// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --include-generated-funcs --replace-value-regex "__omp_offloading_[0-9a-z]+_[0-9a-z]+" "reduction_size[.].+[.]" "pl_cond[.].+[.|,]" --prefix-filecheck-ir-name _
// RUN: %clang_cc1 -disable-noundef-analysis -x c++ -O1 -disable-llvm-optzns -verify -fopenmp -internal-isystem %S/../Headers/Inputs/include -internal-isystem %S/../../lib/Headers/openmp_wrappers -include __clang_openmp_device_functions.h -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc
// RUN: %clang_cc1 -disable-noundef-analysis -x c++ -O1 -disable-llvm-optzns -verify -fopenmp -internal-isystem %S/../Headers/Inputs/include -internal-isystem %S/../../lib/Headers/openmp_wrappers -include __clang_openmp_device_functions.h -triple nvptx64-unknown-unknown -aux-triple powerpc64le-unknown-unknown  -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix=CHECK1
// RUN: %clang_cc1 -disable-noundef-analysis -x c++ -O1 -disable-llvm-optzns -verify -fopenmp -internal-isystem %S/../Headers/Inputs/include -internal-isystem %S/../../lib/Headers/openmp_wrappers -include __clang_openmp_device_functions.h -triple i386-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm-bc %s -o %t-x86-host.bc
// RUN: %clang_cc1 -disable-noundef-analysis -x c++ -O1 -disable-llvm-optzns -verify -fopenmp -internal-isystem %S/../Headers/Inputs/include -internal-isystem %S/../../lib/Headers/openmp_wrappers -include __clang_openmp_device_functions.h -triple nvptx64-unknown-unknown -aux-triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix=CHECK2
// RUN: %clang_cc1 -disable-noundef-analysis -x c++ -O1 -disable-llvm-optzns -verify -fopenmp -internal-isystem %S/../Headers/Inputs/include -internal-isystem %S/../../lib/Headers/openmp_wrappers -include __clang_openmp_device_functions.h -fexceptions -fcxx-exceptions -aux-triple powerpc64le-unknown-unknown -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix=CHECK3
// expected-no-diagnostics
#ifndef HEADER
#define HEADER

#include <complex>

// Verify we do not add tbaa metadata to type punned memory operations:
template <typename T>
void complex_reduction() {
#pragma omp target teams distribute
  for (int ib = 0; ib < 100; ib++) {
    std::complex<T> partial_sum;
    const int istart = ib * 4;
    const int iend = (ib + 1) * 4;
#pragma omp parallel for reduction(+ \
                                   : partial_sum)
    for (int i = istart; i < iend; i++)
      partial_sum += std::complex<T>(i, i);
  }
}

void test() {
  complex_reduction<float>();
  complex_reduction<double>();
}
#endif
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z17complex_reductionIfEvv_l16
// CHECK1-SAME: () #[[ATTR0:[0-9]+]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i8 1, i1 true, i1 true)
// CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
// CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK1:       user_code.entry:
// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK1-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK1-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4, !tbaa [[TBAA8:![0-9]+]]
// CHECK1-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR5:[0-9]+]]
// CHECK1-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 1, i1 true)
// CHECK1-NEXT:    ret void
// CHECK1:       worker.exit:
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@__omp_outlined__
// CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[IB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[REF_TMP:%.*]] = alloca float, align 4
// CHECK1-NEXT:    [[REF_TMP2:%.*]] = alloca float, align 4
// CHECK1-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x i8*], align 8
// CHECK1-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8, !tbaa [[TBAA12:![0-9]+]]
// CHECK1-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[ISTART:%.*]] = call align 16 i8* @__kmpc_alloc_shared(i64 4)
// CHECK1-NEXT:    [[ISTART_ON_STACK:%.*]] = bitcast i8* [[ISTART]] to i32*
// CHECK1-NEXT:    [[IEND:%.*]] = call align 16 i8* @__kmpc_alloc_shared(i64 4)
// CHECK1-NEXT:    [[IEND_ON_STACK:%.*]] = bitcast i8* [[IEND]] to i32*
// CHECK1-NEXT:    [[PARTIAL_SUM:%.*]] = call align 16 i8* @__kmpc_alloc_shared(i64 8)
// CHECK1-NEXT:    [[PARTIAL_SUM_ON_STACK:%.*]] = bitcast i8* [[PARTIAL_SUM]] to %"class.std::complex"*
// CHECK1-NEXT:    [[TMP0:%.*]] = bitcast i32* [[DOTOMP_IV]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP0]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP1:%.*]] = bitcast i32* [[DOTOMP_LB]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP1]]) #[[ATTR5]]
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP2:%.*]] = bitcast i32* [[DOTOMP_UB]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP2]]) #[[ATTR5]]
// CHECK1-NEXT:    store i32 99, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP3:%.*]] = bitcast i32* [[DOTOMP_STRIDE]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP3]]) #[[ATTR5]]
// CHECK1-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP4:%.*]] = bitcast i32* [[DOTOMP_IS_LAST]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP4]]) #[[ATTR5]]
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP5:%.*]] = bitcast i32* [[IB]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP5]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP6:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[TMP6]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP7]], i32 92, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP8]], 99
// CHECK1-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK1:       cond.true:
// CHECK1-NEXT:    br label [[COND_END:%.*]]
// CHECK1:       cond.false:
// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    br label [[COND_END]]
// CHECK1:       cond.end:
// CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP9]], [[COND_FALSE]] ]
// CHECK1-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    store i32 [[TMP10]], i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK1:       omp.inner.for.cond:
// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP11]], [[TMP12]]
// 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:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP13]], 1
// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK1-NEXT:    store i32 [[ADD]], i32* [[IB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP14:%.*]] = bitcast float* [[REF_TMP]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP14]]) #[[ATTR5]]
// CHECK1-NEXT:    store float 0.000000e+00, float* [[REF_TMP]], align 4, !tbaa [[TBAA14:![0-9]+]]
// CHECK1-NEXT:    [[TMP15:%.*]] = bitcast float* [[REF_TMP2]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP15]]) #[[ATTR5]]
// CHECK1-NEXT:    store float 0.000000e+00, float* [[REF_TMP2]], align 4, !tbaa [[TBAA14]]
// CHECK1-NEXT:    call void @_ZNSt7complexIfEC1ERKfS2_(%"class.std::complex"* nonnull align 4 dereferenceable(8) [[PARTIAL_SUM_ON_STACK]], float* nonnull align 4 dereferenceable(4) [[REF_TMP]], float* nonnull align 4 dereferenceable(4) [[REF_TMP2]]) #[[ATTR10:[0-9]+]]
// CHECK1-NEXT:    [[TMP16:%.*]] = bitcast float* [[REF_TMP2]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP16]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP17:%.*]] = bitcast float* [[REF_TMP]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP17]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP18:%.*]] = load i32, i32* [[IB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[MUL3:%.*]] = mul nsw i32 [[TMP18]], 4
// CHECK1-NEXT:    store i32 [[MUL3]], i32* [[ISTART_ON_STACK]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP19:%.*]] = load i32, i32* [[IB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK1-NEXT:    [[MUL5:%.*]] = mul nsw i32 [[ADD4]], 4
// CHECK1-NEXT:    store i32 [[MUL5]], i32* [[IEND_ON_STACK]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK1-NEXT:    [[TMP21:%.*]] = bitcast i32* [[ISTART_ON_STACK]] to i8*
// CHECK1-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK1-NEXT:    [[TMP23:%.*]] = bitcast i32* [[IEND_ON_STACK]] to i8*
// CHECK1-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK1-NEXT:    [[TMP25:%.*]] = bitcast %"class.std::complex"* [[PARTIAL_SUM_ON_STACK]] to i8*
// CHECK1-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP26:%.*]] = bitcast [3 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK1-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP7]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32*, i32*, %"class.std::complex"*)* @__omp_outlined__1 to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined__1_wrapper to i8*), i8** [[TMP26]], i64 3)
// 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:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP27]], 1
// CHECK1-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK1:       omp.inner.for.end:
// CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK1:       omp.loop.exit:
// CHECK1-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP7]])
// CHECK1-NEXT:    [[TMP28:%.*]] = bitcast i32* [[IB]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP28]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP29:%.*]] = bitcast i32* [[DOTOMP_IS_LAST]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP29]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP30:%.*]] = bitcast i32* [[DOTOMP_STRIDE]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP30]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP31:%.*]] = bitcast i32* [[DOTOMP_UB]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP31]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP32:%.*]] = bitcast i32* [[DOTOMP_LB]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP32]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP33:%.*]] = bitcast i32* [[DOTOMP_IV]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP33]]) #[[ATTR5]]
// CHECK1-NEXT:    call void @__kmpc_free_shared(i8* [[PARTIAL_SUM]], i64 8)
// CHECK1-NEXT:    call void @__kmpc_free_shared(i8* [[IEND]], i64 4)
// CHECK1-NEXT:    call void @__kmpc_free_shared(i8* [[ISTART]], i64 4)
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZNSt7complexIfEC1ERKfS2_
// CHECK1-SAME: (%"class.std::complex"* nonnull align 4 dereferenceable(8) [[THIS:%.*]], float* nonnull align 4 dereferenceable(4) [[__RE:%.*]], float* nonnull align 4 dereferenceable(4) [[__IM:%.*]]) unnamed_addr #[[ATTR3:[0-9]+]] comdat align 2 {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex"*, align 8
// CHECK1-NEXT:    [[__RE_ADDR:%.*]] = alloca float*, align 8
// CHECK1-NEXT:    [[__IM_ADDR:%.*]] = alloca float*, align 8
// CHECK1-NEXT:    store %"class.std::complex"* [[THIS]], %"class.std::complex"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    store float* [[__RE]], float** [[__RE_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    store float* [[__IM]], float** [[__IM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[THIS1:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[THIS_ADDR]], align 8
// CHECK1-NEXT:    [[TMP0:%.*]] = load float*, float** [[__RE_ADDR]], align 8
// CHECK1-NEXT:    [[TMP1:%.*]] = load float*, float** [[__IM_ADDR]], align 8
// CHECK1-NEXT:    call void @_ZNSt7complexIfEC2ERKfS2_(%"class.std::complex"* nonnull align 4 dereferenceable(8) [[THIS1]], float* nonnull align 4 dereferenceable(4) [[TMP0]], float* nonnull align 4 dereferenceable(4) [[TMP1]]) #[[ATTR10]]
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@__omp_outlined__1
// CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32* nonnull align 4 dereferenceable(4) [[ISTART:%.*]], i32* nonnull align 4 dereferenceable(4) [[IEND:%.*]], %"class.std::complex"* nonnull align 4 dereferenceable(8) [[PARTIAL_SUM:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[ISTART_ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[IEND_ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[PARTIAL_SUM_ADDR:%.*]] = alloca %"class.std::complex"*, align 8
// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[PARTIAL_SUM5:%.*]] = alloca %"class.std::complex", align 4
// CHECK1-NEXT:    [[REF_TMP:%.*]] = alloca float, align 4
// CHECK1-NEXT:    [[REF_TMP6:%.*]] = alloca float, align 4
// CHECK1-NEXT:    [[I7:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[REF_TMP14:%.*]] = alloca %"class.std::complex", align 4
// CHECK1-NEXT:    [[REF_TMP15:%.*]] = alloca float, align 4
// CHECK1-NEXT:    [[REF_TMP16:%.*]] = alloca float, align 4
// CHECK1-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x i8*], align 8
// CHECK1-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    store i32* [[ISTART]], i32** [[ISTART_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    store i32* [[IEND]], i32** [[IEND_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    store %"class.std::complex"* [[PARTIAL_SUM]], %"class.std::complex"** [[PARTIAL_SUM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[ISTART_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[IEND_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP2:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[PARTIAL_SUM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP3:%.*]] = bitcast i32* [[DOTOMP_IV]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP3]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP4:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP4]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, i32* [[TMP0]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    store i32 [[TMP5]], i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP6:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_1]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP6]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[TMP1]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    store i32 [[TMP7]], i32* [[DOTCAPTURE_EXPR_1]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP8:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_2]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP8]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[SUB:%.*]] = sub i32 [[TMP9]], [[TMP10]]
// CHECK1-NEXT:    [[SUB3:%.*]] = sub i32 [[SUB]], 1
// CHECK1-NEXT:    [[ADD:%.*]] = add i32 [[SUB3]], 1
// CHECK1-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], 1
// CHECK1-NEXT:    [[SUB4:%.*]] = sub i32 [[DIV]], 1
// CHECK1-NEXT:    store i32 [[SUB4]], i32* [[DOTCAPTURE_EXPR_2]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP11:%.*]] = bitcast i32* [[I]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP11]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    store i32 [[TMP12]], i32* [[I]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP13:%.*]] = bitcast i32* [[I]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP13]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP14]], [[TMP15]]
// CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK1:       omp.precond.then:
// CHECK1-NEXT:    [[TMP16:%.*]] = bitcast i32* [[DOTOMP_LB]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP16]]) #[[ATTR5]]
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP17:%.*]] = bitcast i32* [[DOTOMP_UB]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP17]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    store i32 [[TMP18]], i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP19:%.*]] = bitcast i32* [[DOTOMP_STRIDE]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP19]]) #[[ATTR5]]
// CHECK1-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP20:%.*]] = bitcast i32* [[DOTOMP_IS_LAST]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP20]]) #[[ATTR5]]
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP21:%.*]] = bitcast %"class.std::complex"* [[PARTIAL_SUM5]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 8, i8* [[TMP21]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP22:%.*]] = bitcast float* [[REF_TMP]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP22]]) #[[ATTR5]]
// CHECK1-NEXT:    store float 0.000000e+00, float* [[REF_TMP]], align 4, !tbaa [[TBAA14]]
// CHECK1-NEXT:    [[TMP23:%.*]] = bitcast float* [[REF_TMP6]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP23]]) #[[ATTR5]]
// CHECK1-NEXT:    store float 0.000000e+00, float* [[REF_TMP6]], align 4, !tbaa [[TBAA14]]
// CHECK1-NEXT:    call void @_ZNSt7complexIfEC1ERKfS2_(%"class.std::complex"* nonnull align 4 dereferenceable(8) [[PARTIAL_SUM5]], float* nonnull align 4 dereferenceable(4) [[REF_TMP]], float* nonnull align 4 dereferenceable(4) [[REF_TMP6]]) #[[ATTR10]]
// CHECK1-NEXT:    [[TMP24:%.*]] = bitcast float* [[REF_TMP6]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP24]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP25:%.*]] = bitcast float* [[REF_TMP]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP25]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP26:%.*]] = bitcast i32* [[I7]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP26]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP27:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP28:%.*]] = load i32, i32* [[TMP27]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    call void @__kmpc_for_static_init_4u(%struct.ident_t* @[[GLOB3:[0-9]+]], i32 [[TMP28]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK1-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK1:       omp.dispatch.cond:
// CHECK1-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[CMP8:%.*]] = icmp ugt i32 [[TMP29]], [[TMP30]]
// CHECK1-NEXT:    br i1 [[CMP8]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK1:       cond.true:
// CHECK1-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    br label [[COND_END:%.*]]
// CHECK1:       cond.false:
// CHECK1-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    br label [[COND_END]]
// CHECK1:       cond.end:
// CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP31]], [[COND_TRUE]] ], [ [[TMP32]], [[COND_FALSE]] ]
// CHECK1-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    store i32 [[TMP33]], i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[ADD9:%.*]] = add i32 [[TMP35]], 1
// CHECK1-NEXT:    [[CMP10:%.*]] = icmp ult i32 [[TMP34]], [[ADD9]]
// CHECK1-NEXT:    br i1 [[CMP10]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_CLEANUP:%.*]]
// CHECK1:       omp.dispatch.cleanup:
// CHECK1-NEXT:    br label [[OMP_DISPATCH_END:%.*]]
// CHECK1:       omp.dispatch.body:
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK1:       omp.inner.for.cond:
// CHECK1-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[ADD11:%.*]] = add i32 [[TMP37]], 1
// CHECK1-NEXT:    [[CMP12:%.*]] = icmp ult i32 [[TMP36]], [[ADD11]]
// CHECK1-NEXT:    br i1 [[CMP12]], 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:    [[TMP38:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[MUL:%.*]] = mul i32 [[TMP39]], 1
// CHECK1-NEXT:    [[ADD13:%.*]] = add i32 [[TMP38]], [[MUL]]
// CHECK1-NEXT:    store i32 [[ADD13]], i32* [[I7]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP40:%.*]] = bitcast %"class.std::complex"* [[REF_TMP14]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 8, i8* [[TMP40]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP41:%.*]] = bitcast float* [[REF_TMP15]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP41]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP42:%.*]] = load i32, i32* [[I7]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[CONV:%.*]] = sitofp i32 [[TMP42]] to float
// CHECK1-NEXT:    store float [[CONV]], float* [[REF_TMP15]], align 4, !tbaa [[TBAA14]]
// CHECK1-NEXT:    [[TMP43:%.*]] = bitcast float* [[REF_TMP16]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP43]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP44:%.*]] = load i32, i32* [[I7]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[CONV17:%.*]] = sitofp i32 [[TMP44]] to float
// CHECK1-NEXT:    store float [[CONV17]], float* [[REF_TMP16]], align 4, !tbaa [[TBAA14]]
// CHECK1-NEXT:    call void @_ZNSt7complexIfEC1ERKfS2_(%"class.std::complex"* nonnull align 4 dereferenceable(8) [[REF_TMP14]], float* nonnull align 4 dereferenceable(4) [[REF_TMP15]], float* nonnull align 4 dereferenceable(4) [[REF_TMP16]]) #[[ATTR10]]
// CHECK1-NEXT:    [[CALL:%.*]] = call nonnull align 4 dereferenceable(8) %"class.std::complex"* @_ZNSt7complexIfEpLIfEERS0_RKS_IT_E(%"class.std::complex"* nonnull align 4 dereferenceable(8) [[PARTIAL_SUM5]], %"class.std::complex"* nonnull align 4 dereferenceable(8) [[REF_TMP14]]) #[[ATTR10]]
// CHECK1-NEXT:    [[TMP45:%.*]] = bitcast float* [[REF_TMP16]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP45]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP46:%.*]] = bitcast float* [[REF_TMP15]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP46]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP47:%.*]] = bitcast %"class.std::complex"* [[REF_TMP14]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 8, i8* [[TMP47]]) #[[ATTR5]]
// 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:    [[TMP48:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[ADD18:%.*]] = add i32 [[TMP48]], 1
// CHECK1-NEXT:    store i32 [[ADD18]], i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK1:       omp.inner.for.end:
// CHECK1-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK1:       omp.dispatch.inc:
// CHECK1-NEXT:    [[TMP49:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP50:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[ADD19:%.*]] = add i32 [[TMP49]], [[TMP50]]
// CHECK1-NEXT:    store i32 [[ADD19]], i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP51:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP52:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[ADD20:%.*]] = add i32 [[TMP51]], [[TMP52]]
// CHECK1-NEXT:    store i32 [[ADD20]], i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK1:       omp.dispatch.end:
// CHECK1-NEXT:    [[TMP53:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP54:%.*]] = load i32, i32* [[TMP53]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB3]], i32 [[TMP54]])
// CHECK1-NEXT:    [[TMP55:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP56:%.*]] = load i32, i32* [[TMP55]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP57:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0
// CHECK1-NEXT:    [[TMP58:%.*]] = bitcast %"class.std::complex"* [[PARTIAL_SUM5]] to i8*
// CHECK1-NEXT:    store i8* [[TMP58]], i8** [[TMP57]], align 8
// CHECK1-NEXT:    [[TMP59:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8*
// CHECK1-NEXT:    [[TMP60:%.*]] = call i32 @__kmpc_nvptx_parallel_reduce_nowait_v2(%struct.ident_t* @[[GLOB1]], i32 [[TMP56]], i32 1, i64 8, i8* [[TMP59]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func)
// CHECK1-NEXT:    [[TMP61:%.*]] = icmp eq i32 [[TMP60]], 1
// CHECK1-NEXT:    br i1 [[TMP61]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]]
// CHECK1:       .omp.reduction.then:
// CHECK1-NEXT:    [[CALL21:%.*]] = call nonnull align 4 dereferenceable(8) %"class.std::complex"* @_ZNSt7complexIfEpLIfEERS0_RKS_IT_E(%"class.std::complex"* nonnull align 4 dereferenceable(8) [[TMP2]], %"class.std::complex"* nonnull align 4 dereferenceable(8) [[PARTIAL_SUM5]]) #[[ATTR10]]
// CHECK1-NEXT:    call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP56]])
// CHECK1-NEXT:    br label [[DOTOMP_REDUCTION_DONE]]
// CHECK1:       .omp.reduction.done:
// CHECK1-NEXT:    [[TMP62:%.*]] = bitcast i32* [[I7]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP62]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP63:%.*]] = bitcast %"class.std::complex"* [[PARTIAL_SUM5]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 8, i8* [[TMP63]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP64:%.*]] = bitcast i32* [[DOTOMP_IS_LAST]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP64]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP65:%.*]] = bitcast i32* [[DOTOMP_STRIDE]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP65]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP66:%.*]] = bitcast i32* [[DOTOMP_UB]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP66]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP67:%.*]] = bitcast i32* [[DOTOMP_LB]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP67]]) #[[ATTR5]]
// CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK1:       omp.precond.end:
// CHECK1-NEXT:    [[TMP68:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_2]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP68]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP69:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_1]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP69]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP70:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP70]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP71:%.*]] = bitcast i32* [[DOTOMP_IV]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP71]]) #[[ATTR5]]
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZNSt7complexIfEpLIfEERS0_RKS_IT_E
// CHECK1-SAME: (%"class.std::complex"* nonnull align 4 dereferenceable(8) [[THIS:%.*]], %"class.std::complex"* nonnull align 4 dereferenceable(8) [[__C:%.*]]) #[[ATTR4:[0-9]+]] comdat align 2 {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex"*, align 8
// CHECK1-NEXT:    [[__C_ADDR:%.*]] = alloca %"class.std::complex"*, align 8
// CHECK1-NEXT:    store %"class.std::complex"* [[THIS]], %"class.std::complex"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    store %"class.std::complex"* [[__C]], %"class.std::complex"** [[__C_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[THIS1:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[THIS_ADDR]], align 8
// CHECK1-NEXT:    [[TMP0:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[__C_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[CALL:%.*]] = call float @_ZNKSt7complexIfE4realEv(%"class.std::complex"* nonnull align 4 dereferenceable(8) [[TMP0]]) #[[ATTR10]]
// CHECK1-NEXT:    [[__RE_:%.*]] = getelementptr inbounds %"class.std::complex", %"class.std::complex"* [[THIS1]], i32 0, i32 0
// CHECK1-NEXT:    [[TMP1:%.*]] = load float, float* [[__RE_]], align 4, !tbaa [[TBAA16:![0-9]+]]
// CHECK1-NEXT:    [[ADD:%.*]] = fadd float [[TMP1]], [[CALL]]
// CHECK1-NEXT:    store float [[ADD]], float* [[__RE_]], align 4, !tbaa [[TBAA16]]
// CHECK1-NEXT:    [[TMP2:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[__C_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[CALL2:%.*]] = call float @_ZNKSt7complexIfE4imagEv(%"class.std::complex"* nonnull align 4 dereferenceable(8) [[TMP2]]) #[[ATTR10]]
// CHECK1-NEXT:    [[__IM_:%.*]] = getelementptr inbounds %"class.std::complex", %"class.std::complex"* [[THIS1]], i32 0, i32 1
// CHECK1-NEXT:    [[TMP3:%.*]] = load float, float* [[__IM_]], align 4, !tbaa [[TBAA18:![0-9]+]]
// CHECK1-NEXT:    [[ADD3:%.*]] = fadd float [[TMP3]], [[CALL2]]
// CHECK1-NEXT:    store float [[ADD3]], float* [[__IM_]], align 4, !tbaa [[TBAA18]]
// CHECK1-NEXT:    ret %"class.std::complex"* [[THIS1]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func
// CHECK1-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR6:[0-9]+]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i16, align 2
// CHECK1-NEXT:    [[DOTADDR2:%.*]] = alloca i16, align 2
// CHECK1-NEXT:    [[DOTADDR3:%.*]] = alloca i16, align 2
// CHECK1-NEXT:    [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [1 x i8*], align 8
// CHECK1-NEXT:    [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca %"class.std::complex", align 4
// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    store i16 [[TMP1]], i16* [[DOTADDR1]], align 2, !tbaa [[TBAA19:![0-9]+]]
// CHECK1-NEXT:    store i16 [[TMP2]], i16* [[DOTADDR2]], align 2, !tbaa [[TBAA19]]
// CHECK1-NEXT:    store i16 [[TMP3]], i16* [[DOTADDR3]], align 2, !tbaa [[TBAA19]]
// CHECK1-NEXT:    [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [1 x i8*]*
// CHECK1-NEXT:    [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2, !tbaa [[TBAA19]]
// CHECK1-NEXT:    [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2, !tbaa [[TBAA19]]
// CHECK1-NEXT:    [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2, !tbaa [[TBAA19]]
// CHECK1-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK1-NEXT:    [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8
// CHECK1-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0
// CHECK1-NEXT:    [[TMP12:%.*]] = bitcast i8* [[TMP10]] to %"class.std::complex"*
// CHECK1-NEXT:    [[TMP13:%.*]] = getelementptr %"class.std::complex", %"class.std::complex"* [[TMP12]], i64 1
// CHECK1-NEXT:    [[TMP14:%.*]] = bitcast %"class.std::complex"* [[TMP13]] to i8*
// CHECK1-NEXT:    [[TMP15:%.*]] = bitcast %"class.std::complex"* [[TMP12]] to i64*
// CHECK1-NEXT:    [[TMP16:%.*]] = bitcast %"class.std::complex"* [[DOTOMP_REDUCTION_ELEMENT]] to i64*
// CHECK1-NEXT:    [[TMP17:%.*]] = load i64, i64* [[TMP15]], align 4
// CHECK1-NEXT:    [[TMP18:%.*]] = call i32 @__kmpc_get_warp_size()
// CHECK1-NEXT:    [[TMP19:%.*]] = trunc i32 [[TMP18]] to i16
// CHECK1-NEXT:    [[TMP20:%.*]] = call i64 @__kmpc_shuffle_int64(i64 [[TMP17]], i16 [[TMP7]], i16 [[TMP19]])
// CHECK1-NEXT:    store i64 [[TMP20]], i64* [[TMP16]], align 4
// CHECK1-NEXT:    [[TMP21:%.*]] = getelementptr i64, i64* [[TMP15]], i64 1
// CHECK1-NEXT:    [[TMP22:%.*]] = getelementptr i64, i64* [[TMP16]], i64 1
// CHECK1-NEXT:    [[TMP23:%.*]] = bitcast %"class.std::complex"* [[DOTOMP_REDUCTION_ELEMENT]] to i8*
// CHECK1-NEXT:    store i8* [[TMP23]], i8** [[TMP11]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP24:%.*]] = icmp eq i16 [[TMP8]], 0
// CHECK1-NEXT:    [[TMP25:%.*]] = icmp eq i16 [[TMP8]], 1
// CHECK1-NEXT:    [[TMP26:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]]
// CHECK1-NEXT:    [[TMP27:%.*]] = and i1 [[TMP25]], [[TMP26]]
// CHECK1-NEXT:    [[TMP28:%.*]] = icmp eq i16 [[TMP8]], 2
// CHECK1-NEXT:    [[TMP29:%.*]] = and i16 [[TMP6]], 1
// CHECK1-NEXT:    [[TMP30:%.*]] = icmp eq i16 [[TMP29]], 0
// CHECK1-NEXT:    [[TMP31:%.*]] = and i1 [[TMP28]], [[TMP30]]
// CHECK1-NEXT:    [[TMP32:%.*]] = icmp sgt i16 [[TMP7]], 0
// CHECK1-NEXT:    [[TMP33:%.*]] = and i1 [[TMP31]], [[TMP32]]
// CHECK1-NEXT:    [[TMP34:%.*]] = or i1 [[TMP24]], [[TMP27]]
// CHECK1-NEXT:    [[TMP35:%.*]] = or i1 [[TMP34]], [[TMP33]]
// CHECK1-NEXT:    br i1 [[TMP35]], label [[THEN:%.*]], label [[ELSE:%.*]]
// CHECK1:       then:
// CHECK1-NEXT:    [[TMP36:%.*]] = bitcast [1 x i8*]* [[TMP5]] to i8*
// CHECK1-NEXT:    [[TMP37:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8*
// CHECK1-NEXT:    call void @"_omp$reduction$reduction_func"(i8* [[TMP36]], i8* [[TMP37]]) #[[ATTR5]]
// CHECK1-NEXT:    br label [[IFCONT:%.*]]
// CHECK1:       else:
// CHECK1-NEXT:    br label [[IFCONT]]
// CHECK1:       ifcont:
// CHECK1-NEXT:    [[TMP38:%.*]] = icmp eq i16 [[TMP8]], 1
// CHECK1-NEXT:    [[TMP39:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]]
// CHECK1-NEXT:    [[TMP40:%.*]] = and i1 [[TMP38]], [[TMP39]]
// CHECK1-NEXT:    br i1 [[TMP40]], label [[THEN4:%.*]], label [[ELSE5:%.*]]
// CHECK1:       then4:
// CHECK1-NEXT:    [[TMP41:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0
// CHECK1-NEXT:    [[TMP42:%.*]] = load i8*, i8** [[TMP41]], align 8
// CHECK1-NEXT:    [[TMP43:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK1-NEXT:    [[TMP44:%.*]] = load i8*, i8** [[TMP43]], align 8
// CHECK1-NEXT:    [[TMP45:%.*]] = bitcast i8* [[TMP42]] to %"class.std::complex"*
// CHECK1-NEXT:    [[TMP46:%.*]] = bitcast i8* [[TMP44]] to %"class.std::complex"*
// CHECK1-NEXT:    [[TMP47:%.*]] = bitcast %"class.std::complex"* [[TMP46]] to i8*
// CHECK1-NEXT:    [[TMP48:%.*]] = bitcast %"class.std::complex"* [[TMP45]] to i8*
// CHECK1-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP47]], i8* align 4 [[TMP48]], i64 8, i1 false), !tbaa.struct !21
// CHECK1-NEXT:    br label [[IFCONT6:%.*]]
// CHECK1:       else5:
// CHECK1-NEXT:    br label [[IFCONT6]]
// CHECK1:       ifcont6:
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func
// CHECK1-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR6]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCNT_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK1-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK1-NEXT:    [[NVPTX_LANE_ID:%.*]] = and i32 [[TMP4]], 31
// CHECK1-NEXT:    [[TMP5:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK1-NEXT:    [[NVPTX_WARP_ID:%.*]] = ashr i32 [[TMP5]], 5
// CHECK1-NEXT:    [[TMP6:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK1-NEXT:    [[TMP7:%.*]] = bitcast i8* [[TMP6]] to [1 x i8*]*
// CHECK1-NEXT:    store i32 0, i32* [[DOTCNT_ADDR]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    br label [[PRECOND:%.*]]
// CHECK1:       precond:
// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCNT_ADDR]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP9:%.*]] = icmp ult i32 [[TMP8]], 2
// CHECK1-NEXT:    br i1 [[TMP9]], label [[BODY:%.*]], label [[EXIT:%.*]]
// CHECK1:       body:
// CHECK1-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4:[0-9]+]], i32 [[TMP2]])
// CHECK1-NEXT:    [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0
// CHECK1-NEXT:    br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]]
// CHECK1:       then:
// CHECK1-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP7]], i64 0, i64 0
// CHECK1-NEXT:    [[TMP11:%.*]] = load i8*, i8** [[TMP10]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP12:%.*]] = bitcast i8* [[TMP11]] to i32*
// CHECK1-NEXT:    [[TMP13:%.*]] = getelementptr i32, i32* [[TMP12]], i32 [[TMP8]]
// CHECK1-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]]
// CHECK1-NEXT:    [[TMP15:%.*]] = load i32, i32* [[TMP13]], align 4
// CHECK1-NEXT:    store volatile i32 [[TMP15]], i32 addrspace(3)* [[TMP14]], align 4
// CHECK1-NEXT:    br label [[IFCONT:%.*]]
// CHECK1:       else:
// CHECK1-NEXT:    br label [[IFCONT]]
// CHECK1:       ifcont:
// CHECK1-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]])
// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTADDR1]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[TMP3]], [[TMP16]]
// CHECK1-NEXT:    br i1 [[IS_ACTIVE_THREAD]], label [[THEN2:%.*]], label [[ELSE3:%.*]]
// CHECK1:       then2:
// CHECK1-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[TMP3]]
// CHECK1-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP7]], i64 0, i64 0
// CHECK1-NEXT:    [[TMP19:%.*]] = load i8*, i8** [[TMP18]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP20:%.*]] = bitcast i8* [[TMP19]] to i32*
// CHECK1-NEXT:    [[TMP21:%.*]] = getelementptr i32, i32* [[TMP20]], i32 [[TMP8]]
// CHECK1-NEXT:    [[TMP22:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP17]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    store i32 [[TMP22]], i32* [[TMP21]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    br label [[IFCONT4:%.*]]
// CHECK1:       else3:
// CHECK1-NEXT:    br label [[IFCONT4]]
// CHECK1:       ifcont4:
// CHECK1-NEXT:    [[TMP23:%.*]] = add nsw i32 [[TMP8]], 1
// CHECK1-NEXT:    store i32 [[TMP23]], i32* [[DOTCNT_ADDR]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    br label [[PRECOND]]
// CHECK1:       exit:
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@__omp_outlined__1_wrapper
// CHECK1-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR6]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
// CHECK1-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2, !tbaa [[TBAA19]]
// CHECK1-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK1-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
// CHECK1-NEXT:    [[TMP2:%.*]] = load i8**, i8*** [[GLOBAL_ARGS]], align 8
// CHECK1-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 0
// CHECK1-NEXT:    [[TMP4:%.*]] = bitcast i8** [[TMP3]] to i32**
// CHECK1-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[TMP4]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 1
// CHECK1-NEXT:    [[TMP7:%.*]] = bitcast i8** [[TMP6]] to i32**
// CHECK1-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[TMP7]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 2
// CHECK1-NEXT:    [[TMP10:%.*]] = bitcast i8** [[TMP9]] to %"class.std::complex"**
// CHECK1-NEXT:    [[TMP11:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[TMP10]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    call void @__omp_outlined__1(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]], i32* [[TMP5]], i32* [[TMP8]], %"class.std::complex"* [[TMP11]]) #[[ATTR5]]
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z17complex_reductionIdEvv_l16
// CHECK1-SAME: () #[[ATTR0]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 1, i1 true, i1 true)
// CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
// CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK1:       user_code.entry:
// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK1-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK1-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    call void @__omp_outlined__2(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR5]]
// CHECK1-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 1, i1 true)
// CHECK1-NEXT:    ret void
// CHECK1:       worker.exit:
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@__omp_outlined__2
// CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[IB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[REF_TMP:%.*]] = alloca double, align 8
// CHECK1-NEXT:    [[REF_TMP2:%.*]] = alloca double, align 8
// CHECK1-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x i8*], align 8
// CHECK1-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[ISTART:%.*]] = call align 16 i8* @__kmpc_alloc_shared(i64 4)
// CHECK1-NEXT:    [[ISTART_ON_STACK:%.*]] = bitcast i8* [[ISTART]] to i32*
// CHECK1-NEXT:    [[IEND:%.*]] = call align 16 i8* @__kmpc_alloc_shared(i64 4)
// CHECK1-NEXT:    [[IEND_ON_STACK:%.*]] = bitcast i8* [[IEND]] to i32*
// CHECK1-NEXT:    [[PARTIAL_SUM:%.*]] = call align 16 i8* @__kmpc_alloc_shared(i64 16)
// CHECK1-NEXT:    [[PARTIAL_SUM_ON_STACK:%.*]] = bitcast i8* [[PARTIAL_SUM]] to %"class.std::complex.0"*
// CHECK1-NEXT:    [[TMP0:%.*]] = bitcast i32* [[DOTOMP_IV]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP0]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP1:%.*]] = bitcast i32* [[DOTOMP_LB]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP1]]) #[[ATTR5]]
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP2:%.*]] = bitcast i32* [[DOTOMP_UB]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP2]]) #[[ATTR5]]
// CHECK1-NEXT:    store i32 99, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP3:%.*]] = bitcast i32* [[DOTOMP_STRIDE]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP3]]) #[[ATTR5]]
// CHECK1-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP4:%.*]] = bitcast i32* [[DOTOMP_IS_LAST]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP4]]) #[[ATTR5]]
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP5:%.*]] = bitcast i32* [[IB]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP5]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP6:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[TMP6]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP7]], i32 92, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP8]], 99
// CHECK1-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK1:       cond.true:
// CHECK1-NEXT:    br label [[COND_END:%.*]]
// CHECK1:       cond.false:
// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    br label [[COND_END]]
// CHECK1:       cond.end:
// CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP9]], [[COND_FALSE]] ]
// CHECK1-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    store i32 [[TMP10]], i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK1:       omp.inner.for.cond:
// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP11]], [[TMP12]]
// 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:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP13]], 1
// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK1-NEXT:    store i32 [[ADD]], i32* [[IB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP14:%.*]] = bitcast double* [[REF_TMP]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 8, i8* [[TMP14]]) #[[ATTR5]]
// CHECK1-NEXT:    store double 0.000000e+00, double* [[REF_TMP]], align 8, !tbaa [[TBAA22:![0-9]+]]
// CHECK1-NEXT:    [[TMP15:%.*]] = bitcast double* [[REF_TMP2]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 8, i8* [[TMP15]]) #[[ATTR5]]
// CHECK1-NEXT:    store double 0.000000e+00, double* [[REF_TMP2]], align 8, !tbaa [[TBAA22]]
// CHECK1-NEXT:    call void @_ZNSt7complexIdEC1ERKdS2_(%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[PARTIAL_SUM_ON_STACK]], double* nonnull align 8 dereferenceable(8) [[REF_TMP]], double* nonnull align 8 dereferenceable(8) [[REF_TMP2]]) #[[ATTR10]]
// CHECK1-NEXT:    [[TMP16:%.*]] = bitcast double* [[REF_TMP2]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 8, i8* [[TMP16]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP17:%.*]] = bitcast double* [[REF_TMP]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 8, i8* [[TMP17]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP18:%.*]] = load i32, i32* [[IB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[MUL3:%.*]] = mul nsw i32 [[TMP18]], 4
// CHECK1-NEXT:    store i32 [[MUL3]], i32* [[ISTART_ON_STACK]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP19:%.*]] = load i32, i32* [[IB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK1-NEXT:    [[MUL5:%.*]] = mul nsw i32 [[ADD4]], 4
// CHECK1-NEXT:    store i32 [[MUL5]], i32* [[IEND_ON_STACK]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK1-NEXT:    [[TMP21:%.*]] = bitcast i32* [[ISTART_ON_STACK]] to i8*
// CHECK1-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK1-NEXT:    [[TMP23:%.*]] = bitcast i32* [[IEND_ON_STACK]] to i8*
// CHECK1-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK1-NEXT:    [[TMP25:%.*]] = bitcast %"class.std::complex.0"* [[PARTIAL_SUM_ON_STACK]] to i8*
// CHECK1-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP26:%.*]] = bitcast [3 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK1-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP7]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32*, i32*, %"class.std::complex.0"*)* @__omp_outlined__3 to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined__3_wrapper to i8*), i8** [[TMP26]], i64 3)
// 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:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP27]], 1
// CHECK1-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK1:       omp.inner.for.end:
// CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK1:       omp.loop.exit:
// CHECK1-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP7]])
// CHECK1-NEXT:    [[TMP28:%.*]] = bitcast i32* [[IB]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP28]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP29:%.*]] = bitcast i32* [[DOTOMP_IS_LAST]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP29]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP30:%.*]] = bitcast i32* [[DOTOMP_STRIDE]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP30]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP31:%.*]] = bitcast i32* [[DOTOMP_UB]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP31]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP32:%.*]] = bitcast i32* [[DOTOMP_LB]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP32]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP33:%.*]] = bitcast i32* [[DOTOMP_IV]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP33]]) #[[ATTR5]]
// CHECK1-NEXT:    call void @__kmpc_free_shared(i8* [[PARTIAL_SUM]], i64 16)
// CHECK1-NEXT:    call void @__kmpc_free_shared(i8* [[IEND]], i64 4)
// CHECK1-NEXT:    call void @__kmpc_free_shared(i8* [[ISTART]], i64 4)
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZNSt7complexIdEC1ERKdS2_
// CHECK1-SAME: (%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[THIS:%.*]], double* nonnull align 8 dereferenceable(8) [[__RE:%.*]], double* nonnull align 8 dereferenceable(8) [[__IM:%.*]]) unnamed_addr #[[ATTR3]] comdat align 2 {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex.0"*, align 8
// CHECK1-NEXT:    [[__RE_ADDR:%.*]] = alloca double*, align 8
// CHECK1-NEXT:    [[__IM_ADDR:%.*]] = alloca double*, align 8
// CHECK1-NEXT:    store %"class.std::complex.0"* [[THIS]], %"class.std::complex.0"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    store double* [[__RE]], double** [[__RE_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    store double* [[__IM]], double** [[__IM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[THIS1:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[THIS_ADDR]], align 8
// CHECK1-NEXT:    [[TMP0:%.*]] = load double*, double** [[__RE_ADDR]], align 8
// CHECK1-NEXT:    [[TMP1:%.*]] = load double*, double** [[__IM_ADDR]], align 8
// CHECK1-NEXT:    call void @_ZNSt7complexIdEC2ERKdS2_(%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[THIS1]], double* nonnull align 8 dereferenceable(8) [[TMP0]], double* nonnull align 8 dereferenceable(8) [[TMP1]]) #[[ATTR10]]
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@__omp_outlined__3
// CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32* nonnull align 4 dereferenceable(4) [[ISTART:%.*]], i32* nonnull align 4 dereferenceable(4) [[IEND:%.*]], %"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[PARTIAL_SUM:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[ISTART_ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[IEND_ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[PARTIAL_SUM_ADDR:%.*]] = alloca %"class.std::complex.0"*, align 8
// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[PARTIAL_SUM5:%.*]] = alloca %"class.std::complex.0", align 8
// CHECK1-NEXT:    [[REF_TMP:%.*]] = alloca double, align 8
// CHECK1-NEXT:    [[REF_TMP6:%.*]] = alloca double, align 8
// CHECK1-NEXT:    [[I7:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[REF_TMP14:%.*]] = alloca %"class.std::complex.0", align 8
// CHECK1-NEXT:    [[REF_TMP15:%.*]] = alloca double, align 8
// CHECK1-NEXT:    [[REF_TMP16:%.*]] = alloca double, align 8
// CHECK1-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x i8*], align 8
// CHECK1-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    store i32* [[ISTART]], i32** [[ISTART_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    store i32* [[IEND]], i32** [[IEND_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    store %"class.std::complex.0"* [[PARTIAL_SUM]], %"class.std::complex.0"** [[PARTIAL_SUM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[ISTART_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[IEND_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP2:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[PARTIAL_SUM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP3:%.*]] = bitcast i32* [[DOTOMP_IV]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP3]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP4:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP4]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, i32* [[TMP0]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    store i32 [[TMP5]], i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP6:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_1]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP6]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[TMP1]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    store i32 [[TMP7]], i32* [[DOTCAPTURE_EXPR_1]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP8:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_2]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP8]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[SUB:%.*]] = sub i32 [[TMP9]], [[TMP10]]
// CHECK1-NEXT:    [[SUB3:%.*]] = sub i32 [[SUB]], 1
// CHECK1-NEXT:    [[ADD:%.*]] = add i32 [[SUB3]], 1
// CHECK1-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], 1
// CHECK1-NEXT:    [[SUB4:%.*]] = sub i32 [[DIV]], 1
// CHECK1-NEXT:    store i32 [[SUB4]], i32* [[DOTCAPTURE_EXPR_2]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP11:%.*]] = bitcast i32* [[I]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP11]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    store i32 [[TMP12]], i32* [[I]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP13:%.*]] = bitcast i32* [[I]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP13]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP14]], [[TMP15]]
// CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK1:       omp.precond.then:
// CHECK1-NEXT:    [[TMP16:%.*]] = bitcast i32* [[DOTOMP_LB]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP16]]) #[[ATTR5]]
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP17:%.*]] = bitcast i32* [[DOTOMP_UB]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP17]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    store i32 [[TMP18]], i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP19:%.*]] = bitcast i32* [[DOTOMP_STRIDE]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP19]]) #[[ATTR5]]
// CHECK1-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP20:%.*]] = bitcast i32* [[DOTOMP_IS_LAST]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP20]]) #[[ATTR5]]
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP21:%.*]] = bitcast %"class.std::complex.0"* [[PARTIAL_SUM5]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 16, i8* [[TMP21]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP22:%.*]] = bitcast double* [[REF_TMP]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 8, i8* [[TMP22]]) #[[ATTR5]]
// CHECK1-NEXT:    store double 0.000000e+00, double* [[REF_TMP]], align 8, !tbaa [[TBAA22]]
// CHECK1-NEXT:    [[TMP23:%.*]] = bitcast double* [[REF_TMP6]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 8, i8* [[TMP23]]) #[[ATTR5]]
// CHECK1-NEXT:    store double 0.000000e+00, double* [[REF_TMP6]], align 8, !tbaa [[TBAA22]]
// CHECK1-NEXT:    call void @_ZNSt7complexIdEC1ERKdS2_(%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[PARTIAL_SUM5]], double* nonnull align 8 dereferenceable(8) [[REF_TMP]], double* nonnull align 8 dereferenceable(8) [[REF_TMP6]]) #[[ATTR10]]
// CHECK1-NEXT:    [[TMP24:%.*]] = bitcast double* [[REF_TMP6]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 8, i8* [[TMP24]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP25:%.*]] = bitcast double* [[REF_TMP]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 8, i8* [[TMP25]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP26:%.*]] = bitcast i32* [[I7]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP26]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP27:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP28:%.*]] = load i32, i32* [[TMP27]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    call void @__kmpc_for_static_init_4u(%struct.ident_t* @[[GLOB3]], i32 [[TMP28]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK1-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK1:       omp.dispatch.cond:
// CHECK1-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[CMP8:%.*]] = icmp ugt i32 [[TMP29]], [[TMP30]]
// CHECK1-NEXT:    br i1 [[CMP8]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK1:       cond.true:
// CHECK1-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    br label [[COND_END:%.*]]
// CHECK1:       cond.false:
// CHECK1-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    br label [[COND_END]]
// CHECK1:       cond.end:
// CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP31]], [[COND_TRUE]] ], [ [[TMP32]], [[COND_FALSE]] ]
// CHECK1-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    store i32 [[TMP33]], i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[ADD9:%.*]] = add i32 [[TMP35]], 1
// CHECK1-NEXT:    [[CMP10:%.*]] = icmp ult i32 [[TMP34]], [[ADD9]]
// CHECK1-NEXT:    br i1 [[CMP10]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_CLEANUP:%.*]]
// CHECK1:       omp.dispatch.cleanup:
// CHECK1-NEXT:    br label [[OMP_DISPATCH_END:%.*]]
// CHECK1:       omp.dispatch.body:
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK1:       omp.inner.for.cond:
// CHECK1-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[ADD11:%.*]] = add i32 [[TMP37]], 1
// CHECK1-NEXT:    [[CMP12:%.*]] = icmp ult i32 [[TMP36]], [[ADD11]]
// CHECK1-NEXT:    br i1 [[CMP12]], 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:    [[TMP38:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[MUL:%.*]] = mul i32 [[TMP39]], 1
// CHECK1-NEXT:    [[ADD13:%.*]] = add i32 [[TMP38]], [[MUL]]
// CHECK1-NEXT:    store i32 [[ADD13]], i32* [[I7]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP40:%.*]] = bitcast %"class.std::complex.0"* [[REF_TMP14]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 16, i8* [[TMP40]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP41:%.*]] = bitcast double* [[REF_TMP15]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 8, i8* [[TMP41]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP42:%.*]] = load i32, i32* [[I7]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[CONV:%.*]] = sitofp i32 [[TMP42]] to double
// CHECK1-NEXT:    store double [[CONV]], double* [[REF_TMP15]], align 8, !tbaa [[TBAA22]]
// CHECK1-NEXT:    [[TMP43:%.*]] = bitcast double* [[REF_TMP16]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.start.p0i8(i64 8, i8* [[TMP43]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP44:%.*]] = load i32, i32* [[I7]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[CONV17:%.*]] = sitofp i32 [[TMP44]] to double
// CHECK1-NEXT:    store double [[CONV17]], double* [[REF_TMP16]], align 8, !tbaa [[TBAA22]]
// CHECK1-NEXT:    call void @_ZNSt7complexIdEC1ERKdS2_(%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[REF_TMP14]], double* nonnull align 8 dereferenceable(8) [[REF_TMP15]], double* nonnull align 8 dereferenceable(8) [[REF_TMP16]]) #[[ATTR10]]
// CHECK1-NEXT:    [[CALL:%.*]] = call nonnull align 8 dereferenceable(16) %"class.std::complex.0"* @_ZNSt7complexIdEpLIdEERS0_RKS_IT_E(%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[PARTIAL_SUM5]], %"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[REF_TMP14]]) #[[ATTR10]]
// CHECK1-NEXT:    [[TMP45:%.*]] = bitcast double* [[REF_TMP16]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 8, i8* [[TMP45]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP46:%.*]] = bitcast double* [[REF_TMP15]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 8, i8* [[TMP46]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP47:%.*]] = bitcast %"class.std::complex.0"* [[REF_TMP14]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 16, i8* [[TMP47]]) #[[ATTR5]]
// 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:    [[TMP48:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[ADD18:%.*]] = add i32 [[TMP48]], 1
// CHECK1-NEXT:    store i32 [[ADD18]], i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK1:       omp.inner.for.end:
// CHECK1-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK1:       omp.dispatch.inc:
// CHECK1-NEXT:    [[TMP49:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP50:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[ADD19:%.*]] = add i32 [[TMP49]], [[TMP50]]
// CHECK1-NEXT:    store i32 [[ADD19]], i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP51:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP52:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[ADD20:%.*]] = add i32 [[TMP51]], [[TMP52]]
// CHECK1-NEXT:    store i32 [[ADD20]], i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK1:       omp.dispatch.end:
// CHECK1-NEXT:    [[TMP53:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP54:%.*]] = load i32, i32* [[TMP53]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB3]], i32 [[TMP54]])
// CHECK1-NEXT:    [[TMP55:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP56:%.*]] = load i32, i32* [[TMP55]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP57:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0
// CHECK1-NEXT:    [[TMP58:%.*]] = bitcast %"class.std::complex.0"* [[PARTIAL_SUM5]] to i8*
// CHECK1-NEXT:    store i8* [[TMP58]], i8** [[TMP57]], align 8
// CHECK1-NEXT:    [[TMP59:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8*
// CHECK1-NEXT:    [[TMP60:%.*]] = call i32 @__kmpc_nvptx_parallel_reduce_nowait_v2(%struct.ident_t* @[[GLOB1]], i32 [[TMP56]], i32 1, i64 8, i8* [[TMP59]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func5, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func6)
// CHECK1-NEXT:    [[TMP61:%.*]] = icmp eq i32 [[TMP60]], 1
// CHECK1-NEXT:    br i1 [[TMP61]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]]
// CHECK1:       .omp.reduction.then:
// CHECK1-NEXT:    [[CALL21:%.*]] = call nonnull align 8 dereferenceable(16) %"class.std::complex.0"* @_ZNSt7complexIdEpLIdEERS0_RKS_IT_E(%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[TMP2]], %"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[PARTIAL_SUM5]]) #[[ATTR10]]
// CHECK1-NEXT:    call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP56]])
// CHECK1-NEXT:    br label [[DOTOMP_REDUCTION_DONE]]
// CHECK1:       .omp.reduction.done:
// CHECK1-NEXT:    [[TMP62:%.*]] = bitcast i32* [[I7]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP62]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP63:%.*]] = bitcast %"class.std::complex.0"* [[PARTIAL_SUM5]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 16, i8* [[TMP63]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP64:%.*]] = bitcast i32* [[DOTOMP_IS_LAST]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP64]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP65:%.*]] = bitcast i32* [[DOTOMP_STRIDE]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP65]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP66:%.*]] = bitcast i32* [[DOTOMP_UB]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP66]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP67:%.*]] = bitcast i32* [[DOTOMP_LB]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP67]]) #[[ATTR5]]
// CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK1:       omp.precond.end:
// CHECK1-NEXT:    [[TMP68:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_2]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP68]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP69:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_1]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP69]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP70:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP70]]) #[[ATTR5]]
// CHECK1-NEXT:    [[TMP71:%.*]] = bitcast i32* [[DOTOMP_IV]] to i8*
// CHECK1-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP71]]) #[[ATTR5]]
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZNSt7complexIdEpLIdEERS0_RKS_IT_E
// CHECK1-SAME: (%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[THIS:%.*]], %"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[__C:%.*]]) #[[ATTR4]] comdat align 2 {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex.0"*, align 8
// CHECK1-NEXT:    [[__C_ADDR:%.*]] = alloca %"class.std::complex.0"*, align 8
// CHECK1-NEXT:    store %"class.std::complex.0"* [[THIS]], %"class.std::complex.0"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    store %"class.std::complex.0"* [[__C]], %"class.std::complex.0"** [[__C_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[THIS1:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[THIS_ADDR]], align 8
// CHECK1-NEXT:    [[TMP0:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[__C_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[CALL:%.*]] = call double @_ZNKSt7complexIdE4realEv(%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[TMP0]]) #[[ATTR10]]
// CHECK1-NEXT:    [[__RE_:%.*]] = getelementptr inbounds %"class.std::complex.0", %"class.std::complex.0"* [[THIS1]], i32 0, i32 0
// CHECK1-NEXT:    [[TMP1:%.*]] = load double, double* [[__RE_]], align 8, !tbaa [[TBAA24:![0-9]+]]
// CHECK1-NEXT:    [[ADD:%.*]] = fadd double [[TMP1]], [[CALL]]
// CHECK1-NEXT:    store double [[ADD]], double* [[__RE_]], align 8, !tbaa [[TBAA24]]
// CHECK1-NEXT:    [[TMP2:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[__C_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[CALL2:%.*]] = call double @_ZNKSt7complexIdE4imagEv(%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[TMP2]]) #[[ATTR10]]
// CHECK1-NEXT:    [[__IM_:%.*]] = getelementptr inbounds %"class.std::complex.0", %"class.std::complex.0"* [[THIS1]], i32 0, i32 1
// CHECK1-NEXT:    [[TMP3:%.*]] = load double, double* [[__IM_]], align 8, !tbaa [[TBAA26:![0-9]+]]
// CHECK1-NEXT:    [[ADD3:%.*]] = fadd double [[TMP3]], [[CALL2]]
// CHECK1-NEXT:    store double [[ADD3]], double* [[__IM_]], align 8, !tbaa [[TBAA26]]
// CHECK1-NEXT:    ret %"class.std::complex.0"* [[THIS1]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func5
// CHECK1-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR6]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i16, align 2
// CHECK1-NEXT:    [[DOTADDR2:%.*]] = alloca i16, align 2
// CHECK1-NEXT:    [[DOTADDR3:%.*]] = alloca i16, align 2
// CHECK1-NEXT:    [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [1 x i8*], align 8
// CHECK1-NEXT:    [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca %"class.std::complex.0", align 8
// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    store i16 [[TMP1]], i16* [[DOTADDR1]], align 2, !tbaa [[TBAA19]]
// CHECK1-NEXT:    store i16 [[TMP2]], i16* [[DOTADDR2]], align 2, !tbaa [[TBAA19]]
// CHECK1-NEXT:    store i16 [[TMP3]], i16* [[DOTADDR3]], align 2, !tbaa [[TBAA19]]
// CHECK1-NEXT:    [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [1 x i8*]*
// CHECK1-NEXT:    [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2, !tbaa [[TBAA19]]
// CHECK1-NEXT:    [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2, !tbaa [[TBAA19]]
// CHECK1-NEXT:    [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2, !tbaa [[TBAA19]]
// CHECK1-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK1-NEXT:    [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8
// CHECK1-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0
// CHECK1-NEXT:    [[TMP12:%.*]] = bitcast i8* [[TMP10]] to %"class.std::complex.0"*
// CHECK1-NEXT:    [[TMP13:%.*]] = getelementptr %"class.std::complex.0", %"class.std::complex.0"* [[TMP12]], i64 1
// CHECK1-NEXT:    [[TMP14:%.*]] = bitcast %"class.std::complex.0"* [[TMP13]] to i8*
// CHECK1-NEXT:    [[TMP15:%.*]] = bitcast %"class.std::complex.0"* [[TMP12]] to i64*
// CHECK1-NEXT:    [[TMP16:%.*]] = bitcast %"class.std::complex.0"* [[DOTOMP_REDUCTION_ELEMENT]] to i64*
// CHECK1-NEXT:    br label [[DOTSHUFFLE_PRE_COND:%.*]]
// CHECK1:       .shuffle.pre_cond:
// CHECK1-NEXT:    [[TMP17:%.*]] = phi i64* [ [[TMP15]], [[ENTRY:%.*]] ], [ [[TMP29:%.*]], [[DOTSHUFFLE_THEN:%.*]] ]
// CHECK1-NEXT:    [[TMP18:%.*]] = phi i64* [ [[TMP16]], [[ENTRY]] ], [ [[TMP30:%.*]], [[DOTSHUFFLE_THEN]] ]
// CHECK1-NEXT:    [[TMP19:%.*]] = bitcast i64* [[TMP17]] to i8*
// CHECK1-NEXT:    [[TMP20:%.*]] = ptrtoint i8* [[TMP14]] to i64
// CHECK1-NEXT:    [[TMP21:%.*]] = ptrtoint i8* [[TMP19]] to i64
// CHECK1-NEXT:    [[TMP22:%.*]] = sub i64 [[TMP20]], [[TMP21]]
// CHECK1-NEXT:    [[TMP23:%.*]] = sdiv exact i64 [[TMP22]], ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)
// CHECK1-NEXT:    [[TMP24:%.*]] = icmp sgt i64 [[TMP23]], 7
// CHECK1-NEXT:    br i1 [[TMP24]], label [[DOTSHUFFLE_THEN]], label [[DOTSHUFFLE_EXIT:%.*]]
// CHECK1:       .shuffle.then:
// CHECK1-NEXT:    [[TMP25:%.*]] = load i64, i64* [[TMP17]], align 8
// CHECK1-NEXT:    [[TMP26:%.*]] = call i32 @__kmpc_get_warp_size()
// CHECK1-NEXT:    [[TMP27:%.*]] = trunc i32 [[TMP26]] to i16
// CHECK1-NEXT:    [[TMP28:%.*]] = call i64 @__kmpc_shuffle_int64(i64 [[TMP25]], i16 [[TMP7]], i16 [[TMP27]])
// CHECK1-NEXT:    store i64 [[TMP28]], i64* [[TMP18]], align 8
// CHECK1-NEXT:    [[TMP29]] = getelementptr i64, i64* [[TMP17]], i64 1
// CHECK1-NEXT:    [[TMP30]] = getelementptr i64, i64* [[TMP18]], i64 1
// CHECK1-NEXT:    br label [[DOTSHUFFLE_PRE_COND]]
// CHECK1:       .shuffle.exit:
// CHECK1-NEXT:    [[TMP31:%.*]] = bitcast %"class.std::complex.0"* [[DOTOMP_REDUCTION_ELEMENT]] to i8*
// CHECK1-NEXT:    store i8* [[TMP31]], i8** [[TMP11]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP32:%.*]] = icmp eq i16 [[TMP8]], 0
// CHECK1-NEXT:    [[TMP33:%.*]] = icmp eq i16 [[TMP8]], 1
// CHECK1-NEXT:    [[TMP34:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]]
// CHECK1-NEXT:    [[TMP35:%.*]] = and i1 [[TMP33]], [[TMP34]]
// CHECK1-NEXT:    [[TMP36:%.*]] = icmp eq i16 [[TMP8]], 2
// CHECK1-NEXT:    [[TMP37:%.*]] = and i16 [[TMP6]], 1
// CHECK1-NEXT:    [[TMP38:%.*]] = icmp eq i16 [[TMP37]], 0
// CHECK1-NEXT:    [[TMP39:%.*]] = and i1 [[TMP36]], [[TMP38]]
// CHECK1-NEXT:    [[TMP40:%.*]] = icmp sgt i16 [[TMP7]], 0
// CHECK1-NEXT:    [[TMP41:%.*]] = and i1 [[TMP39]], [[TMP40]]
// CHECK1-NEXT:    [[TMP42:%.*]] = or i1 [[TMP32]], [[TMP35]]
// CHECK1-NEXT:    [[TMP43:%.*]] = or i1 [[TMP42]], [[TMP41]]
// CHECK1-NEXT:    br i1 [[TMP43]], label [[THEN:%.*]], label [[ELSE:%.*]]
// CHECK1:       then:
// CHECK1-NEXT:    [[TMP44:%.*]] = bitcast [1 x i8*]* [[TMP5]] to i8*
// CHECK1-NEXT:    [[TMP45:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8*
// CHECK1-NEXT:    call void @"_omp$reduction$reduction_func4"(i8* [[TMP44]], i8* [[TMP45]]) #[[ATTR5]]
// CHECK1-NEXT:    br label [[IFCONT:%.*]]
// CHECK1:       else:
// CHECK1-NEXT:    br label [[IFCONT]]
// CHECK1:       ifcont:
// CHECK1-NEXT:    [[TMP46:%.*]] = icmp eq i16 [[TMP8]], 1
// CHECK1-NEXT:    [[TMP47:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]]
// CHECK1-NEXT:    [[TMP48:%.*]] = and i1 [[TMP46]], [[TMP47]]
// CHECK1-NEXT:    br i1 [[TMP48]], label [[THEN4:%.*]], label [[ELSE5:%.*]]
// CHECK1:       then4:
// CHECK1-NEXT:    [[TMP49:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0
// CHECK1-NEXT:    [[TMP50:%.*]] = load i8*, i8** [[TMP49]], align 8
// CHECK1-NEXT:    [[TMP51:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK1-NEXT:    [[TMP52:%.*]] = load i8*, i8** [[TMP51]], align 8
// CHECK1-NEXT:    [[TMP53:%.*]] = bitcast i8* [[TMP50]] to %"class.std::complex.0"*
// CHECK1-NEXT:    [[TMP54:%.*]] = bitcast i8* [[TMP52]] to %"class.std::complex.0"*
// CHECK1-NEXT:    [[TMP55:%.*]] = bitcast %"class.std::complex.0"* [[TMP54]] to i8*
// CHECK1-NEXT:    [[TMP56:%.*]] = bitcast %"class.std::complex.0"* [[TMP53]] to i8*
// CHECK1-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[TMP55]], i8* align 8 [[TMP56]], i64 16, i1 false), !tbaa.struct !27
// CHECK1-NEXT:    br label [[IFCONT6:%.*]]
// CHECK1:       else5:
// CHECK1-NEXT:    br label [[IFCONT6]]
// CHECK1:       ifcont6:
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func6
// CHECK1-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR6]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCNT_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK1-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK1-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK1-NEXT:    [[NVPTX_LANE_ID:%.*]] = and i32 [[TMP4]], 31
// CHECK1-NEXT:    [[TMP5:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK1-NEXT:    [[NVPTX_WARP_ID:%.*]] = ashr i32 [[TMP5]], 5
// CHECK1-NEXT:    [[TMP6:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK1-NEXT:    [[TMP7:%.*]] = bitcast i8* [[TMP6]] to [1 x i8*]*
// CHECK1-NEXT:    store i32 0, i32* [[DOTCNT_ADDR]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    br label [[PRECOND:%.*]]
// CHECK1:       precond:
// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCNT_ADDR]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[TMP9:%.*]] = icmp ult i32 [[TMP8]], 4
// CHECK1-NEXT:    br i1 [[TMP9]], label [[BODY:%.*]], label [[EXIT:%.*]]
// CHECK1:       body:
// CHECK1-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]])
// CHECK1-NEXT:    [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0
// CHECK1-NEXT:    br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]]
// CHECK1:       then:
// CHECK1-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP7]], i64 0, i64 0
// CHECK1-NEXT:    [[TMP11:%.*]] = load i8*, i8** [[TMP10]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP12:%.*]] = bitcast i8* [[TMP11]] to i32*
// CHECK1-NEXT:    [[TMP13:%.*]] = getelementptr i32, i32* [[TMP12]], i32 [[TMP8]]
// CHECK1-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]]
// CHECK1-NEXT:    [[TMP15:%.*]] = load i32, i32* [[TMP13]], align 4
// CHECK1-NEXT:    store volatile i32 [[TMP15]], i32 addrspace(3)* [[TMP14]], align 4
// CHECK1-NEXT:    br label [[IFCONT:%.*]]
// CHECK1:       else:
// CHECK1-NEXT:    br label [[IFCONT]]
// CHECK1:       ifcont:
// CHECK1-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]])
// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTADDR1]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[TMP3]], [[TMP16]]
// CHECK1-NEXT:    br i1 [[IS_ACTIVE_THREAD]], label [[THEN2:%.*]], label [[ELSE3:%.*]]
// CHECK1:       then2:
// CHECK1-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[TMP3]]
// CHECK1-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP7]], i64 0, i64 0
// CHECK1-NEXT:    [[TMP19:%.*]] = load i8*, i8** [[TMP18]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP20:%.*]] = bitcast i8* [[TMP19]] to i32*
// CHECK1-NEXT:    [[TMP21:%.*]] = getelementptr i32, i32* [[TMP20]], i32 [[TMP8]]
// CHECK1-NEXT:    [[TMP22:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP17]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    store i32 [[TMP22]], i32* [[TMP21]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    br label [[IFCONT4:%.*]]
// CHECK1:       else3:
// CHECK1-NEXT:    br label [[IFCONT4]]
// CHECK1:       ifcont4:
// CHECK1-NEXT:    [[TMP23:%.*]] = add nsw i32 [[TMP8]], 1
// CHECK1-NEXT:    store i32 [[TMP23]], i32* [[DOTCNT_ADDR]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    br label [[PRECOND]]
// CHECK1:       exit:
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@__omp_outlined__3_wrapper
// CHECK1-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR6]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
// CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
// CHECK1-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2, !tbaa [[TBAA19]]
// CHECK1-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4, !tbaa [[TBAA8]]
// CHECK1-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK1-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
// CHECK1-NEXT:    [[TMP2:%.*]] = load i8**, i8*** [[GLOBAL_ARGS]], align 8
// CHECK1-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 0
// CHECK1-NEXT:    [[TMP4:%.*]] = bitcast i8** [[TMP3]] to i32**
// CHECK1-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[TMP4]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 1
// CHECK1-NEXT:    [[TMP7:%.*]] = bitcast i8** [[TMP6]] to i32**
// CHECK1-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[TMP7]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 2
// CHECK1-NEXT:    [[TMP10:%.*]] = bitcast i8** [[TMP9]] to %"class.std::complex.0"**
// CHECK1-NEXT:    [[TMP11:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[TMP10]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    call void @__omp_outlined__3(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]], i32* [[TMP5]], i32* [[TMP8]], %"class.std::complex.0"* [[TMP11]]) #[[ATTR5]]
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZNSt7complexIfEC2ERKfS2_
// CHECK1-SAME: (%"class.std::complex"* nonnull align 4 dereferenceable(8) [[THIS:%.*]], float* nonnull align 4 dereferenceable(4) [[__RE:%.*]], float* nonnull align 4 dereferenceable(4) [[__IM:%.*]]) unnamed_addr #[[ATTR3]] comdat align 2 {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex"*, align 8
// CHECK1-NEXT:    [[__RE_ADDR:%.*]] = alloca float*, align 8
// CHECK1-NEXT:    [[__IM_ADDR:%.*]] = alloca float*, align 8
// CHECK1-NEXT:    store %"class.std::complex"* [[THIS]], %"class.std::complex"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    store float* [[__RE]], float** [[__RE_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    store float* [[__IM]], float** [[__IM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[THIS1:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[THIS_ADDR]], align 8
// CHECK1-NEXT:    [[__RE_:%.*]] = getelementptr inbounds %"class.std::complex", %"class.std::complex"* [[THIS1]], i32 0, i32 0
// CHECK1-NEXT:    [[TMP0:%.*]] = load float*, float** [[__RE_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP1:%.*]] = load float, float* [[TMP0]], align 4, !tbaa [[TBAA14]]
// CHECK1-NEXT:    store float [[TMP1]], float* [[__RE_]], align 4, !tbaa [[TBAA16]]
// CHECK1-NEXT:    [[__IM_:%.*]] = getelementptr inbounds %"class.std::complex", %"class.std::complex"* [[THIS1]], i32 0, i32 1
// CHECK1-NEXT:    [[TMP2:%.*]] = load float*, float** [[__IM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP3:%.*]] = load float, float* [[TMP2]], align 4, !tbaa [[TBAA14]]
// CHECK1-NEXT:    store float [[TMP3]], float* [[__IM_]], align 4, !tbaa [[TBAA18]]
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZNKSt7complexIfE4realEv
// CHECK1-SAME: (%"class.std::complex"* nonnull align 4 dereferenceable(8) [[THIS:%.*]]) #[[ATTR4]] comdat align 2 {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex"*, align 8
// CHECK1-NEXT:    store %"class.std::complex"* [[THIS]], %"class.std::complex"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[THIS1:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[THIS_ADDR]], align 8
// CHECK1-NEXT:    [[__RE_:%.*]] = getelementptr inbounds %"class.std::complex", %"class.std::complex"* [[THIS1]], i32 0, i32 0
// CHECK1-NEXT:    [[TMP0:%.*]] = load float, float* [[__RE_]], align 4, !tbaa [[TBAA16]]
// CHECK1-NEXT:    ret float [[TMP0]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZNKSt7complexIfE4imagEv
// CHECK1-SAME: (%"class.std::complex"* nonnull align 4 dereferenceable(8) [[THIS:%.*]]) #[[ATTR4]] comdat align 2 {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex"*, align 8
// CHECK1-NEXT:    store %"class.std::complex"* [[THIS]], %"class.std::complex"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[THIS1:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[THIS_ADDR]], align 8
// CHECK1-NEXT:    [[__IM_:%.*]] = getelementptr inbounds %"class.std::complex", %"class.std::complex"* [[THIS1]], i32 0, i32 1
// CHECK1-NEXT:    [[TMP0:%.*]] = load float, float* [[__IM_]], align 4, !tbaa [[TBAA18]]
// CHECK1-NEXT:    ret float [[TMP0]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZNSt7complexIdEC2ERKdS2_
// CHECK1-SAME: (%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[THIS:%.*]], double* nonnull align 8 dereferenceable(8) [[__RE:%.*]], double* nonnull align 8 dereferenceable(8) [[__IM:%.*]]) unnamed_addr #[[ATTR3]] comdat align 2 {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex.0"*, align 8
// CHECK1-NEXT:    [[__RE_ADDR:%.*]] = alloca double*, align 8
// CHECK1-NEXT:    [[__IM_ADDR:%.*]] = alloca double*, align 8
// CHECK1-NEXT:    store %"class.std::complex.0"* [[THIS]], %"class.std::complex.0"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    store double* [[__RE]], double** [[__RE_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    store double* [[__IM]], double** [[__IM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[THIS1:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[THIS_ADDR]], align 8
// CHECK1-NEXT:    [[__RE_:%.*]] = getelementptr inbounds %"class.std::complex.0", %"class.std::complex.0"* [[THIS1]], i32 0, i32 0
// CHECK1-NEXT:    [[TMP0:%.*]] = load double*, double** [[__RE_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP1:%.*]] = load double, double* [[TMP0]], align 8, !tbaa [[TBAA22]]
// CHECK1-NEXT:    store double [[TMP1]], double* [[__RE_]], align 8, !tbaa [[TBAA24]]
// CHECK1-NEXT:    [[__IM_:%.*]] = getelementptr inbounds %"class.std::complex.0", %"class.std::complex.0"* [[THIS1]], i32 0, i32 1
// CHECK1-NEXT:    [[TMP2:%.*]] = load double*, double** [[__IM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[TMP3:%.*]] = load double, double* [[TMP2]], align 8, !tbaa [[TBAA22]]
// CHECK1-NEXT:    store double [[TMP3]], double* [[__IM_]], align 8, !tbaa [[TBAA26]]
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZNKSt7complexIdE4realEv
// CHECK1-SAME: (%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[THIS:%.*]]) #[[ATTR4]] comdat align 2 {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex.0"*, align 8
// CHECK1-NEXT:    store %"class.std::complex.0"* [[THIS]], %"class.std::complex.0"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[THIS1:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[THIS_ADDR]], align 8
// CHECK1-NEXT:    [[__RE_:%.*]] = getelementptr inbounds %"class.std::complex.0", %"class.std::complex.0"* [[THIS1]], i32 0, i32 0
// CHECK1-NEXT:    [[TMP0:%.*]] = load double, double* [[__RE_]], align 8, !tbaa [[TBAA24]]
// CHECK1-NEXT:    ret double [[TMP0]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZNKSt7complexIdE4imagEv
// CHECK1-SAME: (%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[THIS:%.*]]) #[[ATTR4]] comdat align 2 {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex.0"*, align 8
// CHECK1-NEXT:    store %"class.std::complex.0"* [[THIS]], %"class.std::complex.0"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK1-NEXT:    [[THIS1:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[THIS_ADDR]], align 8
// CHECK1-NEXT:    [[__IM_:%.*]] = getelementptr inbounds %"class.std::complex.0", %"class.std::complex.0"* [[THIS1]], i32 0, i32 1
// CHECK1-NEXT:    [[TMP0:%.*]] = load double, double* [[__IM_]], align 8, !tbaa [[TBAA26]]
// CHECK1-NEXT:    ret double [[TMP0]]
//
//
// CHECK2-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z17complex_reductionIfEvv_l16
// CHECK2-SAME: () #[[ATTR0:[0-9]+]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i8 1, i1 true, i1 true)
// CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
// CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK2:       user_code.entry:
// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK2-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK2-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4, !tbaa [[TBAA8:![0-9]+]]
// CHECK2-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR5:[0-9]+]]
// CHECK2-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 1, i1 true)
// CHECK2-NEXT:    ret void
// CHECK2:       worker.exit:
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@__omp_outlined__
// CHECK2-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[IB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[REF_TMP:%.*]] = alloca float, align 4
// CHECK2-NEXT:    [[REF_TMP2:%.*]] = alloca float, align 4
// CHECK2-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x i8*], align 8
// CHECK2-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8, !tbaa [[TBAA12:![0-9]+]]
// CHECK2-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[ISTART:%.*]] = call align 16 i8* @__kmpc_alloc_shared(i64 4)
// CHECK2-NEXT:    [[ISTART_ON_STACK:%.*]] = bitcast i8* [[ISTART]] to i32*
// CHECK2-NEXT:    [[IEND:%.*]] = call align 16 i8* @__kmpc_alloc_shared(i64 4)
// CHECK2-NEXT:    [[IEND_ON_STACK:%.*]] = bitcast i8* [[IEND]] to i32*
// CHECK2-NEXT:    [[PARTIAL_SUM:%.*]] = call align 16 i8* @__kmpc_alloc_shared(i64 8)
// CHECK2-NEXT:    [[PARTIAL_SUM_ON_STACK:%.*]] = bitcast i8* [[PARTIAL_SUM]] to %"class.std::complex"*
// CHECK2-NEXT:    [[TMP0:%.*]] = bitcast i32* [[DOTOMP_IV]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP0]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP1:%.*]] = bitcast i32* [[DOTOMP_LB]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP1]]) #[[ATTR5]]
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP2:%.*]] = bitcast i32* [[DOTOMP_UB]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP2]]) #[[ATTR5]]
// CHECK2-NEXT:    store i32 99, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP3:%.*]] = bitcast i32* [[DOTOMP_STRIDE]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP3]]) #[[ATTR5]]
// CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP4:%.*]] = bitcast i32* [[DOTOMP_IS_LAST]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP4]]) #[[ATTR5]]
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP5:%.*]] = bitcast i32* [[IB]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP5]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP6:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[TMP6]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP7]], i32 92, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP8]], 99
// CHECK2-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK2:       cond.true:
// CHECK2-NEXT:    br label [[COND_END:%.*]]
// CHECK2:       cond.false:
// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    br label [[COND_END]]
// CHECK2:       cond.end:
// CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP9]], [[COND_FALSE]] ]
// CHECK2-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    store i32 [[TMP10]], i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK2:       omp.inner.for.cond:
// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP11]], [[TMP12]]
// 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:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP13]], 1
// CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK2-NEXT:    store i32 [[ADD]], i32* [[IB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP14:%.*]] = bitcast float* [[REF_TMP]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP14]]) #[[ATTR5]]
// CHECK2-NEXT:    store float 0.000000e+00, float* [[REF_TMP]], align 4, !tbaa [[TBAA14:![0-9]+]]
// CHECK2-NEXT:    [[TMP15:%.*]] = bitcast float* [[REF_TMP2]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP15]]) #[[ATTR5]]
// CHECK2-NEXT:    store float 0.000000e+00, float* [[REF_TMP2]], align 4, !tbaa [[TBAA14]]
// CHECK2-NEXT:    call void @_ZNSt7complexIfEC1ERKfS2_(%"class.std::complex"* nonnull align 4 dereferenceable(8) [[PARTIAL_SUM_ON_STACK]], float* nonnull align 4 dereferenceable(4) [[REF_TMP]], float* nonnull align 4 dereferenceable(4) [[REF_TMP2]]) #[[ATTR10:[0-9]+]]
// CHECK2-NEXT:    [[TMP16:%.*]] = bitcast float* [[REF_TMP2]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP16]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP17:%.*]] = bitcast float* [[REF_TMP]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP17]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP18:%.*]] = load i32, i32* [[IB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[MUL3:%.*]] = mul nsw i32 [[TMP18]], 4
// CHECK2-NEXT:    store i32 [[MUL3]], i32* [[ISTART_ON_STACK]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP19:%.*]] = load i32, i32* [[IB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK2-NEXT:    [[MUL5:%.*]] = mul nsw i32 [[ADD4]], 4
// CHECK2-NEXT:    store i32 [[MUL5]], i32* [[IEND_ON_STACK]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK2-NEXT:    [[TMP21:%.*]] = bitcast i32* [[ISTART_ON_STACK]] to i8*
// CHECK2-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK2-NEXT:    [[TMP23:%.*]] = bitcast i32* [[IEND_ON_STACK]] to i8*
// CHECK2-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK2-NEXT:    [[TMP25:%.*]] = bitcast %"class.std::complex"* [[PARTIAL_SUM_ON_STACK]] to i8*
// CHECK2-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP26:%.*]] = bitcast [3 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK2-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP7]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32*, i32*, %"class.std::complex"*)* @__omp_outlined__1 to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined__1_wrapper to i8*), i8** [[TMP26]], i64 3)
// 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:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP27]], 1
// CHECK2-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK2:       omp.inner.for.end:
// CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK2:       omp.loop.exit:
// CHECK2-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP7]])
// CHECK2-NEXT:    [[TMP28:%.*]] = bitcast i32* [[IB]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP28]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP29:%.*]] = bitcast i32* [[DOTOMP_IS_LAST]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP29]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP30:%.*]] = bitcast i32* [[DOTOMP_STRIDE]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP30]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP31:%.*]] = bitcast i32* [[DOTOMP_UB]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP31]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP32:%.*]] = bitcast i32* [[DOTOMP_LB]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP32]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP33:%.*]] = bitcast i32* [[DOTOMP_IV]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP33]]) #[[ATTR5]]
// CHECK2-NEXT:    call void @__kmpc_free_shared(i8* [[PARTIAL_SUM]], i64 8)
// CHECK2-NEXT:    call void @__kmpc_free_shared(i8* [[IEND]], i64 4)
// CHECK2-NEXT:    call void @__kmpc_free_shared(i8* [[ISTART]], i64 4)
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZNSt7complexIfEC1ERKfS2_
// CHECK2-SAME: (%"class.std::complex"* nonnull align 4 dereferenceable(8) [[THIS:%.*]], float* nonnull align 4 dereferenceable(4) [[__RE:%.*]], float* nonnull align 4 dereferenceable(4) [[__IM:%.*]]) unnamed_addr #[[ATTR3:[0-9]+]] comdat align 2 {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex"*, align 8
// CHECK2-NEXT:    [[__RE_ADDR:%.*]] = alloca float*, align 8
// CHECK2-NEXT:    [[__IM_ADDR:%.*]] = alloca float*, align 8
// CHECK2-NEXT:    store %"class.std::complex"* [[THIS]], %"class.std::complex"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    store float* [[__RE]], float** [[__RE_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    store float* [[__IM]], float** [[__IM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[THIS1:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[THIS_ADDR]], align 8
// CHECK2-NEXT:    [[TMP0:%.*]] = load float*, float** [[__RE_ADDR]], align 8
// CHECK2-NEXT:    [[TMP1:%.*]] = load float*, float** [[__IM_ADDR]], align 8
// CHECK2-NEXT:    call void @_ZNSt7complexIfEC2ERKfS2_(%"class.std::complex"* nonnull align 4 dereferenceable(8) [[THIS1]], float* nonnull align 4 dereferenceable(4) [[TMP0]], float* nonnull align 4 dereferenceable(4) [[TMP1]]) #[[ATTR10]]
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@__omp_outlined__1
// CHECK2-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32* nonnull align 4 dereferenceable(4) [[ISTART:%.*]], i32* nonnull align 4 dereferenceable(4) [[IEND:%.*]], %"class.std::complex"* nonnull align 4 dereferenceable(8) [[PARTIAL_SUM:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[ISTART_ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[IEND_ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[PARTIAL_SUM_ADDR:%.*]] = alloca %"class.std::complex"*, align 8
// CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[PARTIAL_SUM5:%.*]] = alloca %"class.std::complex", align 4
// CHECK2-NEXT:    [[REF_TMP:%.*]] = alloca float, align 4
// CHECK2-NEXT:    [[REF_TMP6:%.*]] = alloca float, align 4
// CHECK2-NEXT:    [[I7:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[REF_TMP14:%.*]] = alloca %"class.std::complex", align 4
// CHECK2-NEXT:    [[REF_TMP15:%.*]] = alloca float, align 4
// CHECK2-NEXT:    [[REF_TMP16:%.*]] = alloca float, align 4
// CHECK2-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x i8*], align 8
// CHECK2-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    store i32* [[ISTART]], i32** [[ISTART_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    store i32* [[IEND]], i32** [[IEND_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    store %"class.std::complex"* [[PARTIAL_SUM]], %"class.std::complex"** [[PARTIAL_SUM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[ISTART_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[IEND_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP2:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[PARTIAL_SUM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP3:%.*]] = bitcast i32* [[DOTOMP_IV]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP3]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP4:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP4]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, i32* [[TMP0]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    store i32 [[TMP5]], i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP6:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_1]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP6]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[TMP1]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    store i32 [[TMP7]], i32* [[DOTCAPTURE_EXPR_1]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP8:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_2]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP8]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[SUB:%.*]] = sub i32 [[TMP9]], [[TMP10]]
// CHECK2-NEXT:    [[SUB3:%.*]] = sub i32 [[SUB]], 1
// CHECK2-NEXT:    [[ADD:%.*]] = add i32 [[SUB3]], 1
// CHECK2-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], 1
// CHECK2-NEXT:    [[SUB4:%.*]] = sub i32 [[DIV]], 1
// CHECK2-NEXT:    store i32 [[SUB4]], i32* [[DOTCAPTURE_EXPR_2]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP11:%.*]] = bitcast i32* [[I]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP11]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    store i32 [[TMP12]], i32* [[I]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP13:%.*]] = bitcast i32* [[I]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP13]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP14]], [[TMP15]]
// CHECK2-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK2:       omp.precond.then:
// CHECK2-NEXT:    [[TMP16:%.*]] = bitcast i32* [[DOTOMP_LB]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP16]]) #[[ATTR5]]
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP17:%.*]] = bitcast i32* [[DOTOMP_UB]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP17]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    store i32 [[TMP18]], i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP19:%.*]] = bitcast i32* [[DOTOMP_STRIDE]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP19]]) #[[ATTR5]]
// CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP20:%.*]] = bitcast i32* [[DOTOMP_IS_LAST]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP20]]) #[[ATTR5]]
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP21:%.*]] = bitcast %"class.std::complex"* [[PARTIAL_SUM5]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 8, i8* [[TMP21]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP22:%.*]] = bitcast float* [[REF_TMP]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP22]]) #[[ATTR5]]
// CHECK2-NEXT:    store float 0.000000e+00, float* [[REF_TMP]], align 4, !tbaa [[TBAA14]]
// CHECK2-NEXT:    [[TMP23:%.*]] = bitcast float* [[REF_TMP6]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP23]]) #[[ATTR5]]
// CHECK2-NEXT:    store float 0.000000e+00, float* [[REF_TMP6]], align 4, !tbaa [[TBAA14]]
// CHECK2-NEXT:    call void @_ZNSt7complexIfEC1ERKfS2_(%"class.std::complex"* nonnull align 4 dereferenceable(8) [[PARTIAL_SUM5]], float* nonnull align 4 dereferenceable(4) [[REF_TMP]], float* nonnull align 4 dereferenceable(4) [[REF_TMP6]]) #[[ATTR10]]
// CHECK2-NEXT:    [[TMP24:%.*]] = bitcast float* [[REF_TMP6]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP24]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP25:%.*]] = bitcast float* [[REF_TMP]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP25]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP26:%.*]] = bitcast i32* [[I7]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP26]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP27:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP28:%.*]] = load i32, i32* [[TMP27]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    call void @__kmpc_for_static_init_4u(%struct.ident_t* @[[GLOB3:[0-9]+]], i32 [[TMP28]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK2-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK2:       omp.dispatch.cond:
// CHECK2-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[CMP8:%.*]] = icmp ugt i32 [[TMP29]], [[TMP30]]
// CHECK2-NEXT:    br i1 [[CMP8]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK2:       cond.true:
// CHECK2-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    br label [[COND_END:%.*]]
// CHECK2:       cond.false:
// CHECK2-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    br label [[COND_END]]
// CHECK2:       cond.end:
// CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP31]], [[COND_TRUE]] ], [ [[TMP32]], [[COND_FALSE]] ]
// CHECK2-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    store i32 [[TMP33]], i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[ADD9:%.*]] = add i32 [[TMP35]], 1
// CHECK2-NEXT:    [[CMP10:%.*]] = icmp ult i32 [[TMP34]], [[ADD9]]
// CHECK2-NEXT:    br i1 [[CMP10]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_CLEANUP:%.*]]
// CHECK2:       omp.dispatch.cleanup:
// CHECK2-NEXT:    br label [[OMP_DISPATCH_END:%.*]]
// CHECK2:       omp.dispatch.body:
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK2:       omp.inner.for.cond:
// CHECK2-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[ADD11:%.*]] = add i32 [[TMP37]], 1
// CHECK2-NEXT:    [[CMP12:%.*]] = icmp ult i32 [[TMP36]], [[ADD11]]
// CHECK2-NEXT:    br i1 [[CMP12]], 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:    [[TMP38:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[MUL:%.*]] = mul i32 [[TMP39]], 1
// CHECK2-NEXT:    [[ADD13:%.*]] = add i32 [[TMP38]], [[MUL]]
// CHECK2-NEXT:    store i32 [[ADD13]], i32* [[I7]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP40:%.*]] = bitcast %"class.std::complex"* [[REF_TMP14]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 8, i8* [[TMP40]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP41:%.*]] = bitcast float* [[REF_TMP15]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP41]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP42:%.*]] = load i32, i32* [[I7]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[CONV:%.*]] = sitofp i32 [[TMP42]] to float
// CHECK2-NEXT:    store float [[CONV]], float* [[REF_TMP15]], align 4, !tbaa [[TBAA14]]
// CHECK2-NEXT:    [[TMP43:%.*]] = bitcast float* [[REF_TMP16]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP43]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP44:%.*]] = load i32, i32* [[I7]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[CONV17:%.*]] = sitofp i32 [[TMP44]] to float
// CHECK2-NEXT:    store float [[CONV17]], float* [[REF_TMP16]], align 4, !tbaa [[TBAA14]]
// CHECK2-NEXT:    call void @_ZNSt7complexIfEC1ERKfS2_(%"class.std::complex"* nonnull align 4 dereferenceable(8) [[REF_TMP14]], float* nonnull align 4 dereferenceable(4) [[REF_TMP15]], float* nonnull align 4 dereferenceable(4) [[REF_TMP16]]) #[[ATTR10]]
// CHECK2-NEXT:    [[CALL:%.*]] = call nonnull align 4 dereferenceable(8) %"class.std::complex"* @_ZNSt7complexIfEpLIfEERS0_RKS_IT_E(%"class.std::complex"* nonnull align 4 dereferenceable(8) [[PARTIAL_SUM5]], %"class.std::complex"* nonnull align 4 dereferenceable(8) [[REF_TMP14]]) #[[ATTR10]]
// CHECK2-NEXT:    [[TMP45:%.*]] = bitcast float* [[REF_TMP16]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP45]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP46:%.*]] = bitcast float* [[REF_TMP15]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP46]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP47:%.*]] = bitcast %"class.std::complex"* [[REF_TMP14]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 8, i8* [[TMP47]]) #[[ATTR5]]
// 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:    [[TMP48:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[ADD18:%.*]] = add i32 [[TMP48]], 1
// CHECK2-NEXT:    store i32 [[ADD18]], i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK2:       omp.inner.for.end:
// CHECK2-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK2:       omp.dispatch.inc:
// CHECK2-NEXT:    [[TMP49:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP50:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[ADD19:%.*]] = add i32 [[TMP49]], [[TMP50]]
// CHECK2-NEXT:    store i32 [[ADD19]], i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP51:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP52:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[ADD20:%.*]] = add i32 [[TMP51]], [[TMP52]]
// CHECK2-NEXT:    store i32 [[ADD20]], i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK2:       omp.dispatch.end:
// CHECK2-NEXT:    [[TMP53:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP54:%.*]] = load i32, i32* [[TMP53]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB3]], i32 [[TMP54]])
// CHECK2-NEXT:    [[TMP55:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP56:%.*]] = load i32, i32* [[TMP55]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP57:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0
// CHECK2-NEXT:    [[TMP58:%.*]] = bitcast %"class.std::complex"* [[PARTIAL_SUM5]] to i8*
// CHECK2-NEXT:    store i8* [[TMP58]], i8** [[TMP57]], align 8
// CHECK2-NEXT:    [[TMP59:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8*
// CHECK2-NEXT:    [[TMP60:%.*]] = call i32 @__kmpc_nvptx_parallel_reduce_nowait_v2(%struct.ident_t* @[[GLOB1]], i32 [[TMP56]], i32 1, i64 8, i8* [[TMP59]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func)
// CHECK2-NEXT:    [[TMP61:%.*]] = icmp eq i32 [[TMP60]], 1
// CHECK2-NEXT:    br i1 [[TMP61]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]]
// CHECK2:       .omp.reduction.then:
// CHECK2-NEXT:    [[CALL21:%.*]] = call nonnull align 4 dereferenceable(8) %"class.std::complex"* @_ZNSt7complexIfEpLIfEERS0_RKS_IT_E(%"class.std::complex"* nonnull align 4 dereferenceable(8) [[TMP2]], %"class.std::complex"* nonnull align 4 dereferenceable(8) [[PARTIAL_SUM5]]) #[[ATTR10]]
// CHECK2-NEXT:    call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP56]])
// CHECK2-NEXT:    br label [[DOTOMP_REDUCTION_DONE]]
// CHECK2:       .omp.reduction.done:
// CHECK2-NEXT:    [[TMP62:%.*]] = bitcast i32* [[I7]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP62]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP63:%.*]] = bitcast %"class.std::complex"* [[PARTIAL_SUM5]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 8, i8* [[TMP63]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP64:%.*]] = bitcast i32* [[DOTOMP_IS_LAST]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP64]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP65:%.*]] = bitcast i32* [[DOTOMP_STRIDE]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP65]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP66:%.*]] = bitcast i32* [[DOTOMP_UB]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP66]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP67:%.*]] = bitcast i32* [[DOTOMP_LB]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP67]]) #[[ATTR5]]
// CHECK2-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK2:       omp.precond.end:
// CHECK2-NEXT:    [[TMP68:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_2]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP68]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP69:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_1]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP69]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP70:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP70]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP71:%.*]] = bitcast i32* [[DOTOMP_IV]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP71]]) #[[ATTR5]]
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZNSt7complexIfEpLIfEERS0_RKS_IT_E
// CHECK2-SAME: (%"class.std::complex"* nonnull align 4 dereferenceable(8) [[THIS:%.*]], %"class.std::complex"* nonnull align 4 dereferenceable(8) [[__C:%.*]]) #[[ATTR4:[0-9]+]] comdat align 2 {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex"*, align 8
// CHECK2-NEXT:    [[__C_ADDR:%.*]] = alloca %"class.std::complex"*, align 8
// CHECK2-NEXT:    store %"class.std::complex"* [[THIS]], %"class.std::complex"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    store %"class.std::complex"* [[__C]], %"class.std::complex"** [[__C_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[THIS1:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[THIS_ADDR]], align 8
// CHECK2-NEXT:    [[TMP0:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[__C_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[CALL:%.*]] = call float @_ZNKSt7complexIfE4realEv(%"class.std::complex"* nonnull align 4 dereferenceable(8) [[TMP0]]) #[[ATTR10]]
// CHECK2-NEXT:    [[__RE_:%.*]] = getelementptr inbounds %"class.std::complex", %"class.std::complex"* [[THIS1]], i32 0, i32 0
// CHECK2-NEXT:    [[TMP1:%.*]] = load float, float* [[__RE_]], align 4, !tbaa [[TBAA16:![0-9]+]]
// CHECK2-NEXT:    [[ADD:%.*]] = fadd float [[TMP1]], [[CALL]]
// CHECK2-NEXT:    store float [[ADD]], float* [[__RE_]], align 4, !tbaa [[TBAA16]]
// CHECK2-NEXT:    [[TMP2:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[__C_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[CALL2:%.*]] = call float @_ZNKSt7complexIfE4imagEv(%"class.std::complex"* nonnull align 4 dereferenceable(8) [[TMP2]]) #[[ATTR10]]
// CHECK2-NEXT:    [[__IM_:%.*]] = getelementptr inbounds %"class.std::complex", %"class.std::complex"* [[THIS1]], i32 0, i32 1
// CHECK2-NEXT:    [[TMP3:%.*]] = load float, float* [[__IM_]], align 4, !tbaa [[TBAA18:![0-9]+]]
// CHECK2-NEXT:    [[ADD3:%.*]] = fadd float [[TMP3]], [[CALL2]]
// CHECK2-NEXT:    store float [[ADD3]], float* [[__IM_]], align 4, !tbaa [[TBAA18]]
// CHECK2-NEXT:    ret %"class.std::complex"* [[THIS1]]
//
//
// CHECK2-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func
// CHECK2-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR6:[0-9]+]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK2-NEXT:    [[DOTADDR1:%.*]] = alloca i16, align 2
// CHECK2-NEXT:    [[DOTADDR2:%.*]] = alloca i16, align 2
// CHECK2-NEXT:    [[DOTADDR3:%.*]] = alloca i16, align 2
// CHECK2-NEXT:    [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [1 x i8*], align 8
// CHECK2-NEXT:    [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca %"class.std::complex", align 4
// CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    store i16 [[TMP1]], i16* [[DOTADDR1]], align 2, !tbaa [[TBAA19:![0-9]+]]
// CHECK2-NEXT:    store i16 [[TMP2]], i16* [[DOTADDR2]], align 2, !tbaa [[TBAA19]]
// CHECK2-NEXT:    store i16 [[TMP3]], i16* [[DOTADDR3]], align 2, !tbaa [[TBAA19]]
// CHECK2-NEXT:    [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [1 x i8*]*
// CHECK2-NEXT:    [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2, !tbaa [[TBAA19]]
// CHECK2-NEXT:    [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2, !tbaa [[TBAA19]]
// CHECK2-NEXT:    [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2, !tbaa [[TBAA19]]
// CHECK2-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK2-NEXT:    [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8
// CHECK2-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0
// CHECK2-NEXT:    [[TMP12:%.*]] = bitcast i8* [[TMP10]] to %"class.std::complex"*
// CHECK2-NEXT:    [[TMP13:%.*]] = getelementptr %"class.std::complex", %"class.std::complex"* [[TMP12]], i64 1
// CHECK2-NEXT:    [[TMP14:%.*]] = bitcast %"class.std::complex"* [[TMP13]] to i8*
// CHECK2-NEXT:    [[TMP15:%.*]] = bitcast %"class.std::complex"* [[TMP12]] to i64*
// CHECK2-NEXT:    [[TMP16:%.*]] = bitcast %"class.std::complex"* [[DOTOMP_REDUCTION_ELEMENT]] to i64*
// CHECK2-NEXT:    [[TMP17:%.*]] = load i64, i64* [[TMP15]], align 4
// CHECK2-NEXT:    [[TMP18:%.*]] = call i32 @__kmpc_get_warp_size()
// CHECK2-NEXT:    [[TMP19:%.*]] = trunc i32 [[TMP18]] to i16
// CHECK2-NEXT:    [[TMP20:%.*]] = call i64 @__kmpc_shuffle_int64(i64 [[TMP17]], i16 [[TMP7]], i16 [[TMP19]])
// CHECK2-NEXT:    store i64 [[TMP20]], i64* [[TMP16]], align 4
// CHECK2-NEXT:    [[TMP21:%.*]] = getelementptr i64, i64* [[TMP15]], i64 1
// CHECK2-NEXT:    [[TMP22:%.*]] = getelementptr i64, i64* [[TMP16]], i64 1
// CHECK2-NEXT:    [[TMP23:%.*]] = bitcast %"class.std::complex"* [[DOTOMP_REDUCTION_ELEMENT]] to i8*
// CHECK2-NEXT:    store i8* [[TMP23]], i8** [[TMP11]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP24:%.*]] = icmp eq i16 [[TMP8]], 0
// CHECK2-NEXT:    [[TMP25:%.*]] = icmp eq i16 [[TMP8]], 1
// CHECK2-NEXT:    [[TMP26:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]]
// CHECK2-NEXT:    [[TMP27:%.*]] = and i1 [[TMP25]], [[TMP26]]
// CHECK2-NEXT:    [[TMP28:%.*]] = icmp eq i16 [[TMP8]], 2
// CHECK2-NEXT:    [[TMP29:%.*]] = and i16 [[TMP6]], 1
// CHECK2-NEXT:    [[TMP30:%.*]] = icmp eq i16 [[TMP29]], 0
// CHECK2-NEXT:    [[TMP31:%.*]] = and i1 [[TMP28]], [[TMP30]]
// CHECK2-NEXT:    [[TMP32:%.*]] = icmp sgt i16 [[TMP7]], 0
// CHECK2-NEXT:    [[TMP33:%.*]] = and i1 [[TMP31]], [[TMP32]]
// CHECK2-NEXT:    [[TMP34:%.*]] = or i1 [[TMP24]], [[TMP27]]
// CHECK2-NEXT:    [[TMP35:%.*]] = or i1 [[TMP34]], [[TMP33]]
// CHECK2-NEXT:    br i1 [[TMP35]], label [[THEN:%.*]], label [[ELSE:%.*]]
// CHECK2:       then:
// CHECK2-NEXT:    [[TMP36:%.*]] = bitcast [1 x i8*]* [[TMP5]] to i8*
// CHECK2-NEXT:    [[TMP37:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8*
// CHECK2-NEXT:    call void @"_omp$reduction$reduction_func"(i8* [[TMP36]], i8* [[TMP37]]) #[[ATTR5]]
// CHECK2-NEXT:    br label [[IFCONT:%.*]]
// CHECK2:       else:
// CHECK2-NEXT:    br label [[IFCONT]]
// CHECK2:       ifcont:
// CHECK2-NEXT:    [[TMP38:%.*]] = icmp eq i16 [[TMP8]], 1
// CHECK2-NEXT:    [[TMP39:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]]
// CHECK2-NEXT:    [[TMP40:%.*]] = and i1 [[TMP38]], [[TMP39]]
// CHECK2-NEXT:    br i1 [[TMP40]], label [[THEN4:%.*]], label [[ELSE5:%.*]]
// CHECK2:       then4:
// CHECK2-NEXT:    [[TMP41:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0
// CHECK2-NEXT:    [[TMP42:%.*]] = load i8*, i8** [[TMP41]], align 8
// CHECK2-NEXT:    [[TMP43:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK2-NEXT:    [[TMP44:%.*]] = load i8*, i8** [[TMP43]], align 8
// CHECK2-NEXT:    [[TMP45:%.*]] = bitcast i8* [[TMP42]] to %"class.std::complex"*
// CHECK2-NEXT:    [[TMP46:%.*]] = bitcast i8* [[TMP44]] to %"class.std::complex"*
// CHECK2-NEXT:    [[TMP47:%.*]] = bitcast %"class.std::complex"* [[TMP46]] to i8*
// CHECK2-NEXT:    [[TMP48:%.*]] = bitcast %"class.std::complex"* [[TMP45]] to i8*
// CHECK2-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP47]], i8* align 4 [[TMP48]], i64 8, i1 false), !tbaa.struct !21
// CHECK2-NEXT:    br label [[IFCONT6:%.*]]
// CHECK2:       else5:
// CHECK2-NEXT:    br label [[IFCONT6]]
// CHECK2:       ifcont6:
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func
// CHECK2-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR6]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK2-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCNT_ADDR:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK2-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK2-NEXT:    [[NVPTX_LANE_ID:%.*]] = and i32 [[TMP4]], 31
// CHECK2-NEXT:    [[TMP5:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK2-NEXT:    [[NVPTX_WARP_ID:%.*]] = ashr i32 [[TMP5]], 5
// CHECK2-NEXT:    [[TMP6:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK2-NEXT:    [[TMP7:%.*]] = bitcast i8* [[TMP6]] to [1 x i8*]*
// CHECK2-NEXT:    store i32 0, i32* [[DOTCNT_ADDR]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    br label [[PRECOND:%.*]]
// CHECK2:       precond:
// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCNT_ADDR]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP9:%.*]] = icmp ult i32 [[TMP8]], 2
// CHECK2-NEXT:    br i1 [[TMP9]], label [[BODY:%.*]], label [[EXIT:%.*]]
// CHECK2:       body:
// CHECK2-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4:[0-9]+]], i32 [[TMP2]])
// CHECK2-NEXT:    [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0
// CHECK2-NEXT:    br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]]
// CHECK2:       then:
// CHECK2-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP7]], i64 0, i64 0
// CHECK2-NEXT:    [[TMP11:%.*]] = load i8*, i8** [[TMP10]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP12:%.*]] = bitcast i8* [[TMP11]] to i32*
// CHECK2-NEXT:    [[TMP13:%.*]] = getelementptr i32, i32* [[TMP12]], i32 [[TMP8]]
// CHECK2-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]]
// CHECK2-NEXT:    [[TMP15:%.*]] = load i32, i32* [[TMP13]], align 4
// CHECK2-NEXT:    store volatile i32 [[TMP15]], i32 addrspace(3)* [[TMP14]], align 4
// CHECK2-NEXT:    br label [[IFCONT:%.*]]
// CHECK2:       else:
// CHECK2-NEXT:    br label [[IFCONT]]
// CHECK2:       ifcont:
// CHECK2-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]])
// CHECK2-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTADDR1]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[TMP3]], [[TMP16]]
// CHECK2-NEXT:    br i1 [[IS_ACTIVE_THREAD]], label [[THEN2:%.*]], label [[ELSE3:%.*]]
// CHECK2:       then2:
// CHECK2-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[TMP3]]
// CHECK2-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP7]], i64 0, i64 0
// CHECK2-NEXT:    [[TMP19:%.*]] = load i8*, i8** [[TMP18]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP20:%.*]] = bitcast i8* [[TMP19]] to i32*
// CHECK2-NEXT:    [[TMP21:%.*]] = getelementptr i32, i32* [[TMP20]], i32 [[TMP8]]
// CHECK2-NEXT:    [[TMP22:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP17]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    store i32 [[TMP22]], i32* [[TMP21]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    br label [[IFCONT4:%.*]]
// CHECK2:       else3:
// CHECK2-NEXT:    br label [[IFCONT4]]
// CHECK2:       ifcont4:
// CHECK2-NEXT:    [[TMP23:%.*]] = add nsw i32 [[TMP8]], 1
// CHECK2-NEXT:    store i32 [[TMP23]], i32* [[DOTCNT_ADDR]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    br label [[PRECOND]]
// CHECK2:       exit:
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@__omp_outlined__1_wrapper
// CHECK2-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR6]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
// CHECK2-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
// CHECK2-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2, !tbaa [[TBAA19]]
// CHECK2-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK2-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
// CHECK2-NEXT:    [[TMP2:%.*]] = load i8**, i8*** [[GLOBAL_ARGS]], align 8
// CHECK2-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 0
// CHECK2-NEXT:    [[TMP4:%.*]] = bitcast i8** [[TMP3]] to i32**
// CHECK2-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[TMP4]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 1
// CHECK2-NEXT:    [[TMP7:%.*]] = bitcast i8** [[TMP6]] to i32**
// CHECK2-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[TMP7]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 2
// CHECK2-NEXT:    [[TMP10:%.*]] = bitcast i8** [[TMP9]] to %"class.std::complex"**
// CHECK2-NEXT:    [[TMP11:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[TMP10]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    call void @__omp_outlined__1(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]], i32* [[TMP5]], i32* [[TMP8]], %"class.std::complex"* [[TMP11]]) #[[ATTR5]]
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z17complex_reductionIdEvv_l16
// CHECK2-SAME: () #[[ATTR0]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 1, i1 true, i1 true)
// CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
// CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK2:       user_code.entry:
// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK2-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK2-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    call void @__omp_outlined__2(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR5]]
// CHECK2-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 1, i1 true)
// CHECK2-NEXT:    ret void
// CHECK2:       worker.exit:
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@__omp_outlined__2
// CHECK2-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[IB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[REF_TMP:%.*]] = alloca double, align 8
// CHECK2-NEXT:    [[REF_TMP2:%.*]] = alloca double, align 8
// CHECK2-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x i8*], align 8
// CHECK2-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[ISTART:%.*]] = call align 16 i8* @__kmpc_alloc_shared(i64 4)
// CHECK2-NEXT:    [[ISTART_ON_STACK:%.*]] = bitcast i8* [[ISTART]] to i32*
// CHECK2-NEXT:    [[IEND:%.*]] = call align 16 i8* @__kmpc_alloc_shared(i64 4)
// CHECK2-NEXT:    [[IEND_ON_STACK:%.*]] = bitcast i8* [[IEND]] to i32*
// CHECK2-NEXT:    [[PARTIAL_SUM:%.*]] = call align 16 i8* @__kmpc_alloc_shared(i64 16)
// CHECK2-NEXT:    [[PARTIAL_SUM_ON_STACK:%.*]] = bitcast i8* [[PARTIAL_SUM]] to %"class.std::complex.0"*
// CHECK2-NEXT:    [[TMP0:%.*]] = bitcast i32* [[DOTOMP_IV]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP0]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP1:%.*]] = bitcast i32* [[DOTOMP_LB]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP1]]) #[[ATTR5]]
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP2:%.*]] = bitcast i32* [[DOTOMP_UB]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP2]]) #[[ATTR5]]
// CHECK2-NEXT:    store i32 99, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP3:%.*]] = bitcast i32* [[DOTOMP_STRIDE]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP3]]) #[[ATTR5]]
// CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP4:%.*]] = bitcast i32* [[DOTOMP_IS_LAST]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP4]]) #[[ATTR5]]
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP5:%.*]] = bitcast i32* [[IB]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP5]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP6:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[TMP6]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP7]], i32 92, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP8]], 99
// CHECK2-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK2:       cond.true:
// CHECK2-NEXT:    br label [[COND_END:%.*]]
// CHECK2:       cond.false:
// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    br label [[COND_END]]
// CHECK2:       cond.end:
// CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP9]], [[COND_FALSE]] ]
// CHECK2-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    store i32 [[TMP10]], i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK2:       omp.inner.for.cond:
// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP11]], [[TMP12]]
// 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:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP13]], 1
// CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK2-NEXT:    store i32 [[ADD]], i32* [[IB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP14:%.*]] = bitcast double* [[REF_TMP]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 8, i8* [[TMP14]]) #[[ATTR5]]
// CHECK2-NEXT:    store double 0.000000e+00, double* [[REF_TMP]], align 8, !tbaa [[TBAA22:![0-9]+]]
// CHECK2-NEXT:    [[TMP15:%.*]] = bitcast double* [[REF_TMP2]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 8, i8* [[TMP15]]) #[[ATTR5]]
// CHECK2-NEXT:    store double 0.000000e+00, double* [[REF_TMP2]], align 8, !tbaa [[TBAA22]]
// CHECK2-NEXT:    call void @_ZNSt7complexIdEC1ERKdS2_(%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[PARTIAL_SUM_ON_STACK]], double* nonnull align 8 dereferenceable(8) [[REF_TMP]], double* nonnull align 8 dereferenceable(8) [[REF_TMP2]]) #[[ATTR10]]
// CHECK2-NEXT:    [[TMP16:%.*]] = bitcast double* [[REF_TMP2]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 8, i8* [[TMP16]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP17:%.*]] = bitcast double* [[REF_TMP]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 8, i8* [[TMP17]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP18:%.*]] = load i32, i32* [[IB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[MUL3:%.*]] = mul nsw i32 [[TMP18]], 4
// CHECK2-NEXT:    store i32 [[MUL3]], i32* [[ISTART_ON_STACK]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP19:%.*]] = load i32, i32* [[IB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK2-NEXT:    [[MUL5:%.*]] = mul nsw i32 [[ADD4]], 4
// CHECK2-NEXT:    store i32 [[MUL5]], i32* [[IEND_ON_STACK]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK2-NEXT:    [[TMP21:%.*]] = bitcast i32* [[ISTART_ON_STACK]] to i8*
// CHECK2-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK2-NEXT:    [[TMP23:%.*]] = bitcast i32* [[IEND_ON_STACK]] to i8*
// CHECK2-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK2-NEXT:    [[TMP25:%.*]] = bitcast %"class.std::complex.0"* [[PARTIAL_SUM_ON_STACK]] to i8*
// CHECK2-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP26:%.*]] = bitcast [3 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK2-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP7]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32*, i32*, %"class.std::complex.0"*)* @__omp_outlined__3 to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined__3_wrapper to i8*), i8** [[TMP26]], i64 3)
// 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:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP27]], 1
// CHECK2-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK2:       omp.inner.for.end:
// CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK2:       omp.loop.exit:
// CHECK2-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP7]])
// CHECK2-NEXT:    [[TMP28:%.*]] = bitcast i32* [[IB]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP28]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP29:%.*]] = bitcast i32* [[DOTOMP_IS_LAST]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP29]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP30:%.*]] = bitcast i32* [[DOTOMP_STRIDE]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP30]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP31:%.*]] = bitcast i32* [[DOTOMP_UB]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP31]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP32:%.*]] = bitcast i32* [[DOTOMP_LB]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP32]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP33:%.*]] = bitcast i32* [[DOTOMP_IV]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP33]]) #[[ATTR5]]
// CHECK2-NEXT:    call void @__kmpc_free_shared(i8* [[PARTIAL_SUM]], i64 16)
// CHECK2-NEXT:    call void @__kmpc_free_shared(i8* [[IEND]], i64 4)
// CHECK2-NEXT:    call void @__kmpc_free_shared(i8* [[ISTART]], i64 4)
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZNSt7complexIdEC1ERKdS2_
// CHECK2-SAME: (%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[THIS:%.*]], double* nonnull align 8 dereferenceable(8) [[__RE:%.*]], double* nonnull align 8 dereferenceable(8) [[__IM:%.*]]) unnamed_addr #[[ATTR3]] comdat align 2 {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex.0"*, align 8
// CHECK2-NEXT:    [[__RE_ADDR:%.*]] = alloca double*, align 8
// CHECK2-NEXT:    [[__IM_ADDR:%.*]] = alloca double*, align 8
// CHECK2-NEXT:    store %"class.std::complex.0"* [[THIS]], %"class.std::complex.0"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    store double* [[__RE]], double** [[__RE_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    store double* [[__IM]], double** [[__IM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[THIS1:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[THIS_ADDR]], align 8
// CHECK2-NEXT:    [[TMP0:%.*]] = load double*, double** [[__RE_ADDR]], align 8
// CHECK2-NEXT:    [[TMP1:%.*]] = load double*, double** [[__IM_ADDR]], align 8
// CHECK2-NEXT:    call void @_ZNSt7complexIdEC2ERKdS2_(%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[THIS1]], double* nonnull align 8 dereferenceable(8) [[TMP0]], double* nonnull align 8 dereferenceable(8) [[TMP1]]) #[[ATTR10]]
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@__omp_outlined__3
// CHECK2-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32* nonnull align 4 dereferenceable(4) [[ISTART:%.*]], i32* nonnull align 4 dereferenceable(4) [[IEND:%.*]], %"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[PARTIAL_SUM:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[ISTART_ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[IEND_ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[PARTIAL_SUM_ADDR:%.*]] = alloca %"class.std::complex.0"*, align 8
// CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[PARTIAL_SUM5:%.*]] = alloca %"class.std::complex.0", align 8
// CHECK2-NEXT:    [[REF_TMP:%.*]] = alloca double, align 8
// CHECK2-NEXT:    [[REF_TMP6:%.*]] = alloca double, align 8
// CHECK2-NEXT:    [[I7:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[REF_TMP14:%.*]] = alloca %"class.std::complex.0", align 8
// CHECK2-NEXT:    [[REF_TMP15:%.*]] = alloca double, align 8
// CHECK2-NEXT:    [[REF_TMP16:%.*]] = alloca double, align 8
// CHECK2-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x i8*], align 8
// CHECK2-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    store i32* [[ISTART]], i32** [[ISTART_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    store i32* [[IEND]], i32** [[IEND_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    store %"class.std::complex.0"* [[PARTIAL_SUM]], %"class.std::complex.0"** [[PARTIAL_SUM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[ISTART_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[IEND_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP2:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[PARTIAL_SUM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP3:%.*]] = bitcast i32* [[DOTOMP_IV]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP3]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP4:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP4]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, i32* [[TMP0]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    store i32 [[TMP5]], i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP6:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_1]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP6]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[TMP1]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    store i32 [[TMP7]], i32* [[DOTCAPTURE_EXPR_1]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP8:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_2]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP8]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[SUB:%.*]] = sub i32 [[TMP9]], [[TMP10]]
// CHECK2-NEXT:    [[SUB3:%.*]] = sub i32 [[SUB]], 1
// CHECK2-NEXT:    [[ADD:%.*]] = add i32 [[SUB3]], 1
// CHECK2-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], 1
// CHECK2-NEXT:    [[SUB4:%.*]] = sub i32 [[DIV]], 1
// CHECK2-NEXT:    store i32 [[SUB4]], i32* [[DOTCAPTURE_EXPR_2]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP11:%.*]] = bitcast i32* [[I]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP11]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    store i32 [[TMP12]], i32* [[I]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP13:%.*]] = bitcast i32* [[I]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP13]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP14]], [[TMP15]]
// CHECK2-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK2:       omp.precond.then:
// CHECK2-NEXT:    [[TMP16:%.*]] = bitcast i32* [[DOTOMP_LB]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP16]]) #[[ATTR5]]
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP17:%.*]] = bitcast i32* [[DOTOMP_UB]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP17]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    store i32 [[TMP18]], i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP19:%.*]] = bitcast i32* [[DOTOMP_STRIDE]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP19]]) #[[ATTR5]]
// CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP20:%.*]] = bitcast i32* [[DOTOMP_IS_LAST]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP20]]) #[[ATTR5]]
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP21:%.*]] = bitcast %"class.std::complex.0"* [[PARTIAL_SUM5]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 16, i8* [[TMP21]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP22:%.*]] = bitcast double* [[REF_TMP]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 8, i8* [[TMP22]]) #[[ATTR5]]
// CHECK2-NEXT:    store double 0.000000e+00, double* [[REF_TMP]], align 8, !tbaa [[TBAA22]]
// CHECK2-NEXT:    [[TMP23:%.*]] = bitcast double* [[REF_TMP6]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 8, i8* [[TMP23]]) #[[ATTR5]]
// CHECK2-NEXT:    store double 0.000000e+00, double* [[REF_TMP6]], align 8, !tbaa [[TBAA22]]
// CHECK2-NEXT:    call void @_ZNSt7complexIdEC1ERKdS2_(%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[PARTIAL_SUM5]], double* nonnull align 8 dereferenceable(8) [[REF_TMP]], double* nonnull align 8 dereferenceable(8) [[REF_TMP6]]) #[[ATTR10]]
// CHECK2-NEXT:    [[TMP24:%.*]] = bitcast double* [[REF_TMP6]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 8, i8* [[TMP24]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP25:%.*]] = bitcast double* [[REF_TMP]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 8, i8* [[TMP25]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP26:%.*]] = bitcast i32* [[I7]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP26]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP27:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP28:%.*]] = load i32, i32* [[TMP27]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    call void @__kmpc_for_static_init_4u(%struct.ident_t* @[[GLOB3]], i32 [[TMP28]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK2-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK2:       omp.dispatch.cond:
// CHECK2-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[CMP8:%.*]] = icmp ugt i32 [[TMP29]], [[TMP30]]
// CHECK2-NEXT:    br i1 [[CMP8]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK2:       cond.true:
// CHECK2-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    br label [[COND_END:%.*]]
// CHECK2:       cond.false:
// CHECK2-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    br label [[COND_END]]
// CHECK2:       cond.end:
// CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP31]], [[COND_TRUE]] ], [ [[TMP32]], [[COND_FALSE]] ]
// CHECK2-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    store i32 [[TMP33]], i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[ADD9:%.*]] = add i32 [[TMP35]], 1
// CHECK2-NEXT:    [[CMP10:%.*]] = icmp ult i32 [[TMP34]], [[ADD9]]
// CHECK2-NEXT:    br i1 [[CMP10]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_CLEANUP:%.*]]
// CHECK2:       omp.dispatch.cleanup:
// CHECK2-NEXT:    br label [[OMP_DISPATCH_END:%.*]]
// CHECK2:       omp.dispatch.body:
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK2:       omp.inner.for.cond:
// CHECK2-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[ADD11:%.*]] = add i32 [[TMP37]], 1
// CHECK2-NEXT:    [[CMP12:%.*]] = icmp ult i32 [[TMP36]], [[ADD11]]
// CHECK2-NEXT:    br i1 [[CMP12]], 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:    [[TMP38:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[MUL:%.*]] = mul i32 [[TMP39]], 1
// CHECK2-NEXT:    [[ADD13:%.*]] = add i32 [[TMP38]], [[MUL]]
// CHECK2-NEXT:    store i32 [[ADD13]], i32* [[I7]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP40:%.*]] = bitcast %"class.std::complex.0"* [[REF_TMP14]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 16, i8* [[TMP40]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP41:%.*]] = bitcast double* [[REF_TMP15]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 8, i8* [[TMP41]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP42:%.*]] = load i32, i32* [[I7]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[CONV:%.*]] = sitofp i32 [[TMP42]] to double
// CHECK2-NEXT:    store double [[CONV]], double* [[REF_TMP15]], align 8, !tbaa [[TBAA22]]
// CHECK2-NEXT:    [[TMP43:%.*]] = bitcast double* [[REF_TMP16]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.start.p0i8(i64 8, i8* [[TMP43]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP44:%.*]] = load i32, i32* [[I7]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[CONV17:%.*]] = sitofp i32 [[TMP44]] to double
// CHECK2-NEXT:    store double [[CONV17]], double* [[REF_TMP16]], align 8, !tbaa [[TBAA22]]
// CHECK2-NEXT:    call void @_ZNSt7complexIdEC1ERKdS2_(%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[REF_TMP14]], double* nonnull align 8 dereferenceable(8) [[REF_TMP15]], double* nonnull align 8 dereferenceable(8) [[REF_TMP16]]) #[[ATTR10]]
// CHECK2-NEXT:    [[CALL:%.*]] = call nonnull align 8 dereferenceable(16) %"class.std::complex.0"* @_ZNSt7complexIdEpLIdEERS0_RKS_IT_E(%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[PARTIAL_SUM5]], %"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[REF_TMP14]]) #[[ATTR10]]
// CHECK2-NEXT:    [[TMP45:%.*]] = bitcast double* [[REF_TMP16]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 8, i8* [[TMP45]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP46:%.*]] = bitcast double* [[REF_TMP15]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 8, i8* [[TMP46]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP47:%.*]] = bitcast %"class.std::complex.0"* [[REF_TMP14]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 16, i8* [[TMP47]]) #[[ATTR5]]
// 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:    [[TMP48:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[ADD18:%.*]] = add i32 [[TMP48]], 1
// CHECK2-NEXT:    store i32 [[ADD18]], i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK2:       omp.inner.for.end:
// CHECK2-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK2:       omp.dispatch.inc:
// CHECK2-NEXT:    [[TMP49:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP50:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[ADD19:%.*]] = add i32 [[TMP49]], [[TMP50]]
// CHECK2-NEXT:    store i32 [[ADD19]], i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP51:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP52:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[ADD20:%.*]] = add i32 [[TMP51]], [[TMP52]]
// CHECK2-NEXT:    store i32 [[ADD20]], i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK2:       omp.dispatch.end:
// CHECK2-NEXT:    [[TMP53:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP54:%.*]] = load i32, i32* [[TMP53]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB3]], i32 [[TMP54]])
// CHECK2-NEXT:    [[TMP55:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP56:%.*]] = load i32, i32* [[TMP55]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP57:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0
// CHECK2-NEXT:    [[TMP58:%.*]] = bitcast %"class.std::complex.0"* [[PARTIAL_SUM5]] to i8*
// CHECK2-NEXT:    store i8* [[TMP58]], i8** [[TMP57]], align 8
// CHECK2-NEXT:    [[TMP59:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8*
// CHECK2-NEXT:    [[TMP60:%.*]] = call i32 @__kmpc_nvptx_parallel_reduce_nowait_v2(%struct.ident_t* @[[GLOB1]], i32 [[TMP56]], i32 1, i64 8, i8* [[TMP59]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func5, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func6)
// CHECK2-NEXT:    [[TMP61:%.*]] = icmp eq i32 [[TMP60]], 1
// CHECK2-NEXT:    br i1 [[TMP61]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]]
// CHECK2:       .omp.reduction.then:
// CHECK2-NEXT:    [[CALL21:%.*]] = call nonnull align 8 dereferenceable(16) %"class.std::complex.0"* @_ZNSt7complexIdEpLIdEERS0_RKS_IT_E(%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[TMP2]], %"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[PARTIAL_SUM5]]) #[[ATTR10]]
// CHECK2-NEXT:    call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP56]])
// CHECK2-NEXT:    br label [[DOTOMP_REDUCTION_DONE]]
// CHECK2:       .omp.reduction.done:
// CHECK2-NEXT:    [[TMP62:%.*]] = bitcast i32* [[I7]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP62]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP63:%.*]] = bitcast %"class.std::complex.0"* [[PARTIAL_SUM5]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 16, i8* [[TMP63]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP64:%.*]] = bitcast i32* [[DOTOMP_IS_LAST]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP64]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP65:%.*]] = bitcast i32* [[DOTOMP_STRIDE]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP65]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP66:%.*]] = bitcast i32* [[DOTOMP_UB]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP66]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP67:%.*]] = bitcast i32* [[DOTOMP_LB]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP67]]) #[[ATTR5]]
// CHECK2-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK2:       omp.precond.end:
// CHECK2-NEXT:    [[TMP68:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_2]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP68]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP69:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_1]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP69]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP70:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP70]]) #[[ATTR5]]
// CHECK2-NEXT:    [[TMP71:%.*]] = bitcast i32* [[DOTOMP_IV]] to i8*
// CHECK2-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP71]]) #[[ATTR5]]
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZNSt7complexIdEpLIdEERS0_RKS_IT_E
// CHECK2-SAME: (%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[THIS:%.*]], %"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[__C:%.*]]) #[[ATTR4]] comdat align 2 {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex.0"*, align 8
// CHECK2-NEXT:    [[__C_ADDR:%.*]] = alloca %"class.std::complex.0"*, align 8
// CHECK2-NEXT:    store %"class.std::complex.0"* [[THIS]], %"class.std::complex.0"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    store %"class.std::complex.0"* [[__C]], %"class.std::complex.0"** [[__C_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[THIS1:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[THIS_ADDR]], align 8
// CHECK2-NEXT:    [[TMP0:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[__C_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[CALL:%.*]] = call double @_ZNKSt7complexIdE4realEv(%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[TMP0]]) #[[ATTR10]]
// CHECK2-NEXT:    [[__RE_:%.*]] = getelementptr inbounds %"class.std::complex.0", %"class.std::complex.0"* [[THIS1]], i32 0, i32 0
// CHECK2-NEXT:    [[TMP1:%.*]] = load double, double* [[__RE_]], align 8, !tbaa [[TBAA24:![0-9]+]]
// CHECK2-NEXT:    [[ADD:%.*]] = fadd double [[TMP1]], [[CALL]]
// CHECK2-NEXT:    store double [[ADD]], double* [[__RE_]], align 8, !tbaa [[TBAA24]]
// CHECK2-NEXT:    [[TMP2:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[__C_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[CALL2:%.*]] = call double @_ZNKSt7complexIdE4imagEv(%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[TMP2]]) #[[ATTR10]]
// CHECK2-NEXT:    [[__IM_:%.*]] = getelementptr inbounds %"class.std::complex.0", %"class.std::complex.0"* [[THIS1]], i32 0, i32 1
// CHECK2-NEXT:    [[TMP3:%.*]] = load double, double* [[__IM_]], align 8, !tbaa [[TBAA26:![0-9]+]]
// CHECK2-NEXT:    [[ADD3:%.*]] = fadd double [[TMP3]], [[CALL2]]
// CHECK2-NEXT:    store double [[ADD3]], double* [[__IM_]], align 8, !tbaa [[TBAA26]]
// CHECK2-NEXT:    ret %"class.std::complex.0"* [[THIS1]]
//
//
// CHECK2-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func5
// CHECK2-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR6]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK2-NEXT:    [[DOTADDR1:%.*]] = alloca i16, align 2
// CHECK2-NEXT:    [[DOTADDR2:%.*]] = alloca i16, align 2
// CHECK2-NEXT:    [[DOTADDR3:%.*]] = alloca i16, align 2
// CHECK2-NEXT:    [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [1 x i8*], align 8
// CHECK2-NEXT:    [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca %"class.std::complex.0", align 8
// CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    store i16 [[TMP1]], i16* [[DOTADDR1]], align 2, !tbaa [[TBAA19]]
// CHECK2-NEXT:    store i16 [[TMP2]], i16* [[DOTADDR2]], align 2, !tbaa [[TBAA19]]
// CHECK2-NEXT:    store i16 [[TMP3]], i16* [[DOTADDR3]], align 2, !tbaa [[TBAA19]]
// CHECK2-NEXT:    [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [1 x i8*]*
// CHECK2-NEXT:    [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2, !tbaa [[TBAA19]]
// CHECK2-NEXT:    [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2, !tbaa [[TBAA19]]
// CHECK2-NEXT:    [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2, !tbaa [[TBAA19]]
// CHECK2-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK2-NEXT:    [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8
// CHECK2-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0
// CHECK2-NEXT:    [[TMP12:%.*]] = bitcast i8* [[TMP10]] to %"class.std::complex.0"*
// CHECK2-NEXT:    [[TMP13:%.*]] = getelementptr %"class.std::complex.0", %"class.std::complex.0"* [[TMP12]], i64 1
// CHECK2-NEXT:    [[TMP14:%.*]] = bitcast %"class.std::complex.0"* [[TMP13]] to i8*
// CHECK2-NEXT:    [[TMP15:%.*]] = bitcast %"class.std::complex.0"* [[TMP12]] to i64*
// CHECK2-NEXT:    [[TMP16:%.*]] = bitcast %"class.std::complex.0"* [[DOTOMP_REDUCTION_ELEMENT]] to i64*
// CHECK2-NEXT:    br label [[DOTSHUFFLE_PRE_COND:%.*]]
// CHECK2:       .shuffle.pre_cond:
// CHECK2-NEXT:    [[TMP17:%.*]] = phi i64* [ [[TMP15]], [[ENTRY:%.*]] ], [ [[TMP29:%.*]], [[DOTSHUFFLE_THEN:%.*]] ]
// CHECK2-NEXT:    [[TMP18:%.*]] = phi i64* [ [[TMP16]], [[ENTRY]] ], [ [[TMP30:%.*]], [[DOTSHUFFLE_THEN]] ]
// CHECK2-NEXT:    [[TMP19:%.*]] = bitcast i64* [[TMP17]] to i8*
// CHECK2-NEXT:    [[TMP20:%.*]] = ptrtoint i8* [[TMP14]] to i64
// CHECK2-NEXT:    [[TMP21:%.*]] = ptrtoint i8* [[TMP19]] to i64
// CHECK2-NEXT:    [[TMP22:%.*]] = sub i64 [[TMP20]], [[TMP21]]
// CHECK2-NEXT:    [[TMP23:%.*]] = sdiv exact i64 [[TMP22]], ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)
// CHECK2-NEXT:    [[TMP24:%.*]] = icmp sgt i64 [[TMP23]], 7
// CHECK2-NEXT:    br i1 [[TMP24]], label [[DOTSHUFFLE_THEN]], label [[DOTSHUFFLE_EXIT:%.*]]
// CHECK2:       .shuffle.then:
// CHECK2-NEXT:    [[TMP25:%.*]] = load i64, i64* [[TMP17]], align 8
// CHECK2-NEXT:    [[TMP26:%.*]] = call i32 @__kmpc_get_warp_size()
// CHECK2-NEXT:    [[TMP27:%.*]] = trunc i32 [[TMP26]] to i16
// CHECK2-NEXT:    [[TMP28:%.*]] = call i64 @__kmpc_shuffle_int64(i64 [[TMP25]], i16 [[TMP7]], i16 [[TMP27]])
// CHECK2-NEXT:    store i64 [[TMP28]], i64* [[TMP18]], align 8
// CHECK2-NEXT:    [[TMP29]] = getelementptr i64, i64* [[TMP17]], i64 1
// CHECK2-NEXT:    [[TMP30]] = getelementptr i64, i64* [[TMP18]], i64 1
// CHECK2-NEXT:    br label [[DOTSHUFFLE_PRE_COND]]
// CHECK2:       .shuffle.exit:
// CHECK2-NEXT:    [[TMP31:%.*]] = bitcast %"class.std::complex.0"* [[DOTOMP_REDUCTION_ELEMENT]] to i8*
// CHECK2-NEXT:    store i8* [[TMP31]], i8** [[TMP11]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP32:%.*]] = icmp eq i16 [[TMP8]], 0
// CHECK2-NEXT:    [[TMP33:%.*]] = icmp eq i16 [[TMP8]], 1
// CHECK2-NEXT:    [[TMP34:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]]
// CHECK2-NEXT:    [[TMP35:%.*]] = and i1 [[TMP33]], [[TMP34]]
// CHECK2-NEXT:    [[TMP36:%.*]] = icmp eq i16 [[TMP8]], 2
// CHECK2-NEXT:    [[TMP37:%.*]] = and i16 [[TMP6]], 1
// CHECK2-NEXT:    [[TMP38:%.*]] = icmp eq i16 [[TMP37]], 0
// CHECK2-NEXT:    [[TMP39:%.*]] = and i1 [[TMP36]], [[TMP38]]
// CHECK2-NEXT:    [[TMP40:%.*]] = icmp sgt i16 [[TMP7]], 0
// CHECK2-NEXT:    [[TMP41:%.*]] = and i1 [[TMP39]], [[TMP40]]
// CHECK2-NEXT:    [[TMP42:%.*]] = or i1 [[TMP32]], [[TMP35]]
// CHECK2-NEXT:    [[TMP43:%.*]] = or i1 [[TMP42]], [[TMP41]]
// CHECK2-NEXT:    br i1 [[TMP43]], label [[THEN:%.*]], label [[ELSE:%.*]]
// CHECK2:       then:
// CHECK2-NEXT:    [[TMP44:%.*]] = bitcast [1 x i8*]* [[TMP5]] to i8*
// CHECK2-NEXT:    [[TMP45:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8*
// CHECK2-NEXT:    call void @"_omp$reduction$reduction_func4"(i8* [[TMP44]], i8* [[TMP45]]) #[[ATTR5]]
// CHECK2-NEXT:    br label [[IFCONT:%.*]]
// CHECK2:       else:
// CHECK2-NEXT:    br label [[IFCONT]]
// CHECK2:       ifcont:
// CHECK2-NEXT:    [[TMP46:%.*]] = icmp eq i16 [[TMP8]], 1
// CHECK2-NEXT:    [[TMP47:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]]
// CHECK2-NEXT:    [[TMP48:%.*]] = and i1 [[TMP46]], [[TMP47]]
// CHECK2-NEXT:    br i1 [[TMP48]], label [[THEN4:%.*]], label [[ELSE5:%.*]]
// CHECK2:       then4:
// CHECK2-NEXT:    [[TMP49:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0
// CHECK2-NEXT:    [[TMP50:%.*]] = load i8*, i8** [[TMP49]], align 8
// CHECK2-NEXT:    [[TMP51:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK2-NEXT:    [[TMP52:%.*]] = load i8*, i8** [[TMP51]], align 8
// CHECK2-NEXT:    [[TMP53:%.*]] = bitcast i8* [[TMP50]] to %"class.std::complex.0"*
// CHECK2-NEXT:    [[TMP54:%.*]] = bitcast i8* [[TMP52]] to %"class.std::complex.0"*
// CHECK2-NEXT:    [[TMP55:%.*]] = bitcast %"class.std::complex.0"* [[TMP54]] to i8*
// CHECK2-NEXT:    [[TMP56:%.*]] = bitcast %"class.std::complex.0"* [[TMP53]] to i8*
// CHECK2-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[TMP55]], i8* align 8 [[TMP56]], i64 16, i1 false), !tbaa.struct !27
// CHECK2-NEXT:    br label [[IFCONT6:%.*]]
// CHECK2:       else5:
// CHECK2-NEXT:    br label [[IFCONT6]]
// CHECK2:       ifcont6:
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func6
// CHECK2-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR6]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK2-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCNT_ADDR:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK2-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK2-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK2-NEXT:    [[NVPTX_LANE_ID:%.*]] = and i32 [[TMP4]], 31
// CHECK2-NEXT:    [[TMP5:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK2-NEXT:    [[NVPTX_WARP_ID:%.*]] = ashr i32 [[TMP5]], 5
// CHECK2-NEXT:    [[TMP6:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK2-NEXT:    [[TMP7:%.*]] = bitcast i8* [[TMP6]] to [1 x i8*]*
// CHECK2-NEXT:    store i32 0, i32* [[DOTCNT_ADDR]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    br label [[PRECOND:%.*]]
// CHECK2:       precond:
// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCNT_ADDR]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[TMP9:%.*]] = icmp ult i32 [[TMP8]], 4
// CHECK2-NEXT:    br i1 [[TMP9]], label [[BODY:%.*]], label [[EXIT:%.*]]
// CHECK2:       body:
// CHECK2-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]])
// CHECK2-NEXT:    [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0
// CHECK2-NEXT:    br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]]
// CHECK2:       then:
// CHECK2-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP7]], i64 0, i64 0
// CHECK2-NEXT:    [[TMP11:%.*]] = load i8*, i8** [[TMP10]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP12:%.*]] = bitcast i8* [[TMP11]] to i32*
// CHECK2-NEXT:    [[TMP13:%.*]] = getelementptr i32, i32* [[TMP12]], i32 [[TMP8]]
// CHECK2-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]]
// CHECK2-NEXT:    [[TMP15:%.*]] = load i32, i32* [[TMP13]], align 4
// CHECK2-NEXT:    store volatile i32 [[TMP15]], i32 addrspace(3)* [[TMP14]], align 4
// CHECK2-NEXT:    br label [[IFCONT:%.*]]
// CHECK2:       else:
// CHECK2-NEXT:    br label [[IFCONT]]
// CHECK2:       ifcont:
// CHECK2-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]])
// CHECK2-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTADDR1]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[TMP3]], [[TMP16]]
// CHECK2-NEXT:    br i1 [[IS_ACTIVE_THREAD]], label [[THEN2:%.*]], label [[ELSE3:%.*]]
// CHECK2:       then2:
// CHECK2-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[TMP3]]
// CHECK2-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP7]], i64 0, i64 0
// CHECK2-NEXT:    [[TMP19:%.*]] = load i8*, i8** [[TMP18]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP20:%.*]] = bitcast i8* [[TMP19]] to i32*
// CHECK2-NEXT:    [[TMP21:%.*]] = getelementptr i32, i32* [[TMP20]], i32 [[TMP8]]
// CHECK2-NEXT:    [[TMP22:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP17]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    store i32 [[TMP22]], i32* [[TMP21]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    br label [[IFCONT4:%.*]]
// CHECK2:       else3:
// CHECK2-NEXT:    br label [[IFCONT4]]
// CHECK2:       ifcont4:
// CHECK2-NEXT:    [[TMP23:%.*]] = add nsw i32 [[TMP8]], 1
// CHECK2-NEXT:    store i32 [[TMP23]], i32* [[DOTCNT_ADDR]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    br label [[PRECOND]]
// CHECK2:       exit:
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@__omp_outlined__3_wrapper
// CHECK2-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR6]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
// CHECK2-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
// CHECK2-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2, !tbaa [[TBAA19]]
// CHECK2-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4, !tbaa [[TBAA8]]
// CHECK2-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK2-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
// CHECK2-NEXT:    [[TMP2:%.*]] = load i8**, i8*** [[GLOBAL_ARGS]], align 8
// CHECK2-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 0
// CHECK2-NEXT:    [[TMP4:%.*]] = bitcast i8** [[TMP3]] to i32**
// CHECK2-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[TMP4]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 1
// CHECK2-NEXT:    [[TMP7:%.*]] = bitcast i8** [[TMP6]] to i32**
// CHECK2-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[TMP7]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 2
// CHECK2-NEXT:    [[TMP10:%.*]] = bitcast i8** [[TMP9]] to %"class.std::complex.0"**
// CHECK2-NEXT:    [[TMP11:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[TMP10]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    call void @__omp_outlined__3(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]], i32* [[TMP5]], i32* [[TMP8]], %"class.std::complex.0"* [[TMP11]]) #[[ATTR5]]
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZNSt7complexIfEC2ERKfS2_
// CHECK2-SAME: (%"class.std::complex"* nonnull align 4 dereferenceable(8) [[THIS:%.*]], float* nonnull align 4 dereferenceable(4) [[__RE:%.*]], float* nonnull align 4 dereferenceable(4) [[__IM:%.*]]) unnamed_addr #[[ATTR3]] comdat align 2 {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex"*, align 8
// CHECK2-NEXT:    [[__RE_ADDR:%.*]] = alloca float*, align 8
// CHECK2-NEXT:    [[__IM_ADDR:%.*]] = alloca float*, align 8
// CHECK2-NEXT:    store %"class.std::complex"* [[THIS]], %"class.std::complex"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    store float* [[__RE]], float** [[__RE_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    store float* [[__IM]], float** [[__IM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[THIS1:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[THIS_ADDR]], align 8
// CHECK2-NEXT:    [[__RE_:%.*]] = getelementptr inbounds %"class.std::complex", %"class.std::complex"* [[THIS1]], i32 0, i32 0
// CHECK2-NEXT:    [[TMP0:%.*]] = load float*, float** [[__RE_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP1:%.*]] = load float, float* [[TMP0]], align 4, !tbaa [[TBAA14]]
// CHECK2-NEXT:    store float [[TMP1]], float* [[__RE_]], align 4, !tbaa [[TBAA16]]
// CHECK2-NEXT:    [[__IM_:%.*]] = getelementptr inbounds %"class.std::complex", %"class.std::complex"* [[THIS1]], i32 0, i32 1
// CHECK2-NEXT:    [[TMP2:%.*]] = load float*, float** [[__IM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP3:%.*]] = load float, float* [[TMP2]], align 4, !tbaa [[TBAA14]]
// CHECK2-NEXT:    store float [[TMP3]], float* [[__IM_]], align 4, !tbaa [[TBAA18]]
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZNKSt7complexIfE4realEv
// CHECK2-SAME: (%"class.std::complex"* nonnull align 4 dereferenceable(8) [[THIS:%.*]]) #[[ATTR4]] comdat align 2 {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex"*, align 8
// CHECK2-NEXT:    store %"class.std::complex"* [[THIS]], %"class.std::complex"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[THIS1:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[THIS_ADDR]], align 8
// CHECK2-NEXT:    [[__RE_:%.*]] = getelementptr inbounds %"class.std::complex", %"class.std::complex"* [[THIS1]], i32 0, i32 0
// CHECK2-NEXT:    [[TMP0:%.*]] = load float, float* [[__RE_]], align 4, !tbaa [[TBAA16]]
// CHECK2-NEXT:    ret float [[TMP0]]
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZNKSt7complexIfE4imagEv
// CHECK2-SAME: (%"class.std::complex"* nonnull align 4 dereferenceable(8) [[THIS:%.*]]) #[[ATTR4]] comdat align 2 {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex"*, align 8
// CHECK2-NEXT:    store %"class.std::complex"* [[THIS]], %"class.std::complex"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[THIS1:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[THIS_ADDR]], align 8
// CHECK2-NEXT:    [[__IM_:%.*]] = getelementptr inbounds %"class.std::complex", %"class.std::complex"* [[THIS1]], i32 0, i32 1
// CHECK2-NEXT:    [[TMP0:%.*]] = load float, float* [[__IM_]], align 4, !tbaa [[TBAA18]]
// CHECK2-NEXT:    ret float [[TMP0]]
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZNSt7complexIdEC2ERKdS2_
// CHECK2-SAME: (%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[THIS:%.*]], double* nonnull align 8 dereferenceable(8) [[__RE:%.*]], double* nonnull align 8 dereferenceable(8) [[__IM:%.*]]) unnamed_addr #[[ATTR3]] comdat align 2 {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex.0"*, align 8
// CHECK2-NEXT:    [[__RE_ADDR:%.*]] = alloca double*, align 8
// CHECK2-NEXT:    [[__IM_ADDR:%.*]] = alloca double*, align 8
// CHECK2-NEXT:    store %"class.std::complex.0"* [[THIS]], %"class.std::complex.0"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    store double* [[__RE]], double** [[__RE_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    store double* [[__IM]], double** [[__IM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[THIS1:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[THIS_ADDR]], align 8
// CHECK2-NEXT:    [[__RE_:%.*]] = getelementptr inbounds %"class.std::complex.0", %"class.std::complex.0"* [[THIS1]], i32 0, i32 0
// CHECK2-NEXT:    [[TMP0:%.*]] = load double*, double** [[__RE_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP1:%.*]] = load double, double* [[TMP0]], align 8, !tbaa [[TBAA22]]
// CHECK2-NEXT:    store double [[TMP1]], double* [[__RE_]], align 8, !tbaa [[TBAA24]]
// CHECK2-NEXT:    [[__IM_:%.*]] = getelementptr inbounds %"class.std::complex.0", %"class.std::complex.0"* [[THIS1]], i32 0, i32 1
// CHECK2-NEXT:    [[TMP2:%.*]] = load double*, double** [[__IM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[TMP3:%.*]] = load double, double* [[TMP2]], align 8, !tbaa [[TBAA22]]
// CHECK2-NEXT:    store double [[TMP3]], double* [[__IM_]], align 8, !tbaa [[TBAA26]]
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZNKSt7complexIdE4realEv
// CHECK2-SAME: (%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[THIS:%.*]]) #[[ATTR4]] comdat align 2 {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex.0"*, align 8
// CHECK2-NEXT:    store %"class.std::complex.0"* [[THIS]], %"class.std::complex.0"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[THIS1:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[THIS_ADDR]], align 8
// CHECK2-NEXT:    [[__RE_:%.*]] = getelementptr inbounds %"class.std::complex.0", %"class.std::complex.0"* [[THIS1]], i32 0, i32 0
// CHECK2-NEXT:    [[TMP0:%.*]] = load double, double* [[__RE_]], align 8, !tbaa [[TBAA24]]
// CHECK2-NEXT:    ret double [[TMP0]]
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZNKSt7complexIdE4imagEv
// CHECK2-SAME: (%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[THIS:%.*]]) #[[ATTR4]] comdat align 2 {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex.0"*, align 8
// CHECK2-NEXT:    store %"class.std::complex.0"* [[THIS]], %"class.std::complex.0"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK2-NEXT:    [[THIS1:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[THIS_ADDR]], align 8
// CHECK2-NEXT:    [[__IM_:%.*]] = getelementptr inbounds %"class.std::complex.0", %"class.std::complex.0"* [[THIS1]], i32 0, i32 1
// CHECK2-NEXT:    [[TMP0:%.*]] = load double, double* [[__IM_]], align 8, !tbaa [[TBAA26]]
// CHECK2-NEXT:    ret double [[TMP0]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z17complex_reductionIfEvv_l16
// CHECK3-SAME: () #[[ATTR0:[0-9]+]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i8 1, i1 true, i1 true)
// CHECK3-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
// CHECK3-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK3:       user_code.entry:
// CHECK3-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK3-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4, !tbaa [[TBAA8:![0-9]+]]
// CHECK3-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR5:[0-9]+]]
// CHECK3-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 1, i1 true)
// CHECK3-NEXT:    ret void
// CHECK3:       worker.exit:
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@__omp_outlined__
// CHECK3-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[IB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[REF_TMP:%.*]] = alloca float, align 4
// CHECK3-NEXT:    [[REF_TMP2:%.*]] = alloca float, align 4
// CHECK3-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x i8*], align 8
// CHECK3-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8, !tbaa [[TBAA12:![0-9]+]]
// CHECK3-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[ISTART:%.*]] = call align 16 i8* @__kmpc_alloc_shared(i64 4)
// CHECK3-NEXT:    [[ISTART_ON_STACK:%.*]] = bitcast i8* [[ISTART]] to i32*
// CHECK3-NEXT:    [[IEND:%.*]] = call align 16 i8* @__kmpc_alloc_shared(i64 4)
// CHECK3-NEXT:    [[IEND_ON_STACK:%.*]] = bitcast i8* [[IEND]] to i32*
// CHECK3-NEXT:    [[PARTIAL_SUM:%.*]] = call align 16 i8* @__kmpc_alloc_shared(i64 8)
// CHECK3-NEXT:    [[PARTIAL_SUM_ON_STACK:%.*]] = bitcast i8* [[PARTIAL_SUM]] to %"class.std::complex"*
// CHECK3-NEXT:    [[TMP0:%.*]] = bitcast i32* [[DOTOMP_IV]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP0]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP1:%.*]] = bitcast i32* [[DOTOMP_LB]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP1]]) #[[ATTR5]]
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP2:%.*]] = bitcast i32* [[DOTOMP_UB]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP2]]) #[[ATTR5]]
// CHECK3-NEXT:    store i32 99, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP3:%.*]] = bitcast i32* [[DOTOMP_STRIDE]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP3]]) #[[ATTR5]]
// CHECK3-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP4:%.*]] = bitcast i32* [[DOTOMP_IS_LAST]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP4]]) #[[ATTR5]]
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP5:%.*]] = bitcast i32* [[IB]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP5]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP6:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[TMP6]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP7]], i32 92, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP8]], 99
// CHECK3-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK3:       cond.true:
// CHECK3-NEXT:    br label [[COND_END:%.*]]
// CHECK3:       cond.false:
// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    br label [[COND_END]]
// CHECK3:       cond.end:
// CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP9]], [[COND_FALSE]] ]
// CHECK3-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    store i32 [[TMP10]], i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK3:       omp.inner.for.cond:
// CHECK3-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP11]], [[TMP12]]
// 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:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP13]], 1
// CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK3-NEXT:    store i32 [[ADD]], i32* [[IB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP14:%.*]] = bitcast float* [[REF_TMP]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP14]]) #[[ATTR5]]
// CHECK3-NEXT:    store float 0.000000e+00, float* [[REF_TMP]], align 4, !tbaa [[TBAA14:![0-9]+]]
// CHECK3-NEXT:    [[TMP15:%.*]] = bitcast float* [[REF_TMP2]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP15]]) #[[ATTR5]]
// CHECK3-NEXT:    store float 0.000000e+00, float* [[REF_TMP2]], align 4, !tbaa [[TBAA14]]
// CHECK3-NEXT:    call void @_ZNSt7complexIfEC1ERKfS2_(%"class.std::complex"* nonnull align 4 dereferenceable(8) [[PARTIAL_SUM_ON_STACK]], float* nonnull align 4 dereferenceable(4) [[REF_TMP]], float* nonnull align 4 dereferenceable(4) [[REF_TMP2]]) #[[ATTR10:[0-9]+]]
// CHECK3-NEXT:    [[TMP16:%.*]] = bitcast float* [[REF_TMP2]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP16]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP17:%.*]] = bitcast float* [[REF_TMP]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP17]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP18:%.*]] = load i32, i32* [[IB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[MUL3:%.*]] = mul nsw i32 [[TMP18]], 4
// CHECK3-NEXT:    store i32 [[MUL3]], i32* [[ISTART_ON_STACK]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP19:%.*]] = load i32, i32* [[IB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK3-NEXT:    [[MUL5:%.*]] = mul nsw i32 [[ADD4]], 4
// CHECK3-NEXT:    store i32 [[MUL5]], i32* [[IEND_ON_STACK]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK3-NEXT:    [[TMP21:%.*]] = bitcast i32* [[ISTART_ON_STACK]] to i8*
// CHECK3-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK3-NEXT:    [[TMP23:%.*]] = bitcast i32* [[IEND_ON_STACK]] to i8*
// CHECK3-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK3-NEXT:    [[TMP25:%.*]] = bitcast %"class.std::complex"* [[PARTIAL_SUM_ON_STACK]] to i8*
// CHECK3-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP26:%.*]] = bitcast [3 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK3-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP7]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32*, i32*, %"class.std::complex"*)* @__omp_outlined__1 to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined__1_wrapper to i8*), i8** [[TMP26]], i64 3)
// 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:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP27]], 1
// CHECK3-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK3:       omp.inner.for.end:
// CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK3:       omp.loop.exit:
// CHECK3-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP7]])
// CHECK3-NEXT:    [[TMP28:%.*]] = bitcast i32* [[IB]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP28]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP29:%.*]] = bitcast i32* [[DOTOMP_IS_LAST]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP29]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP30:%.*]] = bitcast i32* [[DOTOMP_STRIDE]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP30]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP31:%.*]] = bitcast i32* [[DOTOMP_UB]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP31]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP32:%.*]] = bitcast i32* [[DOTOMP_LB]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP32]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP33:%.*]] = bitcast i32* [[DOTOMP_IV]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP33]]) #[[ATTR5]]
// CHECK3-NEXT:    call void @__kmpc_free_shared(i8* [[PARTIAL_SUM]], i64 8)
// CHECK3-NEXT:    call void @__kmpc_free_shared(i8* [[IEND]], i64 4)
// CHECK3-NEXT:    call void @__kmpc_free_shared(i8* [[ISTART]], i64 4)
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZNSt7complexIfEC1ERKfS2_
// CHECK3-SAME: (%"class.std::complex"* nonnull align 4 dereferenceable(8) [[THIS:%.*]], float* nonnull align 4 dereferenceable(4) [[__RE:%.*]], float* nonnull align 4 dereferenceable(4) [[__IM:%.*]]) unnamed_addr #[[ATTR3:[0-9]+]] comdat align 2 {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex"*, align 8
// CHECK3-NEXT:    [[__RE_ADDR:%.*]] = alloca float*, align 8
// CHECK3-NEXT:    [[__IM_ADDR:%.*]] = alloca float*, align 8
// CHECK3-NEXT:    store %"class.std::complex"* [[THIS]], %"class.std::complex"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    store float* [[__RE]], float** [[__RE_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    store float* [[__IM]], float** [[__IM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[THIS1:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[THIS_ADDR]], align 8
// CHECK3-NEXT:    [[TMP0:%.*]] = load float*, float** [[__RE_ADDR]], align 8
// CHECK3-NEXT:    [[TMP1:%.*]] = load float*, float** [[__IM_ADDR]], align 8
// CHECK3-NEXT:    call void @_ZNSt7complexIfEC2ERKfS2_(%"class.std::complex"* nonnull align 4 dereferenceable(8) [[THIS1]], float* nonnull align 4 dereferenceable(4) [[TMP0]], float* nonnull align 4 dereferenceable(4) [[TMP1]]) #[[ATTR10]]
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@__omp_outlined__1
// CHECK3-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32* nonnull align 4 dereferenceable(4) [[ISTART:%.*]], i32* nonnull align 4 dereferenceable(4) [[IEND:%.*]], %"class.std::complex"* nonnull align 4 dereferenceable(8) [[PARTIAL_SUM:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK3-NEXT:    [[ISTART_ADDR:%.*]] = alloca i32*, align 8
// CHECK3-NEXT:    [[IEND_ADDR:%.*]] = alloca i32*, align 8
// CHECK3-NEXT:    [[PARTIAL_SUM_ADDR:%.*]] = alloca %"class.std::complex"*, align 8
// CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[PARTIAL_SUM5:%.*]] = alloca %"class.std::complex", align 4
// CHECK3-NEXT:    [[REF_TMP:%.*]] = alloca float, align 4
// CHECK3-NEXT:    [[REF_TMP6:%.*]] = alloca float, align 4
// CHECK3-NEXT:    [[I7:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[REF_TMP14:%.*]] = alloca %"class.std::complex", align 4
// CHECK3-NEXT:    [[REF_TMP15:%.*]] = alloca float, align 4
// CHECK3-NEXT:    [[REF_TMP16:%.*]] = alloca float, align 4
// CHECK3-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x i8*], align 8
// CHECK3-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    store i32* [[ISTART]], i32** [[ISTART_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    store i32* [[IEND]], i32** [[IEND_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    store %"class.std::complex"* [[PARTIAL_SUM]], %"class.std::complex"** [[PARTIAL_SUM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[ISTART_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[IEND_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP2:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[PARTIAL_SUM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP3:%.*]] = bitcast i32* [[DOTOMP_IV]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP3]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP4:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP4]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[TMP0]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    store i32 [[TMP5]], i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP6:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_1]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP6]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[TMP1]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    store i32 [[TMP7]], i32* [[DOTCAPTURE_EXPR_1]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP8:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_2]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP8]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[SUB:%.*]] = sub i32 [[TMP9]], [[TMP10]]
// CHECK3-NEXT:    [[SUB3:%.*]] = sub i32 [[SUB]], 1
// CHECK3-NEXT:    [[ADD:%.*]] = add i32 [[SUB3]], 1
// CHECK3-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], 1
// CHECK3-NEXT:    [[SUB4:%.*]] = sub i32 [[DIV]], 1
// CHECK3-NEXT:    store i32 [[SUB4]], i32* [[DOTCAPTURE_EXPR_2]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP11:%.*]] = bitcast i32* [[I]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP11]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    store i32 [[TMP12]], i32* [[I]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP13:%.*]] = bitcast i32* [[I]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP13]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP14]], [[TMP15]]
// CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK3:       omp.precond.then:
// CHECK3-NEXT:    [[TMP16:%.*]] = bitcast i32* [[DOTOMP_LB]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP16]]) #[[ATTR5]]
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP17:%.*]] = bitcast i32* [[DOTOMP_UB]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP17]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    store i32 [[TMP18]], i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP19:%.*]] = bitcast i32* [[DOTOMP_STRIDE]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP19]]) #[[ATTR5]]
// CHECK3-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP20:%.*]] = bitcast i32* [[DOTOMP_IS_LAST]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP20]]) #[[ATTR5]]
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP21:%.*]] = bitcast %"class.std::complex"* [[PARTIAL_SUM5]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 8, i8* [[TMP21]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP22:%.*]] = bitcast float* [[REF_TMP]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP22]]) #[[ATTR5]]
// CHECK3-NEXT:    store float 0.000000e+00, float* [[REF_TMP]], align 4, !tbaa [[TBAA14]]
// CHECK3-NEXT:    [[TMP23:%.*]] = bitcast float* [[REF_TMP6]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP23]]) #[[ATTR5]]
// CHECK3-NEXT:    store float 0.000000e+00, float* [[REF_TMP6]], align 4, !tbaa [[TBAA14]]
// CHECK3-NEXT:    call void @_ZNSt7complexIfEC1ERKfS2_(%"class.std::complex"* nonnull align 4 dereferenceable(8) [[PARTIAL_SUM5]], float* nonnull align 4 dereferenceable(4) [[REF_TMP]], float* nonnull align 4 dereferenceable(4) [[REF_TMP6]]) #[[ATTR10]]
// CHECK3-NEXT:    [[TMP24:%.*]] = bitcast float* [[REF_TMP6]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP24]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP25:%.*]] = bitcast float* [[REF_TMP]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP25]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP26:%.*]] = bitcast i32* [[I7]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP26]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP27:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK3-NEXT:    [[TMP28:%.*]] = load i32, i32* [[TMP27]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    call void @__kmpc_for_static_init_4u(%struct.ident_t* @[[GLOB3:[0-9]+]], i32 [[TMP28]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK3-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK3:       omp.dispatch.cond:
// CHECK3-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[CMP8:%.*]] = icmp ugt i32 [[TMP29]], [[TMP30]]
// CHECK3-NEXT:    br i1 [[CMP8]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK3:       cond.true:
// CHECK3-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    br label [[COND_END:%.*]]
// CHECK3:       cond.false:
// CHECK3-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    br label [[COND_END]]
// CHECK3:       cond.end:
// CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP31]], [[COND_TRUE]] ], [ [[TMP32]], [[COND_FALSE]] ]
// CHECK3-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    store i32 [[TMP33]], i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[ADD9:%.*]] = add i32 [[TMP35]], 1
// CHECK3-NEXT:    [[CMP10:%.*]] = icmp ult i32 [[TMP34]], [[ADD9]]
// CHECK3-NEXT:    br i1 [[CMP10]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_CLEANUP:%.*]]
// CHECK3:       omp.dispatch.cleanup:
// CHECK3-NEXT:    br label [[OMP_DISPATCH_END:%.*]]
// CHECK3:       omp.dispatch.body:
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK3:       omp.inner.for.cond:
// CHECK3-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[ADD11:%.*]] = add i32 [[TMP37]], 1
// CHECK3-NEXT:    [[CMP12:%.*]] = icmp ult i32 [[TMP36]], [[ADD11]]
// CHECK3-NEXT:    br i1 [[CMP12]], 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:    [[TMP38:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[MUL:%.*]] = mul i32 [[TMP39]], 1
// CHECK3-NEXT:    [[ADD13:%.*]] = add i32 [[TMP38]], [[MUL]]
// CHECK3-NEXT:    store i32 [[ADD13]], i32* [[I7]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP40:%.*]] = bitcast %"class.std::complex"* [[REF_TMP14]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 8, i8* [[TMP40]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP41:%.*]] = bitcast float* [[REF_TMP15]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP41]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP42:%.*]] = load i32, i32* [[I7]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[CONV:%.*]] = sitofp i32 [[TMP42]] to float
// CHECK3-NEXT:    store float [[CONV]], float* [[REF_TMP15]], align 4, !tbaa [[TBAA14]]
// CHECK3-NEXT:    [[TMP43:%.*]] = bitcast float* [[REF_TMP16]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP43]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP44:%.*]] = load i32, i32* [[I7]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[CONV17:%.*]] = sitofp i32 [[TMP44]] to float
// CHECK3-NEXT:    store float [[CONV17]], float* [[REF_TMP16]], align 4, !tbaa [[TBAA14]]
// CHECK3-NEXT:    call void @_ZNSt7complexIfEC1ERKfS2_(%"class.std::complex"* nonnull align 4 dereferenceable(8) [[REF_TMP14]], float* nonnull align 4 dereferenceable(4) [[REF_TMP15]], float* nonnull align 4 dereferenceable(4) [[REF_TMP16]]) #[[ATTR10]]
// CHECK3-NEXT:    [[CALL:%.*]] = call nonnull align 4 dereferenceable(8) %"class.std::complex"* @_ZNSt7complexIfEpLIfEERS0_RKS_IT_E(%"class.std::complex"* nonnull align 4 dereferenceable(8) [[PARTIAL_SUM5]], %"class.std::complex"* nonnull align 4 dereferenceable(8) [[REF_TMP14]]) #[[ATTR10]]
// CHECK3-NEXT:    [[TMP45:%.*]] = bitcast float* [[REF_TMP16]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP45]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP46:%.*]] = bitcast float* [[REF_TMP15]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP46]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP47:%.*]] = bitcast %"class.std::complex"* [[REF_TMP14]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 8, i8* [[TMP47]]) #[[ATTR5]]
// 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:    [[TMP48:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[ADD18:%.*]] = add i32 [[TMP48]], 1
// CHECK3-NEXT:    store i32 [[ADD18]], i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK3:       omp.inner.for.end:
// CHECK3-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK3:       omp.dispatch.inc:
// CHECK3-NEXT:    [[TMP49:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP50:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[ADD19:%.*]] = add i32 [[TMP49]], [[TMP50]]
// CHECK3-NEXT:    store i32 [[ADD19]], i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP51:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP52:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[ADD20:%.*]] = add i32 [[TMP51]], [[TMP52]]
// CHECK3-NEXT:    store i32 [[ADD20]], i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK3:       omp.dispatch.end:
// CHECK3-NEXT:    [[TMP53:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK3-NEXT:    [[TMP54:%.*]] = load i32, i32* [[TMP53]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB3]], i32 [[TMP54]])
// CHECK3-NEXT:    [[TMP55:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK3-NEXT:    [[TMP56:%.*]] = load i32, i32* [[TMP55]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP57:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0
// CHECK3-NEXT:    [[TMP58:%.*]] = bitcast %"class.std::complex"* [[PARTIAL_SUM5]] to i8*
// CHECK3-NEXT:    store i8* [[TMP58]], i8** [[TMP57]], align 8
// CHECK3-NEXT:    [[TMP59:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8*
// CHECK3-NEXT:    [[TMP60:%.*]] = call i32 @__kmpc_nvptx_parallel_reduce_nowait_v2(%struct.ident_t* @[[GLOB1]], i32 [[TMP56]], i32 1, i64 8, i8* [[TMP59]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func)
// CHECK3-NEXT:    [[TMP61:%.*]] = icmp eq i32 [[TMP60]], 1
// CHECK3-NEXT:    br i1 [[TMP61]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]]
// CHECK3:       .omp.reduction.then:
// CHECK3-NEXT:    [[CALL21:%.*]] = call nonnull align 4 dereferenceable(8) %"class.std::complex"* @_ZNSt7complexIfEpLIfEERS0_RKS_IT_E(%"class.std::complex"* nonnull align 4 dereferenceable(8) [[TMP2]], %"class.std::complex"* nonnull align 4 dereferenceable(8) [[PARTIAL_SUM5]]) #[[ATTR10]]
// CHECK3-NEXT:    call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP56]])
// CHECK3-NEXT:    br label [[DOTOMP_REDUCTION_DONE]]
// CHECK3:       .omp.reduction.done:
// CHECK3-NEXT:    [[TMP62:%.*]] = bitcast i32* [[I7]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP62]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP63:%.*]] = bitcast %"class.std::complex"* [[PARTIAL_SUM5]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 8, i8* [[TMP63]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP64:%.*]] = bitcast i32* [[DOTOMP_IS_LAST]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP64]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP65:%.*]] = bitcast i32* [[DOTOMP_STRIDE]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP65]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP66:%.*]] = bitcast i32* [[DOTOMP_UB]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP66]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP67:%.*]] = bitcast i32* [[DOTOMP_LB]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP67]]) #[[ATTR5]]
// CHECK3-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK3:       omp.precond.end:
// CHECK3-NEXT:    [[TMP68:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_2]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP68]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP69:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_1]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP69]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP70:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP70]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP71:%.*]] = bitcast i32* [[DOTOMP_IV]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP71]]) #[[ATTR5]]
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZNSt7complexIfEpLIfEERS0_RKS_IT_E
// CHECK3-SAME: (%"class.std::complex"* nonnull align 4 dereferenceable(8) [[THIS:%.*]], %"class.std::complex"* nonnull align 4 dereferenceable(8) [[__C:%.*]]) #[[ATTR4:[0-9]+]] comdat align 2 {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex"*, align 8
// CHECK3-NEXT:    [[__C_ADDR:%.*]] = alloca %"class.std::complex"*, align 8
// CHECK3-NEXT:    store %"class.std::complex"* [[THIS]], %"class.std::complex"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    store %"class.std::complex"* [[__C]], %"class.std::complex"** [[__C_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[THIS1:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[THIS_ADDR]], align 8
// CHECK3-NEXT:    [[TMP0:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[__C_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[CALL:%.*]] = call float @_ZNKSt7complexIfE4realEv(%"class.std::complex"* nonnull align 4 dereferenceable(8) [[TMP0]]) #[[ATTR10]]
// CHECK3-NEXT:    [[__RE_:%.*]] = getelementptr inbounds %"class.std::complex", %"class.std::complex"* [[THIS1]], i32 0, i32 0
// CHECK3-NEXT:    [[TMP1:%.*]] = load float, float* [[__RE_]], align 4, !tbaa [[TBAA16:![0-9]+]]
// CHECK3-NEXT:    [[ADD:%.*]] = fadd float [[TMP1]], [[CALL]]
// CHECK3-NEXT:    store float [[ADD]], float* [[__RE_]], align 4, !tbaa [[TBAA16]]
// CHECK3-NEXT:    [[TMP2:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[__C_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[CALL2:%.*]] = call float @_ZNKSt7complexIfE4imagEv(%"class.std::complex"* nonnull align 4 dereferenceable(8) [[TMP2]]) #[[ATTR10]]
// CHECK3-NEXT:    [[__IM_:%.*]] = getelementptr inbounds %"class.std::complex", %"class.std::complex"* [[THIS1]], i32 0, i32 1
// CHECK3-NEXT:    [[TMP3:%.*]] = load float, float* [[__IM_]], align 4, !tbaa [[TBAA18:![0-9]+]]
// CHECK3-NEXT:    [[ADD3:%.*]] = fadd float [[TMP3]], [[CALL2]]
// CHECK3-NEXT:    store float [[ADD3]], float* [[__IM_]], align 4, !tbaa [[TBAA18]]
// CHECK3-NEXT:    ret %"class.std::complex"* [[THIS1]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func
// CHECK3-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR6:[0-9]+]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK3-NEXT:    [[DOTADDR1:%.*]] = alloca i16, align 2
// CHECK3-NEXT:    [[DOTADDR2:%.*]] = alloca i16, align 2
// CHECK3-NEXT:    [[DOTADDR3:%.*]] = alloca i16, align 2
// CHECK3-NEXT:    [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [1 x i8*], align 8
// CHECK3-NEXT:    [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca %"class.std::complex", align 4
// CHECK3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    store i16 [[TMP1]], i16* [[DOTADDR1]], align 2, !tbaa [[TBAA19:![0-9]+]]
// CHECK3-NEXT:    store i16 [[TMP2]], i16* [[DOTADDR2]], align 2, !tbaa [[TBAA19]]
// CHECK3-NEXT:    store i16 [[TMP3]], i16* [[DOTADDR3]], align 2, !tbaa [[TBAA19]]
// CHECK3-NEXT:    [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [1 x i8*]*
// CHECK3-NEXT:    [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2, !tbaa [[TBAA19]]
// CHECK3-NEXT:    [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2, !tbaa [[TBAA19]]
// CHECK3-NEXT:    [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2, !tbaa [[TBAA19]]
// CHECK3-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK3-NEXT:    [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8
// CHECK3-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0
// CHECK3-NEXT:    [[TMP12:%.*]] = bitcast i8* [[TMP10]] to %"class.std::complex"*
// CHECK3-NEXT:    [[TMP13:%.*]] = getelementptr %"class.std::complex", %"class.std::complex"* [[TMP12]], i64 1
// CHECK3-NEXT:    [[TMP14:%.*]] = bitcast %"class.std::complex"* [[TMP13]] to i8*
// CHECK3-NEXT:    [[TMP15:%.*]] = bitcast %"class.std::complex"* [[TMP12]] to i64*
// CHECK3-NEXT:    [[TMP16:%.*]] = bitcast %"class.std::complex"* [[DOTOMP_REDUCTION_ELEMENT]] to i64*
// CHECK3-NEXT:    [[TMP17:%.*]] = load i64, i64* [[TMP15]], align 4
// CHECK3-NEXT:    [[TMP18:%.*]] = call i32 @__kmpc_get_warp_size()
// CHECK3-NEXT:    [[TMP19:%.*]] = trunc i32 [[TMP18]] to i16
// CHECK3-NEXT:    [[TMP20:%.*]] = call i64 @__kmpc_shuffle_int64(i64 [[TMP17]], i16 [[TMP7]], i16 [[TMP19]])
// CHECK3-NEXT:    store i64 [[TMP20]], i64* [[TMP16]], align 4
// CHECK3-NEXT:    [[TMP21:%.*]] = getelementptr i64, i64* [[TMP15]], i64 1
// CHECK3-NEXT:    [[TMP22:%.*]] = getelementptr i64, i64* [[TMP16]], i64 1
// CHECK3-NEXT:    [[TMP23:%.*]] = bitcast %"class.std::complex"* [[DOTOMP_REDUCTION_ELEMENT]] to i8*
// CHECK3-NEXT:    store i8* [[TMP23]], i8** [[TMP11]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP24:%.*]] = icmp eq i16 [[TMP8]], 0
// CHECK3-NEXT:    [[TMP25:%.*]] = icmp eq i16 [[TMP8]], 1
// CHECK3-NEXT:    [[TMP26:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]]
// CHECK3-NEXT:    [[TMP27:%.*]] = and i1 [[TMP25]], [[TMP26]]
// CHECK3-NEXT:    [[TMP28:%.*]] = icmp eq i16 [[TMP8]], 2
// CHECK3-NEXT:    [[TMP29:%.*]] = and i16 [[TMP6]], 1
// CHECK3-NEXT:    [[TMP30:%.*]] = icmp eq i16 [[TMP29]], 0
// CHECK3-NEXT:    [[TMP31:%.*]] = and i1 [[TMP28]], [[TMP30]]
// CHECK3-NEXT:    [[TMP32:%.*]] = icmp sgt i16 [[TMP7]], 0
// CHECK3-NEXT:    [[TMP33:%.*]] = and i1 [[TMP31]], [[TMP32]]
// CHECK3-NEXT:    [[TMP34:%.*]] = or i1 [[TMP24]], [[TMP27]]
// CHECK3-NEXT:    [[TMP35:%.*]] = or i1 [[TMP34]], [[TMP33]]
// CHECK3-NEXT:    br i1 [[TMP35]], label [[THEN:%.*]], label [[ELSE:%.*]]
// CHECK3:       then:
// CHECK3-NEXT:    [[TMP36:%.*]] = bitcast [1 x i8*]* [[TMP5]] to i8*
// CHECK3-NEXT:    [[TMP37:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8*
// CHECK3-NEXT:    call void @"_omp$reduction$reduction_func"(i8* [[TMP36]], i8* [[TMP37]]) #[[ATTR5]]
// CHECK3-NEXT:    br label [[IFCONT:%.*]]
// CHECK3:       else:
// CHECK3-NEXT:    br label [[IFCONT]]
// CHECK3:       ifcont:
// CHECK3-NEXT:    [[TMP38:%.*]] = icmp eq i16 [[TMP8]], 1
// CHECK3-NEXT:    [[TMP39:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]]
// CHECK3-NEXT:    [[TMP40:%.*]] = and i1 [[TMP38]], [[TMP39]]
// CHECK3-NEXT:    br i1 [[TMP40]], label [[THEN4:%.*]], label [[ELSE5:%.*]]
// CHECK3:       then4:
// CHECK3-NEXT:    [[TMP41:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0
// CHECK3-NEXT:    [[TMP42:%.*]] = load i8*, i8** [[TMP41]], align 8
// CHECK3-NEXT:    [[TMP43:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK3-NEXT:    [[TMP44:%.*]] = load i8*, i8** [[TMP43]], align 8
// CHECK3-NEXT:    [[TMP45:%.*]] = bitcast i8* [[TMP42]] to %"class.std::complex"*
// CHECK3-NEXT:    [[TMP46:%.*]] = bitcast i8* [[TMP44]] to %"class.std::complex"*
// CHECK3-NEXT:    [[TMP47:%.*]] = bitcast %"class.std::complex"* [[TMP46]] to i8*
// CHECK3-NEXT:    [[TMP48:%.*]] = bitcast %"class.std::complex"* [[TMP45]] to i8*
// CHECK3-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP47]], i8* align 4 [[TMP48]], i64 8, i1 false), !tbaa.struct !21
// CHECK3-NEXT:    br label [[IFCONT6:%.*]]
// CHECK3:       else5:
// CHECK3-NEXT:    br label [[IFCONT6]]
// CHECK3:       ifcont6:
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func
// CHECK3-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR6]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK3-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCNT_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK3-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK3-NEXT:    [[NVPTX_LANE_ID:%.*]] = and i32 [[TMP4]], 31
// CHECK3-NEXT:    [[TMP5:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK3-NEXT:    [[NVPTX_WARP_ID:%.*]] = ashr i32 [[TMP5]], 5
// CHECK3-NEXT:    [[TMP6:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK3-NEXT:    [[TMP7:%.*]] = bitcast i8* [[TMP6]] to [1 x i8*]*
// CHECK3-NEXT:    store i32 0, i32* [[DOTCNT_ADDR]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    br label [[PRECOND:%.*]]
// CHECK3:       precond:
// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCNT_ADDR]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP9:%.*]] = icmp ult i32 [[TMP8]], 2
// CHECK3-NEXT:    br i1 [[TMP9]], label [[BODY:%.*]], label [[EXIT:%.*]]
// CHECK3:       body:
// CHECK3-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4:[0-9]+]], i32 [[TMP2]])
// CHECK3-NEXT:    [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0
// CHECK3-NEXT:    br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]]
// CHECK3:       then:
// CHECK3-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP7]], i64 0, i64 0
// CHECK3-NEXT:    [[TMP11:%.*]] = load i8*, i8** [[TMP10]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP12:%.*]] = bitcast i8* [[TMP11]] to i32*
// CHECK3-NEXT:    [[TMP13:%.*]] = getelementptr i32, i32* [[TMP12]], i32 [[TMP8]]
// CHECK3-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]]
// CHECK3-NEXT:    [[TMP15:%.*]] = load i32, i32* [[TMP13]], align 4
// CHECK3-NEXT:    store volatile i32 [[TMP15]], i32 addrspace(3)* [[TMP14]], align 4
// CHECK3-NEXT:    br label [[IFCONT:%.*]]
// CHECK3:       else:
// CHECK3-NEXT:    br label [[IFCONT]]
// CHECK3:       ifcont:
// CHECK3-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]])
// CHECK3-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTADDR1]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[TMP3]], [[TMP16]]
// CHECK3-NEXT:    br i1 [[IS_ACTIVE_THREAD]], label [[THEN2:%.*]], label [[ELSE3:%.*]]
// CHECK3:       then2:
// CHECK3-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[TMP3]]
// CHECK3-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP7]], i64 0, i64 0
// CHECK3-NEXT:    [[TMP19:%.*]] = load i8*, i8** [[TMP18]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP20:%.*]] = bitcast i8* [[TMP19]] to i32*
// CHECK3-NEXT:    [[TMP21:%.*]] = getelementptr i32, i32* [[TMP20]], i32 [[TMP8]]
// CHECK3-NEXT:    [[TMP22:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP17]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    store i32 [[TMP22]], i32* [[TMP21]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    br label [[IFCONT4:%.*]]
// CHECK3:       else3:
// CHECK3-NEXT:    br label [[IFCONT4]]
// CHECK3:       ifcont4:
// CHECK3-NEXT:    [[TMP23:%.*]] = add nsw i32 [[TMP8]], 1
// CHECK3-NEXT:    store i32 [[TMP23]], i32* [[DOTCNT_ADDR]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    br label [[PRECOND]]
// CHECK3:       exit:
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@__omp_outlined__1_wrapper
// CHECK3-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR6]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
// CHECK3-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
// CHECK3-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2, !tbaa [[TBAA19]]
// CHECK3-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK3-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
// CHECK3-NEXT:    [[TMP2:%.*]] = load i8**, i8*** [[GLOBAL_ARGS]], align 8
// CHECK3-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 0
// CHECK3-NEXT:    [[TMP4:%.*]] = bitcast i8** [[TMP3]] to i32**
// CHECK3-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[TMP4]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 1
// CHECK3-NEXT:    [[TMP7:%.*]] = bitcast i8** [[TMP6]] to i32**
// CHECK3-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[TMP7]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 2
// CHECK3-NEXT:    [[TMP10:%.*]] = bitcast i8** [[TMP9]] to %"class.std::complex"**
// CHECK3-NEXT:    [[TMP11:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[TMP10]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    call void @__omp_outlined__1(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]], i32* [[TMP5]], i32* [[TMP8]], %"class.std::complex"* [[TMP11]]) #[[ATTR5]]
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z17complex_reductionIdEvv_l16
// CHECK3-SAME: () #[[ATTR0]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 1, i1 true, i1 true)
// CHECK3-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
// CHECK3-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK3:       user_code.entry:
// CHECK3-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK3-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    call void @__omp_outlined__2(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR5]]
// CHECK3-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 1, i1 true)
// CHECK3-NEXT:    ret void
// CHECK3:       worker.exit:
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@__omp_outlined__2
// CHECK3-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[IB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[REF_TMP:%.*]] = alloca double, align 8
// CHECK3-NEXT:    [[REF_TMP2:%.*]] = alloca double, align 8
// CHECK3-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x i8*], align 8
// CHECK3-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[ISTART:%.*]] = call align 16 i8* @__kmpc_alloc_shared(i64 4)
// CHECK3-NEXT:    [[ISTART_ON_STACK:%.*]] = bitcast i8* [[ISTART]] to i32*
// CHECK3-NEXT:    [[IEND:%.*]] = call align 16 i8* @__kmpc_alloc_shared(i64 4)
// CHECK3-NEXT:    [[IEND_ON_STACK:%.*]] = bitcast i8* [[IEND]] to i32*
// CHECK3-NEXT:    [[PARTIAL_SUM:%.*]] = call align 16 i8* @__kmpc_alloc_shared(i64 16)
// CHECK3-NEXT:    [[PARTIAL_SUM_ON_STACK:%.*]] = bitcast i8* [[PARTIAL_SUM]] to %"class.std::complex.0"*
// CHECK3-NEXT:    [[TMP0:%.*]] = bitcast i32* [[DOTOMP_IV]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP0]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP1:%.*]] = bitcast i32* [[DOTOMP_LB]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP1]]) #[[ATTR5]]
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP2:%.*]] = bitcast i32* [[DOTOMP_UB]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP2]]) #[[ATTR5]]
// CHECK3-NEXT:    store i32 99, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP3:%.*]] = bitcast i32* [[DOTOMP_STRIDE]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP3]]) #[[ATTR5]]
// CHECK3-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP4:%.*]] = bitcast i32* [[DOTOMP_IS_LAST]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP4]]) #[[ATTR5]]
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP5:%.*]] = bitcast i32* [[IB]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP5]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP6:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[TMP6]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP7]], i32 92, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP8]], 99
// CHECK3-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK3:       cond.true:
// CHECK3-NEXT:    br label [[COND_END:%.*]]
// CHECK3:       cond.false:
// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    br label [[COND_END]]
// CHECK3:       cond.end:
// CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP9]], [[COND_FALSE]] ]
// CHECK3-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    store i32 [[TMP10]], i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK3:       omp.inner.for.cond:
// CHECK3-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP11]], [[TMP12]]
// 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:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP13]], 1
// CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK3-NEXT:    store i32 [[ADD]], i32* [[IB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP14:%.*]] = bitcast double* [[REF_TMP]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 8, i8* [[TMP14]]) #[[ATTR5]]
// CHECK3-NEXT:    store double 0.000000e+00, double* [[REF_TMP]], align 8, !tbaa [[TBAA22:![0-9]+]]
// CHECK3-NEXT:    [[TMP15:%.*]] = bitcast double* [[REF_TMP2]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 8, i8* [[TMP15]]) #[[ATTR5]]
// CHECK3-NEXT:    store double 0.000000e+00, double* [[REF_TMP2]], align 8, !tbaa [[TBAA22]]
// CHECK3-NEXT:    call void @_ZNSt7complexIdEC1ERKdS2_(%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[PARTIAL_SUM_ON_STACK]], double* nonnull align 8 dereferenceable(8) [[REF_TMP]], double* nonnull align 8 dereferenceable(8) [[REF_TMP2]]) #[[ATTR10]]
// CHECK3-NEXT:    [[TMP16:%.*]] = bitcast double* [[REF_TMP2]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 8, i8* [[TMP16]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP17:%.*]] = bitcast double* [[REF_TMP]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 8, i8* [[TMP17]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP18:%.*]] = load i32, i32* [[IB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[MUL3:%.*]] = mul nsw i32 [[TMP18]], 4
// CHECK3-NEXT:    store i32 [[MUL3]], i32* [[ISTART_ON_STACK]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP19:%.*]] = load i32, i32* [[IB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK3-NEXT:    [[MUL5:%.*]] = mul nsw i32 [[ADD4]], 4
// CHECK3-NEXT:    store i32 [[MUL5]], i32* [[IEND_ON_STACK]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK3-NEXT:    [[TMP21:%.*]] = bitcast i32* [[ISTART_ON_STACK]] to i8*
// CHECK3-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK3-NEXT:    [[TMP23:%.*]] = bitcast i32* [[IEND_ON_STACK]] to i8*
// CHECK3-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK3-NEXT:    [[TMP25:%.*]] = bitcast %"class.std::complex.0"* [[PARTIAL_SUM_ON_STACK]] to i8*
// CHECK3-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP26:%.*]] = bitcast [3 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK3-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP7]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32*, i32*, %"class.std::complex.0"*)* @__omp_outlined__3 to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined__3_wrapper to i8*), i8** [[TMP26]], i64 3)
// 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:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP27]], 1
// CHECK3-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK3:       omp.inner.for.end:
// CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK3:       omp.loop.exit:
// CHECK3-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP7]])
// CHECK3-NEXT:    [[TMP28:%.*]] = bitcast i32* [[IB]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP28]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP29:%.*]] = bitcast i32* [[DOTOMP_IS_LAST]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP29]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP30:%.*]] = bitcast i32* [[DOTOMP_STRIDE]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP30]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP31:%.*]] = bitcast i32* [[DOTOMP_UB]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP31]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP32:%.*]] = bitcast i32* [[DOTOMP_LB]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP32]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP33:%.*]] = bitcast i32* [[DOTOMP_IV]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP33]]) #[[ATTR5]]
// CHECK3-NEXT:    call void @__kmpc_free_shared(i8* [[PARTIAL_SUM]], i64 16)
// CHECK3-NEXT:    call void @__kmpc_free_shared(i8* [[IEND]], i64 4)
// CHECK3-NEXT:    call void @__kmpc_free_shared(i8* [[ISTART]], i64 4)
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZNSt7complexIdEC1ERKdS2_
// CHECK3-SAME: (%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[THIS:%.*]], double* nonnull align 8 dereferenceable(8) [[__RE:%.*]], double* nonnull align 8 dereferenceable(8) [[__IM:%.*]]) unnamed_addr #[[ATTR3]] comdat align 2 {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex.0"*, align 8
// CHECK3-NEXT:    [[__RE_ADDR:%.*]] = alloca double*, align 8
// CHECK3-NEXT:    [[__IM_ADDR:%.*]] = alloca double*, align 8
// CHECK3-NEXT:    store %"class.std::complex.0"* [[THIS]], %"class.std::complex.0"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    store double* [[__RE]], double** [[__RE_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    store double* [[__IM]], double** [[__IM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[THIS1:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[THIS_ADDR]], align 8
// CHECK3-NEXT:    [[TMP0:%.*]] = load double*, double** [[__RE_ADDR]], align 8
// CHECK3-NEXT:    [[TMP1:%.*]] = load double*, double** [[__IM_ADDR]], align 8
// CHECK3-NEXT:    call void @_ZNSt7complexIdEC2ERKdS2_(%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[THIS1]], double* nonnull align 8 dereferenceable(8) [[TMP0]], double* nonnull align 8 dereferenceable(8) [[TMP1]]) #[[ATTR10]]
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@__omp_outlined__3
// CHECK3-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32* nonnull align 4 dereferenceable(4) [[ISTART:%.*]], i32* nonnull align 4 dereferenceable(4) [[IEND:%.*]], %"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[PARTIAL_SUM:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK3-NEXT:    [[ISTART_ADDR:%.*]] = alloca i32*, align 8
// CHECK3-NEXT:    [[IEND_ADDR:%.*]] = alloca i32*, align 8
// CHECK3-NEXT:    [[PARTIAL_SUM_ADDR:%.*]] = alloca %"class.std::complex.0"*, align 8
// CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[PARTIAL_SUM5:%.*]] = alloca %"class.std::complex.0", align 8
// CHECK3-NEXT:    [[REF_TMP:%.*]] = alloca double, align 8
// CHECK3-NEXT:    [[REF_TMP6:%.*]] = alloca double, align 8
// CHECK3-NEXT:    [[I7:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[REF_TMP14:%.*]] = alloca %"class.std::complex.0", align 8
// CHECK3-NEXT:    [[REF_TMP15:%.*]] = alloca double, align 8
// CHECK3-NEXT:    [[REF_TMP16:%.*]] = alloca double, align 8
// CHECK3-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x i8*], align 8
// CHECK3-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    store i32* [[ISTART]], i32** [[ISTART_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    store i32* [[IEND]], i32** [[IEND_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    store %"class.std::complex.0"* [[PARTIAL_SUM]], %"class.std::complex.0"** [[PARTIAL_SUM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[ISTART_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[IEND_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP2:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[PARTIAL_SUM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP3:%.*]] = bitcast i32* [[DOTOMP_IV]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP3]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP4:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP4]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[TMP0]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    store i32 [[TMP5]], i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP6:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_1]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP6]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[TMP1]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    store i32 [[TMP7]], i32* [[DOTCAPTURE_EXPR_1]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP8:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_2]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP8]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[SUB:%.*]] = sub i32 [[TMP9]], [[TMP10]]
// CHECK3-NEXT:    [[SUB3:%.*]] = sub i32 [[SUB]], 1
// CHECK3-NEXT:    [[ADD:%.*]] = add i32 [[SUB3]], 1
// CHECK3-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], 1
// CHECK3-NEXT:    [[SUB4:%.*]] = sub i32 [[DIV]], 1
// CHECK3-NEXT:    store i32 [[SUB4]], i32* [[DOTCAPTURE_EXPR_2]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP11:%.*]] = bitcast i32* [[I]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP11]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    store i32 [[TMP12]], i32* [[I]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP13:%.*]] = bitcast i32* [[I]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP13]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP14]], [[TMP15]]
// CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK3:       omp.precond.then:
// CHECK3-NEXT:    [[TMP16:%.*]] = bitcast i32* [[DOTOMP_LB]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP16]]) #[[ATTR5]]
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP17:%.*]] = bitcast i32* [[DOTOMP_UB]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP17]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    store i32 [[TMP18]], i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP19:%.*]] = bitcast i32* [[DOTOMP_STRIDE]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP19]]) #[[ATTR5]]
// CHECK3-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP20:%.*]] = bitcast i32* [[DOTOMP_IS_LAST]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP20]]) #[[ATTR5]]
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP21:%.*]] = bitcast %"class.std::complex.0"* [[PARTIAL_SUM5]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 16, i8* [[TMP21]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP22:%.*]] = bitcast double* [[REF_TMP]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 8, i8* [[TMP22]]) #[[ATTR5]]
// CHECK3-NEXT:    store double 0.000000e+00, double* [[REF_TMP]], align 8, !tbaa [[TBAA22]]
// CHECK3-NEXT:    [[TMP23:%.*]] = bitcast double* [[REF_TMP6]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 8, i8* [[TMP23]]) #[[ATTR5]]
// CHECK3-NEXT:    store double 0.000000e+00, double* [[REF_TMP6]], align 8, !tbaa [[TBAA22]]
// CHECK3-NEXT:    call void @_ZNSt7complexIdEC1ERKdS2_(%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[PARTIAL_SUM5]], double* nonnull align 8 dereferenceable(8) [[REF_TMP]], double* nonnull align 8 dereferenceable(8) [[REF_TMP6]]) #[[ATTR10]]
// CHECK3-NEXT:    [[TMP24:%.*]] = bitcast double* [[REF_TMP6]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 8, i8* [[TMP24]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP25:%.*]] = bitcast double* [[REF_TMP]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 8, i8* [[TMP25]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP26:%.*]] = bitcast i32* [[I7]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* [[TMP26]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP27:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK3-NEXT:    [[TMP28:%.*]] = load i32, i32* [[TMP27]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    call void @__kmpc_for_static_init_4u(%struct.ident_t* @[[GLOB3]], i32 [[TMP28]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK3-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK3:       omp.dispatch.cond:
// CHECK3-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[CMP8:%.*]] = icmp ugt i32 [[TMP29]], [[TMP30]]
// CHECK3-NEXT:    br i1 [[CMP8]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK3:       cond.true:
// CHECK3-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    br label [[COND_END:%.*]]
// CHECK3:       cond.false:
// CHECK3-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    br label [[COND_END]]
// CHECK3:       cond.end:
// CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP31]], [[COND_TRUE]] ], [ [[TMP32]], [[COND_FALSE]] ]
// CHECK3-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    store i32 [[TMP33]], i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[ADD9:%.*]] = add i32 [[TMP35]], 1
// CHECK3-NEXT:    [[CMP10:%.*]] = icmp ult i32 [[TMP34]], [[ADD9]]
// CHECK3-NEXT:    br i1 [[CMP10]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_CLEANUP:%.*]]
// CHECK3:       omp.dispatch.cleanup:
// CHECK3-NEXT:    br label [[OMP_DISPATCH_END:%.*]]
// CHECK3:       omp.dispatch.body:
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK3:       omp.inner.for.cond:
// CHECK3-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[ADD11:%.*]] = add i32 [[TMP37]], 1
// CHECK3-NEXT:    [[CMP12:%.*]] = icmp ult i32 [[TMP36]], [[ADD11]]
// CHECK3-NEXT:    br i1 [[CMP12]], 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:    [[TMP38:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[MUL:%.*]] = mul i32 [[TMP39]], 1
// CHECK3-NEXT:    [[ADD13:%.*]] = add i32 [[TMP38]], [[MUL]]
// CHECK3-NEXT:    store i32 [[ADD13]], i32* [[I7]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP40:%.*]] = bitcast %"class.std::complex.0"* [[REF_TMP14]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 16, i8* [[TMP40]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP41:%.*]] = bitcast double* [[REF_TMP15]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 8, i8* [[TMP41]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP42:%.*]] = load i32, i32* [[I7]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[CONV:%.*]] = sitofp i32 [[TMP42]] to double
// CHECK3-NEXT:    store double [[CONV]], double* [[REF_TMP15]], align 8, !tbaa [[TBAA22]]
// CHECK3-NEXT:    [[TMP43:%.*]] = bitcast double* [[REF_TMP16]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.start.p0i8(i64 8, i8* [[TMP43]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP44:%.*]] = load i32, i32* [[I7]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[CONV17:%.*]] = sitofp i32 [[TMP44]] to double
// CHECK3-NEXT:    store double [[CONV17]], double* [[REF_TMP16]], align 8, !tbaa [[TBAA22]]
// CHECK3-NEXT:    call void @_ZNSt7complexIdEC1ERKdS2_(%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[REF_TMP14]], double* nonnull align 8 dereferenceable(8) [[REF_TMP15]], double* nonnull align 8 dereferenceable(8) [[REF_TMP16]]) #[[ATTR10]]
// CHECK3-NEXT:    [[CALL:%.*]] = call nonnull align 8 dereferenceable(16) %"class.std::complex.0"* @_ZNSt7complexIdEpLIdEERS0_RKS_IT_E(%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[PARTIAL_SUM5]], %"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[REF_TMP14]]) #[[ATTR10]]
// CHECK3-NEXT:    [[TMP45:%.*]] = bitcast double* [[REF_TMP16]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 8, i8* [[TMP45]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP46:%.*]] = bitcast double* [[REF_TMP15]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 8, i8* [[TMP46]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP47:%.*]] = bitcast %"class.std::complex.0"* [[REF_TMP14]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 16, i8* [[TMP47]]) #[[ATTR5]]
// 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:    [[TMP48:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[ADD18:%.*]] = add i32 [[TMP48]], 1
// CHECK3-NEXT:    store i32 [[ADD18]], i32* [[DOTOMP_IV]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK3:       omp.inner.for.end:
// CHECK3-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK3:       omp.dispatch.inc:
// CHECK3-NEXT:    [[TMP49:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP50:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[ADD19:%.*]] = add i32 [[TMP49]], [[TMP50]]
// CHECK3-NEXT:    store i32 [[ADD19]], i32* [[DOTOMP_LB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP51:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP52:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[ADD20:%.*]] = add i32 [[TMP51]], [[TMP52]]
// CHECK3-NEXT:    store i32 [[ADD20]], i32* [[DOTOMP_UB]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK3:       omp.dispatch.end:
// CHECK3-NEXT:    [[TMP53:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK3-NEXT:    [[TMP54:%.*]] = load i32, i32* [[TMP53]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB3]], i32 [[TMP54]])
// CHECK3-NEXT:    [[TMP55:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK3-NEXT:    [[TMP56:%.*]] = load i32, i32* [[TMP55]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP57:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0
// CHECK3-NEXT:    [[TMP58:%.*]] = bitcast %"class.std::complex.0"* [[PARTIAL_SUM5]] to i8*
// CHECK3-NEXT:    store i8* [[TMP58]], i8** [[TMP57]], align 8
// CHECK3-NEXT:    [[TMP59:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8*
// CHECK3-NEXT:    [[TMP60:%.*]] = call i32 @__kmpc_nvptx_parallel_reduce_nowait_v2(%struct.ident_t* @[[GLOB1]], i32 [[TMP56]], i32 1, i64 8, i8* [[TMP59]], void (i8*, i16, i16, i16)* @_omp_reduction_shuffle_and_reduce_func5, void (i8*, i32)* @_omp_reduction_inter_warp_copy_func6)
// CHECK3-NEXT:    [[TMP61:%.*]] = icmp eq i32 [[TMP60]], 1
// CHECK3-NEXT:    br i1 [[TMP61]], label [[DOTOMP_REDUCTION_THEN:%.*]], label [[DOTOMP_REDUCTION_DONE:%.*]]
// CHECK3:       .omp.reduction.then:
// CHECK3-NEXT:    [[CALL21:%.*]] = call nonnull align 8 dereferenceable(16) %"class.std::complex.0"* @_ZNSt7complexIdEpLIdEERS0_RKS_IT_E(%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[TMP2]], %"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[PARTIAL_SUM5]]) #[[ATTR10]]
// CHECK3-NEXT:    call void @__kmpc_nvptx_end_reduce_nowait(i32 [[TMP56]])
// CHECK3-NEXT:    br label [[DOTOMP_REDUCTION_DONE]]
// CHECK3:       .omp.reduction.done:
// CHECK3-NEXT:    [[TMP62:%.*]] = bitcast i32* [[I7]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP62]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP63:%.*]] = bitcast %"class.std::complex.0"* [[PARTIAL_SUM5]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 16, i8* [[TMP63]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP64:%.*]] = bitcast i32* [[DOTOMP_IS_LAST]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP64]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP65:%.*]] = bitcast i32* [[DOTOMP_STRIDE]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP65]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP66:%.*]] = bitcast i32* [[DOTOMP_UB]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP66]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP67:%.*]] = bitcast i32* [[DOTOMP_LB]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP67]]) #[[ATTR5]]
// CHECK3-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK3:       omp.precond.end:
// CHECK3-NEXT:    [[TMP68:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_2]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP68]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP69:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_1]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP69]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP70:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR_]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP70]]) #[[ATTR5]]
// CHECK3-NEXT:    [[TMP71:%.*]] = bitcast i32* [[DOTOMP_IV]] to i8*
// CHECK3-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* [[TMP71]]) #[[ATTR5]]
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZNSt7complexIdEpLIdEERS0_RKS_IT_E
// CHECK3-SAME: (%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[THIS:%.*]], %"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[__C:%.*]]) #[[ATTR4]] comdat align 2 {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex.0"*, align 8
// CHECK3-NEXT:    [[__C_ADDR:%.*]] = alloca %"class.std::complex.0"*, align 8
// CHECK3-NEXT:    store %"class.std::complex.0"* [[THIS]], %"class.std::complex.0"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    store %"class.std::complex.0"* [[__C]], %"class.std::complex.0"** [[__C_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[THIS1:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[THIS_ADDR]], align 8
// CHECK3-NEXT:    [[TMP0:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[__C_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[CALL:%.*]] = call double @_ZNKSt7complexIdE4realEv(%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[TMP0]]) #[[ATTR10]]
// CHECK3-NEXT:    [[__RE_:%.*]] = getelementptr inbounds %"class.std::complex.0", %"class.std::complex.0"* [[THIS1]], i32 0, i32 0
// CHECK3-NEXT:    [[TMP1:%.*]] = load double, double* [[__RE_]], align 8, !tbaa [[TBAA24:![0-9]+]]
// CHECK3-NEXT:    [[ADD:%.*]] = fadd double [[TMP1]], [[CALL]]
// CHECK3-NEXT:    store double [[ADD]], double* [[__RE_]], align 8, !tbaa [[TBAA24]]
// CHECK3-NEXT:    [[TMP2:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[__C_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[CALL2:%.*]] = call double @_ZNKSt7complexIdE4imagEv(%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[TMP2]]) #[[ATTR10]]
// CHECK3-NEXT:    [[__IM_:%.*]] = getelementptr inbounds %"class.std::complex.0", %"class.std::complex.0"* [[THIS1]], i32 0, i32 1
// CHECK3-NEXT:    [[TMP3:%.*]] = load double, double* [[__IM_]], align 8, !tbaa [[TBAA26:![0-9]+]]
// CHECK3-NEXT:    [[ADD3:%.*]] = fadd double [[TMP3]], [[CALL2]]
// CHECK3-NEXT:    store double [[ADD3]], double* [[__IM_]], align 8, !tbaa [[TBAA26]]
// CHECK3-NEXT:    ret %"class.std::complex.0"* [[THIS1]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_omp_reduction_shuffle_and_reduce_func5
// CHECK3-SAME: (i8* [[TMP0:%.*]], i16 signext [[TMP1:%.*]], i16 signext [[TMP2:%.*]], i16 signext [[TMP3:%.*]]) #[[ATTR6]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK3-NEXT:    [[DOTADDR1:%.*]] = alloca i16, align 2
// CHECK3-NEXT:    [[DOTADDR2:%.*]] = alloca i16, align 2
// CHECK3-NEXT:    [[DOTADDR3:%.*]] = alloca i16, align 2
// CHECK3-NEXT:    [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST:%.*]] = alloca [1 x i8*], align 8
// CHECK3-NEXT:    [[DOTOMP_REDUCTION_ELEMENT:%.*]] = alloca %"class.std::complex.0", align 8
// CHECK3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    store i16 [[TMP1]], i16* [[DOTADDR1]], align 2, !tbaa [[TBAA19]]
// CHECK3-NEXT:    store i16 [[TMP2]], i16* [[DOTADDR2]], align 2, !tbaa [[TBAA19]]
// CHECK3-NEXT:    store i16 [[TMP3]], i16* [[DOTADDR3]], align 2, !tbaa [[TBAA19]]
// CHECK3-NEXT:    [[TMP4:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [1 x i8*]*
// CHECK3-NEXT:    [[TMP6:%.*]] = load i16, i16* [[DOTADDR1]], align 2, !tbaa [[TBAA19]]
// CHECK3-NEXT:    [[TMP7:%.*]] = load i16, i16* [[DOTADDR2]], align 2, !tbaa [[TBAA19]]
// CHECK3-NEXT:    [[TMP8:%.*]] = load i16, i16* [[DOTADDR3]], align 2, !tbaa [[TBAA19]]
// CHECK3-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK3-NEXT:    [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8
// CHECK3-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0
// CHECK3-NEXT:    [[TMP12:%.*]] = bitcast i8* [[TMP10]] to %"class.std::complex.0"*
// CHECK3-NEXT:    [[TMP13:%.*]] = getelementptr %"class.std::complex.0", %"class.std::complex.0"* [[TMP12]], i64 1
// CHECK3-NEXT:    [[TMP14:%.*]] = bitcast %"class.std::complex.0"* [[TMP13]] to i8*
// CHECK3-NEXT:    [[TMP15:%.*]] = bitcast %"class.std::complex.0"* [[TMP12]] to i64*
// CHECK3-NEXT:    [[TMP16:%.*]] = bitcast %"class.std::complex.0"* [[DOTOMP_REDUCTION_ELEMENT]] to i64*
// CHECK3-NEXT:    br label [[DOTSHUFFLE_PRE_COND:%.*]]
// CHECK3:       .shuffle.pre_cond:
// CHECK3-NEXT:    [[TMP17:%.*]] = phi i64* [ [[TMP15]], [[ENTRY:%.*]] ], [ [[TMP29:%.*]], [[DOTSHUFFLE_THEN:%.*]] ]
// CHECK3-NEXT:    [[TMP18:%.*]] = phi i64* [ [[TMP16]], [[ENTRY]] ], [ [[TMP30:%.*]], [[DOTSHUFFLE_THEN]] ]
// CHECK3-NEXT:    [[TMP19:%.*]] = bitcast i64* [[TMP17]] to i8*
// CHECK3-NEXT:    [[TMP20:%.*]] = ptrtoint i8* [[TMP14]] to i64
// CHECK3-NEXT:    [[TMP21:%.*]] = ptrtoint i8* [[TMP19]] to i64
// CHECK3-NEXT:    [[TMP22:%.*]] = sub i64 [[TMP20]], [[TMP21]]
// CHECK3-NEXT:    [[TMP23:%.*]] = sdiv exact i64 [[TMP22]], ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)
// CHECK3-NEXT:    [[TMP24:%.*]] = icmp sgt i64 [[TMP23]], 7
// CHECK3-NEXT:    br i1 [[TMP24]], label [[DOTSHUFFLE_THEN]], label [[DOTSHUFFLE_EXIT:%.*]]
// CHECK3:       .shuffle.then:
// CHECK3-NEXT:    [[TMP25:%.*]] = load i64, i64* [[TMP17]], align 8
// CHECK3-NEXT:    [[TMP26:%.*]] = call i32 @__kmpc_get_warp_size()
// CHECK3-NEXT:    [[TMP27:%.*]] = trunc i32 [[TMP26]] to i16
// CHECK3-NEXT:    [[TMP28:%.*]] = call i64 @__kmpc_shuffle_int64(i64 [[TMP25]], i16 [[TMP7]], i16 [[TMP27]])
// CHECK3-NEXT:    store i64 [[TMP28]], i64* [[TMP18]], align 8
// CHECK3-NEXT:    [[TMP29]] = getelementptr i64, i64* [[TMP17]], i64 1
// CHECK3-NEXT:    [[TMP30]] = getelementptr i64, i64* [[TMP18]], i64 1
// CHECK3-NEXT:    br label [[DOTSHUFFLE_PRE_COND]]
// CHECK3:       .shuffle.exit:
// CHECK3-NEXT:    [[TMP31:%.*]] = bitcast %"class.std::complex.0"* [[DOTOMP_REDUCTION_ELEMENT]] to i8*
// CHECK3-NEXT:    store i8* [[TMP31]], i8** [[TMP11]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP32:%.*]] = icmp eq i16 [[TMP8]], 0
// CHECK3-NEXT:    [[TMP33:%.*]] = icmp eq i16 [[TMP8]], 1
// CHECK3-NEXT:    [[TMP34:%.*]] = icmp ult i16 [[TMP6]], [[TMP7]]
// CHECK3-NEXT:    [[TMP35:%.*]] = and i1 [[TMP33]], [[TMP34]]
// CHECK3-NEXT:    [[TMP36:%.*]] = icmp eq i16 [[TMP8]], 2
// CHECK3-NEXT:    [[TMP37:%.*]] = and i16 [[TMP6]], 1
// CHECK3-NEXT:    [[TMP38:%.*]] = icmp eq i16 [[TMP37]], 0
// CHECK3-NEXT:    [[TMP39:%.*]] = and i1 [[TMP36]], [[TMP38]]
// CHECK3-NEXT:    [[TMP40:%.*]] = icmp sgt i16 [[TMP7]], 0
// CHECK3-NEXT:    [[TMP41:%.*]] = and i1 [[TMP39]], [[TMP40]]
// CHECK3-NEXT:    [[TMP42:%.*]] = or i1 [[TMP32]], [[TMP35]]
// CHECK3-NEXT:    [[TMP43:%.*]] = or i1 [[TMP42]], [[TMP41]]
// CHECK3-NEXT:    br i1 [[TMP43]], label [[THEN:%.*]], label [[ELSE:%.*]]
// CHECK3:       then:
// CHECK3-NEXT:    [[TMP44:%.*]] = bitcast [1 x i8*]* [[TMP5]] to i8*
// CHECK3-NEXT:    [[TMP45:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]] to i8*
// CHECK3-NEXT:    call void @"_omp$reduction$reduction_func4"(i8* [[TMP44]], i8* [[TMP45]]) #[[ATTR5]]
// CHECK3-NEXT:    br label [[IFCONT:%.*]]
// CHECK3:       else:
// CHECK3-NEXT:    br label [[IFCONT]]
// CHECK3:       ifcont:
// CHECK3-NEXT:    [[TMP46:%.*]] = icmp eq i16 [[TMP8]], 1
// CHECK3-NEXT:    [[TMP47:%.*]] = icmp uge i16 [[TMP6]], [[TMP7]]
// CHECK3-NEXT:    [[TMP48:%.*]] = and i1 [[TMP46]], [[TMP47]]
// CHECK3-NEXT:    br i1 [[TMP48]], label [[THEN4:%.*]], label [[ELSE5:%.*]]
// CHECK3:       then4:
// CHECK3-NEXT:    [[TMP49:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_REDUCTION_REMOTE_REDUCE_LIST]], i64 0, i64 0
// CHECK3-NEXT:    [[TMP50:%.*]] = load i8*, i8** [[TMP49]], align 8
// CHECK3-NEXT:    [[TMP51:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK3-NEXT:    [[TMP52:%.*]] = load i8*, i8** [[TMP51]], align 8
// CHECK3-NEXT:    [[TMP53:%.*]] = bitcast i8* [[TMP50]] to %"class.std::complex.0"*
// CHECK3-NEXT:    [[TMP54:%.*]] = bitcast i8* [[TMP52]] to %"class.std::complex.0"*
// CHECK3-NEXT:    [[TMP55:%.*]] = bitcast %"class.std::complex.0"* [[TMP54]] to i8*
// CHECK3-NEXT:    [[TMP56:%.*]] = bitcast %"class.std::complex.0"* [[TMP53]] to i8*
// CHECK3-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[TMP55]], i8* align 8 [[TMP56]], i64 16, i1 false), !tbaa.struct !27
// CHECK3-NEXT:    br label [[IFCONT6:%.*]]
// CHECK3:       else5:
// CHECK3-NEXT:    br label [[IFCONT6]]
// CHECK3:       ifcont6:
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_omp_reduction_inter_warp_copy_func6
// CHECK3-SAME: (i8* [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR6]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK3-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCNT_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK3-NEXT:    store i8* [[TMP0]], i8** [[DOTADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK3-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK3-NEXT:    [[NVPTX_LANE_ID:%.*]] = and i32 [[TMP4]], 31
// CHECK3-NEXT:    [[TMP5:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK3-NEXT:    [[NVPTX_WARP_ID:%.*]] = ashr i32 [[TMP5]], 5
// CHECK3-NEXT:    [[TMP6:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK3-NEXT:    [[TMP7:%.*]] = bitcast i8* [[TMP6]] to [1 x i8*]*
// CHECK3-NEXT:    store i32 0, i32* [[DOTCNT_ADDR]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    br label [[PRECOND:%.*]]
// CHECK3:       precond:
// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCNT_ADDR]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[TMP9:%.*]] = icmp ult i32 [[TMP8]], 4
// CHECK3-NEXT:    br i1 [[TMP9]], label [[BODY:%.*]], label [[EXIT:%.*]]
// CHECK3:       body:
// CHECK3-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]])
// CHECK3-NEXT:    [[WARP_MASTER:%.*]] = icmp eq i32 [[NVPTX_LANE_ID]], 0
// CHECK3-NEXT:    br i1 [[WARP_MASTER]], label [[THEN:%.*]], label [[ELSE:%.*]]
// CHECK3:       then:
// CHECK3-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP7]], i64 0, i64 0
// CHECK3-NEXT:    [[TMP11:%.*]] = load i8*, i8** [[TMP10]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP12:%.*]] = bitcast i8* [[TMP11]] to i32*
// CHECK3-NEXT:    [[TMP13:%.*]] = getelementptr i32, i32* [[TMP12]], i32 [[TMP8]]
// CHECK3-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[NVPTX_WARP_ID]]
// CHECK3-NEXT:    [[TMP15:%.*]] = load i32, i32* [[TMP13]], align 4
// CHECK3-NEXT:    store volatile i32 [[TMP15]], i32 addrspace(3)* [[TMP14]], align 4
// CHECK3-NEXT:    br label [[IFCONT:%.*]]
// CHECK3:       else:
// CHECK3-NEXT:    br label [[IFCONT]]
// CHECK3:       ifcont:
// CHECK3-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]])
// CHECK3-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTADDR1]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    [[IS_ACTIVE_THREAD:%.*]] = icmp ult i32 [[TMP3]], [[TMP16]]
// CHECK3-NEXT:    br i1 [[IS_ACTIVE_THREAD]], label [[THEN2:%.*]], label [[ELSE3:%.*]]
// CHECK3:       then2:
// CHECK3-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [32 x i32], [32 x i32] addrspace(3)* @__openmp_nvptx_data_transfer_temporary_storage, i64 0, i32 [[TMP3]]
// CHECK3-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP7]], i64 0, i64 0
// CHECK3-NEXT:    [[TMP19:%.*]] = load i8*, i8** [[TMP18]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP20:%.*]] = bitcast i8* [[TMP19]] to i32*
// CHECK3-NEXT:    [[TMP21:%.*]] = getelementptr i32, i32* [[TMP20]], i32 [[TMP8]]
// CHECK3-NEXT:    [[TMP22:%.*]] = load volatile i32, i32 addrspace(3)* [[TMP17]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    store i32 [[TMP22]], i32* [[TMP21]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    br label [[IFCONT4:%.*]]
// CHECK3:       else3:
// CHECK3-NEXT:    br label [[IFCONT4]]
// CHECK3:       ifcont4:
// CHECK3-NEXT:    [[TMP23:%.*]] = add nsw i32 [[TMP8]], 1
// CHECK3-NEXT:    store i32 [[TMP23]], i32* [[DOTCNT_ADDR]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    br label [[PRECOND]]
// CHECK3:       exit:
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@__omp_outlined__3_wrapper
// CHECK3-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR6]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
// CHECK3-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
// CHECK3-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2, !tbaa [[TBAA19]]
// CHECK3-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4, !tbaa [[TBAA8]]
// CHECK3-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK3-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
// CHECK3-NEXT:    [[TMP2:%.*]] = load i8**, i8*** [[GLOBAL_ARGS]], align 8
// CHECK3-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 0
// CHECK3-NEXT:    [[TMP4:%.*]] = bitcast i8** [[TMP3]] to i32**
// CHECK3-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[TMP4]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 1
// CHECK3-NEXT:    [[TMP7:%.*]] = bitcast i8** [[TMP6]] to i32**
// CHECK3-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[TMP7]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 2
// CHECK3-NEXT:    [[TMP10:%.*]] = bitcast i8** [[TMP9]] to %"class.std::complex.0"**
// CHECK3-NEXT:    [[TMP11:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[TMP10]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    call void @__omp_outlined__3(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]], i32* [[TMP5]], i32* [[TMP8]], %"class.std::complex.0"* [[TMP11]]) #[[ATTR5]]
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZNSt7complexIfEC2ERKfS2_
// CHECK3-SAME: (%"class.std::complex"* nonnull align 4 dereferenceable(8) [[THIS:%.*]], float* nonnull align 4 dereferenceable(4) [[__RE:%.*]], float* nonnull align 4 dereferenceable(4) [[__IM:%.*]]) unnamed_addr #[[ATTR3]] comdat align 2 {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex"*, align 8
// CHECK3-NEXT:    [[__RE_ADDR:%.*]] = alloca float*, align 8
// CHECK3-NEXT:    [[__IM_ADDR:%.*]] = alloca float*, align 8
// CHECK3-NEXT:    store %"class.std::complex"* [[THIS]], %"class.std::complex"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    store float* [[__RE]], float** [[__RE_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    store float* [[__IM]], float** [[__IM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[THIS1:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[THIS_ADDR]], align 8
// CHECK3-NEXT:    [[__RE_:%.*]] = getelementptr inbounds %"class.std::complex", %"class.std::complex"* [[THIS1]], i32 0, i32 0
// CHECK3-NEXT:    [[TMP0:%.*]] = load float*, float** [[__RE_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP1:%.*]] = load float, float* [[TMP0]], align 4, !tbaa [[TBAA14]]
// CHECK3-NEXT:    store float [[TMP1]], float* [[__RE_]], align 4, !tbaa [[TBAA16]]
// CHECK3-NEXT:    [[__IM_:%.*]] = getelementptr inbounds %"class.std::complex", %"class.std::complex"* [[THIS1]], i32 0, i32 1
// CHECK3-NEXT:    [[TMP2:%.*]] = load float*, float** [[__IM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP3:%.*]] = load float, float* [[TMP2]], align 4, !tbaa [[TBAA14]]
// CHECK3-NEXT:    store float [[TMP3]], float* [[__IM_]], align 4, !tbaa [[TBAA18]]
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZNKSt7complexIfE4realEv
// CHECK3-SAME: (%"class.std::complex"* nonnull align 4 dereferenceable(8) [[THIS:%.*]]) #[[ATTR4]] comdat align 2 {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex"*, align 8
// CHECK3-NEXT:    store %"class.std::complex"* [[THIS]], %"class.std::complex"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[THIS1:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[THIS_ADDR]], align 8
// CHECK3-NEXT:    [[__RE_:%.*]] = getelementptr inbounds %"class.std::complex", %"class.std::complex"* [[THIS1]], i32 0, i32 0
// CHECK3-NEXT:    [[TMP0:%.*]] = load float, float* [[__RE_]], align 4, !tbaa [[TBAA16]]
// CHECK3-NEXT:    ret float [[TMP0]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZNKSt7complexIfE4imagEv
// CHECK3-SAME: (%"class.std::complex"* nonnull align 4 dereferenceable(8) [[THIS:%.*]]) #[[ATTR4]] comdat align 2 {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex"*, align 8
// CHECK3-NEXT:    store %"class.std::complex"* [[THIS]], %"class.std::complex"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[THIS1:%.*]] = load %"class.std::complex"*, %"class.std::complex"** [[THIS_ADDR]], align 8
// CHECK3-NEXT:    [[__IM_:%.*]] = getelementptr inbounds %"class.std::complex", %"class.std::complex"* [[THIS1]], i32 0, i32 1
// CHECK3-NEXT:    [[TMP0:%.*]] = load float, float* [[__IM_]], align 4, !tbaa [[TBAA18]]
// CHECK3-NEXT:    ret float [[TMP0]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZNSt7complexIdEC2ERKdS2_
// CHECK3-SAME: (%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[THIS:%.*]], double* nonnull align 8 dereferenceable(8) [[__RE:%.*]], double* nonnull align 8 dereferenceable(8) [[__IM:%.*]]) unnamed_addr #[[ATTR3]] comdat align 2 {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex.0"*, align 8
// CHECK3-NEXT:    [[__RE_ADDR:%.*]] = alloca double*, align 8
// CHECK3-NEXT:    [[__IM_ADDR:%.*]] = alloca double*, align 8
// CHECK3-NEXT:    store %"class.std::complex.0"* [[THIS]], %"class.std::complex.0"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    store double* [[__RE]], double** [[__RE_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    store double* [[__IM]], double** [[__IM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[THIS1:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[THIS_ADDR]], align 8
// CHECK3-NEXT:    [[__RE_:%.*]] = getelementptr inbounds %"class.std::complex.0", %"class.std::complex.0"* [[THIS1]], i32 0, i32 0
// CHECK3-NEXT:    [[TMP0:%.*]] = load double*, double** [[__RE_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP1:%.*]] = load double, double* [[TMP0]], align 8, !tbaa [[TBAA22]]
// CHECK3-NEXT:    store double [[TMP1]], double* [[__RE_]], align 8, !tbaa [[TBAA24]]
// CHECK3-NEXT:    [[__IM_:%.*]] = getelementptr inbounds %"class.std::complex.0", %"class.std::complex.0"* [[THIS1]], i32 0, i32 1
// CHECK3-NEXT:    [[TMP2:%.*]] = load double*, double** [[__IM_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[TMP3:%.*]] = load double, double* [[TMP2]], align 8, !tbaa [[TBAA22]]
// CHECK3-NEXT:    store double [[TMP3]], double* [[__IM_]], align 8, !tbaa [[TBAA26]]
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZNKSt7complexIdE4realEv
// CHECK3-SAME: (%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[THIS:%.*]]) #[[ATTR4]] comdat align 2 {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex.0"*, align 8
// CHECK3-NEXT:    store %"class.std::complex.0"* [[THIS]], %"class.std::complex.0"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[THIS1:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[THIS_ADDR]], align 8
// CHECK3-NEXT:    [[__RE_:%.*]] = getelementptr inbounds %"class.std::complex.0", %"class.std::complex.0"* [[THIS1]], i32 0, i32 0
// CHECK3-NEXT:    [[TMP0:%.*]] = load double, double* [[__RE_]], align 8, !tbaa [[TBAA24]]
// CHECK3-NEXT:    ret double [[TMP0]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZNKSt7complexIdE4imagEv
// CHECK3-SAME: (%"class.std::complex.0"* nonnull align 8 dereferenceable(16) [[THIS:%.*]]) #[[ATTR4]] comdat align 2 {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[THIS_ADDR:%.*]] = alloca %"class.std::complex.0"*, align 8
// CHECK3-NEXT:    store %"class.std::complex.0"* [[THIS]], %"class.std::complex.0"** [[THIS_ADDR]], align 8, !tbaa [[TBAA12]]
// CHECK3-NEXT:    [[THIS1:%.*]] = load %"class.std::complex.0"*, %"class.std::complex.0"** [[THIS_ADDR]], align 8
// CHECK3-NEXT:    [[__IM_:%.*]] = getelementptr inbounds %"class.std::complex.0", %"class.std::complex.0"* [[THIS1]], i32 0, i32 1
// CHECK3-NEXT:    [[TMP0:%.*]] = load double, double* [[__IM_]], align 8, !tbaa [[TBAA26]]
// CHECK3-NEXT:    ret double [[TMP0]]
//