%.*]] = add nsw i32 [[TMP4]], [[TMP3]], !dbg [[DBG221]]
-// SIMD4-NEXT:    store i32 [[ADD1]], i32* [[RES]], align 4, !dbg [[DBG221]]
-// SIMD4-NEXT:    [[TMP5:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S5:%.*]], %struct.S5* @gs3, i32 0, i32 0), align 4, !dbg [[DBG222:![0-9]+]]
-// SIMD4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG223:![0-9]+]]
-// SIMD4-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP6]], [[TMP5]], !dbg [[DBG223]]
-// SIMD4-NEXT:    store i32 [[ADD2]], i32* [[RES]], align 4, !dbg [[DBG223]]
-// SIMD4-NEXT:    [[TMP7:%.*]] = load i32, i32* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 1, i32 0), align 4, !dbg [[DBG224:![0-9]+]]
-// SIMD4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG225:![0-9]+]]
-// SIMD4-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP8]], [[TMP7]], !dbg [[DBG225]]
-// SIMD4-NEXT:    store i32 [[ADD3]], i32* [[RES]], align 4, !dbg [[DBG225]]
-// SIMD4-NEXT:    [[TMP9:%.*]] = load i32, i32* @_ZN2STIiE2stE, align 4, !dbg [[DBG226:![0-9]+]]
-// SIMD4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG227:![0-9]+]]
-// SIMD4-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP10]], [[TMP9]], !dbg [[DBG227]]
-// SIMD4-NEXT:    store i32 [[ADD4]], i32* [[RES]], align 4, !dbg [[DBG227]]
-// SIMD4-NEXT:    [[TMP11:%.*]] = load float, float* @_ZN2STIfE2stE, align 4, !dbg [[DBG228:![0-9]+]]
-// SIMD4-NEXT:    [[CONV:%.*]] = fptosi float [[TMP11]] to i32, !dbg [[DBG228]]
-// SIMD4-NEXT:    [[TMP12:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG229:![0-9]+]]
-// SIMD4-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP12]], [[CONV]], !dbg [[DBG229]]
-// SIMD4-NEXT:    store i32 [[ADD5]], i32* [[RES]], align 4, !dbg [[DBG229]]
-// SIMD4-NEXT:    [[TMP13:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S4:%.*]], %struct.S4* @_ZN2STI2S4E2stE, i32 0, i32 0), align 4, !dbg [[DBG230:![0-9]+]]
-// SIMD4-NEXT:    [[TMP14:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG231:![0-9]+]]
-// SIMD4-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP14]], [[TMP13]], !dbg [[DBG231]]
-// SIMD4-NEXT:    store i32 [[ADD6]], i32* [[RES]], align 4, !dbg [[DBG231]]
-// SIMD4-NEXT:    [[TMP15:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG232:![0-9]+]]
-// SIMD4-NEXT:    ret i32 [[TMP15]], !dbg [[DBG233:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[RES]], metadata [[META213:![0-9]+]], metadata !DIExpression()), !dbg [[DBG214:![0-9]+]]
+// SIMD4-NEXT:    [[TMP0:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S3:%.*]], %struct.S3* @_ZN6Static1sE, i32 0, i32 0), align 4, !dbg [[DBG215:![0-9]+]]
+// SIMD4-NEXT:    store i32 [[TMP0]], i32* [[RES]], align 4, !dbg [[DBG216:![0-9]+]]
+// SIMD4-NEXT:    [[TMP1:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S1:%.*]], %struct.S1* @_ZL3gs1, i32 0, i32 0), align 4, !dbg [[DBG217:![0-9]+]]
+// SIMD4-NEXT:    [[TMP2:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG218:![0-9]+]]
+// SIMD4-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], [[TMP1]], !dbg [[DBG218]]
+// SIMD4-NEXT:    store i32 [[ADD]], i32* [[RES]], align 4, !dbg [[DBG218]]
+// SIMD4-NEXT:    [[TMP3:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S2:%.*]], %struct.S2* @_ZL3gs2, i32 0, i32 0), align 8, !dbg [[DBG219:![0-9]+]]
+// SIMD4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG220:![0-9]+]]
+// SIMD4-NEXT:    [[ADD1:%.*]] = add nsw i32 [[TMP4]], [[TMP3]], !dbg [[DBG220]]
+// SIMD4-NEXT:    store i32 [[ADD1]], i32* [[RES]], align 4, !dbg [[DBG220]]
+// SIMD4-NEXT:    [[TMP5:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S5:%.*]], %struct.S5* @gs3, i32 0, i32 0), align 4, !dbg [[DBG221:![0-9]+]]
+// SIMD4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG222:![0-9]+]]
+// SIMD4-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP6]], [[TMP5]], !dbg [[DBG222]]
+// SIMD4-NEXT:    store i32 [[ADD2]], i32* [[RES]], align 4, !dbg [[DBG222]]
+// SIMD4-NEXT:    [[TMP7:%.*]] = load i32, i32* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 0, i64 1, i64 1, i32 0), align 4, !dbg [[DBG223:![0-9]+]]
+// SIMD4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG224:![0-9]+]]
+// SIMD4-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP8]], [[TMP7]], !dbg [[DBG224]]
+// SIMD4-NEXT:    store i32 [[ADD3]], i32* [[RES]], align 4, !dbg [[DBG224]]
+// SIMD4-NEXT:    [[TMP9:%.*]] = load i32, i32* @_ZN2STIiE2stE, align 4, !dbg [[DBG225:![0-9]+]]
+// SIMD4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG226:![0-9]+]]
+// SIMD4-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP10]], [[TMP9]], !dbg [[DBG226]]
+// SIMD4-NEXT:    store i32 [[ADD4]], i32* [[RES]], align 4, !dbg [[DBG226]]
+// SIMD4-NEXT:    [[TMP11:%.*]] = load float, float* @_ZN2STIfE2stE, align 4, !dbg [[DBG227:![0-9]+]]
+// SIMD4-NEXT:    [[CONV:%.*]] = fptosi float [[TMP11]] to i32, !dbg [[DBG227]]
+// SIMD4-NEXT:    [[TMP12:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG228:![0-9]+]]
+// SIMD4-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP12]], [[CONV]], !dbg [[DBG228]]
+// SIMD4-NEXT:    store i32 [[ADD5]], i32* [[RES]], align 4, !dbg [[DBG228]]
+// SIMD4-NEXT:    [[TMP13:%.*]] = load i32, i32* getelementptr inbounds ([[STRUCT_S4:%.*]], %struct.S4* @_ZN2STI2S4E2stE, i32 0, i32 0), align 4, !dbg [[DBG229:![0-9]+]]
+// SIMD4-NEXT:    [[TMP14:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG230:![0-9]+]]
+// SIMD4-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP14]], [[TMP13]], !dbg [[DBG230]]
+// SIMD4-NEXT:    store i32 [[ADD6]], i32* [[RES]], align 4, !dbg [[DBG230]]
+// SIMD4-NEXT:    [[TMP15:%.*]] = load i32, i32* [[RES]], align 4, !dbg [[DBG231:![0-9]+]]
+// SIMD4-NEXT:    ret i32 [[TMP15]], !dbg [[DBG232:![0-9]+]]
 //
 //
 // SIMD4-LABEL: define {{[^@]+}}@__cxx_global_var_init.3
-// SIMD4-SAME: () #[[ATTR0]] comdat($_ZN2STI2S4E2stE) !dbg [[DBG234:![0-9]+]] {
+// SIMD4-SAME: () #[[ATTR0]] comdat($_ZN2STI2S4E2stE) !dbg [[DBG233:![0-9]+]] {
 // SIMD4-NEXT:  entry:
-// SIMD4-NEXT:    [[TMP0:%.*]] = load i8, i8* bitcast (i64* @_ZGVN2STI2S4E2stE to i8*), align 8, !dbg [[DBG235:![0-9]+]]
-// SIMD4-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0, !dbg [[DBG235]]
-// SIMD4-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !dbg [[DBG235]]
+// SIMD4-NEXT:    [[TMP0:%.*]] = load i8, i8* bitcast (i64* @_ZGVN2STI2S4E2stE to i8*), align 8, !dbg [[DBG234:![0-9]+]]
+// SIMD4-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0, !dbg [[DBG234]]
+// SIMD4-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !dbg [[DBG234]]
 // SIMD4:       init.check:
-// SIMD4-NEXT:    call void @_ZN2S4C1Ei(%struct.S4* nonnull align 4 dereferenceable(8) @_ZN2STI2S4E2stE, i32 23), !dbg [[DBG236:![0-9]+]]
-// SIMD4-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.S4*)* @_ZN2S4D1Ev to void (i8*)*), i8* bitcast (%struct.S4* @_ZN2STI2S4E2stE to i8*), i8* @__dso_handle) #[[ATTR3]], !dbg [[DBG235]]
-// SIMD4-NEXT:    store i8 1, i8* bitcast (i64* @_ZGVN2STI2S4E2stE to i8*), align 8, !dbg [[DBG235]]
-// SIMD4-NEXT:    br label [[INIT_END]], !dbg [[DBG235]]
+// SIMD4-NEXT:    call void @_ZN2S4C1Ei(%struct.S4* nonnull align 4 dereferenceable(8) @_ZN2STI2S4E2stE, i32 23), !dbg [[DBG235:![0-9]+]]
+// SIMD4-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.S4*)* @_ZN2S4D1Ev to void (i8*)*), i8* bitcast (%struct.S4* @_ZN2STI2S4E2stE to i8*), i8* @__dso_handle) #[[ATTR3]], !dbg [[DBG234]]
+// SIMD4-NEXT:    store i8 1, i8* bitcast (i64* @_ZGVN2STI2S4E2stE to i8*), align 8, !dbg [[DBG234]]
+// SIMD4-NEXT:    br label [[INIT_END]], !dbg [[DBG234]]
 // SIMD4:       init.end:
-// SIMD4-NEXT:    ret void, !dbg [[DBG238:![0-9]+]]
+// SIMD4-NEXT:    ret void, !dbg [[DBG237:![0-9]+]]
 //
 //
 // SIMD4-LABEL: define {{[^@]+}}@_ZN2S4C1Ei
-// SIMD4-SAME: (%struct.S4* nonnull align 4 dereferenceable(8) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 !dbg [[DBG239:![0-9]+]] {
+// SIMD4-SAME: (%struct.S4* nonnull align 4 dereferenceable(8) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 !dbg [[DBG238:![0-9]+]] {
 // SIMD4-NEXT:  entry:
 // SIMD4-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S4*, align 8
 // SIMD4-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
 // SIMD4-NEXT:    store %struct.S4* [[THIS]], %struct.S4** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.S4** [[THIS_ADDR]], metadata [[META240:![0-9]+]], metadata !DIExpression()), !dbg [[DBG242:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.S4** [[THIS_ADDR]], metadata [[META239:![0-9]+]], metadata !DIExpression()), !dbg [[DBG241:![0-9]+]]
 // SIMD4-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META243:![0-9]+]], metadata !DIExpression()), !dbg [[DBG244:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META242:![0-9]+]], metadata !DIExpression()), !dbg [[DBG243:![0-9]+]]
 // SIMD4-NEXT:    [[THIS1:%.*]] = load %struct.S4*, %struct.S4** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG245:![0-9]+]]
-// SIMD4-NEXT:    call void @_ZN2S4C2Ei(%struct.S4* nonnull align 4 dereferenceable(8) [[THIS1]], i32 [[TMP0]]), !dbg [[DBG245]]
-// SIMD4-NEXT:    ret void, !dbg [[DBG246:![0-9]+]]
+// SIMD4-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG244:![0-9]+]]
+// SIMD4-NEXT:    call void @_ZN2S4C2Ei(%struct.S4* nonnull align 4 dereferenceable(8) [[THIS1]], i32 [[TMP0]]), !dbg [[DBG244]]
+// SIMD4-NEXT:    ret void, !dbg [[DBG245:![0-9]+]]
 //
 //
 // SIMD4-LABEL: define {{[^@]+}}@_ZN2S4D1Ev
-// SIMD4-SAME: (%struct.S4* nonnull align 4 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG247:![0-9]+]] {
+// SIMD4-SAME: (%struct.S4* nonnull align 4 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG246:![0-9]+]] {
 // SIMD4-NEXT:  entry:
 // SIMD4-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S4*, align 8
 // SIMD4-NEXT:    store %struct.S4* [[THIS]], %struct.S4** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.S4** [[THIS_ADDR]], metadata [[META248:![0-9]+]], metadata !DIExpression()), !dbg [[DBG249:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.S4** [[THIS_ADDR]], metadata [[META247:![0-9]+]], metadata !DIExpression()), !dbg [[DBG248:![0-9]+]]
 // SIMD4-NEXT:    [[THIS1:%.*]] = load %struct.S4*, %struct.S4** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    call void @_ZN2S4D2Ev(%struct.S4* nonnull align 4 dereferenceable(8) [[THIS1]]) #[[ATTR3]], !dbg [[DBG250:![0-9]+]]
-// SIMD4-NEXT:    ret void, !dbg [[DBG251:![0-9]+]]
+// SIMD4-NEXT:    call void @_ZN2S4D2Ev(%struct.S4* nonnull align 4 dereferenceable(8) [[THIS1]]) #[[ATTR3]], !dbg [[DBG249:![0-9]+]]
+// SIMD4-NEXT:    ret void, !dbg [[DBG250:![0-9]+]]
 //
 //
 // SIMD4-LABEL: define {{[^@]+}}@_ZN2S1C2Ei
-// SIMD4-SAME: (%struct.S1* nonnull align 4 dereferenceable(4) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG252:![0-9]+]] {
+// SIMD4-SAME: (%struct.S1* nonnull align 4 dereferenceable(4) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG251:![0-9]+]] {
 // SIMD4-NEXT:  entry:
 // SIMD4-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S1*, align 8
 // SIMD4-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
 // SIMD4-NEXT:    store %struct.S1* [[THIS]], %struct.S1** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.S1** [[THIS_ADDR]], metadata [[META253:![0-9]+]], metadata !DIExpression()), !dbg [[DBG254:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.S1** [[THIS_ADDR]], metadata [[META252:![0-9]+]], metadata !DIExpression()), !dbg [[DBG253:![0-9]+]]
 // SIMD4-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META255:![0-9]+]], metadata !DIExpression()), !dbg [[DBG256:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META254:![0-9]+]], metadata !DIExpression()), !dbg [[DBG255:![0-9]+]]
 // SIMD4-NEXT:    [[THIS1:%.*]] = load %struct.S1*, %struct.S1** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    [[A2:%.*]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[THIS1]], i32 0, i32 0, !dbg [[DBG257:![0-9]+]]
-// SIMD4-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG258:![0-9]+]]
-// SIMD4-NEXT:    store i32 [[TMP0]], i32* [[A2]], align 4, !dbg [[DBG257]]
-// SIMD4-NEXT:    ret void, !dbg [[DBG259:![0-9]+]]
+// SIMD4-NEXT:    [[A2:%.*]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[THIS1]], i32 0, i32 0, !dbg [[DBG256:![0-9]+]]
+// SIMD4-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG257:![0-9]+]]
+// SIMD4-NEXT:    store i32 [[TMP0]], i32* [[A2]], align 4, !dbg [[DBG256]]
+// SIMD4-NEXT:    ret void, !dbg [[DBG258:![0-9]+]]
 //
 //
 // SIMD4-LABEL: define {{[^@]+}}@_ZN2S1D2Ev
-// SIMD4-SAME: (%struct.S1* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG260:![0-9]+]] {
+// SIMD4-SAME: (%struct.S1* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG259:![0-9]+]] {
 // SIMD4-NEXT:  entry:
 // SIMD4-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S1*, align 8
 // SIMD4-NEXT:    store %struct.S1* [[THIS]], %struct.S1** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.S1** [[THIS_ADDR]], metadata [[META261:![0-9]+]], metadata !DIExpression()), !dbg [[DBG262:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.S1** [[THIS_ADDR]], metadata [[META260:![0-9]+]], metadata !DIExpression()), !dbg [[DBG261:![0-9]+]]
 // SIMD4-NEXT:    [[THIS1:%.*]] = load %struct.S1*, %struct.S1** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[THIS1]], i32 0, i32 0, !dbg [[DBG263:![0-9]+]]
-// SIMD4-NEXT:    store i32 0, i32* [[A]], align 4, !dbg [[DBG265:![0-9]+]]
-// SIMD4-NEXT:    ret void, !dbg [[DBG266:![0-9]+]]
+// SIMD4-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[THIS1]], i32 0, i32 0, !dbg [[DBG262:![0-9]+]]
+// SIMD4-NEXT:    store i32 0, i32* [[A]], align 4, !dbg [[DBG264:![0-9]+]]
+// SIMD4-NEXT:    ret void, !dbg [[DBG265:![0-9]+]]
 //
 //
 // SIMD4-LABEL: define {{[^@]+}}@_ZN2S2C2Ei
-// SIMD4-SAME: (%struct.S2* nonnull align 8 dereferenceable(16) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG267:![0-9]+]] {
+// SIMD4-SAME: (%struct.S2* nonnull align 8 dereferenceable(16) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG266:![0-9]+]] {
 // SIMD4-NEXT:  entry:
 // SIMD4-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S2*, align 8
 // SIMD4-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
 // SIMD4-NEXT:    store %struct.S2* [[THIS]], %struct.S2** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.S2** [[THIS_ADDR]], metadata [[META268:![0-9]+]], metadata !DIExpression()), !dbg [[DBG269:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.S2** [[THIS_ADDR]], metadata [[META267:![0-9]+]], metadata !DIExpression()), !dbg [[DBG268:![0-9]+]]
 // SIMD4-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META270:![0-9]+]], metadata !DIExpression()), !dbg [[DBG271:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META269:![0-9]+]], metadata !DIExpression()), !dbg [[DBG270:![0-9]+]]
 // SIMD4-NEXT:    [[THIS1:%.*]] = load %struct.S2*, %struct.S2** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    [[A2:%.*]] = getelementptr inbounds [[STRUCT_S2:%.*]], %struct.S2* [[THIS1]], i32 0, i32 0, !dbg [[DBG272:![0-9]+]]
-// SIMD4-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG273:![0-9]+]]
-// SIMD4-NEXT:    store i32 [[TMP0]], i32* [[A2]], align 8, !dbg [[DBG272]]
-// SIMD4-NEXT:    ret void, !dbg [[DBG274:![0-9]+]]
+// SIMD4-NEXT:    [[A2:%.*]] = getelementptr inbounds [[STRUCT_S2:%.*]], %struct.S2* [[THIS1]], i32 0, i32 0, !dbg [[DBG271:![0-9]+]]
+// SIMD4-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG272:![0-9]+]]
+// SIMD4-NEXT:    store i32 [[TMP0]], i32* [[A2]], align 8, !dbg [[DBG271]]
+// SIMD4-NEXT:    ret void, !dbg [[DBG273:![0-9]+]]
 //
 //
 // SIMD4-LABEL: define {{[^@]+}}@_ZN2S2D2Ev
-// SIMD4-SAME: (%struct.S2* nonnull align 8 dereferenceable(16) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG275:![0-9]+]] {
+// SIMD4-SAME: (%struct.S2* nonnull align 8 dereferenceable(16) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG274:![0-9]+]] {
 // SIMD4-NEXT:  entry:
 // SIMD4-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S2*, align 8
 // SIMD4-NEXT:    store %struct.S2* [[THIS]], %struct.S2** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.S2** [[THIS_ADDR]], metadata [[META276:![0-9]+]], metadata !DIExpression()), !dbg [[DBG277:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.S2** [[THIS_ADDR]], metadata [[META275:![0-9]+]], metadata !DIExpression()), !dbg [[DBG276:![0-9]+]]
 // SIMD4-NEXT:    [[THIS1:%.*]] = load %struct.S2*, %struct.S2** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_S2:%.*]], %struct.S2* [[THIS1]], i32 0, i32 0, !dbg [[DBG278:![0-9]+]]
-// SIMD4-NEXT:    store i32 0, i32* [[A]], align 8, !dbg [[DBG280:![0-9]+]]
-// SIMD4-NEXT:    ret void, !dbg [[DBG281:![0-9]+]]
+// SIMD4-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_S2:%.*]], %struct.S2* [[THIS1]], i32 0, i32 0, !dbg [[DBG277:![0-9]+]]
+// SIMD4-NEXT:    store i32 0, i32* [[A]], align 8, !dbg [[DBG279:![0-9]+]]
+// SIMD4-NEXT:    ret void, !dbg [[DBG280:![0-9]+]]
 //
 //
 // SIMD4-LABEL: define {{[^@]+}}@_ZZ4mainEN5SmainC2Ei
-// SIMD4-SAME: (%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR2]] align 2 !dbg [[DBG282:![0-9]+]] {
+// SIMD4-SAME: (%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR2]] align 2 !dbg [[DBG281:![0-9]+]] {
 // SIMD4-NEXT:  entry:
 // SIMD4-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.Smain*, align 8
 // SIMD4-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
 // SIMD4-NEXT:    store %struct.Smain* [[THIS]], %struct.Smain** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.Smain** [[THIS_ADDR]], metadata [[META283:![0-9]+]], metadata !DIExpression()), !dbg [[DBG284:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.Smain** [[THIS_ADDR]], metadata [[META282:![0-9]+]], metadata !DIExpression()), !dbg [[DBG283:![0-9]+]]
 // SIMD4-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META285:![0-9]+]], metadata !DIExpression()), !dbg [[DBG286:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META284:![0-9]+]], metadata !DIExpression()), !dbg [[DBG285:![0-9]+]]
 // SIMD4-NEXT:    [[THIS1:%.*]] = load %struct.Smain*, %struct.Smain** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    [[A2:%.*]] = getelementptr inbounds [[STRUCT_SMAIN:%.*]], %struct.Smain* [[THIS1]], i32 0, i32 0, !dbg [[DBG287:![0-9]+]]
-// SIMD4-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG288:![0-9]+]]
-// SIMD4-NEXT:    store i32 [[TMP0]], i32* [[A2]], align 8, !dbg [[DBG287]]
-// SIMD4-NEXT:    ret void, !dbg [[DBG289:![0-9]+]]
+// SIMD4-NEXT:    [[A2:%.*]] = getelementptr inbounds [[STRUCT_SMAIN:%.*]], %struct.Smain* [[THIS1]], i32 0, i32 0, !dbg [[DBG286:![0-9]+]]
+// SIMD4-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG287:![0-9]+]]
+// SIMD4-NEXT:    store i32 [[TMP0]], i32* [[A2]], align 8, !dbg [[DBG286]]
+// SIMD4-NEXT:    ret void, !dbg [[DBG288:![0-9]+]]
 //
 //
 // SIMD4-LABEL: define {{[^@]+}}@_ZZ4mainEN5SmainD2Ev
-// SIMD4-SAME: (%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] align 2 !dbg [[DBG290:![0-9]+]] {
+// SIMD4-SAME: (%struct.Smain* nonnull align 8 dereferenceable(24) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] align 2 !dbg [[DBG289:![0-9]+]] {
 // SIMD4-NEXT:  entry:
 // SIMD4-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.Smain*, align 8
 // SIMD4-NEXT:    store %struct.Smain* [[THIS]], %struct.Smain** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.Smain** [[THIS_ADDR]], metadata [[META291:![0-9]+]], metadata !DIExpression()), !dbg [[DBG292:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.Smain** [[THIS_ADDR]], metadata [[META290:![0-9]+]], metadata !DIExpression()), !dbg [[DBG291:![0-9]+]]
 // SIMD4-NEXT:    [[THIS1:%.*]] = load %struct.Smain*, %struct.Smain** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_SMAIN:%.*]], %struct.Smain* [[THIS1]], i32 0, i32 0, !dbg [[DBG293:![0-9]+]]
-// SIMD4-NEXT:    store i32 0, i32* [[A]], align 8, !dbg [[DBG295:![0-9]+]]
-// SIMD4-NEXT:    ret void, !dbg [[DBG296:![0-9]+]]
+// SIMD4-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_SMAIN:%.*]], %struct.Smain* [[THIS1]], i32 0, i32 0, !dbg [[DBG292:![0-9]+]]
+// SIMD4-NEXT:    store i32 0, i32* [[A]], align 8, !dbg [[DBG294:![0-9]+]]
+// SIMD4-NEXT:    ret void, !dbg [[DBG295:![0-9]+]]
 //
 //
 // SIMD4-LABEL: define {{[^@]+}}@_ZN2S4C2Ei
-// SIMD4-SAME: (%struct.S4* nonnull align 4 dereferenceable(8) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG297:![0-9]+]] {
+// SIMD4-SAME: (%struct.S4* nonnull align 4 dereferenceable(8) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG296:![0-9]+]] {
 // SIMD4-NEXT:  entry:
 // SIMD4-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S4*, align 8
 // SIMD4-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
 // SIMD4-NEXT:    store %struct.S4* [[THIS]], %struct.S4** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.S4** [[THIS_ADDR]], metadata [[META298:![0-9]+]], metadata !DIExpression()), !dbg [[DBG299:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.S4** [[THIS_ADDR]], metadata [[META297:![0-9]+]], metadata !DIExpression()), !dbg [[DBG298:![0-9]+]]
 // SIMD4-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META300:![0-9]+]], metadata !DIExpression()), !dbg [[DBG301:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META299:![0-9]+]], metadata !DIExpression()), !dbg [[DBG300:![0-9]+]]
 // SIMD4-NEXT:    [[THIS1:%.*]] = load %struct.S4*, %struct.S4** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    [[A2:%.*]] = getelementptr inbounds [[STRUCT_S4:%.*]], %struct.S4* [[THIS1]], i32 0, i32 0, !dbg [[DBG302:![0-9]+]]
-// SIMD4-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG303:![0-9]+]]
-// SIMD4-NEXT:    store i32 [[TMP0]], i32* [[A2]], align 4, !dbg [[DBG302]]
-// SIMD4-NEXT:    ret void, !dbg [[DBG304:![0-9]+]]
+// SIMD4-NEXT:    [[A2:%.*]] = getelementptr inbounds [[STRUCT_S4:%.*]], %struct.S4* [[THIS1]], i32 0, i32 0, !dbg [[DBG301:![0-9]+]]
+// SIMD4-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG302:![0-9]+]]
+// SIMD4-NEXT:    store i32 [[TMP0]], i32* [[A2]], align 4, !dbg [[DBG301]]
+// SIMD4-NEXT:    ret void, !dbg [[DBG303:![0-9]+]]
 //
 //
 // SIMD4-LABEL: define {{[^@]+}}@_ZN2S4D2Ev
-// SIMD4-SAME: (%struct.S4* nonnull align 4 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG305:![0-9]+]] {
+// SIMD4-SAME: (%struct.S4* nonnull align 4 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 !dbg [[DBG304:![0-9]+]] {
 // SIMD4-NEXT:  entry:
 // SIMD4-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S4*, align 8
 // SIMD4-NEXT:    store %struct.S4* [[THIS]], %struct.S4** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.S4** [[THIS_ADDR]], metadata [[META306:![0-9]+]], metadata !DIExpression()), !dbg [[DBG307:![0-9]+]]
+// SIMD4-NEXT:    call void @llvm.dbg.declare(metadata %struct.S4** [[THIS_ADDR]], metadata [[META305:![0-9]+]], metadata !DIExpression()), !dbg [[DBG306:![0-9]+]]
 // SIMD4-NEXT:    [[THIS1:%.*]] = load %struct.S4*, %struct.S4** [[THIS_ADDR]], align 8
-// SIMD4-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_S4:%.*]], %struct.S4* [[THIS1]], i32 0, i32 0, !dbg [[DBG308:![0-9]+]]
-// SIMD4-NEXT:    store i32 0, i32* [[A]], align 4, !dbg [[DBG310:![0-9]+]]
-// SIMD4-NEXT:    ret void, !dbg [[DBG311:![0-9]+]]
+// SIMD4-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_S4:%.*]], %struct.S4* [[THIS1]], i32 0, i32 0, !dbg [[DBG307:![0-9]+]]
+// SIMD4-NEXT:    store i32 0, i32* [[A]], align 4, !dbg [[DBG309:![0-9]+]]
+// SIMD4-NEXT:    ret void, !dbg [[DBG310:![0-9]+]]
 //
 //
 // SIMD4-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_threadprivate_codegen.cpp
-// SIMD4-SAME: () #[[ATTR0]] !dbg [[DBG312:![0-9]+]] {
+// SIMD4-SAME: () #[[ATTR0]] !dbg [[DBG311:![0-9]+]] {
 // SIMD4-NEXT:  entry:
-// SIMD4-NEXT:    call void @__cxx_global_var_init(), !dbg [[DBG314:![0-9]+]]
-// SIMD4-NEXT:    call void @__cxx_global_var_init.1(), !dbg [[DBG314]]
-// SIMD4-NEXT:    call void @__cxx_global_var_init.2(), !dbg [[DBG314]]
+// SIMD4-NEXT:    call void @__cxx_global_var_init(), !dbg [[DBG313:![0-9]+]]
+// SIMD4-NEXT:    call void @__cxx_global_var_init.1(), !dbg [[DBG313]]
+// SIMD4-NEXT:    call void @__cxx_global_var_init.2(), !dbg [[DBG313]]
 // SIMD4-NEXT:    ret void
 //
 //
 // DEBUG1-LABEL: define {{[^@]+}}@.__kmpc_global_ctor_.
-// DEBUG1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0:[0-9]+]] !dbg [[DBG115:![0-9]+]] {
+// DEBUG1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0:[0-9]+]] !dbg [[DBG116:![0-9]+]] {
 // DEBUG1-NEXT:  entry:
 // DEBUG1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // DEBUG1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
-// DEBUG1-NEXT:    call void @llvm.dbg.declare(metadata i8** [[DOTADDR]], metadata [[META117:![0-9]+]], metadata !DIExpression()), !dbg [[DBG119:![0-9]+]]
-// DEBUG1-NEXT:    [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !dbg [[DBG120:![0-9]+]]
-// DEBUG1-NEXT:    [[TMP2:%.*]] = bitcast i8* [[TMP1]] to %struct.S1*, !dbg [[DBG120]]
-// DEBUG1-NEXT:    call void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) [[TMP2]], i32 5), !dbg [[DBG121:![0-9]+]]
-// DEBUG1-NEXT:    [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !dbg [[DBG120]]
-// DEBUG1-NEXT:    ret i8* [[TMP3]], !dbg [[DBG120]]
+// DEBUG1-NEXT:    call void @llvm.dbg.declare(metadata i8** [[DOTADDR]], metadata [[META118:![0-9]+]], metadata !DIExpression()), !dbg [[DBG120:![0-9]+]]
+// DEBUG1-NEXT:    [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !dbg [[DBG121:![0-9]+]]
+// DEBUG1-NEXT:    [[TMP2:%.*]] = bitcast i8* [[TMP1]] to %struct.S1*, !dbg [[DBG121]]
+// DEBUG1-NEXT:    call void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) [[TMP2]], i32 5), !dbg [[DBG122:![0-9]+]]
+// DEBUG1-NEXT:    [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !dbg [[DBG121]]
+// DEBUG1-NEXT:    ret i8* [[TMP3]], !dbg [[DBG121]]
 //
 //
 // DEBUG1-LABEL: define {{[^@]+}}@_ZN2S1C1Ei
-// DEBUG1-SAME: (%struct.S1* nonnull align 4 dereferenceable(4) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR2:[0-9]+]] comdat align 2 !dbg [[DBG122:![0-9]+]] {
+// DEBUG1-SAME: (%struct.S1* nonnull align 4 dereferenceable(4) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR2:[0-9]+]] comdat align 2 !dbg [[DBG123:![0-9]+]] {
 // DEBUG1-NEXT:  entry:
 // DEBUG1-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S1*, align 8
 // DEBUG1-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
 // DEBUG1-NEXT:    store %struct.S1* [[THIS]], %struct.S1** [[THIS_ADDR]], align 8
-// DEBUG1-NEXT:    call void @llvm.dbg.declare(metadata %struct.S1** [[THIS_ADDR]], metadata [[META123:![0-9]+]], metadata !DIExpression()), !dbg [[DBG125:![0-9]+]]
+// DEBUG1-NEXT:    call void @llvm.dbg.declare(metadata %struct.S1** [[THIS_ADDR]], metadata [[META124:![0-9]+]], metadata !DIExpression()), !dbg [[DBG126:![0-9]+]]
 // DEBUG1-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
-// DEBUG1-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META126:![0-9]+]], metadata !DIExpression()), !dbg [[DBG127:![0-9]+]]
+// DEBUG1-NEXT:    call void @llvm.dbg.declare(metadata i32* [[A_ADDR]], metadata [[META127:![0-9]+]], metadata !DIExpression()), !dbg [[DBG128:![0-9]+]]
 // DEBUG1-NEXT:    [[THIS1:%.*]] = load %struct.S1*, %struct.S1** [[THIS_ADDR]], align 8
-// DEBUG1-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG128:![0-9]+]]
-// DEBUG1-NEXT:    call void @_ZN2S1C2Ei(%struct.S1* nonnull align 4 dereferenceable(4) [[THIS1]], i32 [[TMP0]]), !dbg [[DBG128]]
-// DEBUG1-NEXT:    ret void, !dbg [[DBG129:![0-9]+]]
+// DEBUG1-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4, !dbg [[DBG129:![0-9]+]]
+// DEBUG1-NEXT:    call void @_ZN2S1C2Ei(%struct.S1* nonnull align 4 dereferenceable(4) [[THIS1]], i32 [[TMP0]]), !dbg [[DBG129]]
+// DEBUG1-NEXT:    ret void, !dbg [[DBG130:![0-9]+]]
 //
 //
 // DEBUG1-LABEL: define {{[^@]+}}@.__kmpc_global_dtor_.
-// DEBUG1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] !dbg [[DBG130:![0-9]+]] {
+// DEBUG1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] !dbg [[DBG131:![0-9]+]] {
 // DEBUG1-NEXT:  entry:
 // DEBUG1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // DEBUG1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
-// DEBUG1-NEXT:    call void @llvm.dbg.declare(metadata i8** [[DOTADDR]], metadata [[META131:![0-9]+]], metadata !DIExpression()), !dbg [[DBG132:![0-9]+]]
-// DEBUG1-NEXT:    [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !dbg [[DBG132]]
-// DEBUG1-NEXT:    [[TMP2:%.*]] = bitcast i8* [[TMP1]] to %struct.S1*, !dbg [[DBG132]]
-// DEBUG1-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[TMP2]]) #[[ATTR4:[0-9]+]], !dbg [[DBG132]]
-// DEBUG1-NEXT:    ret void, !dbg [[DBG133:![0-9]+]]
+// DEBUG1-NEXT:    call void @llvm.dbg.declare(metadata i8** [[DOTADDR]], metadata [[META132:![0-9]+]], metadata !DIExpression()), !dbg [[DBG133:![0-9]+]]
+// DEBUG1-NEXT:    [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !dbg [[DBG133]]
+// DEBUG1-NEXT:    [[TMP2:%.*]] = bitcast i8* [[TMP1]] to %struct.S1*, !dbg [[DBG133]]
+// DEBUG1-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[TMP2]]) #[[ATTR4:[0-9]+]], !dbg [[DBG133]]
+// DEBUG1-NEXT:    ret void, !dbg [[DBG134:![0-9]+]]
 //
 //
 // DEBUG1-LABEL: define {{[^@]+}}@_ZN2S1D1Ev
-// DEBUG1-SAME: (%struct.S1* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR3:[0-9]+]] comdat align 2 !dbg [[DBG134:![0-9]+]] {
+// DEBUG1-SAME: (%struct.S1* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR3:[0-9]+]] comdat align 2 !dbg [[DBG135:![0-9]+]] {
 // DEBUG1-NEXT:  entry:
 // DEBUG1-NEXT:    [[THIS_ADDR:%.*]] = alloca %struct.S1*, align 8
 // DEBUG1-NEXT:    store %struct.S1* [[THIS]], %struct.S1** [[THIS_ADDR]], align 8
-// DEBUG1-NEXT:    call void @llvm.dbg.declare(metadata %struct.S1** [[THIS_ADDR]], metadata [[META135:![0-9]+]], metadata !DIExpression()), !dbg [[DBG136:![0-9]+]]
+// DEBUG1-NEXT:    call void @llvm.dbg.declare(metadata %struct.S1** [[THIS_ADDR]], metadata [[META136:![0-9]+]], metadata !DIExpression()), !dbg [[DBG137:![0-9]+]]
 // DEBUG1-NEXT:    [[THIS1:%.*]] = load %struct.S1*, %struct.S1** [[THIS_ADDR]], align 8
-// DEBUG1-NEXT:    call void @_ZN2S1D2Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[THIS1]]) #[[ATTR4]], !dbg [[DBG137:![0-9]+]]
-// DEBUG1-NEXT:    ret void, !dbg [[DBG138:![0-9]+]]
+// DEBUG1-NEXT:    call void @_ZN2S1D2Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[THIS1]]) #[[ATTR4]], !dbg [[DBG138:![0-9]+]]
+// DEBUG1-NEXT:    ret void, !dbg [[DBG139:![0-9]+]]
 //
 //
 // DEBUG1-LABEL: define {{[^@]+}}@.__omp_threadprivate_init_.
-// DEBUG1-SAME: () #[[ATTR0]] !dbg [[DBG139:![0-9]+]] {
+// DEBUG1-SAME: () #[[ATTR0]] !dbg [[DBG140:![0-9]+]] {
 // DEBUG1-NEXT:  entry:
-// DEBUG1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]]), !dbg [[DBG140:![0-9]+]]
-// DEBUG1-NEXT:    call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB1]], i8* bitcast (%struct.S1* @_ZL3gs1 to i8*), i8* (i8*)* @.__kmpc_global_ctor_., i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_.), !dbg [[DBG140]]
-// DEBUG1-NEXT:    ret void, !dbg [[DBG140]]
+// DEBUG1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]]), !dbg [[DBG141:![0-9]+]]
+// DEBUG1-NEXT:    call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB1]], i8* bitcast (%struct.S1* @_ZL3gs1 to i8*), i8* (i8*)* @.__kmpc_global_ctor_., i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_.), !dbg [[DBG141]]
+// DEBUG1-NEXT:    ret void, !dbg [[DBG141]]
 //
 //
 // DEBUG1-LABEL: define {{[^@]+}}@.__kmpc_global_ctor_..1
-// DEBUG1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) !dbg [[DBG141:![0-9]+]] {
+// DEBUG1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) !dbg [[DBG142:![0-9]+]] {
 // DEBUG1-NEXT:  entry:
 // DEBUG1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // DEBUG1-NEXT:    [[ARRAYINIT_ENDOFINIT:%.*]] = alloca [3 x %struct.S1]*, align 8
 // DEBUG1-NEXT:    [[ARRAYINIT_ENDOFINIT2:%.*]] = alloca %struct.S1*, align 8
 // DEBUG1-NEXT:    [[EXN_SLOT:%.*]] = alloca i8*, align 8
 // DEBUG1-NEXT:    [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
 // DEBUG1-NEXT:    [[ARRAYINIT_ENDOFINIT9:%.*]] = alloca %struct.S1*, align 8
 // DEBUG1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
-// DEBUG1-NEXT:    call void @llvm.dbg.declare(metadata i8** [[DOTADDR]], metadata [[META142:![0-9]+]], metadata !DIExpression()), !dbg [[DBG143:![0-9]+]]
-// DEBUG1-NEXT:    [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !dbg [[DBG144:![0-9]+]]
-// DEBUG1-NEXT:    [[TMP2:%.*]] = bitcast i8* [[TMP1]] to [2 x [3 x %struct.S1]]*, !dbg [[DBG144]]
-// DEBUG1-NEXT:    [[ARRAYINIT_BEGIN:%.*]] = getelementptr inbounds [2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* [[TMP2]], i64 0, i64 0, !dbg [[DBG145:![0-9]+]]
-// DEBUG1-NEXT:    store [3 x %struct.S1]* [[ARRAYINIT_BEGIN]], [3 x %struct.S1]** [[ARRAYINIT_ENDOFINIT]], align 8, !dbg [[DBG145]]
-// DEBUG1-NEXT:    [[ARRAYINIT_BEGIN1:%.*]] = getelementptr inbounds [3 x %struct.S1], [3 x %struct.S1]* [[ARRAYINIT_BEGIN]], i64 0, i64 0, !dbg [[DBG146:![0-9]+]]
-// DEBUG1-NEXT:    store %struct.S1* [[ARRAYINIT_BEGIN1]], %struct.S1** [[ARRAYINIT_ENDOFINIT2]], align 8, !dbg [[DBG146]]
+// DEBUG1-NEXT:    call void @llvm.dbg.declare(metadata i8** [[DOTADDR]], metadata [[META143:![0-9]+]], metadata !DIExpression()), !dbg [[DBG144:![0-9]+]]
+// DEBUG1-NEXT:    [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !dbg [[DBG145:![0-9]+]]
+// DEBUG1-NEXT:    [[TMP2:%.*]] = bitcast i8* [[TMP1]] to [2 x [3 x %struct.S1]]*, !dbg [[DBG145]]
+// DEBUG1-NEXT:    [[ARRAYINIT_BEGIN:%.*]] = getelementptr inbounds [2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* [[TMP2]], i64 0, i64 0, !dbg [[DBG146:![0-9]+]]
+// DEBUG1-NEXT:    store [3 x %struct.S1]* [[ARRAYINIT_BEGIN]], [3 x %struct.S1]** [[ARRAYINIT_ENDOFINIT]], align 8, !dbg [[DBG146]]
+// DEBUG1-NEXT:    [[ARRAYINIT_BEGIN1:%.*]] = getelementptr inbounds [3 x %struct.S1], [3 x %struct.S1]* [[ARRAYINIT_BEGIN]], i64 0, i64 0, !dbg [[DBG147:![0-9]+]]
+// DEBUG1-NEXT:    store %struct.S1* [[ARRAYINIT_BEGIN1]], %struct.S1** [[ARRAYINIT_ENDOFINIT2]], align 8, !dbg [[DBG147]]
 // DEBUG1-NEXT:    invoke void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYINIT_BEGIN1]], i32 1)
-// DEBUG1-NEXT:    to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]], !dbg [[DBG147:![0-9]+]]
+// DEBUG1-NEXT:    to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]], !dbg [[DBG148:![0-9]+]]
 // DEBUG1:       invoke.cont:
-// DEBUG1-NEXT:    [[ARRAYINIT_ELEMENT:%.*]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[ARRAYINIT_BEGIN1]], i64 1, !dbg [[DBG146]]
-// DEBUG1-NEXT:    store %struct.S1* [[ARRAYINIT_ELEMENT]], %struct.S1** [[ARRAYINIT_ENDOFINIT2]], align 8, !dbg [[DBG146]]
+// DEBUG1-NEXT:    [[ARRAYINIT_ELEMENT:%.*]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[ARRAYINIT_BEGIN1]], i64 1, !dbg [[DBG147]]
+// DEBUG1-NEXT:    store %struct.S1* [[ARRAYINIT_ELEMENT]], %struct.S1** [[ARRAYINIT_ENDOFINIT2]], align 8, !dbg [[DBG147]]
 // DEBUG1-NEXT:    invoke void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYINIT_ELEMENT]], i32 2)
-// DEBUG1-NEXT:    to label [[INVOKE_CONT3:%.*]] unwind label [[LPAD]], !dbg [[DBG148:![0-9]+]]
+// DEBUG1-NEXT:    to label [[INVOKE_CONT3:%.*]] unwind label [[LPAD]], !dbg [[DBG149:![0-9]+]]
 // DEBUG1:       invoke.cont3:
-// DEBUG1-NEXT:    [[ARRAYINIT_ELEMENT4:%.*]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYINIT_ELEMENT]], i64 1, !dbg [[DBG146]]
-// DEBUG1-NEXT:    store %struct.S1* [[ARRAYINIT_ELEMENT4]], %struct.S1** [[ARRAYINIT_ENDOFINIT2]], align 8, !dbg [[DBG146]]
+// DEBUG1-NEXT:    [[ARRAYINIT_ELEMENT4:%.*]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYINIT_ELEMENT]], i64 1, !dbg [[DBG147]]
+// DEBUG1-NEXT:    store %struct.S1* [[ARRAYINIT_ELEMENT4]], %struct.S1** [[ARRAYINIT_ENDOFINIT2]], align 8, !dbg [[DBG147]]
 // DEBUG1-NEXT:    invoke void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYINIT_ELEMENT4]], i32 3)
-// DEBUG1-NEXT:    to label [[INVOKE_CONT5:%.*]] unwind label [[LPAD]], !dbg [[DBG149:![0-9]+]]
+// DEBUG1-NEXT:    to label [[INVOKE_CONT5:%.*]] unwind label [[LPAD]], !dbg [[DBG150:![0-9]+]]
 // DEBUG1:       invoke.cont5:
-// DEBUG1-NEXT:    [[ARRAYINIT_ELEMENT7:%.*]] = getelementptr inbounds [3 x %struct.S1], [3 x %struct.S1]* [[ARRAYINIT_BEGIN]], i64 1, !dbg [[DBG145]]
-// DEBUG1-NEXT:    store [3 x %struct.S1]* [[ARRAYINIT_ELEMENT7]], [3 x %struct.S1]** [[ARRAYINIT_ENDOFINIT]], align 8, !dbg [[DBG145]]
-// DEBUG1-NEXT:    [[ARRAYINIT_BEGIN8:%.*]] = getelementptr inbounds [3 x %struct.S1], [3 x %struct.S1]* [[ARRAYINIT_ELEMENT7]], i64 0, i64 0, !dbg [[DBG150:![0-9]+]]
-// DEBUG1-NEXT:    store %struct.S1* [[ARRAYINIT_BEGIN8]], %struct.S1** [[ARRAYINIT_ENDOFINIT9]], align 8, !dbg [[DBG150]]
+// DEBUG1-NEXT:    [[ARRAYINIT_ELEMENT7:%.*]] = getelementptr inbounds [3 x %struct.S1], [3 x %struct.S1]* [[ARRAYINIT_BEGIN]], i64 1, !dbg [[DBG146]]
+// DEBUG1-NEXT:    store [3 x %struct.S1]* [[ARRAYINIT_ELEMENT7]], [3 x %struct.S1]** [[ARRAYINIT_ENDOFINIT]], align 8, !dbg [[DBG146]]
+// DEBUG1-NEXT:    [[ARRAYINIT_BEGIN8:%.*]] = getelementptr inbounds [3 x %struct.S1], [3 x %struct.S1]* [[ARRAYINIT_ELEMENT7]], i64 0, i64 0, !dbg [[DBG151:![0-9]+]]
+// DEBUG1-NEXT:    store %struct.S1* [[ARRAYINIT_BEGIN8]], %struct.S1** [[ARRAYINIT_ENDOFINIT9]], align 8, !dbg [[DBG151]]
 // DEBUG1-NEXT:    invoke void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYINIT_BEGIN8]], i32 4)
-// DEBUG1-NEXT:    to label [[INVOKE_CONT11:%.*]] unwind label [[LPAD10:%.*]], !dbg [[DBG151:![0-9]+]]
+// DEBUG1-NEXT:    to label [[INVOKE_CONT11:%.*]] unwind label [[LPAD10:%.*]], !dbg [[DBG152:![0-9]+]]
 // DEBUG1:       invoke.cont11:
-// DEBUG1-NEXT:    [[ARRAYINIT_ELEMENT12:%.*]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYINIT_BEGIN8]], i64 1, !dbg [[DBG150]]
-// DEBUG1-NEXT:    store %struct.S1* [[ARRAYINIT_ELEMENT12]], %struct.S1** [[ARRAYINIT_ENDOFINIT9]], align 8, !dbg [[DBG150]]
+// DEBUG1-NEXT:    [[ARRAYINIT_ELEMENT12:%.*]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYINIT_BEGIN8]], i64 1, !dbg [[DBG151]]
+// DEBUG1-NEXT:    store %struct.S1* [[ARRAYINIT_ELEMENT12]], %struct.S1** [[ARRAYINIT_ENDOFINIT9]], align 8, !dbg [[DBG151]]
 // DEBUG1-NEXT:    invoke void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYINIT_ELEMENT12]], i32 5)
-// DEBUG1-NEXT:    to label [[INVOKE_CONT13:%.*]] unwind label [[LPAD10]], !dbg [[DBG152:![0-9]+]]
+// DEBUG1-NEXT:    to label [[INVOKE_CONT13:%.*]] unwind label [[LPAD10]], !dbg [[DBG153:![0-9]+]]
 // DEBUG1:       invoke.cont13:
-// DEBUG1-NEXT:    [[ARRAYINIT_ELEMENT14:%.*]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYINIT_ELEMENT12]], i64 1, !dbg [[DBG150]]
-// DEBUG1-NEXT:    store %struct.S1* [[ARRAYINIT_ELEMENT14]], %struct.S1** [[ARRAYINIT_ENDOFINIT9]], align 8, !dbg [[DBG150]]
+// DEBUG1-NEXT:    [[ARRAYINIT_ELEMENT14:%.*]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYINIT_ELEMENT12]], i64 1, !dbg [[DBG151]]
+// DEBUG1-NEXT:    store %struct.S1* [[ARRAYINIT_ELEMENT14]], %struct.S1** [[ARRAYINIT_ENDOFINIT9]], align 8, !dbg [[DBG151]]
 // DEBUG1-NEXT:    invoke void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYINIT_ELEMENT14]], i32 6)
-// DEBUG1-NEXT:    to label [[INVOKE_CONT15:%.*]] unwind label [[LPAD10]], !dbg [[DBG153:![0-9]+]]
+// DEBUG1-NEXT:    to label [[INVOKE_CONT15:%.*]] unwind label [[LPAD10]], !dbg [[DBG154:![0-9]+]]
 // DEBUG1:       invoke.cont15:
-// DEBUG1-NEXT:    [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !dbg [[DBG144]]
-// DEBUG1-NEXT:    ret i8* [[TMP3]], !dbg [[DBG144]]
+// DEBUG1-NEXT:    [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !dbg [[DBG145]]
+// DEBUG1-NEXT:    ret i8* [[TMP3]], !dbg [[DBG145]]
 // DEBUG1:       lpad:
 // DEBUG1-NEXT:    [[TMP4:%.*]] = landingpad { i8*, i32 }
-// DEBUG1-NEXT:    cleanup, !dbg [[DBG143]]
-// DEBUG1-NEXT:    [[TMP5:%.*]] = extractvalue { i8*, i32 } [[TMP4]], 0, !dbg [[DBG143]]
-// DEBUG1-NEXT:    store i8* [[TMP5]], i8** [[EXN_SLOT]], align 8, !dbg [[DBG143]]
-// DEBUG1-NEXT:    [[TMP6:%.*]] = extractvalue { i8*, i32 } [[TMP4]], 1, !dbg [[DBG143]]
-// DEBUG1-NEXT:    store i32 [[TMP6]], i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG143]]
-// DEBUG1-NEXT:    [[TMP7:%.*]] = load %struct.S1*, %struct.S1** [[ARRAYINIT_ENDOFINIT2]], align 8, !dbg [[DBG146]]
-// DEBUG1-NEXT:    [[ARRAYDESTROY_ISEMPTY:%.*]] = icmp eq %struct.S1* [[ARRAYINIT_BEGIN1]], [[TMP7]], !dbg [[DBG146]]
-// DEBUG1-NEXT:    br i1 [[ARRAYDESTROY_ISEMPTY]], label [[ARRAYDESTROY_DONE6:%.*]], label [[ARRAYDESTROY_BODY:%.*]], !dbg [[DBG146]]
+// DEBUG1-NEXT:    cleanup, !dbg [[DBG144]]
+// DEBUG1-NEXT:    [[TMP5:%.*]] = extractvalue { i8*, i32 } [[TMP4]], 0, !dbg [[DBG144]]
+// DEBUG1-NEXT:    store i8* [[TMP5]], i8** [[EXN_SLOT]], align 8, !dbg [[DBG144]]
+// DEBUG1-NEXT:    [[TMP6:%.*]] = extractvalue { i8*, i32 } [[TMP4]], 1, !dbg [[DBG144]]
+// DEBUG1-NEXT:    store i32 [[TMP6]], i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG144]]
+// DEBUG1-NEXT:    [[TMP7:%.*]] = load %struct.S1*, %struct.S1** [[ARRAYINIT_ENDOFINIT2]], align 8, !dbg [[DBG147]]
+// DEBUG1-NEXT:    [[ARRAYDESTROY_ISEMPTY:%.*]] = icmp eq %struct.S1* [[ARRAYINIT_BEGIN1]], [[TMP7]], !dbg [[DBG147]]
+// DEBUG1-NEXT:    br i1 [[ARRAYDESTROY_ISEMPTY]], label [[ARRAYDESTROY_DONE6:%.*]], label [[ARRAYDESTROY_BODY:%.*]], !dbg [[DBG147]]
 // DEBUG1:       arraydestroy.body:
-// DEBUG1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ [[TMP7]], [[LPAD]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG146]]
-// DEBUG1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG146]]
-// DEBUG1-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]], !dbg [[DBG146]]
-// DEBUG1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT]], [[ARRAYINIT_BEGIN1]], !dbg [[DBG146]]
-// DEBUG1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE6]], label [[ARRAYDESTROY_BODY]], !dbg [[DBG146]]
+// DEBUG1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ [[TMP7]], [[LPAD]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG147]]
+// DEBUG1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG147]]
+// DEBUG1-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]], !dbg [[DBG147]]
+// DEBUG1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT]], [[ARRAYINIT_BEGIN1]], !dbg [[DBG147]]
+// DEBUG1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE6]], label [[ARRAYDESTROY_BODY]], !dbg [[DBG147]]
 // DEBUG1:       arraydestroy.done6:
-// DEBUG1-NEXT:    br label [[EHCLEANUP:%.*]], !dbg [[DBG146]]
+// DEBUG1-NEXT:    br label [[EHCLEANUP:%.*]], !dbg [[DBG147]]
 // DEBUG1:       lpad10:
 // DEBUG1-NEXT:    [[TMP8:%.*]] = landingpad { i8*, i32 }
-// DEBUG1-NEXT:    cleanup, !dbg [[DBG143]]
-// DEBUG1-NEXT:    [[TMP9:%.*]] = extractvalue { i8*, i32 } [[TMP8]], 0, !dbg [[DBG143]]
-// DEBUG1-NEXT:    store i8* [[TMP9]], i8** [[EXN_SLOT]], align 8, !dbg [[DBG143]]
-// DEBUG1-NEXT:    [[TMP10:%.*]] = extractvalue { i8*, i32 } [[TMP8]], 1, !dbg [[DBG143]]
-// DEBUG1-NEXT:    store i32 [[TMP10]], i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG143]]
-// DEBUG1-NEXT:    [[TMP11:%.*]] = load %struct.S1*, %struct.S1** [[ARRAYINIT_ENDOFINIT9]], align 8, !dbg [[DBG150]]
-// DEBUG1-NEXT:    [[ARRAYDESTROY_ISEMPTY16:%.*]] = icmp eq %struct.S1* [[ARRAYINIT_BEGIN8]], [[TMP11]], !dbg [[DBG150]]
-// DEBUG1-NEXT:    br i1 [[ARRAYDESTROY_ISEMPTY16]], label [[ARRAYDESTROY_DONE21:%.*]], label [[ARRAYDESTROY_BODY17:%.*]], !dbg [[DBG150]]
+// DEBUG1-NEXT:    cleanup, !dbg [[DBG144]]
+// DEBUG1-NEXT:    [[TMP9:%.*]] = extractvalue { i8*, i32 } [[TMP8]], 0, !dbg [[DBG144]]
+// DEBUG1-NEXT:    store i8* [[TMP9]], i8** [[EXN_SLOT]], align 8, !dbg [[DBG144]]
+// DEBUG1-NEXT:    [[TMP10:%.*]] = extractvalue { i8*, i32 } [[TMP8]], 1, !dbg [[DBG144]]
+// DEBUG1-NEXT:    store i32 [[TMP10]], i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG144]]
+// DEBUG1-NEXT:    [[TMP11:%.*]] = load %struct.S1*, %struct.S1** [[ARRAYINIT_ENDOFINIT9]], align 8, !dbg [[DBG151]]
+// DEBUG1-NEXT:    [[ARRAYDESTROY_ISEMPTY16:%.*]] = icmp eq %struct.S1* [[ARRAYINIT_BEGIN8]], [[TMP11]], !dbg [[DBG151]]
+// DEBUG1-NEXT:    br i1 [[ARRAYDESTROY_ISEMPTY16]], label [[ARRAYDESTROY_DONE21:%.*]], label [[ARRAYDESTROY_BODY17:%.*]], !dbg [[DBG151]]
 // DEBUG1:       arraydestroy.body17:
-// DEBUG1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST18:%.*]] = phi %struct.S1* [ [[TMP11]], [[LPAD10]] ], [ [[ARRAYDESTROY_ELEMENT19:%.*]], [[ARRAYDESTROY_BODY17]] ], !dbg [[DBG150]]
-// DEBUG1-NEXT:    [[ARRAYDESTROY_ELEMENT19]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST18]], i64 -1, !dbg [[DBG150]]
-// DEBUG1-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT19]]) #[[ATTR4]], !dbg [[DBG150]]
-// DEBUG1-NEXT:    [[ARRAYDESTROY_DONE20:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT19]], [[ARRAYINIT_BEGIN8]], !dbg [[DBG150]]
-// DEBUG1-NEXT:    br i1 [[ARRAYDESTROY_DONE20]], label [[ARRAYDESTROY_DONE21]], label [[ARRAYDESTROY_BODY17]], !dbg [[DBG150]]
+// DEBUG1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST18:%.*]] = phi %struct.S1* [ [[TMP11]], [[LPAD10]] ], [ [[ARRAYDESTROY_ELEMENT19:%.*]], [[ARRAYDESTROY_BODY17]] ], !dbg [[DBG151]]
+// DEBUG1-NEXT:    [[ARRAYDESTROY_ELEMENT19]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST18]], i64 -1, !dbg [[DBG151]]
+// DEBUG1-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT19]]) #[[ATTR4]], !dbg [[DBG151]]
+// DEBUG1-NEXT:    [[ARRAYDESTROY_DONE20:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT19]], [[ARRAYINIT_BEGIN8]], !dbg [[DBG151]]
+// DEBUG1-NEXT:    br i1 [[ARRAYDESTROY_DONE20]], label [[ARRAYDESTROY_DONE21]], label [[ARRAYDESTROY_BODY17]], !dbg [[DBG151]]
 // DEBUG1:       arraydestroy.done21:
-// DEBUG1-NEXT:    br label [[EHCLEANUP]], !dbg [[DBG150]]
+// DEBUG1-NEXT:    br label [[EHCLEANUP]], !dbg [[DBG151]]
 // DEBUG1:       ehcleanup:
-// DEBUG1-NEXT:    [[TMP12:%.*]] = load [3 x %struct.S1]*, [3 x %struct.S1]** [[ARRAYINIT_ENDOFINIT]], align 8, !dbg [[DBG145]]
-// DEBUG1-NEXT:    [[PAD_ARRAYBEGIN:%.*]] = getelementptr inbounds [3 x %struct.S1], [3 x %struct.S1]* [[ARRAYINIT_BEGIN]], i64 0, i64 0, !dbg [[DBG145]]
-// DEBUG1-NEXT:    [[PAD_ARRAYEND:%.*]] = getelementptr inbounds [3 x %struct.S1], [3 x %struct.S1]* [[TMP12]], i64 0, i64 0, !dbg [[DBG145]]
-// DEBUG1-NEXT:    [[ARRAYDESTROY_ISEMPTY22:%.*]] = icmp eq %struct.S1* [[PAD_ARRAYBEGIN]], [[PAD_ARRAYEND]], !dbg [[DBG145]]
-// DEBUG1-NEXT:    br i1 [[ARRAYDESTROY_ISEMPTY22]], label [[ARRAYDESTROY_DONE27:%.*]], label [[ARRAYDESTROY_BODY23:%.*]], !dbg [[DBG145]]
+// DEBUG1-NEXT:    [[TMP12:%.*]] = load [3 x %struct.S1]*, [3 x %struct.S1]** [[ARRAYINIT_ENDOFINIT]], align 8, !dbg [[DBG146]]
+// DEBUG1-NEXT:    [[PAD_ARRAYBEGIN:%.*]] = getelementptr inbounds [3 x %struct.S1], [3 x %struct.S1]* [[ARRAYINIT_BEGIN]], i64 0, i64 0, !dbg [[DBG146]]
+// DEBUG1-NEXT:    [[PAD_ARRAYEND:%.*]] = getelementptr inbounds [3 x %struct.S1], [3 x %struct.S1]* [[TMP12]], i64 0, i64 0, !dbg [[DBG146]]
+// DEBUG1-NEXT:    [[ARRAYDESTROY_ISEMPTY22:%.*]] = icmp eq %struct.S1* [[PAD_ARRAYBEGIN]], [[PAD_ARRAYEND]], !dbg [[DBG146]]
+// DEBUG1-NEXT:    br i1 [[ARRAYDESTROY_ISEMPTY22]], label [[ARRAYDESTROY_DONE27:%.*]], label [[ARRAYDESTROY_BODY23:%.*]], !dbg [[DBG146]]
 // DEBUG1:       arraydestroy.body23:
-// DEBUG1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST24:%.*]] = phi %struct.S1* [ [[PAD_ARRAYEND]], [[EHCLEANUP]] ], [ [[ARRAYDESTROY_ELEMENT25:%.*]], [[ARRAYDESTROY_BODY23]] ], !dbg [[DBG145]]
-// DEBUG1-NEXT:    [[ARRAYDESTROY_ELEMENT25]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST24]], i64 -1, !dbg [[DBG145]]
-// DEBUG1-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT25]]) #[[ATTR4]], !dbg [[DBG145]]
-// DEBUG1-NEXT:    [[ARRAYDESTROY_DONE26:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT25]], [[PAD_ARRAYBEGIN]], !dbg [[DBG145]]
-// DEBUG1-NEXT:    br i1 [[ARRAYDESTROY_DONE26]], label [[ARRAYDESTROY_DONE27]], label [[ARRAYDESTROY_BODY23]], !dbg [[DBG145]]
+// DEBUG1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST24:%.*]] = phi %struct.S1* [ [[PAD_ARRAYEND]], [[EHCLEANUP]] ], [ [[ARRAYDESTROY_ELEMENT25:%.*]], [[ARRAYDESTROY_BODY23]] ], !dbg [[DBG146]]
+// DEBUG1-NEXT:    [[ARRAYDESTROY_ELEMENT25]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST24]], i64 -1, !dbg [[DBG146]]
+// DEBUG1-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT25]]) #[[ATTR4]], !dbg [[DBG146]]
+// DEBUG1-NEXT:    [[ARRAYDESTROY_DONE26:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT25]], [[PAD_ARRAYBEGIN]], !dbg [[DBG146]]
+// DEBUG1-NEXT:    br i1 [[ARRAYDESTROY_DONE26]], label [[ARRAYDESTROY_DONE27]], label [[ARRAYDESTROY_BODY23]], !dbg [[DBG146]]
 // DEBUG1:       arraydestroy.done27:
-// DEBUG1-NEXT:    br label [[EH_RESUME:%.*]], !dbg [[DBG145]]
+// DEBUG1-NEXT:    br label [[EH_RESUME:%.*]], !dbg [[DBG146]]
 // DEBUG1:       eh.resume:
-// DEBUG1-NEXT:    [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8, !dbg [[DBG145]]
-// DEBUG1-NEXT:    [[SEL:%.*]] = load i32, i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG145]]
-// DEBUG1-NEXT:    [[LPAD_VAL:%.*]] = insertvalue { i8*, i32 } undef, i8* [[EXN]], 0, !dbg [[DBG145]]
-// DEBUG1-NEXT:    [[LPAD_VAL28:%.*]] = insertvalue { i8*, i32 } [[LPAD_VAL]], i32 [[SEL]], 1, !dbg [[DBG145]]
-// DEBUG1-NEXT:    resume { i8*, i32 } [[LPAD_VAL28]], !dbg [[DBG145]]
+// DEBUG1-NEXT:    [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8, !dbg [[DBG146]]
+// DEBUG1-NEXT:    [[SEL:%.*]] = load i32, i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG146]]
+// DEBUG1-NEXT:    [[LPAD_VAL:%.*]] = insertvalue { i8*, i32 } undef, i8* [[EXN]], 0, !dbg [[DBG146]]
+// DEBUG1-NEXT:    [[LPAD_VAL28:%.*]] = insertvalue { i8*, i32 } [[LPAD_VAL]], i32 [[SEL]], 1, !dbg [[DBG146]]
+// DEBUG1-NEXT:    resume { i8*, i32 } [[LPAD_VAL28]], !dbg [[DBG146]]
 //
 //
 // DEBUG1-LABEL: define {{[^@]+}}@.__kmpc_global_dtor_..2
-// DEBUG1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] !dbg [[DBG154:![0-9]+]] {
+// DEBUG1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] !dbg [[DBG155:![0-9]+]] {
 // DEBUG1-NEXT:  entry:
 // DEBUG1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // DEBUG1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
-// DEBUG1-NEXT:    call void @llvm.dbg.declare(metadata i8** [[DOTADDR]], metadata [[META155:![0-9]+]], metadata !DIExpression()), !dbg [[DBG156:![0-9]+]]
-// DEBUG1-NEXT:    [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !dbg [[DBG156]]
-// DEBUG1-NEXT:    [[ARRAY_BEGIN:%.*]] = bitcast i8* [[TMP1]] to %struct.S1*, !dbg [[DBG156]]
-// DEBUG1-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[ARRAY_BEGIN]], i64 6, !dbg [[DBG156]]
-// DEBUG1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]], !dbg [[DBG156]]
+// DEBUG1-NEXT:    call void @llvm.dbg.declare(metadata i8** [[DOTADDR]], metadata [[META156:![0-9]+]], metadata !DIExpression()), !dbg [[DBG157:![0-9]+]]
+// DEBUG1-NEXT:    [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !dbg [[DBG157]]
+// DEBUG1-NEXT:    [[ARRAY_BEGIN:%.*]] = bitcast i8* [[TMP1]] to %struct.S1*, !dbg [[DBG157]]
+// DEBUG1-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[ARRAY_BEGIN]], i64 6, !dbg [[DBG157]]
+// DEBUG1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]], !dbg [[DBG157]]
 // DEBUG1:       arraydestroy.body:
-// DEBUG1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ [[TMP2]], [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG156]]
-// DEBUG1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG156]]
-// DEBUG1-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]], !dbg [[DBG156]]
-// DEBUG1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]], !dbg [[DBG156]]
-// DEBUG1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]], !dbg [[DBG156]]
+// DEBUG1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ [[TMP2]], [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG157]]
+// DEBUG1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG157]]
+// DEBUG1-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]], !dbg [[DBG157]]
+// DEBUG1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]], !dbg [[DBG157]]
+// DEBUG1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]], !dbg [[DBG157]]
 // DEBUG1:       arraydestroy.done1:
-// DEBUG1-NEXT:    ret void, !dbg [[DBG157:![0-9]+]]
+// DEBUG1-NEXT:    ret void, !dbg [[DBG158:![0-9]+]]
 //
 //
 // DEBUG1-LABEL: define {{[^@]+}}@.__omp_threadprivate_init_..3
-// DEBUG1-SAME: () #[[ATTR0]] !dbg [[DBG158:![0-9]+]] {
+// DEBUG1-SAME: () #[[ATTR0]] !dbg [[DBG159:![0-9]+]] {
 // DEBUG1-NEXT:  entry:
-// DEBUG1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3:[0-9]+]]), !dbg [[DBG159:![0-9]+]]
-// DEBUG1-NEXT:    call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB3]], i8* bitcast ([2 x [3 x %struct.S1]]* @arr_x to i8*), i8* (i8*)* @.__kmpc_global_ctor_..1, i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_..2), !dbg [[DBG159]]
-// DEBUG1-NEXT:    ret void, !dbg [[DBG159]]
+// DEBUG1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3:[0-9]+]]), !dbg [[DBG160:![0-9]+]]
+// DEBUG1-NEXT:    call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB3]], i8* bitcast ([2 x [3 x %struct.S1]]* @arr_x to i8*), i8* (i8*)* @.__kmpc_global_ctor_..1, i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_..2), !dbg [[DBG160]]
+// DEBUG1-NEXT:    ret void, !dbg [[DBG160]]
 //
 //
 // DEBUG1-LABEL: define {{[^@]+}}@__cxx_global_var_init
-// DEBUG1-SAME: () #[[ATTR0]] !dbg [[DBG160:![0-9]+]] {
+// DEBUG1-SAME: () #[[ATTR0]] !dbg [[DBG161:![0-9]+]] {
 // DEBUG1-NEXT:  entry:
 // DEBUG1-NEXT:    call void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) @_ZL3gs1, i32 5), !dbg [[DBG164:![0-9]+]]
 // DEBUG1-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.S1*)* @_ZN2S1D1Ev to void (i8*)*), i8* bitcast (%struct.S1* @_ZL3gs1 to i8*), i8* @__dso_handle) #[[ATTR4]], !dbg [[DBG166:![0-9]+]]
 // DEBUG1-NEXT:    ret void, !dbg [[DBG167:![0-9]+]]
 //
 //
 // DEBUG1-LABEL: define {{[^@]+}}@_ZN2S1C2Ei
 // DEBUG1-SAME: (%struct.S1* nonnull align 4 dereferenceable(4) [[THIS:%.*]], i32 [[A:%.*]]) unnamed_addr #[[ATTR3]] comdat align 2 !dbg [[DBG168:![0-9]+]] {

 // DEBUG1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // DEBUG1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] !dbg [[DBG229:![0-9]+]] {
 // DEBUG1-NEXT:  entry:
 // DEBUG1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // DEBUG1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // DEBUG1-NEXT:    call void @llvm.dbg.declare(metadata i8** [[DOTADDR]], metadata [[META232:![0-9]+]], metadata !DIExpression()), !dbg [[DBG233:![0-9]+]]
 // DEBUG1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]], !dbg [[DBG233]]
 // DEBUG1:       arraydestroy.body:
-// DEBUG1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ getelementptr inbounds ([[STRUCT_S1:%.*]], %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0), i64 6), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG233]]
-// DEBUG1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG233]]
+// DEBUG1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 1, i64 0, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG233]]
+// DEBUG1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG233]]
 // DEBUG1-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]], !dbg [[DBG233]]
 // DEBUG1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0), !dbg [[DBG233]]
 // DEBUG1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]], !dbg [[DBG233]]
 // DEBUG1:       arraydestroy.done1:
 // DEBUG1-NEXT:    ret void, !dbg [[DBG233]]
 //
 //
 // DEBUG1-LABEL: define {{[^@]+}}@main
-// DEBUG1-SAME: () #[[ATTR5:[0-9]+]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) !dbg [[DBG53:![0-9]+]] {
+// DEBUG1-SAME: () #[[ATTR5:[0-9]+]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) !dbg [[DBG52:![0-9]+]] {
 // DEBUG1-NEXT:  entry:
 // DEBUG1-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
 // DEBUG1-NEXT:    [[RES:%.*]] = alloca i32, align 4
 // DEBUG1-NEXT:    [[EXN_SLOT:%.*]] = alloca i8*, align 8
 // DEBUG1-NEXT:    [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
 // DEBUG1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB9:[0-9]+]])
 // DEBUG1-NEXT:    store i32 0, i32* [[RETVAL]], align 4
 // DEBUG1-NEXT:    call void @llvm.dbg.declare(metadata i32* [[RES]], metadata [[META234:![0-9]+]], metadata !DIExpression()), !dbg [[DBG235:![0-9]+]]

 // DEBUG1-NEXT:    call void @.__omp_threadprivate_init_.(), !dbg [[DBG376]]
 // DEBUG1-NEXT:    call void @__cxx_global_var_init.4(), !dbg [[DBG376]]
 // DEBUG1-NEXT:    call void @__cxx_global_var_init.5(), !dbg [[DBG376]]
 // DEBUG1-NEXT:    call void @.__omp_threadprivate_init_..3(), !dbg [[DBG376]]
 // DEBUG1-NEXT:    ret void
 //
 //
 // DEBUG2-LABEL: define {{[^@]+}}@__cxx_global_var_init
-// DEBUG2-SAME: () #[[ATTR0:[0-9]+]] !dbg [[DBG115:![0-9]+]] {
+// DEBUG2-SAME: () #[[ATTR0:[0-9]+]] !dbg [[DBG116:![0-9]+]] {
 // DEBUG2-NEXT:  entry:
 // DEBUG2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]]), !dbg [[DBG119:![0-9]+]]
 // DEBUG2-NEXT:    call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB1]], i8* bitcast (%struct.S1* @_ZL3gs1 to i8*), i8* (i8*)* @.__kmpc_global_ctor_., i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_.), !dbg [[DBG119]]
 // DEBUG2-NEXT:    call void @_ZN2S1C1Ei(%struct.S1* nonnull align 4 dereferenceable(4) @_ZL3gs1, i32 5), !dbg [[DBG120:![0-9]+]]
 // DEBUG2-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.S1*)* @_ZN2S1D1Ev to void (i8*)*), i8* bitcast (%struct.S1* @_ZL3gs1 to i8*), i8* @__dso_handle) #[[ATTR4:[0-9]+]], !dbg [[DBG119]]
 // DEBUG2-NEXT:    ret void, !dbg [[DBG122:![0-9]+]]
 //
 //

 // DEBUG2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
 // DEBUG2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] !dbg [[DBG195:![0-9]+]] {
 // DEBUG2-NEXT:  entry:
 // DEBUG2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
 // DEBUG2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8
 // DEBUG2-NEXT:    call void @llvm.dbg.declare(metadata i8** [[DOTADDR]], metadata [[META198:![0-9]+]], metadata !DIExpression()), !dbg [[DBG199:![0-9]+]]
 // DEBUG2-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]], !dbg [[DBG199]]
 // DEBUG2:       arraydestroy.body:
-// DEBUG2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ getelementptr inbounds ([[STRUCT_S1:%.*]], %struct.S1* getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0), i64 6), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG199]]
-// DEBUG2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG199]]
+// DEBUG2-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S1* [ getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i64 1, i64 0, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG199]]
+// DEBUG2-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S1:%.*]], %struct.S1* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG199]]
 // DEBUG2-NEXT:    call void @_ZN2S1D1Ev(%struct.S1* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR4]], !dbg [[DBG199]]
 // DEBUG2-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S1* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x [3 x %struct.S1]], [2 x [3 x %struct.S1]]* @arr_x, i32 0, i32 0, i32 0), !dbg [[DBG199]]
 // DEBUG2-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]], !dbg [[DBG199]]
 // DEBUG2:       arraydestroy.done1:
 // DEBUG2-NEXT:    ret void, !dbg [[DBG199]]
 //
 //
 // DEBUG2-LABEL: define {{[^@]+}}@main
-// DEBUG2-SAME: () #[[ATTR5:[0-9]+]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) !dbg [[DBG53:![0-9]+]] {
+// DEBUG2-SAME: () #[[ATTR5:[0-9]+]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) !dbg [[DBG52:![0-9]+]] {
 // DEBUG2-NEXT:  entry:
 // DEBUG2-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
 // DEBUG2-NEXT:    [[RES:%.*]] = alloca i32, align 4
 // DEBUG2-NEXT:    [[EXN_SLOT:%.*]] = alloca i8*, align 8
 // DEBUG2-NEXT:    [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
 // DEBUG2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB9:[0-9]+]])
 // DEBUG2-NEXT:    store i32 0, i32* [[RETVAL]], align 4
 // DEBUG2-NEXT:    call void @llvm.dbg.declare(metadata i32* [[RES]], metadata [[META200:![0-9]+]], metadata !DIExpression()), !dbg [[DBG201:![0-9]+]]

llvm/lib/IR/ConstantFold.cpp

return ConstantExpr::getGetElementPtr(
         NewIndices, InBounds && GEP->isInBounds(), GEP->getInRangeIndex());
   }
 
   gep_type_iterator LastI = gep_type_end(GEP);
   for (gep_type_iterator I = gep_type_begin(GEP), E = gep_type_end(GEP);
        I != E; ++I)
     LastI = I;
 
-  // We cannot combine indices if doing so would take us outside of an
-  // array or vector.  Doing otherwise could trick us if we evaluated such a
-  // GEP as part of a load.
-  //
-  // e.g. Consider if the original GEP was:
-  // i8* getelementptr ({ [2 x i8], i32, i8, [3 x i8] }* @main.c,
-  //                    i32 0, i32 0, i64 0)
-  //
-  // If we then tried to offset it by '8' to get to the third element,
-  // an i8, we should *not* get:
-  // i8* getelementptr ({ [2 x i8], i32, i8, [3 x i8] }* @main.c,
-  //                    i32 0, i32 0, i64 8)
-  //
-  // This GEP tries to index array element '8  which runs out-of-bounds.
-  // Subsequent evaluation would get confused and produce erroneous results.
-  //
-  // The following prohibits such a GEP from being formed by checking to see
-  // if the index is in-range with respect to an array.
+  // We can't combine GEPs if the last index is a struct type.
   if (!LastI.isSequential())
     return nullptr;
+  // We could perform the transform with non-constant index, but prefer leaving
+  // it as GEP of GEP rather than GEP of add for now.
   ConstantInt *CI = dyn_cast<ConstantInt>(Idx0);
   if (!CI)
     return nullptr;
-  if (LastI.isBoundedSequential() &&
-      !isIndexInRangeOfArrayType(LastI.getSequentialNumElements(), CI))
-    return nullptr;
 
   // TODO: This code may be extended to handle vectors as well.
   auto *LastIdx = cast<Constant>(GEP->getOperand(GEP->getNumOperands()-1));
   Type *LastIdxTy = LastIdx->getType();
   if (LastIdxTy->isVectorTy())
     return nullptr;
 
   SmallVector<Value*, 16> NewIndices;

llvm/test/Transforms/SCCP/apint-bigint2.ll

;
   %F = getelementptr [6 x i101], [6 x i101]* @Y, i32 0, i32 5
   %G = getelementptr i101, i101* %F, i101 %DD
   %L3 = load i101, i101* %G
   ret i101 %L3
 }
 
 define void @index_too_large() {
 ; CHECK-LABEL: @index_too_large(
-; CHECK-NEXT:    store i101* getelementptr (i101, i101* getelementptr ([6 x i101], [6 x i101]* @Y, i32 0, i32 -1), i101 9224497936761618431), i101** undef, align 8
+; CHECK-NEXT:    store i101* getelementptr ([6 x i101], [6 x i101]* @Y, i101 1537416322793603071, i101 4), i101** undef, align 8
 ; CHECK-NEXT:    ret void
 ;
   %ptr1 = getelementptr [6 x i101], [6 x i101]* @Y, i32 0, i32 -1
   %ptr2 = getelementptr i101, i101* %ptr1, i101 9224497936761618431
   store i101* %ptr2, i101** undef
   ret void
 }
 

llvm/test/Transforms/SCCP/replace-dereferenceable-ptr-with-undereferenceable.ll

 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt -passes=ipsccp -S %s | FileCheck %s
 
 @y = common global [1 x i32] zeroinitializer, align 4
 @x = common global [1 x i32] zeroinitializer, align 4
 
 define i32 @eq_undereferenceable(i32* %p) {
 ; CHECK-LABEL: @eq_undereferenceable(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    store i32 1, i32* getelementptr inbounds ([1 x i32], [1 x i32]* @y, i64 0, i64 0), align 4
-; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32* [[P:%.*]], getelementptr inbounds (i32, i32* getelementptr inbounds ([1 x i32], [1 x i32]* @x, i64 0, i64 0), i64 1)
+; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32* [[P:%.*]], getelementptr inbounds ([1 x i32], [1 x i32]* @x, i64 1, i64 0)
 ; CHECK-NEXT:    br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_END:%.*]]
 ; CHECK:       if.then:
-; CHECK-NEXT:    store i32 2, i32* getelementptr inbounds (i32, i32* getelementptr inbounds ([1 x i32], [1 x i32]* @x, i64 0, i64 0), i64 1), align 4
+; CHECK-NEXT:    store i32 2, i32* getelementptr inbounds ([1 x i32], [1 x i32]* @x, i64 1, i64 0), align 4
 ; CHECK-NEXT:    br label [[IF_END]]
 ; CHECK:       if.end:
 ; CHECK-NEXT:    [[TMP0:%.*]] = load i32, i32* getelementptr inbounds ([1 x i32], [1 x i32]* @y, i64 0, i64 0), align 4
 ; CHECK-NEXT:    ret i32 [[TMP0]]
 ;
 entry:
   store i32 1, i32* getelementptr inbounds ([1 x i32], [1 x i32]* @y, i64 0, i64 0), align 4
   %cmp = icmp eq i32* %p, getelementptr inbounds (i32, i32* getelementptr inbounds ([1 x i32], [1 x i32]* @x, i64 0, i64 0), i64 1)

 if.end:                                           ; preds = %if.then, %entry
   %0 = load i32, i32* getelementptr inbounds ([1 x i32], [1 x i32]* @y, i64 0, i64 0), align 4
   ret i32 %0
 }
 
 define i1 @eq_undereferenceable_cmp_simp(i32* %p) {
 ; CHECK-LABEL: @eq_undereferenceable_cmp_simp(
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[CMP_0:%.*]] = icmp eq i32* [[P:%.*]], getelementptr inbounds (i32, i32* getelementptr inbounds ([1 x i32], [1 x i32]* @x, i64 0, i64 0), i64 1)
+; CHECK-NEXT:    [[CMP_0:%.*]] = icmp eq i32* [[P:%.*]], getelementptr inbounds ([1 x i32], [1 x i32]* @x, i64 1, i64 0)
 ; CHECK-NEXT:    br i1 [[CMP_0]], label [[IF_THEN:%.*]], label [[IF_END:%.*]]
 ; CHECK:       if.then:
-; CHECK-NEXT:    store i32 2, i32* getelementptr inbounds (i32, i32* getelementptr inbounds ([1 x i32], [1 x i32]* @x, i64 0, i64 0), i64 1), align 4
+; CHECK-NEXT:    store i32 2, i32* getelementptr inbounds ([1 x i32], [1 x i32]* @x, i64 1, i64 0), align 4
 ; CHECK-NEXT:    ret i1 true
 ; CHECK:       if.end:
-; CHECK-NEXT:    [[CMP_2:%.*]] = icmp eq i32* [[P]], getelementptr inbounds (i32, i32* getelementptr inbounds ([1 x i32], [1 x i32]* @x, i64 0, i64 0), i64 1)
+; CHECK-NEXT:    [[CMP_2:%.*]] = icmp eq i32* [[P]], getelementptr inbounds ([1 x i32], [1 x i32]* @x, i64 1, i64 0)
 ; CHECK-NEXT:    ret i1 [[CMP_2]]
 ;
 entry:
   %cmp.0 = icmp eq i32* %p, getelementptr inbounds (i32, i32* getelementptr inbounds ([1 x i32], [1 x i32]* @x, i64 0, i64 0), i64 1)
   br i1 %cmp.0, label %if.then, label %if.end
 
 if.then:                                          ; preds = %entry
   store i32 2, i32* %p, align 4
   %cmp.1 = icmp eq i32* %p, getelementptr inbounds (i32, i32* getelementptr inbounds ([1 x i32], [1 x i32]* @x, i64 0, i64 0), i64 1)
   ret i1 %cmp.1
 
 if.end:                                           ; preds = %if.then, %entry
   %cmp.2 = icmp eq i32* %p, getelementptr inbounds (i32, i32* getelementptr inbounds ([1 x i32], [1 x i32]* @x, i64 0, i64 0), i64 1)
   ret i1 %cmp.2
 }