// 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 _
// Test target codegen - host bc file has to be created first.
// RUN: %clang_cc1 -no-enable-noundef-analysis -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc
// RUN: %clang_cc1 -no-enable-noundef-analysis -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple nvptx64-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=CHECK45-64
// RUN: %clang_cc1 -no-enable-noundef-analysis -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple i386-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm-bc %s -o %t-x86-host.bc
// RUN: %clang_cc1 -no-enable-noundef-analysis -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix=CHECK45-32
// RUN: %clang_cc1 -no-enable-noundef-analysis -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -fexceptions -fcxx-exceptions -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix=CHECK45-32-EX

// RUN: %clang_cc1 -no-enable-noundef-analysis -verify -fopenmp -fopenmp-cuda-mode -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc
// RUN: %clang_cc1 -no-enable-noundef-analysis -verify -fopenmp -fopenmp-cuda-mode -x c++ -triple nvptx64-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=CHECK-64
// RUN: %clang_cc1 -no-enable-noundef-analysis -verify -fopenmp -fopenmp-cuda-mode -x c++ -triple i386-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm-bc %s -o %t-x86-host.bc
// RUN: %clang_cc1 -no-enable-noundef-analysis -verify -fopenmp -fopenmp-cuda-mode -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix=CHECK-32
// RUN: %clang_cc1 -no-enable-noundef-analysis -verify -fopenmp -fopenmp-cuda-mode -fexceptions -fcxx-exceptions -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix=CHECK-32-EX

// expected-no-diagnostics
#ifndef HEADER
#define HEADER

// Check that the execution mode of all 2 target regions on the gpu is set to NonSPMD Mode.

#define N 1000
#define M 10

template<typename tx>
tx ftemplate(int n) {
  tx a[N];
  short aa[N];
  tx b[10];
  tx c[M][M];
  tx f = n;
  tx l;
  int k;

#pragma omp target teams distribute simd lastprivate(l) dist_schedule(static,128)
  for(int i = 0; i < n; i++) {
    a[i] = 1;
    l = i;
  }

  #pragma omp target teams distribute simd map(tofrom: aa) num_teams(M) thread_limit(64)
  for(int i = 0; i < n; i++) {
    aa[i] += 1;
  }

#pragma omp target teams distribute simd map(tofrom:a, aa, b) if(target: n>40)
  for(int i = 0; i < 10; i++) {
    b[i] += 1;
  }

#pragma omp target teams distribute simd collapse(2) firstprivate(f) private(k)
  for(int i = 0; i < M; i++) {
    for(int j = 0; j < M; j++) {
      k = M;
      c[i][j] = i + j * f + k;
    }
  }

  return a[0];
}

int bar(int n){
  int a = 0;

  a += ftemplate<int>(n);

  return a;
}

#endif
// CHECK45-64-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34
// CHECK45-64-SAME: (i64 [[N:%.*]], ptr nonnull align 4 dereferenceable(4000) [[A:%.*]], i64 [[L:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK45-64-NEXT:  entry:
// CHECK45-64-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK45-64-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
// CHECK45-64-NEXT:    [[L_ADDR:%.*]] = alloca i64, align 8
// CHECK45-64-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK45-64-NEXT:    [[L_CASTED:%.*]] = alloca i64, align 8
// CHECK45-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
// CHECK45-64-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
// CHECK45-64-NEXT:    store i64 [[L]], ptr [[L_ADDR]], align 8
// CHECK45-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
// CHECK45-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
// CHECK45-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK45-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK45-64:       user_code.entry:
// CHECK45-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK45-64-NEXT:    [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK45-64-NEXT:    store i32 [[TMP3]], ptr [[N_CASTED]], align 4
// CHECK45-64-NEXT:    [[TMP4:%.*]] = load i64, ptr [[N_CASTED]], align 8
// CHECK45-64-NEXT:    [[TMP5:%.*]] = load i32, ptr [[L_ADDR]], align 4
// CHECK45-64-NEXT:    store i32 [[TMP5]], ptr [[L_CASTED]], align 4
// CHECK45-64-NEXT:    [[TMP6:%.*]] = load i64, ptr [[L_CASTED]], align 8
// CHECK45-64-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK45-64-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK45-64-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_omp_outlined(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i64 [[TMP4]], ptr [[TMP0]], i64 [[TMP6]]) #[[ATTR3:[0-9]+]]
// CHECK45-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK45-64-NEXT:    ret void
// CHECK45-64:       worker.exit:
// CHECK45-64-NEXT:    ret void
//
//
// CHECK45-64-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_omp_outlined
// CHECK45-64-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], ptr nonnull align 4 dereferenceable(4000) [[A:%.*]], i64 [[L:%.*]]) #[[ATTR1:[0-9]+]] {
// CHECK45-64-NEXT:  entry:
// CHECK45-64-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK45-64-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK45-64-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK45-64-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
// CHECK45-64-NEXT:    [[L_ADDR:%.*]] = alloca i64, align 8
// CHECK45-64-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[I4:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK45-64-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK45-64-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
// CHECK45-64-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
// CHECK45-64-NEXT:    store i64 [[L]], ptr [[L_ADDR]], align 8
// CHECK45-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
// CHECK45-64-NEXT:    [[L1:%.*]] = call align 8 ptr @__kmpc_alloc_shared(i64 4)
// CHECK45-64-NEXT:    [[TMP1:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK45-64-NEXT:    store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK45-64-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK45-64-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK45-64-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK45-64-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK45-64-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
// CHECK45-64-NEXT:    store i32 0, ptr [[I]], align 4
// CHECK45-64-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK45-64-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK45-64-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK45-64:       omp.precond.then:
// CHECK45-64-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK45-64-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
// CHECK45-64-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-64-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-64-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK45-64-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4
// CHECK45-64-NEXT:    call void @__kmpc_distribute_static_init_4(ptr @[[GLOB2:[0-9]+]], i32 [[TMP6]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 128)
// CHECK45-64-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK45-64:       omp.dispatch.cond:
// CHECK45-64-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
// CHECK45-64-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK45-64-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK45-64:       cond.true:
// CHECK45-64-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
// CHECK45-64-NEXT:    br label [[COND_END:%.*]]
// CHECK45-64:       cond.false:
// CHECK45-64-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    br label [[COND_END]]
// CHECK45-64:       cond.end:
// CHECK45-64-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK45-64-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK45-64-NEXT:    store i32 [[TMP11]], ptr [[DOTOMP_IV]], align 4
// CHECK45-64-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK45-64-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP12]], [[TMP13]]
// CHECK45-64-NEXT:    br i1 [[CMP6]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK45-64:       omp.dispatch.body:
// CHECK45-64-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK45-64:       omp.inner.for.cond:
// CHECK45-64-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12:![0-9]+]]
// CHECK45-64-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-64-NEXT:    [[CMP7:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK45-64-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK45-64:       omp.inner.for.body:
// CHECK45-64-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-64-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP16]], 1
// CHECK45-64-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK45-64-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-64-NEXT:    [[TMP17:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-64-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP17]] to i64
// CHECK45-64-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], ptr [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK45-64-NEXT:    store i32 1, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-64-NEXT:    [[TMP18:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-64-NEXT:    store i32 [[TMP18]], ptr [[L_ADDR]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-64-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK45-64:       omp.body.continue:
// CHECK45-64-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK45-64:       omp.inner.for.inc:
// CHECK45-64-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-64-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK45-64-NEXT:    store i32 [[ADD8]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-64-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP13:![0-9]+]]
// CHECK45-64:       omp.inner.for.end:
// CHECK45-64-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK45-64:       omp.dispatch.inc:
// CHECK45-64-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK45-64-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-64-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK45-64-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_LB]], align 4
// CHECK45-64-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-64-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK45-64-NEXT:    store i32 [[ADD10]], ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK45-64:       omp.dispatch.end:
// CHECK45-64-NEXT:    [[TMP24:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK45-64-NEXT:    [[TMP25:%.*]] = load i32, ptr [[TMP24]], align 4
// CHECK45-64-NEXT:    call void @__kmpc_distribute_static_fini(ptr @[[GLOB2]], i32 [[TMP25]])
// CHECK45-64-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-64-NEXT:    [[TMP27:%.*]] = icmp ne i32 [[TMP26]], 0
// CHECK45-64-NEXT:    br i1 [[TMP27]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK45-64:       .omp.final.then:
// CHECK45-64-NEXT:    [[TMP28:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK45-64-NEXT:    [[SUB11:%.*]] = sub nsw i32 [[TMP28]], 0
// CHECK45-64-NEXT:    [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK45-64-NEXT:    [[MUL13:%.*]] = mul nsw i32 [[DIV12]], 1
// CHECK45-64-NEXT:    [[ADD14:%.*]] = add nsw i32 0, [[MUL13]]
// CHECK45-64-NEXT:    store i32 [[ADD14]], ptr [[I4]], align 4
// CHECK45-64-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK45-64:       .omp.final.done:
// CHECK45-64-NEXT:    [[TMP29:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-64-NEXT:    [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
// CHECK45-64-NEXT:    br i1 [[TMP30]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK45-64:       .omp.lastprivate.then:
// CHECK45-64-NEXT:    [[TMP31:%.*]] = load i32, ptr [[L_ADDR]], align 4
// CHECK45-64-NEXT:    store i32 [[TMP31]], ptr [[L_ADDR]], align 4
// CHECK45-64-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK45-64:       .omp.lastprivate.done:
// CHECK45-64-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK45-64:       omp.precond.end:
// CHECK45-64-NEXT:    call void @__kmpc_free_shared(ptr [[L1]], i64 4)
// CHECK45-64-NEXT:    ret void
//
//
// CHECK45-64-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l40
// CHECK45-64-SAME: (i64 [[N:%.*]], ptr nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR5:[0-9]+]] {
// CHECK45-64-NEXT:  entry:
// CHECK45-64-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK45-64-NEXT:    [[AA_ADDR:%.*]] = alloca ptr, align 8
// CHECK45-64-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK45-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
// CHECK45-64-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 8
// CHECK45-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 8
// CHECK45-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
// CHECK45-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK45-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK45-64:       user_code.entry:
// CHECK45-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK45-64-NEXT:    [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK45-64-NEXT:    store i32 [[TMP3]], ptr [[N_CASTED]], align 4
// CHECK45-64-NEXT:    [[TMP4:%.*]] = load i64, ptr [[N_CASTED]], align 8
// CHECK45-64-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK45-64-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK45-64-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l40_omp_outlined(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i64 [[TMP4]], ptr [[TMP0]]) #[[ATTR3]]
// CHECK45-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK45-64-NEXT:    ret void
// CHECK45-64:       worker.exit:
// CHECK45-64-NEXT:    ret void
//
//
// CHECK45-64-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l40_omp_outlined
// CHECK45-64-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], ptr nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR1]] {
// CHECK45-64-NEXT:  entry:
// CHECK45-64-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK45-64-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK45-64-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK45-64-NEXT:    [[AA_ADDR:%.*]] = alloca ptr, align 8
// CHECK45-64-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK45-64-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK45-64-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
// CHECK45-64-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 8
// CHECK45-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 8
// CHECK45-64-NEXT:    [[TMP1:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK45-64-NEXT:    store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK45-64-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK45-64-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK45-64-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK45-64-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK45-64-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK45-64-NEXT:    store i32 0, ptr [[I]], align 4
// CHECK45-64-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK45-64-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK45-64-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK45-64:       omp.precond.then:
// CHECK45-64-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK45-64-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK45-64-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-64-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-64-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK45-64-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK45-64-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4
// CHECK45-64-NEXT:    call void @__kmpc_distribute_static_init_4(ptr @[[GLOB2]], i32 [[TMP6]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK45-64-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK45-64:       omp.dispatch.cond:
// CHECK45-64-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK45-64-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK45-64-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK45-64:       cond.true:
// CHECK45-64-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK45-64-NEXT:    br label [[COND_END:%.*]]
// CHECK45-64:       cond.false:
// CHECK45-64-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    br label [[COND_END]]
// CHECK45-64:       cond.end:
// CHECK45-64-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK45-64-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK45-64-NEXT:    store i32 [[TMP11]], ptr [[DOTOMP_IV]], align 4
// CHECK45-64-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK45-64-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP12]], [[TMP13]]
// CHECK45-64-NEXT:    br i1 [[CMP5]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK45-64:       omp.dispatch.body:
// CHECK45-64-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK45-64:       omp.inner.for.cond:
// CHECK45-64-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16:![0-9]+]]
// CHECK45-64-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK45-64-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK45-64-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK45-64:       omp.inner.for.body:
// CHECK45-64-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK45-64-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP16]], 1
// CHECK45-64-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK45-64-NEXT:    store i32 [[ADD]], ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK45-64-NEXT:    [[TMP17:%.*]] = load i32, ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK45-64-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP17]] to i64
// CHECK45-64-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i16], ptr [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK45-64-NEXT:    [[TMP18:%.*]] = load i16, ptr [[ARRAYIDX]], align 2, !llvm.access.group [[ACC_GRP16]]
// CHECK45-64-NEXT:    [[CONV:%.*]] = sext i16 [[TMP18]] to i32
// CHECK45-64-NEXT:    [[ADD7:%.*]] = add nsw i32 [[CONV]], 1
// CHECK45-64-NEXT:    [[CONV8:%.*]] = trunc i32 [[ADD7]] to i16
// CHECK45-64-NEXT:    store i16 [[CONV8]], ptr [[ARRAYIDX]], align 2, !llvm.access.group [[ACC_GRP16]]
// CHECK45-64-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK45-64:       omp.body.continue:
// CHECK45-64-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK45-64:       omp.inner.for.inc:
// CHECK45-64-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK45-64-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK45-64-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK45-64-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP17:![0-9]+]]
// CHECK45-64:       omp.inner.for.end:
// CHECK45-64-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK45-64:       omp.dispatch.inc:
// CHECK45-64-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK45-64-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-64-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK45-64-NEXT:    store i32 [[ADD10]], ptr [[DOTOMP_LB]], align 4
// CHECK45-64-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-64-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK45-64-NEXT:    store i32 [[ADD11]], ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK45-64:       omp.dispatch.end:
// CHECK45-64-NEXT:    [[TMP24:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK45-64-NEXT:    [[TMP25:%.*]] = load i32, ptr [[TMP24]], align 4
// CHECK45-64-NEXT:    call void @__kmpc_distribute_static_fini(ptr @[[GLOB2]], i32 [[TMP25]])
// CHECK45-64-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-64-NEXT:    [[TMP27:%.*]] = icmp ne i32 [[TMP26]], 0
// CHECK45-64-NEXT:    br i1 [[TMP27]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK45-64:       .omp.final.then:
// CHECK45-64-NEXT:    [[TMP28:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK45-64-NEXT:    [[SUB12:%.*]] = sub nsw i32 [[TMP28]], 0
// CHECK45-64-NEXT:    [[DIV13:%.*]] = sdiv i32 [[SUB12]], 1
// CHECK45-64-NEXT:    [[MUL14:%.*]] = mul nsw i32 [[DIV13]], 1
// CHECK45-64-NEXT:    [[ADD15:%.*]] = add nsw i32 0, [[MUL14]]
// CHECK45-64-NEXT:    store i32 [[ADD15]], ptr [[I3]], align 4
// CHECK45-64-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK45-64:       .omp.final.done:
// CHECK45-64-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK45-64:       omp.precond.end:
// CHECK45-64-NEXT:    ret void
//
//
// CHECK45-64-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l45
// CHECK45-64-SAME: (ptr nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK45-64-NEXT:  entry:
// CHECK45-64-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
// CHECK45-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
// CHECK45-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
// CHECK45-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
// CHECK45-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK45-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK45-64:       user_code.entry:
// CHECK45-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK45-64-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK45-64-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK45-64-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l45_omp_outlined(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]]) #[[ATTR3]]
// CHECK45-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK45-64-NEXT:    ret void
// CHECK45-64:       worker.exit:
// CHECK45-64-NEXT:    ret void
//
//
// CHECK45-64-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l45_omp_outlined
// CHECK45-64-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], ptr nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR1]] {
// CHECK45-64-NEXT:  entry:
// CHECK45-64-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK45-64-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK45-64-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
// CHECK45-64-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK45-64-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK45-64-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
// CHECK45-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
// CHECK45-64-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK45-64-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-64-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-64-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK45-64-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK45-64-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK45-64-NEXT:    call void @__kmpc_distribute_static_init_4(ptr @[[GLOB2]], i32 [[TMP2]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK45-64-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK45-64:       omp.dispatch.cond:
// CHECK45-64-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9
// CHECK45-64-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK45-64:       cond.true:
// CHECK45-64-NEXT:    br label [[COND_END:%.*]]
// CHECK45-64:       cond.false:
// CHECK45-64-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    br label [[COND_END]]
// CHECK45-64:       cond.end:
// CHECK45-64-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK45-64-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK45-64-NEXT:    store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4
// CHECK45-64-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK45-64-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]]
// CHECK45-64-NEXT:    br i1 [[CMP1]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK45-64:       omp.dispatch.body:
// CHECK45-64-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK45-64:       omp.inner.for.cond:
// CHECK45-64-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19:![0-9]+]]
// CHECK45-64-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK45-64-NEXT:    [[CMP2:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]]
// CHECK45-64-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK45-64:       omp.inner.for.body:
// CHECK45-64-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK45-64-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1
// CHECK45-64-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK45-64-NEXT:    store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK45-64-NEXT:    [[TMP11:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK45-64-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP11]] to i64
// CHECK45-64-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK45-64-NEXT:    [[TMP12:%.*]] = load i32, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK45-64-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP12]], 1
// CHECK45-64-NEXT:    store i32 [[ADD3]], ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK45-64-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK45-64:       omp.body.continue:
// CHECK45-64-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK45-64:       omp.inner.for.inc:
// CHECK45-64-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK45-64-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK45-64-NEXT:    store i32 [[ADD4]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK45-64-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP20:![0-9]+]]
// CHECK45-64:       omp.inner.for.end:
// CHECK45-64-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK45-64:       omp.dispatch.inc:
// CHECK45-64-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK45-64-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-64-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP14]], [[TMP15]]
// CHECK45-64-NEXT:    store i32 [[ADD5]], ptr [[DOTOMP_LB]], align 4
// CHECK45-64-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-64-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP16]], [[TMP17]]
// CHECK45-64-NEXT:    store i32 [[ADD6]], ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK45-64:       omp.dispatch.end:
// CHECK45-64-NEXT:    call void @__kmpc_distribute_static_fini(ptr @[[GLOB2]], i32 [[TMP2]])
// CHECK45-64-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-64-NEXT:    [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
// CHECK45-64-NEXT:    br i1 [[TMP19]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK45-64:       .omp.final.then:
// CHECK45-64-NEXT:    store i32 10, ptr [[I]], align 4
// CHECK45-64-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK45-64:       .omp.final.done:
// CHECK45-64-NEXT:    ret void
//
//
// CHECK45-64-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l50
// CHECK45-64-SAME: (ptr nonnull align 4 dereferenceable(400) [[C:%.*]], i64 [[F:%.*]]) #[[ATTR0]] {
// CHECK45-64-NEXT:  entry:
// CHECK45-64-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
// CHECK45-64-NEXT:    [[F_ADDR:%.*]] = alloca i64, align 8
// CHECK45-64-NEXT:    [[F_CASTED:%.*]] = alloca i64, align 8
// CHECK45-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
// CHECK45-64-NEXT:    store i64 [[F]], ptr [[F_ADDR]], align 8
// CHECK45-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 8
// CHECK45-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
// CHECK45-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK45-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK45-64:       user_code.entry:
// CHECK45-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK45-64-NEXT:    [[TMP3:%.*]] = load i32, ptr [[F_ADDR]], align 4
// CHECK45-64-NEXT:    store i32 [[TMP3]], ptr [[F_CASTED]], align 4
// CHECK45-64-NEXT:    [[TMP4:%.*]] = load i64, ptr [[F_CASTED]], align 8
// CHECK45-64-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK45-64-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK45-64-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l50_omp_outlined(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]], i64 [[TMP4]]) #[[ATTR3]]
// CHECK45-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK45-64-NEXT:    ret void
// CHECK45-64:       worker.exit:
// CHECK45-64-NEXT:    ret void
//
//
// CHECK45-64-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l50_omp_outlined
// CHECK45-64-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], ptr nonnull align 4 dereferenceable(400) [[C:%.*]], i64 [[F:%.*]]) #[[ATTR1]] {
// CHECK45-64-NEXT:  entry:
// CHECK45-64-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK45-64-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK45-64-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
// CHECK45-64-NEXT:    [[F_ADDR:%.*]] = alloca i64, align 8
// CHECK45-64-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK45-64-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK45-64-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK45-64-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
// CHECK45-64-NEXT:    store i64 [[F]], ptr [[F_ADDR]], align 8
// CHECK45-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 8
// CHECK45-64-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK45-64-NEXT:    store i32 99, ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-64-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-64-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK45-64-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK45-64-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK45-64-NEXT:    call void @__kmpc_distribute_static_init_4(ptr @[[GLOB2]], i32 [[TMP2]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK45-64-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK45-64:       omp.dispatch.cond:
// CHECK45-64-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 99
// CHECK45-64-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK45-64:       cond.true:
// CHECK45-64-NEXT:    br label [[COND_END:%.*]]
// CHECK45-64:       cond.false:
// CHECK45-64-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    br label [[COND_END]]
// CHECK45-64:       cond.end:
// CHECK45-64-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK45-64-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK45-64-NEXT:    store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4
// CHECK45-64-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK45-64-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    [[CMP2:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]]
// CHECK45-64-NEXT:    br i1 [[CMP2]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK45-64:       omp.dispatch.body:
// CHECK45-64-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK45-64:       omp.inner.for.cond:
// CHECK45-64-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22:![0-9]+]]
// CHECK45-64-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-64-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]]
// CHECK45-64-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK45-64:       omp.inner.for.body:
// CHECK45-64-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-64-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP10]], 10
// CHECK45-64-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK45-64-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK45-64-NEXT:    store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-64-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-64-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-64-NEXT:    [[DIV4:%.*]] = sdiv i32 [[TMP12]], 10
// CHECK45-64-NEXT:    [[MUL5:%.*]] = mul nsw i32 [[DIV4]], 10
// CHECK45-64-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP11]], [[MUL5]]
// CHECK45-64-NEXT:    [[MUL6:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK45-64-NEXT:    [[ADD7:%.*]] = add nsw i32 0, [[MUL6]]
// CHECK45-64-NEXT:    store i32 [[ADD7]], ptr [[J]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-64-NEXT:    store i32 10, ptr [[K]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-64-NEXT:    [[TMP13:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-64-NEXT:    [[TMP14:%.*]] = load i32, ptr [[J]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-64-NEXT:    [[TMP15:%.*]] = load i32, ptr [[F_ADDR]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-64-NEXT:    [[MUL8:%.*]] = mul nsw i32 [[TMP14]], [[TMP15]]
// CHECK45-64-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP13]], [[MUL8]]
// CHECK45-64-NEXT:    [[TMP16:%.*]] = load i32, ptr [[K]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-64-NEXT:    [[ADD10:%.*]] = add nsw i32 [[ADD9]], [[TMP16]]
// CHECK45-64-NEXT:    [[TMP17:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-64-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP17]] to i64
// CHECK45-64-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK45-64-NEXT:    [[TMP18:%.*]] = load i32, ptr [[J]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-64-NEXT:    [[IDXPROM11:%.*]] = sext i32 [[TMP18]] to i64
// CHECK45-64-NEXT:    [[ARRAYIDX12:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX]], i64 0, i64 [[IDXPROM11]]
// CHECK45-64-NEXT:    store i32 [[ADD10]], ptr [[ARRAYIDX12]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-64-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK45-64:       omp.body.continue:
// CHECK45-64-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK45-64:       omp.inner.for.inc:
// CHECK45-64-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-64-NEXT:    [[ADD13:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK45-64-NEXT:    store i32 [[ADD13]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-64-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP23:![0-9]+]]
// CHECK45-64:       omp.inner.for.end:
// CHECK45-64-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK45-64:       omp.dispatch.inc:
// CHECK45-64-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK45-64-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-64-NEXT:    [[ADD14:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK45-64-NEXT:    store i32 [[ADD14]], ptr [[DOTOMP_LB]], align 4
// CHECK45-64-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-64-NEXT:    [[ADD15:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK45-64-NEXT:    store i32 [[ADD15]], ptr [[DOTOMP_UB]], align 4
// CHECK45-64-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK45-64:       omp.dispatch.end:
// CHECK45-64-NEXT:    call void @__kmpc_distribute_static_fini(ptr @[[GLOB2]], i32 [[TMP2]])
// CHECK45-64-NEXT:    [[TMP24:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-64-NEXT:    [[TMP25:%.*]] = icmp ne i32 [[TMP24]], 0
// CHECK45-64-NEXT:    br i1 [[TMP25]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK45-64:       .omp.final.then:
// CHECK45-64-NEXT:    store i32 10, ptr [[I]], align 4
// CHECK45-64-NEXT:    store i32 10, ptr [[J]], align 4
// CHECK45-64-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK45-64:       .omp.final.done:
// CHECK45-64-NEXT:    ret void
//
//
// CHECK45-32-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34
// CHECK45-32-SAME: (i32 [[N:%.*]], ptr nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK45-32-NEXT:  entry:
// CHECK45-32-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[L_CASTED:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK45-32-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
// CHECK45-32-NEXT:    store i32 [[L]], ptr [[L_ADDR]], align 4
// CHECK45-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
// CHECK45-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
// CHECK45-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK45-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK45-32:       user_code.entry:
// CHECK45-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK45-32-NEXT:    [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK45-32-NEXT:    store i32 [[TMP3]], ptr [[N_CASTED]], align 4
// CHECK45-32-NEXT:    [[TMP4:%.*]] = load i32, ptr [[N_CASTED]], align 4
// CHECK45-32-NEXT:    [[TMP5:%.*]] = load i32, ptr [[L_ADDR]], align 4
// CHECK45-32-NEXT:    store i32 [[TMP5]], ptr [[L_CASTED]], align 4
// CHECK45-32-NEXT:    [[TMP6:%.*]] = load i32, ptr [[L_CASTED]], align 4
// CHECK45-32-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK45-32-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK45-32-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_omp_outlined(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP4]], ptr [[TMP0]], i32 [[TMP6]]) #[[ATTR3:[0-9]+]]
// CHECK45-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK45-32-NEXT:    ret void
// CHECK45-32:       worker.exit:
// CHECK45-32-NEXT:    ret void
//
//
// CHECK45-32-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_omp_outlined
// CHECK45-32-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], ptr nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR1:[0-9]+]] {
// CHECK45-32-NEXT:  entry:
// CHECK45-32-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[I4:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK45-32-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK45-32-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK45-32-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
// CHECK45-32-NEXT:    store i32 [[L]], ptr [[L_ADDR]], align 4
// CHECK45-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
// CHECK45-32-NEXT:    [[L1:%.*]] = call align 8 ptr @__kmpc_alloc_shared(i32 4)
// CHECK45-32-NEXT:    [[TMP1:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK45-32-NEXT:    store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK45-32-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK45-32-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK45-32-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK45-32-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK45-32-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
// CHECK45-32-NEXT:    store i32 0, ptr [[I]], align 4
// CHECK45-32-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK45-32-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK45-32-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK45-32:       omp.precond.then:
// CHECK45-32-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK45-32-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
// CHECK45-32-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-32-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-32-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK45-32-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4
// CHECK45-32-NEXT:    call void @__kmpc_distribute_static_init_4(ptr @[[GLOB2:[0-9]+]], i32 [[TMP6]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 128)
// CHECK45-32-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK45-32:       omp.dispatch.cond:
// CHECK45-32-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
// CHECK45-32-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK45-32-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK45-32:       cond.true:
// CHECK45-32-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
// CHECK45-32-NEXT:    br label [[COND_END:%.*]]
// CHECK45-32:       cond.false:
// CHECK45-32-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    br label [[COND_END]]
// CHECK45-32:       cond.end:
// CHECK45-32-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK45-32-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK45-32-NEXT:    store i32 [[TMP11]], ptr [[DOTOMP_IV]], align 4
// CHECK45-32-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK45-32-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP12]], [[TMP13]]
// CHECK45-32-NEXT:    br i1 [[CMP6]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK45-32:       omp.dispatch.body:
// CHECK45-32-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK45-32:       omp.inner.for.cond:
// CHECK45-32-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12:![0-9]+]]
// CHECK45-32-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-32-NEXT:    [[CMP7:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK45-32-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK45-32:       omp.inner.for.body:
// CHECK45-32-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-32-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP16]], 1
// CHECK45-32-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK45-32-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-32-NEXT:    [[TMP17:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-32-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], ptr [[TMP0]], i32 0, i32 [[TMP17]]
// CHECK45-32-NEXT:    store i32 1, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-32-NEXT:    [[TMP18:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-32-NEXT:    store i32 [[TMP18]], ptr [[L_ADDR]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-32-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK45-32:       omp.body.continue:
// CHECK45-32-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK45-32:       omp.inner.for.inc:
// CHECK45-32-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-32-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK45-32-NEXT:    store i32 [[ADD8]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-32-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP13:![0-9]+]]
// CHECK45-32:       omp.inner.for.end:
// CHECK45-32-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK45-32:       omp.dispatch.inc:
// CHECK45-32-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK45-32-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-32-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK45-32-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_LB]], align 4
// CHECK45-32-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-32-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK45-32-NEXT:    store i32 [[ADD10]], ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK45-32:       omp.dispatch.end:
// CHECK45-32-NEXT:    [[TMP24:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK45-32-NEXT:    [[TMP25:%.*]] = load i32, ptr [[TMP24]], align 4
// CHECK45-32-NEXT:    call void @__kmpc_distribute_static_fini(ptr @[[GLOB2]], i32 [[TMP25]])
// CHECK45-32-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-32-NEXT:    [[TMP27:%.*]] = icmp ne i32 [[TMP26]], 0
// CHECK45-32-NEXT:    br i1 [[TMP27]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK45-32:       .omp.final.then:
// CHECK45-32-NEXT:    [[TMP28:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK45-32-NEXT:    [[SUB11:%.*]] = sub nsw i32 [[TMP28]], 0
// CHECK45-32-NEXT:    [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK45-32-NEXT:    [[MUL13:%.*]] = mul nsw i32 [[DIV12]], 1
// CHECK45-32-NEXT:    [[ADD14:%.*]] = add nsw i32 0, [[MUL13]]
// CHECK45-32-NEXT:    store i32 [[ADD14]], ptr [[I4]], align 4
// CHECK45-32-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK45-32:       .omp.final.done:
// CHECK45-32-NEXT:    [[TMP29:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-32-NEXT:    [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
// CHECK45-32-NEXT:    br i1 [[TMP30]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK45-32:       .omp.lastprivate.then:
// CHECK45-32-NEXT:    [[TMP31:%.*]] = load i32, ptr [[L_ADDR]], align 4
// CHECK45-32-NEXT:    store i32 [[TMP31]], ptr [[L_ADDR]], align 4
// CHECK45-32-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK45-32:       .omp.lastprivate.done:
// CHECK45-32-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK45-32:       omp.precond.end:
// CHECK45-32-NEXT:    call void @__kmpc_free_shared(ptr [[L1]], i32 4)
// CHECK45-32-NEXT:    ret void
//
//
// CHECK45-32-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l40
// CHECK45-32-SAME: (i32 [[N:%.*]], ptr nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR5:[0-9]+]] {
// CHECK45-32-NEXT:  entry:
// CHECK45-32-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[AA_ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK45-32-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 4
// CHECK45-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
// CHECK45-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
// CHECK45-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK45-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK45-32:       user_code.entry:
// CHECK45-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK45-32-NEXT:    [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK45-32-NEXT:    store i32 [[TMP3]], ptr [[N_CASTED]], align 4
// CHECK45-32-NEXT:    [[TMP4:%.*]] = load i32, ptr [[N_CASTED]], align 4
// CHECK45-32-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK45-32-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK45-32-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l40_omp_outlined(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP4]], ptr [[TMP0]]) #[[ATTR3]]
// CHECK45-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK45-32-NEXT:    ret void
// CHECK45-32:       worker.exit:
// CHECK45-32-NEXT:    ret void
//
//
// CHECK45-32-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l40_omp_outlined
// CHECK45-32-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], ptr nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR1]] {
// CHECK45-32-NEXT:  entry:
// CHECK45-32-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[AA_ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK45-32-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK45-32-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK45-32-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 4
// CHECK45-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
// CHECK45-32-NEXT:    [[TMP1:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK45-32-NEXT:    store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK45-32-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK45-32-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK45-32-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK45-32-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK45-32-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK45-32-NEXT:    store i32 0, ptr [[I]], align 4
// CHECK45-32-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK45-32-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK45-32-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK45-32:       omp.precond.then:
// CHECK45-32-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK45-32-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK45-32-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-32-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-32-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK45-32-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK45-32-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4
// CHECK45-32-NEXT:    call void @__kmpc_distribute_static_init_4(ptr @[[GLOB2]], i32 [[TMP6]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK45-32-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK45-32:       omp.dispatch.cond:
// CHECK45-32-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK45-32-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK45-32-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK45-32:       cond.true:
// CHECK45-32-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK45-32-NEXT:    br label [[COND_END:%.*]]
// CHECK45-32:       cond.false:
// CHECK45-32-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    br label [[COND_END]]
// CHECK45-32:       cond.end:
// CHECK45-32-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK45-32-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK45-32-NEXT:    store i32 [[TMP11]], ptr [[DOTOMP_IV]], align 4
// CHECK45-32-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK45-32-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP12]], [[TMP13]]
// CHECK45-32-NEXT:    br i1 [[CMP5]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK45-32:       omp.dispatch.body:
// CHECK45-32-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK45-32:       omp.inner.for.cond:
// CHECK45-32-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16:![0-9]+]]
// CHECK45-32-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK45-32-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK45-32-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK45-32:       omp.inner.for.body:
// CHECK45-32-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK45-32-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP16]], 1
// CHECK45-32-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK45-32-NEXT:    store i32 [[ADD]], ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK45-32-NEXT:    [[TMP17:%.*]] = load i32, ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK45-32-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i16], ptr [[TMP0]], i32 0, i32 [[TMP17]]
// CHECK45-32-NEXT:    [[TMP18:%.*]] = load i16, ptr [[ARRAYIDX]], align 2, !llvm.access.group [[ACC_GRP16]]
// CHECK45-32-NEXT:    [[CONV:%.*]] = sext i16 [[TMP18]] to i32
// CHECK45-32-NEXT:    [[ADD7:%.*]] = add nsw i32 [[CONV]], 1
// CHECK45-32-NEXT:    [[CONV8:%.*]] = trunc i32 [[ADD7]] to i16
// CHECK45-32-NEXT:    store i16 [[CONV8]], ptr [[ARRAYIDX]], align 2, !llvm.access.group [[ACC_GRP16]]
// CHECK45-32-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK45-32:       omp.body.continue:
// CHECK45-32-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK45-32:       omp.inner.for.inc:
// CHECK45-32-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK45-32-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK45-32-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK45-32-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP17:![0-9]+]]
// CHECK45-32:       omp.inner.for.end:
// CHECK45-32-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK45-32:       omp.dispatch.inc:
// CHECK45-32-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK45-32-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-32-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK45-32-NEXT:    store i32 [[ADD10]], ptr [[DOTOMP_LB]], align 4
// CHECK45-32-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-32-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK45-32-NEXT:    store i32 [[ADD11]], ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK45-32:       omp.dispatch.end:
// CHECK45-32-NEXT:    [[TMP24:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK45-32-NEXT:    [[TMP25:%.*]] = load i32, ptr [[TMP24]], align 4
// CHECK45-32-NEXT:    call void @__kmpc_distribute_static_fini(ptr @[[GLOB2]], i32 [[TMP25]])
// CHECK45-32-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-32-NEXT:    [[TMP27:%.*]] = icmp ne i32 [[TMP26]], 0
// CHECK45-32-NEXT:    br i1 [[TMP27]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK45-32:       .omp.final.then:
// CHECK45-32-NEXT:    [[TMP28:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK45-32-NEXT:    [[SUB12:%.*]] = sub nsw i32 [[TMP28]], 0
// CHECK45-32-NEXT:    [[DIV13:%.*]] = sdiv i32 [[SUB12]], 1
// CHECK45-32-NEXT:    [[MUL14:%.*]] = mul nsw i32 [[DIV13]], 1
// CHECK45-32-NEXT:    [[ADD15:%.*]] = add nsw i32 0, [[MUL14]]
// CHECK45-32-NEXT:    store i32 [[ADD15]], ptr [[I3]], align 4
// CHECK45-32-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK45-32:       .omp.final.done:
// CHECK45-32-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK45-32:       omp.precond.end:
// CHECK45-32-NEXT:    ret void
//
//
// CHECK45-32-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l45
// CHECK45-32-SAME: (ptr nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK45-32-NEXT:  entry:
// CHECK45-32-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
// CHECK45-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
// CHECK45-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
// CHECK45-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK45-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK45-32:       user_code.entry:
// CHECK45-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK45-32-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK45-32-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK45-32-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l45_omp_outlined(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]]) #[[ATTR3]]
// CHECK45-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK45-32-NEXT:    ret void
// CHECK45-32:       worker.exit:
// CHECK45-32-NEXT:    ret void
//
//
// CHECK45-32-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l45_omp_outlined
// CHECK45-32-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], ptr nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR1]] {
// CHECK45-32-NEXT:  entry:
// CHECK45-32-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK45-32-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK45-32-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
// CHECK45-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
// CHECK45-32-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK45-32-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-32-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-32-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK45-32-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK45-32-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK45-32-NEXT:    call void @__kmpc_distribute_static_init_4(ptr @[[GLOB2]], i32 [[TMP2]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK45-32-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK45-32:       omp.dispatch.cond:
// CHECK45-32-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9
// CHECK45-32-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK45-32:       cond.true:
// CHECK45-32-NEXT:    br label [[COND_END:%.*]]
// CHECK45-32:       cond.false:
// CHECK45-32-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    br label [[COND_END]]
// CHECK45-32:       cond.end:
// CHECK45-32-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK45-32-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK45-32-NEXT:    store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4
// CHECK45-32-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK45-32-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]]
// CHECK45-32-NEXT:    br i1 [[CMP1]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK45-32:       omp.dispatch.body:
// CHECK45-32-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK45-32:       omp.inner.for.cond:
// CHECK45-32-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19:![0-9]+]]
// CHECK45-32-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK45-32-NEXT:    [[CMP2:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]]
// CHECK45-32-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK45-32:       omp.inner.for.body:
// CHECK45-32-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK45-32-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1
// CHECK45-32-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK45-32-NEXT:    store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK45-32-NEXT:    [[TMP11:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK45-32-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i32 0, i32 [[TMP11]]
// CHECK45-32-NEXT:    [[TMP12:%.*]] = load i32, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK45-32-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP12]], 1
// CHECK45-32-NEXT:    store i32 [[ADD3]], ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK45-32-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK45-32:       omp.body.continue:
// CHECK45-32-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK45-32:       omp.inner.for.inc:
// CHECK45-32-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK45-32-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK45-32-NEXT:    store i32 [[ADD4]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK45-32-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP20:![0-9]+]]
// CHECK45-32:       omp.inner.for.end:
// CHECK45-32-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK45-32:       omp.dispatch.inc:
// CHECK45-32-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK45-32-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-32-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP14]], [[TMP15]]
// CHECK45-32-NEXT:    store i32 [[ADD5]], ptr [[DOTOMP_LB]], align 4
// CHECK45-32-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-32-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP16]], [[TMP17]]
// CHECK45-32-NEXT:    store i32 [[ADD6]], ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK45-32:       omp.dispatch.end:
// CHECK45-32-NEXT:    call void @__kmpc_distribute_static_fini(ptr @[[GLOB2]], i32 [[TMP2]])
// CHECK45-32-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-32-NEXT:    [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
// CHECK45-32-NEXT:    br i1 [[TMP19]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK45-32:       .omp.final.then:
// CHECK45-32-NEXT:    store i32 10, ptr [[I]], align 4
// CHECK45-32-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK45-32:       .omp.final.done:
// CHECK45-32-NEXT:    ret void
//
//
// CHECK45-32-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l50
// CHECK45-32-SAME: (ptr nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK45-32-NEXT:  entry:
// CHECK45-32-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[F_CASTED:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
// CHECK45-32-NEXT:    store i32 [[F]], ptr [[F_ADDR]], align 4
// CHECK45-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 4
// CHECK45-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
// CHECK45-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK45-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK45-32:       user_code.entry:
// CHECK45-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK45-32-NEXT:    [[TMP3:%.*]] = load i32, ptr [[F_ADDR]], align 4
// CHECK45-32-NEXT:    store i32 [[TMP3]], ptr [[F_CASTED]], align 4
// CHECK45-32-NEXT:    [[TMP4:%.*]] = load i32, ptr [[F_CASTED]], align 4
// CHECK45-32-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK45-32-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK45-32-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l50_omp_outlined(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]], i32 [[TMP4]]) #[[ATTR3]]
// CHECK45-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK45-32-NEXT:    ret void
// CHECK45-32:       worker.exit:
// CHECK45-32-NEXT:    ret void
//
//
// CHECK45-32-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l50_omp_outlined
// CHECK45-32-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], ptr nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR1]] {
// CHECK45-32-NEXT:  entry:
// CHECK45-32-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK45-32-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK45-32-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK45-32-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
// CHECK45-32-NEXT:    store i32 [[F]], ptr [[F_ADDR]], align 4
// CHECK45-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 4
// CHECK45-32-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK45-32-NEXT:    store i32 99, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-32-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-32-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK45-32-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK45-32-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK45-32-NEXT:    call void @__kmpc_distribute_static_init_4(ptr @[[GLOB2]], i32 [[TMP2]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK45-32-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK45-32:       omp.dispatch.cond:
// CHECK45-32-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 99
// CHECK45-32-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK45-32:       cond.true:
// CHECK45-32-NEXT:    br label [[COND_END:%.*]]
// CHECK45-32:       cond.false:
// CHECK45-32-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    br label [[COND_END]]
// CHECK45-32:       cond.end:
// CHECK45-32-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK45-32-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK45-32-NEXT:    store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4
// CHECK45-32-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK45-32-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    [[CMP2:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]]
// CHECK45-32-NEXT:    br i1 [[CMP2]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK45-32:       omp.dispatch.body:
// CHECK45-32-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK45-32:       omp.inner.for.cond:
// CHECK45-32-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22:![0-9]+]]
// CHECK45-32-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]]
// CHECK45-32-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK45-32:       omp.inner.for.body:
// CHECK45-32-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP10]], 10
// CHECK45-32-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK45-32-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK45-32-NEXT:    store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-NEXT:    [[DIV4:%.*]] = sdiv i32 [[TMP12]], 10
// CHECK45-32-NEXT:    [[MUL5:%.*]] = mul nsw i32 [[DIV4]], 10
// CHECK45-32-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP11]], [[MUL5]]
// CHECK45-32-NEXT:    [[MUL6:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK45-32-NEXT:    [[ADD7:%.*]] = add nsw i32 0, [[MUL6]]
// CHECK45-32-NEXT:    store i32 [[ADD7]], ptr [[J]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-NEXT:    store i32 10, ptr [[K]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-NEXT:    [[TMP13:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-NEXT:    [[TMP14:%.*]] = load i32, ptr [[J]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-NEXT:    [[TMP15:%.*]] = load i32, ptr [[F_ADDR]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-NEXT:    [[MUL8:%.*]] = mul nsw i32 [[TMP14]], [[TMP15]]
// CHECK45-32-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP13]], [[MUL8]]
// CHECK45-32-NEXT:    [[TMP16:%.*]] = load i32, ptr [[K]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-NEXT:    [[ADD10:%.*]] = add nsw i32 [[ADD9]], [[TMP16]]
// CHECK45-32-NEXT:    [[TMP17:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[TMP0]], i32 0, i32 [[TMP17]]
// CHECK45-32-NEXT:    [[TMP18:%.*]] = load i32, ptr [[J]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX]], i32 0, i32 [[TMP18]]
// CHECK45-32-NEXT:    store i32 [[ADD10]], ptr [[ARRAYIDX11]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK45-32:       omp.body.continue:
// CHECK45-32-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK45-32:       omp.inner.for.inc:
// CHECK45-32-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK45-32-NEXT:    store i32 [[ADD12]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP23:![0-9]+]]
// CHECK45-32:       omp.inner.for.end:
// CHECK45-32-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK45-32:       omp.dispatch.inc:
// CHECK45-32-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK45-32-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-32-NEXT:    [[ADD13:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK45-32-NEXT:    store i32 [[ADD13]], ptr [[DOTOMP_LB]], align 4
// CHECK45-32-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-32-NEXT:    [[ADD14:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK45-32-NEXT:    store i32 [[ADD14]], ptr [[DOTOMP_UB]], align 4
// CHECK45-32-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK45-32:       omp.dispatch.end:
// CHECK45-32-NEXT:    call void @__kmpc_distribute_static_fini(ptr @[[GLOB2]], i32 [[TMP2]])
// CHECK45-32-NEXT:    [[TMP24:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-32-NEXT:    [[TMP25:%.*]] = icmp ne i32 [[TMP24]], 0
// CHECK45-32-NEXT:    br i1 [[TMP25]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK45-32:       .omp.final.then:
// CHECK45-32-NEXT:    store i32 10, ptr [[I]], align 4
// CHECK45-32-NEXT:    store i32 10, ptr [[J]], align 4
// CHECK45-32-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK45-32:       .omp.final.done:
// CHECK45-32-NEXT:    ret void
//
//
// CHECK45-32-EX-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34
// CHECK45-32-EX-SAME: (i32 [[N:%.*]], ptr nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK45-32-EX-NEXT:  entry:
// CHECK45-32-EX-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-EX-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[L_CASTED:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK45-32-EX-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
// CHECK45-32-EX-NEXT:    store i32 [[L]], ptr [[L_ADDR]], align 4
// CHECK45-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
// CHECK45-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK45-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK45-32-EX:       user_code.entry:
// CHECK45-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK45-32-EX-NEXT:    [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK45-32-EX-NEXT:    store i32 [[TMP3]], ptr [[N_CASTED]], align 4
// CHECK45-32-EX-NEXT:    [[TMP4:%.*]] = load i32, ptr [[N_CASTED]], align 4
// CHECK45-32-EX-NEXT:    [[TMP5:%.*]] = load i32, ptr [[L_ADDR]], align 4
// CHECK45-32-EX-NEXT:    store i32 [[TMP5]], ptr [[L_CASTED]], align 4
// CHECK45-32-EX-NEXT:    [[TMP6:%.*]] = load i32, ptr [[L_CASTED]], align 4
// CHECK45-32-EX-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK45-32-EX-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK45-32-EX-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_omp_outlined(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP4]], ptr [[TMP0]], i32 [[TMP6]]) #[[ATTR3:[0-9]+]]
// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK45-32-EX-NEXT:    ret void
// CHECK45-32-EX:       worker.exit:
// CHECK45-32-EX-NEXT:    ret void
//
//
// CHECK45-32-EX-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_omp_outlined
// CHECK45-32-EX-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], ptr nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR1:[0-9]+]] {
// CHECK45-32-EX-NEXT:  entry:
// CHECK45-32-EX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-EX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-EX-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-EX-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[I4:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK45-32-EX-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK45-32-EX-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK45-32-EX-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
// CHECK45-32-EX-NEXT:    store i32 [[L]], ptr [[L_ADDR]], align 4
// CHECK45-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
// CHECK45-32-EX-NEXT:    [[L1:%.*]] = call align 8 ptr @__kmpc_alloc_shared(i32 4)
// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK45-32-EX-NEXT:    store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK45-32-EX-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK45-32-EX-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK45-32-EX-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK45-32-EX-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK45-32-EX-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
// CHECK45-32-EX-NEXT:    store i32 0, ptr [[I]], align 4
// CHECK45-32-EX-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK45-32-EX-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK45-32-EX-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK45-32-EX:       omp.precond.then:
// CHECK45-32-EX-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK45-32-EX-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
// CHECK45-32-EX-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-32-EX-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-32-EX-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK45-32-EX-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4
// CHECK45-32-EX-NEXT:    call void @__kmpc_distribute_static_init_4(ptr @[[GLOB2:[0-9]+]], i32 [[TMP6]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 128)
// CHECK45-32-EX-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK45-32-EX:       omp.dispatch.cond:
// CHECK45-32-EX-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
// CHECK45-32-EX-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK45-32-EX-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK45-32-EX:       cond.true:
// CHECK45-32-EX-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
// CHECK45-32-EX-NEXT:    br label [[COND_END:%.*]]
// CHECK45-32-EX:       cond.false:
// CHECK45-32-EX-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    br label [[COND_END]]
// CHECK45-32-EX:       cond.end:
// CHECK45-32-EX-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK45-32-EX-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK45-32-EX-NEXT:    store i32 [[TMP11]], ptr [[DOTOMP_IV]], align 4
// CHECK45-32-EX-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK45-32-EX-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP12]], [[TMP13]]
// CHECK45-32-EX-NEXT:    br i1 [[CMP6]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK45-32-EX:       omp.dispatch.body:
// CHECK45-32-EX-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK45-32-EX:       omp.inner.for.cond:
// CHECK45-32-EX-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12:![0-9]+]]
// CHECK45-32-EX-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-32-EX-NEXT:    [[CMP7:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK45-32-EX-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK45-32-EX:       omp.inner.for.body:
// CHECK45-32-EX-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-32-EX-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP16]], 1
// CHECK45-32-EX-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK45-32-EX-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-32-EX-NEXT:    [[TMP17:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-32-EX-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], ptr [[TMP0]], i32 0, i32 [[TMP17]]
// CHECK45-32-EX-NEXT:    store i32 1, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-32-EX-NEXT:    [[TMP18:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-32-EX-NEXT:    store i32 [[TMP18]], ptr [[L_ADDR]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-32-EX-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK45-32-EX:       omp.body.continue:
// CHECK45-32-EX-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK45-32-EX:       omp.inner.for.inc:
// CHECK45-32-EX-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-32-EX-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK45-32-EX-NEXT:    store i32 [[ADD8]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK45-32-EX-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP13:![0-9]+]]
// CHECK45-32-EX:       omp.inner.for.end:
// CHECK45-32-EX-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK45-32-EX:       omp.dispatch.inc:
// CHECK45-32-EX-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK45-32-EX-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-32-EX-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK45-32-EX-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_LB]], align 4
// CHECK45-32-EX-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-32-EX-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK45-32-EX-NEXT:    store i32 [[ADD10]], ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK45-32-EX:       omp.dispatch.end:
// CHECK45-32-EX-NEXT:    [[TMP24:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK45-32-EX-NEXT:    [[TMP25:%.*]] = load i32, ptr [[TMP24]], align 4
// CHECK45-32-EX-NEXT:    call void @__kmpc_distribute_static_fini(ptr @[[GLOB2]], i32 [[TMP25]])
// CHECK45-32-EX-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-32-EX-NEXT:    [[TMP27:%.*]] = icmp ne i32 [[TMP26]], 0
// CHECK45-32-EX-NEXT:    br i1 [[TMP27]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK45-32-EX:       .omp.final.then:
// CHECK45-32-EX-NEXT:    [[TMP28:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK45-32-EX-NEXT:    [[SUB11:%.*]] = sub nsw i32 [[TMP28]], 0
// CHECK45-32-EX-NEXT:    [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK45-32-EX-NEXT:    [[MUL13:%.*]] = mul nsw i32 [[DIV12]], 1
// CHECK45-32-EX-NEXT:    [[ADD14:%.*]] = add nsw i32 0, [[MUL13]]
// CHECK45-32-EX-NEXT:    store i32 [[ADD14]], ptr [[I4]], align 4
// CHECK45-32-EX-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK45-32-EX:       .omp.final.done:
// CHECK45-32-EX-NEXT:    [[TMP29:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-32-EX-NEXT:    [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
// CHECK45-32-EX-NEXT:    br i1 [[TMP30]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK45-32-EX:       .omp.lastprivate.then:
// CHECK45-32-EX-NEXT:    [[TMP31:%.*]] = load i32, ptr [[L_ADDR]], align 4
// CHECK45-32-EX-NEXT:    store i32 [[TMP31]], ptr [[L_ADDR]], align 4
// CHECK45-32-EX-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK45-32-EX:       .omp.lastprivate.done:
// CHECK45-32-EX-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK45-32-EX:       omp.precond.end:
// CHECK45-32-EX-NEXT:    call void @__kmpc_free_shared(ptr [[L1]], i32 4)
// CHECK45-32-EX-NEXT:    ret void
//
//
// CHECK45-32-EX-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l40
// CHECK45-32-EX-SAME: (i32 [[N:%.*]], ptr nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR5:[0-9]+]] {
// CHECK45-32-EX-NEXT:  entry:
// CHECK45-32-EX-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[AA_ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-EX-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK45-32-EX-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 4
// CHECK45-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
// CHECK45-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK45-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK45-32-EX:       user_code.entry:
// CHECK45-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK45-32-EX-NEXT:    [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK45-32-EX-NEXT:    store i32 [[TMP3]], ptr [[N_CASTED]], align 4
// CHECK45-32-EX-NEXT:    [[TMP4:%.*]] = load i32, ptr [[N_CASTED]], align 4
// CHECK45-32-EX-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK45-32-EX-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK45-32-EX-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l40_omp_outlined(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP4]], ptr [[TMP0]]) #[[ATTR3]]
// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK45-32-EX-NEXT:    ret void
// CHECK45-32-EX:       worker.exit:
// CHECK45-32-EX-NEXT:    ret void
//
//
// CHECK45-32-EX-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l40_omp_outlined
// CHECK45-32-EX-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], ptr nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR1]] {
// CHECK45-32-EX-NEXT:  entry:
// CHECK45-32-EX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-EX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-EX-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[AA_ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-EX-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK45-32-EX-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK45-32-EX-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK45-32-EX-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 4
// CHECK45-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK45-32-EX-NEXT:    store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK45-32-EX-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK45-32-EX-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK45-32-EX-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK45-32-EX-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK45-32-EX-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK45-32-EX-NEXT:    store i32 0, ptr [[I]], align 4
// CHECK45-32-EX-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK45-32-EX-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK45-32-EX-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK45-32-EX:       omp.precond.then:
// CHECK45-32-EX-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK45-32-EX-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK45-32-EX-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-32-EX-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-32-EX-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK45-32-EX-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK45-32-EX-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4
// CHECK45-32-EX-NEXT:    call void @__kmpc_distribute_static_init_4(ptr @[[GLOB2]], i32 [[TMP6]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK45-32-EX-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK45-32-EX:       omp.dispatch.cond:
// CHECK45-32-EX-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK45-32-EX-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK45-32-EX-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK45-32-EX:       cond.true:
// CHECK45-32-EX-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK45-32-EX-NEXT:    br label [[COND_END:%.*]]
// CHECK45-32-EX:       cond.false:
// CHECK45-32-EX-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    br label [[COND_END]]
// CHECK45-32-EX:       cond.end:
// CHECK45-32-EX-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK45-32-EX-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK45-32-EX-NEXT:    store i32 [[TMP11]], ptr [[DOTOMP_IV]], align 4
// CHECK45-32-EX-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK45-32-EX-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP12]], [[TMP13]]
// CHECK45-32-EX-NEXT:    br i1 [[CMP5]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK45-32-EX:       omp.dispatch.body:
// CHECK45-32-EX-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK45-32-EX:       omp.inner.for.cond:
// CHECK45-32-EX-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16:![0-9]+]]
// CHECK45-32-EX-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK45-32-EX-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK45-32-EX-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK45-32-EX:       omp.inner.for.body:
// CHECK45-32-EX-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK45-32-EX-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP16]], 1
// CHECK45-32-EX-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK45-32-EX-NEXT:    store i32 [[ADD]], ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK45-32-EX-NEXT:    [[TMP17:%.*]] = load i32, ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK45-32-EX-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i16], ptr [[TMP0]], i32 0, i32 [[TMP17]]
// CHECK45-32-EX-NEXT:    [[TMP18:%.*]] = load i16, ptr [[ARRAYIDX]], align 2, !llvm.access.group [[ACC_GRP16]]
// CHECK45-32-EX-NEXT:    [[CONV:%.*]] = sext i16 [[TMP18]] to i32
// CHECK45-32-EX-NEXT:    [[ADD7:%.*]] = add nsw i32 [[CONV]], 1
// CHECK45-32-EX-NEXT:    [[CONV8:%.*]] = trunc i32 [[ADD7]] to i16
// CHECK45-32-EX-NEXT:    store i16 [[CONV8]], ptr [[ARRAYIDX]], align 2, !llvm.access.group [[ACC_GRP16]]
// CHECK45-32-EX-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK45-32-EX:       omp.body.continue:
// CHECK45-32-EX-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK45-32-EX:       omp.inner.for.inc:
// CHECK45-32-EX-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK45-32-EX-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK45-32-EX-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK45-32-EX-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP17:![0-9]+]]
// CHECK45-32-EX:       omp.inner.for.end:
// CHECK45-32-EX-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK45-32-EX:       omp.dispatch.inc:
// CHECK45-32-EX-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK45-32-EX-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-32-EX-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK45-32-EX-NEXT:    store i32 [[ADD10]], ptr [[DOTOMP_LB]], align 4
// CHECK45-32-EX-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-32-EX-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK45-32-EX-NEXT:    store i32 [[ADD11]], ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK45-32-EX:       omp.dispatch.end:
// CHECK45-32-EX-NEXT:    [[TMP24:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK45-32-EX-NEXT:    [[TMP25:%.*]] = load i32, ptr [[TMP24]], align 4
// CHECK45-32-EX-NEXT:    call void @__kmpc_distribute_static_fini(ptr @[[GLOB2]], i32 [[TMP25]])
// CHECK45-32-EX-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-32-EX-NEXT:    [[TMP27:%.*]] = icmp ne i32 [[TMP26]], 0
// CHECK45-32-EX-NEXT:    br i1 [[TMP27]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK45-32-EX:       .omp.final.then:
// CHECK45-32-EX-NEXT:    [[TMP28:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK45-32-EX-NEXT:    [[SUB12:%.*]] = sub nsw i32 [[TMP28]], 0
// CHECK45-32-EX-NEXT:    [[DIV13:%.*]] = sdiv i32 [[SUB12]], 1
// CHECK45-32-EX-NEXT:    [[MUL14:%.*]] = mul nsw i32 [[DIV13]], 1
// CHECK45-32-EX-NEXT:    [[ADD15:%.*]] = add nsw i32 0, [[MUL14]]
// CHECK45-32-EX-NEXT:    store i32 [[ADD15]], ptr [[I3]], align 4
// CHECK45-32-EX-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK45-32-EX:       .omp.final.done:
// CHECK45-32-EX-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK45-32-EX:       omp.precond.end:
// CHECK45-32-EX-NEXT:    ret void
//
//
// CHECK45-32-EX-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l45
// CHECK45-32-EX-SAME: (ptr nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK45-32-EX-NEXT:  entry:
// CHECK45-32-EX-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
// CHECK45-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
// CHECK45-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK45-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK45-32-EX:       user_code.entry:
// CHECK45-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK45-32-EX-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK45-32-EX-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK45-32-EX-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l45_omp_outlined(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]]) #[[ATTR3]]
// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK45-32-EX-NEXT:    ret void
// CHECK45-32-EX:       worker.exit:
// CHECK45-32-EX-NEXT:    ret void
//
//
// CHECK45-32-EX-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l45_omp_outlined
// CHECK45-32-EX-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], ptr nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR1]] {
// CHECK45-32-EX-NEXT:  entry:
// CHECK45-32-EX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-EX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-EX-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-EX-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK45-32-EX-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK45-32-EX-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
// CHECK45-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
// CHECK45-32-EX-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK45-32-EX-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-32-EX-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-32-EX-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK45-32-EX-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK45-32-EX-NEXT:    call void @__kmpc_distribute_static_init_4(ptr @[[GLOB2]], i32 [[TMP2]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK45-32-EX-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK45-32-EX:       omp.dispatch.cond:
// CHECK45-32-EX-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9
// CHECK45-32-EX-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK45-32-EX:       cond.true:
// CHECK45-32-EX-NEXT:    br label [[COND_END:%.*]]
// CHECK45-32-EX:       cond.false:
// CHECK45-32-EX-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    br label [[COND_END]]
// CHECK45-32-EX:       cond.end:
// CHECK45-32-EX-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK45-32-EX-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK45-32-EX-NEXT:    store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4
// CHECK45-32-EX-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK45-32-EX-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]]
// CHECK45-32-EX-NEXT:    br i1 [[CMP1]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK45-32-EX:       omp.dispatch.body:
// CHECK45-32-EX-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK45-32-EX:       omp.inner.for.cond:
// CHECK45-32-EX-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19:![0-9]+]]
// CHECK45-32-EX-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK45-32-EX-NEXT:    [[CMP2:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]]
// CHECK45-32-EX-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK45-32-EX:       omp.inner.for.body:
// CHECK45-32-EX-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK45-32-EX-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1
// CHECK45-32-EX-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK45-32-EX-NEXT:    store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK45-32-EX-NEXT:    [[TMP11:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK45-32-EX-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i32 0, i32 [[TMP11]]
// CHECK45-32-EX-NEXT:    [[TMP12:%.*]] = load i32, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK45-32-EX-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP12]], 1
// CHECK45-32-EX-NEXT:    store i32 [[ADD3]], ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK45-32-EX-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK45-32-EX:       omp.body.continue:
// CHECK45-32-EX-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK45-32-EX:       omp.inner.for.inc:
// CHECK45-32-EX-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK45-32-EX-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK45-32-EX-NEXT:    store i32 [[ADD4]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK45-32-EX-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP20:![0-9]+]]
// CHECK45-32-EX:       omp.inner.for.end:
// CHECK45-32-EX-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK45-32-EX:       omp.dispatch.inc:
// CHECK45-32-EX-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK45-32-EX-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-32-EX-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP14]], [[TMP15]]
// CHECK45-32-EX-NEXT:    store i32 [[ADD5]], ptr [[DOTOMP_LB]], align 4
// CHECK45-32-EX-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-32-EX-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP16]], [[TMP17]]
// CHECK45-32-EX-NEXT:    store i32 [[ADD6]], ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK45-32-EX:       omp.dispatch.end:
// CHECK45-32-EX-NEXT:    call void @__kmpc_distribute_static_fini(ptr @[[GLOB2]], i32 [[TMP2]])
// CHECK45-32-EX-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-32-EX-NEXT:    [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
// CHECK45-32-EX-NEXT:    br i1 [[TMP19]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK45-32-EX:       .omp.final.then:
// CHECK45-32-EX-NEXT:    store i32 10, ptr [[I]], align 4
// CHECK45-32-EX-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK45-32-EX:       .omp.final.done:
// CHECK45-32-EX-NEXT:    ret void
//
//
// CHECK45-32-EX-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l50
// CHECK45-32-EX-SAME: (ptr nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK45-32-EX-NEXT:  entry:
// CHECK45-32-EX-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-EX-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[F_CASTED:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
// CHECK45-32-EX-NEXT:    store i32 [[F]], ptr [[F_ADDR]], align 4
// CHECK45-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 4
// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
// CHECK45-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK45-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK45-32-EX:       user_code.entry:
// CHECK45-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK45-32-EX-NEXT:    [[TMP3:%.*]] = load i32, ptr [[F_ADDR]], align 4
// CHECK45-32-EX-NEXT:    store i32 [[TMP3]], ptr [[F_CASTED]], align 4
// CHECK45-32-EX-NEXT:    [[TMP4:%.*]] = load i32, ptr [[F_CASTED]], align 4
// CHECK45-32-EX-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK45-32-EX-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK45-32-EX-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l50_omp_outlined(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]], i32 [[TMP4]]) #[[ATTR3]]
// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK45-32-EX-NEXT:    ret void
// CHECK45-32-EX:       worker.exit:
// CHECK45-32-EX-NEXT:    ret void
//
//
// CHECK45-32-EX-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l50_omp_outlined
// CHECK45-32-EX-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], ptr nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR1]] {
// CHECK45-32-EX-NEXT:  entry:
// CHECK45-32-EX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-EX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-EX-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
// CHECK45-32-EX-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK45-32-EX-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK45-32-EX-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK45-32-EX-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
// CHECK45-32-EX-NEXT:    store i32 [[F]], ptr [[F_ADDR]], align 4
// CHECK45-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 4
// CHECK45-32-EX-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK45-32-EX-NEXT:    store i32 99, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-32-EX-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-32-EX-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK45-32-EX-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK45-32-EX-NEXT:    call void @__kmpc_distribute_static_init_4(ptr @[[GLOB2]], i32 [[TMP2]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK45-32-EX-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK45-32-EX:       omp.dispatch.cond:
// CHECK45-32-EX-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 99
// CHECK45-32-EX-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK45-32-EX:       cond.true:
// CHECK45-32-EX-NEXT:    br label [[COND_END:%.*]]
// CHECK45-32-EX:       cond.false:
// CHECK45-32-EX-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    br label [[COND_END]]
// CHECK45-32-EX:       cond.end:
// CHECK45-32-EX-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK45-32-EX-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK45-32-EX-NEXT:    store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4
// CHECK45-32-EX-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK45-32-EX-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    [[CMP2:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]]
// CHECK45-32-EX-NEXT:    br i1 [[CMP2]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK45-32-EX:       omp.dispatch.body:
// CHECK45-32-EX-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK45-32-EX:       omp.inner.for.cond:
// CHECK45-32-EX-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22:![0-9]+]]
// CHECK45-32-EX-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-EX-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]]
// CHECK45-32-EX-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK45-32-EX:       omp.inner.for.body:
// CHECK45-32-EX-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-EX-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP10]], 10
// CHECK45-32-EX-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK45-32-EX-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK45-32-EX-NEXT:    store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-EX-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-EX-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-EX-NEXT:    [[DIV4:%.*]] = sdiv i32 [[TMP12]], 10
// CHECK45-32-EX-NEXT:    [[MUL5:%.*]] = mul nsw i32 [[DIV4]], 10
// CHECK45-32-EX-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP11]], [[MUL5]]
// CHECK45-32-EX-NEXT:    [[MUL6:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK45-32-EX-NEXT:    [[ADD7:%.*]] = add nsw i32 0, [[MUL6]]
// CHECK45-32-EX-NEXT:    store i32 [[ADD7]], ptr [[J]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-EX-NEXT:    store i32 10, ptr [[K]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-EX-NEXT:    [[TMP13:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-EX-NEXT:    [[TMP14:%.*]] = load i32, ptr [[J]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-EX-NEXT:    [[TMP15:%.*]] = load i32, ptr [[F_ADDR]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-EX-NEXT:    [[MUL8:%.*]] = mul nsw i32 [[TMP14]], [[TMP15]]
// CHECK45-32-EX-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP13]], [[MUL8]]
// CHECK45-32-EX-NEXT:    [[TMP16:%.*]] = load i32, ptr [[K]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-EX-NEXT:    [[ADD10:%.*]] = add nsw i32 [[ADD9]], [[TMP16]]
// CHECK45-32-EX-NEXT:    [[TMP17:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-EX-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[TMP0]], i32 0, i32 [[TMP17]]
// CHECK45-32-EX-NEXT:    [[TMP18:%.*]] = load i32, ptr [[J]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-EX-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX]], i32 0, i32 [[TMP18]]
// CHECK45-32-EX-NEXT:    store i32 [[ADD10]], ptr [[ARRAYIDX11]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-EX-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK45-32-EX:       omp.body.continue:
// CHECK45-32-EX-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK45-32-EX:       omp.inner.for.inc:
// CHECK45-32-EX-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-EX-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK45-32-EX-NEXT:    store i32 [[ADD12]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK45-32-EX-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP23:![0-9]+]]
// CHECK45-32-EX:       omp.inner.for.end:
// CHECK45-32-EX-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK45-32-EX:       omp.dispatch.inc:
// CHECK45-32-EX-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK45-32-EX-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-32-EX-NEXT:    [[ADD13:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK45-32-EX-NEXT:    store i32 [[ADD13]], ptr [[DOTOMP_LB]], align 4
// CHECK45-32-EX-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK45-32-EX-NEXT:    [[ADD14:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK45-32-EX-NEXT:    store i32 [[ADD14]], ptr [[DOTOMP_UB]], align 4
// CHECK45-32-EX-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK45-32-EX:       omp.dispatch.end:
// CHECK45-32-EX-NEXT:    call void @__kmpc_distribute_static_fini(ptr @[[GLOB2]], i32 [[TMP2]])
// CHECK45-32-EX-NEXT:    [[TMP24:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK45-32-EX-NEXT:    [[TMP25:%.*]] = icmp ne i32 [[TMP24]], 0
// CHECK45-32-EX-NEXT:    br i1 [[TMP25]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK45-32-EX:       .omp.final.then:
// CHECK45-32-EX-NEXT:    store i32 10, ptr [[I]], align 4
// CHECK45-32-EX-NEXT:    store i32 10, ptr [[J]], align 4
// CHECK45-32-EX-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK45-32-EX:       .omp.final.done:
// CHECK45-32-EX-NEXT:    ret void
//
//
// CHECK-64-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34
// CHECK-64-SAME: (i64 [[N:%.*]], ptr nonnull align 4 dereferenceable(4000) [[A:%.*]], i64 [[L:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK-64-NEXT:  entry:
// CHECK-64-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK-64-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
// CHECK-64-NEXT:    [[L_ADDR:%.*]] = alloca i64, align 8
// CHECK-64-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK-64-NEXT:    [[L_CASTED:%.*]] = alloca i64, align 8
// CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
// CHECK-64-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
// CHECK-64-NEXT:    store i64 [[L]], ptr [[L_ADDR]], align 8
// CHECK-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
// CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK-64:       user_code.entry:
// CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-64-NEXT:    [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK-64-NEXT:    store i32 [[TMP3]], ptr [[N_CASTED]], align 4
// CHECK-64-NEXT:    [[TMP4:%.*]] = load i64, ptr [[N_CASTED]], align 8
// CHECK-64-NEXT:    [[TMP5:%.*]] = load i32, ptr [[L_ADDR]], align 4
// CHECK-64-NEXT:    store i32 [[TMP5]], ptr [[L_CASTED]], align 4
// CHECK-64-NEXT:    [[TMP6:%.*]] = load i64, ptr [[L_CASTED]], align 8
// CHECK-64-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK-64-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK-64-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_omp_outlined(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i64 [[TMP4]], ptr [[TMP0]], i64 [[TMP6]]) #[[ATTR3:[0-9]+]]
// CHECK-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK-64-NEXT:    ret void
// CHECK-64:       worker.exit:
// CHECK-64-NEXT:    ret void
//
//
// CHECK-64-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_omp_outlined
// CHECK-64-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], ptr nonnull align 4 dereferenceable(4000) [[A:%.*]], i64 [[L:%.*]]) #[[ATTR1:[0-9]+]] {
// CHECK-64-NEXT:  entry:
// CHECK-64-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-64-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-64-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK-64-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
// CHECK-64-NEXT:    [[L_ADDR:%.*]] = alloca i64, align 8
// CHECK-64-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[I4:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-64-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-64-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
// CHECK-64-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
// CHECK-64-NEXT:    store i64 [[L]], ptr [[L_ADDR]], align 8
// CHECK-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
// CHECK-64-NEXT:    [[L1:%.*]] = call align 8 ptr @__kmpc_alloc_shared(i64 4)
// CHECK-64-NEXT:    [[TMP1:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK-64-NEXT:    store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK-64-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK-64-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK-64-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK-64-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK-64-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
// CHECK-64-NEXT:    store i32 0, ptr [[I]], align 4
// CHECK-64-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK-64-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK-64-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK-64:       omp.precond.then:
// CHECK-64-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-64-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
// CHECK-64-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-64-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-64-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-64-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4
// CHECK-64-NEXT:    call void @__kmpc_distribute_static_init_4(ptr @[[GLOB2:[0-9]+]], i32 [[TMP6]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 128)
// CHECK-64-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK-64:       omp.dispatch.cond:
// CHECK-64-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
// CHECK-64-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK-64-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-64:       cond.true:
// CHECK-64-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
// CHECK-64-NEXT:    br label [[COND_END:%.*]]
// CHECK-64:       cond.false:
// CHECK-64-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    br label [[COND_END]]
// CHECK-64:       cond.end:
// CHECK-64-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK-64-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-64-NEXT:    store i32 [[TMP11]], ptr [[DOTOMP_IV]], align 4
// CHECK-64-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-64-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP12]], [[TMP13]]
// CHECK-64-NEXT:    br i1 [[CMP6]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK-64:       omp.dispatch.body:
// CHECK-64-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-64:       omp.inner.for.cond:
// CHECK-64-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12:![0-9]+]]
// CHECK-64-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-64-NEXT:    [[CMP7:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK-64-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-64:       omp.inner.for.body:
// CHECK-64-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-64-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP16]], 1
// CHECK-64-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-64-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-64-NEXT:    [[TMP17:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-64-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP17]] to i64
// CHECK-64-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], ptr [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK-64-NEXT:    store i32 1, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-64-NEXT:    [[TMP18:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-64-NEXT:    store i32 [[TMP18]], ptr [[L_ADDR]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-64-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-64:       omp.body.continue:
// CHECK-64-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-64:       omp.inner.for.inc:
// CHECK-64-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-64-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK-64-NEXT:    store i32 [[ADD8]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-64-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP13:![0-9]+]]
// CHECK-64:       omp.inner.for.end:
// CHECK-64-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK-64:       omp.dispatch.inc:
// CHECK-64-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-64-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-64-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK-64-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_LB]], align 4
// CHECK-64-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-64-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK-64-NEXT:    store i32 [[ADD10]], ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK-64:       omp.dispatch.end:
// CHECK-64-NEXT:    [[TMP24:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-64-NEXT:    [[TMP25:%.*]] = load i32, ptr [[TMP24]], align 4
// CHECK-64-NEXT:    call void @__kmpc_distribute_static_fini(ptr @[[GLOB2]], i32 [[TMP25]])
// CHECK-64-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-64-NEXT:    [[TMP27:%.*]] = icmp ne i32 [[TMP26]], 0
// CHECK-64-NEXT:    br i1 [[TMP27]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK-64:       .omp.final.then:
// CHECK-64-NEXT:    [[TMP28:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK-64-NEXT:    [[SUB11:%.*]] = sub nsw i32 [[TMP28]], 0
// CHECK-64-NEXT:    [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK-64-NEXT:    [[MUL13:%.*]] = mul nsw i32 [[DIV12]], 1
// CHECK-64-NEXT:    [[ADD14:%.*]] = add nsw i32 0, [[MUL13]]
// CHECK-64-NEXT:    store i32 [[ADD14]], ptr [[I4]], align 4
// CHECK-64-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK-64:       .omp.final.done:
// CHECK-64-NEXT:    [[TMP29:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-64-NEXT:    [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
// CHECK-64-NEXT:    br i1 [[TMP30]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK-64:       .omp.lastprivate.then:
// CHECK-64-NEXT:    [[TMP31:%.*]] = load i32, ptr [[L_ADDR]], align 4
// CHECK-64-NEXT:    store i32 [[TMP31]], ptr [[L_ADDR]], align 4
// CHECK-64-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK-64:       .omp.lastprivate.done:
// CHECK-64-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK-64:       omp.precond.end:
// CHECK-64-NEXT:    call void @__kmpc_free_shared(ptr [[L1]], i64 4)
// CHECK-64-NEXT:    ret void
//
//
// CHECK-64-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l40
// CHECK-64-SAME: (i64 [[N:%.*]], ptr nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR5:[0-9]+]] {
// CHECK-64-NEXT:  entry:
// CHECK-64-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK-64-NEXT:    [[AA_ADDR:%.*]] = alloca ptr, align 8
// CHECK-64-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
// CHECK-64-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 8
// CHECK-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 8
// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
// CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK-64:       user_code.entry:
// CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-64-NEXT:    [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK-64-NEXT:    store i32 [[TMP3]], ptr [[N_CASTED]], align 4
// CHECK-64-NEXT:    [[TMP4:%.*]] = load i64, ptr [[N_CASTED]], align 8
// CHECK-64-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK-64-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK-64-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l40_omp_outlined(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i64 [[TMP4]], ptr [[TMP0]]) #[[ATTR3]]
// CHECK-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK-64-NEXT:    ret void
// CHECK-64:       worker.exit:
// CHECK-64-NEXT:    ret void
//
//
// CHECK-64-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l40_omp_outlined
// CHECK-64-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], ptr nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR1]] {
// CHECK-64-NEXT:  entry:
// CHECK-64-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-64-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-64-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK-64-NEXT:    [[AA_ADDR:%.*]] = alloca ptr, align 8
// CHECK-64-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-64-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-64-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
// CHECK-64-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 8
// CHECK-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 8
// CHECK-64-NEXT:    [[TMP1:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK-64-NEXT:    store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK-64-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK-64-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK-64-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK-64-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK-64-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK-64-NEXT:    store i32 0, ptr [[I]], align 4
// CHECK-64-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK-64-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK-64-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK-64:       omp.precond.then:
// CHECK-64-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-64-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK-64-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-64-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-64-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-64-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-64-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4
// CHECK-64-NEXT:    call void @__kmpc_distribute_static_init_4(ptr @[[GLOB2]], i32 [[TMP6]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK-64-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK-64:       omp.dispatch.cond:
// CHECK-64-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK-64-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK-64-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-64:       cond.true:
// CHECK-64-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK-64-NEXT:    br label [[COND_END:%.*]]
// CHECK-64:       cond.false:
// CHECK-64-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    br label [[COND_END]]
// CHECK-64:       cond.end:
// CHECK-64-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK-64-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-64-NEXT:    store i32 [[TMP11]], ptr [[DOTOMP_IV]], align 4
// CHECK-64-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-64-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP12]], [[TMP13]]
// CHECK-64-NEXT:    br i1 [[CMP5]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK-64:       omp.dispatch.body:
// CHECK-64-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-64:       omp.inner.for.cond:
// CHECK-64-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16:![0-9]+]]
// CHECK-64-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK-64-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK-64-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-64:       omp.inner.for.body:
// CHECK-64-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK-64-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP16]], 1
// CHECK-64-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-64-NEXT:    store i32 [[ADD]], ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK-64-NEXT:    [[TMP17:%.*]] = load i32, ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK-64-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP17]] to i64
// CHECK-64-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i16], ptr [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK-64-NEXT:    [[TMP18:%.*]] = load i16, ptr [[ARRAYIDX]], align 2, !llvm.access.group [[ACC_GRP16]]
// CHECK-64-NEXT:    [[CONV:%.*]] = sext i16 [[TMP18]] to i32
// CHECK-64-NEXT:    [[ADD7:%.*]] = add nsw i32 [[CONV]], 1
// CHECK-64-NEXT:    [[CONV8:%.*]] = trunc i32 [[ADD7]] to i16
// CHECK-64-NEXT:    store i16 [[CONV8]], ptr [[ARRAYIDX]], align 2, !llvm.access.group [[ACC_GRP16]]
// CHECK-64-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-64:       omp.body.continue:
// CHECK-64-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-64:       omp.inner.for.inc:
// CHECK-64-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK-64-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK-64-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK-64-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP17:![0-9]+]]
// CHECK-64:       omp.inner.for.end:
// CHECK-64-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK-64:       omp.dispatch.inc:
// CHECK-64-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-64-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-64-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK-64-NEXT:    store i32 [[ADD10]], ptr [[DOTOMP_LB]], align 4
// CHECK-64-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-64-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK-64-NEXT:    store i32 [[ADD11]], ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK-64:       omp.dispatch.end:
// CHECK-64-NEXT:    [[TMP24:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-64-NEXT:    [[TMP25:%.*]] = load i32, ptr [[TMP24]], align 4
// CHECK-64-NEXT:    call void @__kmpc_distribute_static_fini(ptr @[[GLOB2]], i32 [[TMP25]])
// CHECK-64-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-64-NEXT:    [[TMP27:%.*]] = icmp ne i32 [[TMP26]], 0
// CHECK-64-NEXT:    br i1 [[TMP27]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK-64:       .omp.final.then:
// CHECK-64-NEXT:    [[TMP28:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK-64-NEXT:    [[SUB12:%.*]] = sub nsw i32 [[TMP28]], 0
// CHECK-64-NEXT:    [[DIV13:%.*]] = sdiv i32 [[SUB12]], 1
// CHECK-64-NEXT:    [[MUL14:%.*]] = mul nsw i32 [[DIV13]], 1
// CHECK-64-NEXT:    [[ADD15:%.*]] = add nsw i32 0, [[MUL14]]
// CHECK-64-NEXT:    store i32 [[ADD15]], ptr [[I3]], align 4
// CHECK-64-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK-64:       .omp.final.done:
// CHECK-64-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK-64:       omp.precond.end:
// CHECK-64-NEXT:    ret void
//
//
// CHECK-64-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l45
// CHECK-64-SAME: (ptr nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK-64-NEXT:  entry:
// CHECK-64-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
// CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
// CHECK-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
// CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK-64:       user_code.entry:
// CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-64-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK-64-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK-64-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l45_omp_outlined(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]]) #[[ATTR3]]
// CHECK-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK-64-NEXT:    ret void
// CHECK-64:       worker.exit:
// CHECK-64-NEXT:    ret void
//
//
// CHECK-64-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l45_omp_outlined
// CHECK-64-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], ptr nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR1]] {
// CHECK-64-NEXT:  entry:
// CHECK-64-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-64-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-64-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
// CHECK-64-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-64-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-64-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
// CHECK-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
// CHECK-64-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-64-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-64-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-64-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-64-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-64-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK-64-NEXT:    call void @__kmpc_distribute_static_init_4(ptr @[[GLOB2]], i32 [[TMP2]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK-64-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK-64:       omp.dispatch.cond:
// CHECK-64-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9
// CHECK-64-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-64:       cond.true:
// CHECK-64-NEXT:    br label [[COND_END:%.*]]
// CHECK-64:       cond.false:
// CHECK-64-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    br label [[COND_END]]
// CHECK-64:       cond.end:
// CHECK-64-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK-64-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-64-NEXT:    store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4
// CHECK-64-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-64-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]]
// CHECK-64-NEXT:    br i1 [[CMP1]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK-64:       omp.dispatch.body:
// CHECK-64-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-64:       omp.inner.for.cond:
// CHECK-64-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19:![0-9]+]]
// CHECK-64-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK-64-NEXT:    [[CMP2:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]]
// CHECK-64-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-64:       omp.inner.for.body:
// CHECK-64-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK-64-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1
// CHECK-64-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-64-NEXT:    store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK-64-NEXT:    [[TMP11:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK-64-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP11]] to i64
// CHECK-64-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK-64-NEXT:    [[TMP12:%.*]] = load i32, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK-64-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP12]], 1
// CHECK-64-NEXT:    store i32 [[ADD3]], ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK-64-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-64:       omp.body.continue:
// CHECK-64-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-64:       omp.inner.for.inc:
// CHECK-64-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK-64-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK-64-NEXT:    store i32 [[ADD4]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK-64-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP20:![0-9]+]]
// CHECK-64:       omp.inner.for.end:
// CHECK-64-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK-64:       omp.dispatch.inc:
// CHECK-64-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-64-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-64-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP14]], [[TMP15]]
// CHECK-64-NEXT:    store i32 [[ADD5]], ptr [[DOTOMP_LB]], align 4
// CHECK-64-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-64-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP16]], [[TMP17]]
// CHECK-64-NEXT:    store i32 [[ADD6]], ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK-64:       omp.dispatch.end:
// CHECK-64-NEXT:    call void @__kmpc_distribute_static_fini(ptr @[[GLOB2]], i32 [[TMP2]])
// CHECK-64-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-64-NEXT:    [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
// CHECK-64-NEXT:    br i1 [[TMP19]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK-64:       .omp.final.then:
// CHECK-64-NEXT:    store i32 10, ptr [[I]], align 4
// CHECK-64-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK-64:       .omp.final.done:
// CHECK-64-NEXT:    ret void
//
//
// CHECK-64-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l50
// CHECK-64-SAME: (ptr nonnull align 4 dereferenceable(400) [[C:%.*]], i64 [[F:%.*]]) #[[ATTR0]] {
// CHECK-64-NEXT:  entry:
// CHECK-64-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
// CHECK-64-NEXT:    [[F_ADDR:%.*]] = alloca i64, align 8
// CHECK-64-NEXT:    [[F_CASTED:%.*]] = alloca i64, align 8
// CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
// CHECK-64-NEXT:    store i64 [[F]], ptr [[F_ADDR]], align 8
// CHECK-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 8
// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
// CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK-64:       user_code.entry:
// CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-64-NEXT:    [[TMP3:%.*]] = load i32, ptr [[F_ADDR]], align 4
// CHECK-64-NEXT:    store i32 [[TMP3]], ptr [[F_CASTED]], align 4
// CHECK-64-NEXT:    [[TMP4:%.*]] = load i64, ptr [[F_CASTED]], align 8
// CHECK-64-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK-64-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK-64-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l50_omp_outlined(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]], i64 [[TMP4]]) #[[ATTR3]]
// CHECK-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK-64-NEXT:    ret void
// CHECK-64:       worker.exit:
// CHECK-64-NEXT:    ret void
//
//
// CHECK-64-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l50_omp_outlined
// CHECK-64-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], ptr nonnull align 4 dereferenceable(400) [[C:%.*]], i64 [[F:%.*]]) #[[ATTR1]] {
// CHECK-64-NEXT:  entry:
// CHECK-64-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-64-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-64-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
// CHECK-64-NEXT:    [[F_ADDR:%.*]] = alloca i64, align 8
// CHECK-64-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK-64-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-64-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-64-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
// CHECK-64-NEXT:    store i64 [[F]], ptr [[F_ADDR]], align 8
// CHECK-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 8
// CHECK-64-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-64-NEXT:    store i32 99, ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-64-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-64-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-64-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-64-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK-64-NEXT:    call void @__kmpc_distribute_static_init_4(ptr @[[GLOB2]], i32 [[TMP2]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK-64-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK-64:       omp.dispatch.cond:
// CHECK-64-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 99
// CHECK-64-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-64:       cond.true:
// CHECK-64-NEXT:    br label [[COND_END:%.*]]
// CHECK-64:       cond.false:
// CHECK-64-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    br label [[COND_END]]
// CHECK-64:       cond.end:
// CHECK-64-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK-64-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-64-NEXT:    store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4
// CHECK-64-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-64-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    [[CMP2:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]]
// CHECK-64-NEXT:    br i1 [[CMP2]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK-64:       omp.dispatch.body:
// CHECK-64-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-64:       omp.inner.for.cond:
// CHECK-64-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22:![0-9]+]]
// CHECK-64-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-64-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]]
// CHECK-64-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-64:       omp.inner.for.body:
// CHECK-64-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-64-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP10]], 10
// CHECK-64-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK-64-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-64-NEXT:    store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-64-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-64-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-64-NEXT:    [[DIV4:%.*]] = sdiv i32 [[TMP12]], 10
// CHECK-64-NEXT:    [[MUL5:%.*]] = mul nsw i32 [[DIV4]], 10
// CHECK-64-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP11]], [[MUL5]]
// CHECK-64-NEXT:    [[MUL6:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK-64-NEXT:    [[ADD7:%.*]] = add nsw i32 0, [[MUL6]]
// CHECK-64-NEXT:    store i32 [[ADD7]], ptr [[J]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-64-NEXT:    store i32 10, ptr [[K]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-64-NEXT:    [[TMP13:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-64-NEXT:    [[TMP14:%.*]] = load i32, ptr [[J]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-64-NEXT:    [[TMP15:%.*]] = load i32, ptr [[F_ADDR]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-64-NEXT:    [[MUL8:%.*]] = mul nsw i32 [[TMP14]], [[TMP15]]
// CHECK-64-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP13]], [[MUL8]]
// CHECK-64-NEXT:    [[TMP16:%.*]] = load i32, ptr [[K]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-64-NEXT:    [[ADD10:%.*]] = add nsw i32 [[ADD9]], [[TMP16]]
// CHECK-64-NEXT:    [[TMP17:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-64-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP17]] to i64
// CHECK-64-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK-64-NEXT:    [[TMP18:%.*]] = load i32, ptr [[J]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-64-NEXT:    [[IDXPROM11:%.*]] = sext i32 [[TMP18]] to i64
// CHECK-64-NEXT:    [[ARRAYIDX12:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX]], i64 0, i64 [[IDXPROM11]]
// CHECK-64-NEXT:    store i32 [[ADD10]], ptr [[ARRAYIDX12]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-64-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-64:       omp.body.continue:
// CHECK-64-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-64:       omp.inner.for.inc:
// CHECK-64-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-64-NEXT:    [[ADD13:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK-64-NEXT:    store i32 [[ADD13]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-64-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP23:![0-9]+]]
// CHECK-64:       omp.inner.for.end:
// CHECK-64-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK-64:       omp.dispatch.inc:
// CHECK-64-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-64-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-64-NEXT:    [[ADD14:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK-64-NEXT:    store i32 [[ADD14]], ptr [[DOTOMP_LB]], align 4
// CHECK-64-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-64-NEXT:    [[ADD15:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK-64-NEXT:    store i32 [[ADD15]], ptr [[DOTOMP_UB]], align 4
// CHECK-64-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK-64:       omp.dispatch.end:
// CHECK-64-NEXT:    call void @__kmpc_distribute_static_fini(ptr @[[GLOB2]], i32 [[TMP2]])
// CHECK-64-NEXT:    [[TMP24:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-64-NEXT:    [[TMP25:%.*]] = icmp ne i32 [[TMP24]], 0
// CHECK-64-NEXT:    br i1 [[TMP25]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK-64:       .omp.final.then:
// CHECK-64-NEXT:    store i32 10, ptr [[I]], align 4
// CHECK-64-NEXT:    store i32 10, ptr [[J]], align 4
// CHECK-64-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK-64:       .omp.final.done:
// CHECK-64-NEXT:    ret void
//
//
// CHECK-32-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34
// CHECK-32-SAME: (i32 [[N:%.*]], ptr nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK-32-NEXT:  entry:
// CHECK-32-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[L_CASTED:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK-32-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
// CHECK-32-NEXT:    store i32 [[L]], ptr [[L_ADDR]], align 4
// CHECK-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
// CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK-32:       user_code.entry:
// CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-32-NEXT:    [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK-32-NEXT:    store i32 [[TMP3]], ptr [[N_CASTED]], align 4
// CHECK-32-NEXT:    [[TMP4:%.*]] = load i32, ptr [[N_CASTED]], align 4
// CHECK-32-NEXT:    [[TMP5:%.*]] = load i32, ptr [[L_ADDR]], align 4
// CHECK-32-NEXT:    store i32 [[TMP5]], ptr [[L_CASTED]], align 4
// CHECK-32-NEXT:    [[TMP6:%.*]] = load i32, ptr [[L_CASTED]], align 4
// CHECK-32-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK-32-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK-32-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_omp_outlined(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP4]], ptr [[TMP0]], i32 [[TMP6]]) #[[ATTR3:[0-9]+]]
// CHECK-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK-32-NEXT:    ret void
// CHECK-32:       worker.exit:
// CHECK-32-NEXT:    ret void
//
//
// CHECK-32-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_omp_outlined
// CHECK-32-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], ptr nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR1:[0-9]+]] {
// CHECK-32-NEXT:  entry:
// CHECK-32-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[I4:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK-32-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK-32-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK-32-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
// CHECK-32-NEXT:    store i32 [[L]], ptr [[L_ADDR]], align 4
// CHECK-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
// CHECK-32-NEXT:    [[L1:%.*]] = call align 8 ptr @__kmpc_alloc_shared(i32 4)
// CHECK-32-NEXT:    [[TMP1:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK-32-NEXT:    store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK-32-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK-32-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK-32-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK-32-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK-32-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
// CHECK-32-NEXT:    store i32 0, ptr [[I]], align 4
// CHECK-32-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK-32-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK-32-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK-32:       omp.precond.then:
// CHECK-32-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-32-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
// CHECK-32-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-32-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-32-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK-32-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4
// CHECK-32-NEXT:    call void @__kmpc_distribute_static_init_4(ptr @[[GLOB2:[0-9]+]], i32 [[TMP6]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 128)
// CHECK-32-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK-32:       omp.dispatch.cond:
// CHECK-32-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
// CHECK-32-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK-32-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-32:       cond.true:
// CHECK-32-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
// CHECK-32-NEXT:    br label [[COND_END:%.*]]
// CHECK-32:       cond.false:
// CHECK-32-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    br label [[COND_END]]
// CHECK-32:       cond.end:
// CHECK-32-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK-32-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-32-NEXT:    store i32 [[TMP11]], ptr [[DOTOMP_IV]], align 4
// CHECK-32-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-32-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP12]], [[TMP13]]
// CHECK-32-NEXT:    br i1 [[CMP6]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK-32:       omp.dispatch.body:
// CHECK-32-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-32:       omp.inner.for.cond:
// CHECK-32-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12:![0-9]+]]
// CHECK-32-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-32-NEXT:    [[CMP7:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK-32-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-32:       omp.inner.for.body:
// CHECK-32-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-32-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP16]], 1
// CHECK-32-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-32-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-32-NEXT:    [[TMP17:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-32-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], ptr [[TMP0]], i32 0, i32 [[TMP17]]
// CHECK-32-NEXT:    store i32 1, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-32-NEXT:    [[TMP18:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-32-NEXT:    store i32 [[TMP18]], ptr [[L_ADDR]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-32-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-32:       omp.body.continue:
// CHECK-32-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-32:       omp.inner.for.inc:
// CHECK-32-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-32-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK-32-NEXT:    store i32 [[ADD8]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-32-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP13:![0-9]+]]
// CHECK-32:       omp.inner.for.end:
// CHECK-32-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK-32:       omp.dispatch.inc:
// CHECK-32-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-32-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-32-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK-32-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_LB]], align 4
// CHECK-32-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-32-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK-32-NEXT:    store i32 [[ADD10]], ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK-32:       omp.dispatch.end:
// CHECK-32-NEXT:    [[TMP24:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK-32-NEXT:    [[TMP25:%.*]] = load i32, ptr [[TMP24]], align 4
// CHECK-32-NEXT:    call void @__kmpc_distribute_static_fini(ptr @[[GLOB2]], i32 [[TMP25]])
// CHECK-32-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-32-NEXT:    [[TMP27:%.*]] = icmp ne i32 [[TMP26]], 0
// CHECK-32-NEXT:    br i1 [[TMP27]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK-32:       .omp.final.then:
// CHECK-32-NEXT:    [[TMP28:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK-32-NEXT:    [[SUB11:%.*]] = sub nsw i32 [[TMP28]], 0
// CHECK-32-NEXT:    [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK-32-NEXT:    [[MUL13:%.*]] = mul nsw i32 [[DIV12]], 1
// CHECK-32-NEXT:    [[ADD14:%.*]] = add nsw i32 0, [[MUL13]]
// CHECK-32-NEXT:    store i32 [[ADD14]], ptr [[I4]], align 4
// CHECK-32-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK-32:       .omp.final.done:
// CHECK-32-NEXT:    [[TMP29:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-32-NEXT:    [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
// CHECK-32-NEXT:    br i1 [[TMP30]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK-32:       .omp.lastprivate.then:
// CHECK-32-NEXT:    [[TMP31:%.*]] = load i32, ptr [[L_ADDR]], align 4
// CHECK-32-NEXT:    store i32 [[TMP31]], ptr [[L_ADDR]], align 4
// CHECK-32-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK-32:       .omp.lastprivate.done:
// CHECK-32-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK-32:       omp.precond.end:
// CHECK-32-NEXT:    call void @__kmpc_free_shared(ptr [[L1]], i32 4)
// CHECK-32-NEXT:    ret void
//
//
// CHECK-32-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l40
// CHECK-32-SAME: (i32 [[N:%.*]], ptr nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR5:[0-9]+]] {
// CHECK-32-NEXT:  entry:
// CHECK-32-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[AA_ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK-32-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 4
// CHECK-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
// CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK-32:       user_code.entry:
// CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-32-NEXT:    [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK-32-NEXT:    store i32 [[TMP3]], ptr [[N_CASTED]], align 4
// CHECK-32-NEXT:    [[TMP4:%.*]] = load i32, ptr [[N_CASTED]], align 4
// CHECK-32-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK-32-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK-32-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l40_omp_outlined(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP4]], ptr [[TMP0]]) #[[ATTR3]]
// CHECK-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK-32-NEXT:    ret void
// CHECK-32:       worker.exit:
// CHECK-32-NEXT:    ret void
//
//
// CHECK-32-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l40_omp_outlined
// CHECK-32-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], ptr nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR1]] {
// CHECK-32-NEXT:  entry:
// CHECK-32-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[AA_ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK-32-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK-32-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK-32-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 4
// CHECK-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
// CHECK-32-NEXT:    [[TMP1:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK-32-NEXT:    store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK-32-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK-32-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK-32-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK-32-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK-32-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK-32-NEXT:    store i32 0, ptr [[I]], align 4
// CHECK-32-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK-32-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK-32-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK-32:       omp.precond.then:
// CHECK-32-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-32-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK-32-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-32-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-32-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-32-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK-32-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4
// CHECK-32-NEXT:    call void @__kmpc_distribute_static_init_4(ptr @[[GLOB2]], i32 [[TMP6]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK-32-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK-32:       omp.dispatch.cond:
// CHECK-32-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK-32-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK-32-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-32:       cond.true:
// CHECK-32-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK-32-NEXT:    br label [[COND_END:%.*]]
// CHECK-32:       cond.false:
// CHECK-32-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    br label [[COND_END]]
// CHECK-32:       cond.end:
// CHECK-32-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK-32-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-32-NEXT:    store i32 [[TMP11]], ptr [[DOTOMP_IV]], align 4
// CHECK-32-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-32-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP12]], [[TMP13]]
// CHECK-32-NEXT:    br i1 [[CMP5]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK-32:       omp.dispatch.body:
// CHECK-32-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-32:       omp.inner.for.cond:
// CHECK-32-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16:![0-9]+]]
// CHECK-32-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK-32-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK-32-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-32:       omp.inner.for.body:
// CHECK-32-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK-32-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP16]], 1
// CHECK-32-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-32-NEXT:    store i32 [[ADD]], ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK-32-NEXT:    [[TMP17:%.*]] = load i32, ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK-32-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i16], ptr [[TMP0]], i32 0, i32 [[TMP17]]
// CHECK-32-NEXT:    [[TMP18:%.*]] = load i16, ptr [[ARRAYIDX]], align 2, !llvm.access.group [[ACC_GRP16]]
// CHECK-32-NEXT:    [[CONV:%.*]] = sext i16 [[TMP18]] to i32
// CHECK-32-NEXT:    [[ADD7:%.*]] = add nsw i32 [[CONV]], 1
// CHECK-32-NEXT:    [[CONV8:%.*]] = trunc i32 [[ADD7]] to i16
// CHECK-32-NEXT:    store i16 [[CONV8]], ptr [[ARRAYIDX]], align 2, !llvm.access.group [[ACC_GRP16]]
// CHECK-32-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-32:       omp.body.continue:
// CHECK-32-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-32:       omp.inner.for.inc:
// CHECK-32-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK-32-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK-32-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK-32-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP17:![0-9]+]]
// CHECK-32:       omp.inner.for.end:
// CHECK-32-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK-32:       omp.dispatch.inc:
// CHECK-32-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-32-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-32-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK-32-NEXT:    store i32 [[ADD10]], ptr [[DOTOMP_LB]], align 4
// CHECK-32-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-32-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK-32-NEXT:    store i32 [[ADD11]], ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK-32:       omp.dispatch.end:
// CHECK-32-NEXT:    [[TMP24:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK-32-NEXT:    [[TMP25:%.*]] = load i32, ptr [[TMP24]], align 4
// CHECK-32-NEXT:    call void @__kmpc_distribute_static_fini(ptr @[[GLOB2]], i32 [[TMP25]])
// CHECK-32-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-32-NEXT:    [[TMP27:%.*]] = icmp ne i32 [[TMP26]], 0
// CHECK-32-NEXT:    br i1 [[TMP27]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK-32:       .omp.final.then:
// CHECK-32-NEXT:    [[TMP28:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK-32-NEXT:    [[SUB12:%.*]] = sub nsw i32 [[TMP28]], 0
// CHECK-32-NEXT:    [[DIV13:%.*]] = sdiv i32 [[SUB12]], 1
// CHECK-32-NEXT:    [[MUL14:%.*]] = mul nsw i32 [[DIV13]], 1
// CHECK-32-NEXT:    [[ADD15:%.*]] = add nsw i32 0, [[MUL14]]
// CHECK-32-NEXT:    store i32 [[ADD15]], ptr [[I3]], align 4
// CHECK-32-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK-32:       .omp.final.done:
// CHECK-32-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK-32:       omp.precond.end:
// CHECK-32-NEXT:    ret void
//
//
// CHECK-32-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l45
// CHECK-32-SAME: (ptr nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK-32-NEXT:  entry:
// CHECK-32-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
// CHECK-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
// CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK-32:       user_code.entry:
// CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-32-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK-32-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK-32-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l45_omp_outlined(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]]) #[[ATTR3]]
// CHECK-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK-32-NEXT:    ret void
// CHECK-32:       worker.exit:
// CHECK-32-NEXT:    ret void
//
//
// CHECK-32-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l45_omp_outlined
// CHECK-32-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], ptr nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR1]] {
// CHECK-32-NEXT:  entry:
// CHECK-32-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK-32-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK-32-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
// CHECK-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
// CHECK-32-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-32-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-32-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-32-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-32-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK-32-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK-32-NEXT:    call void @__kmpc_distribute_static_init_4(ptr @[[GLOB2]], i32 [[TMP2]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK-32-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK-32:       omp.dispatch.cond:
// CHECK-32-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9
// CHECK-32-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-32:       cond.true:
// CHECK-32-NEXT:    br label [[COND_END:%.*]]
// CHECK-32:       cond.false:
// CHECK-32-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    br label [[COND_END]]
// CHECK-32:       cond.end:
// CHECK-32-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK-32-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-32-NEXT:    store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4
// CHECK-32-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-32-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]]
// CHECK-32-NEXT:    br i1 [[CMP1]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK-32:       omp.dispatch.body:
// CHECK-32-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-32:       omp.inner.for.cond:
// CHECK-32-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19:![0-9]+]]
// CHECK-32-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK-32-NEXT:    [[CMP2:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]]
// CHECK-32-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-32:       omp.inner.for.body:
// CHECK-32-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK-32-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1
// CHECK-32-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-32-NEXT:    store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK-32-NEXT:    [[TMP11:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK-32-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i32 0, i32 [[TMP11]]
// CHECK-32-NEXT:    [[TMP12:%.*]] = load i32, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK-32-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP12]], 1
// CHECK-32-NEXT:    store i32 [[ADD3]], ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK-32-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-32:       omp.body.continue:
// CHECK-32-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-32:       omp.inner.for.inc:
// CHECK-32-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK-32-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK-32-NEXT:    store i32 [[ADD4]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK-32-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP20:![0-9]+]]
// CHECK-32:       omp.inner.for.end:
// CHECK-32-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK-32:       omp.dispatch.inc:
// CHECK-32-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-32-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-32-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP14]], [[TMP15]]
// CHECK-32-NEXT:    store i32 [[ADD5]], ptr [[DOTOMP_LB]], align 4
// CHECK-32-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-32-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP16]], [[TMP17]]
// CHECK-32-NEXT:    store i32 [[ADD6]], ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK-32:       omp.dispatch.end:
// CHECK-32-NEXT:    call void @__kmpc_distribute_static_fini(ptr @[[GLOB2]], i32 [[TMP2]])
// CHECK-32-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-32-NEXT:    [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
// CHECK-32-NEXT:    br i1 [[TMP19]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK-32:       .omp.final.then:
// CHECK-32-NEXT:    store i32 10, ptr [[I]], align 4
// CHECK-32-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK-32:       .omp.final.done:
// CHECK-32-NEXT:    ret void
//
//
// CHECK-32-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l50
// CHECK-32-SAME: (ptr nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK-32-NEXT:  entry:
// CHECK-32-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[F_CASTED:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
// CHECK-32-NEXT:    store i32 [[F]], ptr [[F_ADDR]], align 4
// CHECK-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 4
// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
// CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK-32:       user_code.entry:
// CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-32-NEXT:    [[TMP3:%.*]] = load i32, ptr [[F_ADDR]], align 4
// CHECK-32-NEXT:    store i32 [[TMP3]], ptr [[F_CASTED]], align 4
// CHECK-32-NEXT:    [[TMP4:%.*]] = load i32, ptr [[F_CASTED]], align 4
// CHECK-32-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK-32-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK-32-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l50_omp_outlined(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]], i32 [[TMP4]]) #[[ATTR3]]
// CHECK-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK-32-NEXT:    ret void
// CHECK-32:       worker.exit:
// CHECK-32-NEXT:    ret void
//
//
// CHECK-32-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l50_omp_outlined
// CHECK-32-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], ptr nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR1]] {
// CHECK-32-NEXT:  entry:
// CHECK-32-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK-32-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK-32-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK-32-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
// CHECK-32-NEXT:    store i32 [[F]], ptr [[F_ADDR]], align 4
// CHECK-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 4
// CHECK-32-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-32-NEXT:    store i32 99, ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-32-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-32-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-32-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK-32-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK-32-NEXT:    call void @__kmpc_distribute_static_init_4(ptr @[[GLOB2]], i32 [[TMP2]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK-32-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK-32:       omp.dispatch.cond:
// CHECK-32-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 99
// CHECK-32-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-32:       cond.true:
// CHECK-32-NEXT:    br label [[COND_END:%.*]]
// CHECK-32:       cond.false:
// CHECK-32-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    br label [[COND_END]]
// CHECK-32:       cond.end:
// CHECK-32-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK-32-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-32-NEXT:    store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4
// CHECK-32-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-32-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    [[CMP2:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]]
// CHECK-32-NEXT:    br i1 [[CMP2]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK-32:       omp.dispatch.body:
// CHECK-32-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-32:       omp.inner.for.cond:
// CHECK-32-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22:![0-9]+]]
// CHECK-32-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]]
// CHECK-32-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-32:       omp.inner.for.body:
// CHECK-32-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP10]], 10
// CHECK-32-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK-32-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-32-NEXT:    store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-NEXT:    [[DIV4:%.*]] = sdiv i32 [[TMP12]], 10
// CHECK-32-NEXT:    [[MUL5:%.*]] = mul nsw i32 [[DIV4]], 10
// CHECK-32-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP11]], [[MUL5]]
// CHECK-32-NEXT:    [[MUL6:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK-32-NEXT:    [[ADD7:%.*]] = add nsw i32 0, [[MUL6]]
// CHECK-32-NEXT:    store i32 [[ADD7]], ptr [[J]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-NEXT:    store i32 10, ptr [[K]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-NEXT:    [[TMP13:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-NEXT:    [[TMP14:%.*]] = load i32, ptr [[J]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-NEXT:    [[TMP15:%.*]] = load i32, ptr [[F_ADDR]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-NEXT:    [[MUL8:%.*]] = mul nsw i32 [[TMP14]], [[TMP15]]
// CHECK-32-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP13]], [[MUL8]]
// CHECK-32-NEXT:    [[TMP16:%.*]] = load i32, ptr [[K]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-NEXT:    [[ADD10:%.*]] = add nsw i32 [[ADD9]], [[TMP16]]
// CHECK-32-NEXT:    [[TMP17:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[TMP0]], i32 0, i32 [[TMP17]]
// CHECK-32-NEXT:    [[TMP18:%.*]] = load i32, ptr [[J]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX]], i32 0, i32 [[TMP18]]
// CHECK-32-NEXT:    store i32 [[ADD10]], ptr [[ARRAYIDX11]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-32:       omp.body.continue:
// CHECK-32-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-32:       omp.inner.for.inc:
// CHECK-32-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK-32-NEXT:    store i32 [[ADD12]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP23:![0-9]+]]
// CHECK-32:       omp.inner.for.end:
// CHECK-32-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK-32:       omp.dispatch.inc:
// CHECK-32-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-32-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-32-NEXT:    [[ADD13:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK-32-NEXT:    store i32 [[ADD13]], ptr [[DOTOMP_LB]], align 4
// CHECK-32-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-32-NEXT:    [[ADD14:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK-32-NEXT:    store i32 [[ADD14]], ptr [[DOTOMP_UB]], align 4
// CHECK-32-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK-32:       omp.dispatch.end:
// CHECK-32-NEXT:    call void @__kmpc_distribute_static_fini(ptr @[[GLOB2]], i32 [[TMP2]])
// CHECK-32-NEXT:    [[TMP24:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-32-NEXT:    [[TMP25:%.*]] = icmp ne i32 [[TMP24]], 0
// CHECK-32-NEXT:    br i1 [[TMP25]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK-32:       .omp.final.then:
// CHECK-32-NEXT:    store i32 10, ptr [[I]], align 4
// CHECK-32-NEXT:    store i32 10, ptr [[J]], align 4
// CHECK-32-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK-32:       .omp.final.done:
// CHECK-32-NEXT:    ret void
//
//
// CHECK-32-EX-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34
// CHECK-32-EX-SAME: (i32 [[N:%.*]], ptr nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK-32-EX-NEXT:  entry:
// CHECK-32-EX-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-EX-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[L_CASTED:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK-32-EX-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
// CHECK-32-EX-NEXT:    store i32 [[L]], ptr [[L_ADDR]], align 4
// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
// CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK-32-EX:       user_code.entry:
// CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-32-EX-NEXT:    [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK-32-EX-NEXT:    store i32 [[TMP3]], ptr [[N_CASTED]], align 4
// CHECK-32-EX-NEXT:    [[TMP4:%.*]] = load i32, ptr [[N_CASTED]], align 4
// CHECK-32-EX-NEXT:    [[TMP5:%.*]] = load i32, ptr [[L_ADDR]], align 4
// CHECK-32-EX-NEXT:    store i32 [[TMP5]], ptr [[L_CASTED]], align 4
// CHECK-32-EX-NEXT:    [[TMP6:%.*]] = load i32, ptr [[L_CASTED]], align 4
// CHECK-32-EX-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK-32-EX-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK-32-EX-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_omp_outlined(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP4]], ptr [[TMP0]], i32 [[TMP6]]) #[[ATTR3:[0-9]+]]
// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK-32-EX-NEXT:    ret void
// CHECK-32-EX:       worker.exit:
// CHECK-32-EX-NEXT:    ret void
//
//
// CHECK-32-EX-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_omp_outlined
// CHECK-32-EX-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], ptr nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR1:[0-9]+]] {
// CHECK-32-EX-NEXT:  entry:
// CHECK-32-EX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-EX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-EX-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-EX-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[I4:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK-32-EX-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK-32-EX-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK-32-EX-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
// CHECK-32-EX-NEXT:    store i32 [[L]], ptr [[L_ADDR]], align 4
// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
// CHECK-32-EX-NEXT:    [[L1:%.*]] = call align 8 ptr @__kmpc_alloc_shared(i32 4)
// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK-32-EX-NEXT:    store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK-32-EX-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK-32-EX-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK-32-EX-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK-32-EX-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK-32-EX-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
// CHECK-32-EX-NEXT:    store i32 0, ptr [[I]], align 4
// CHECK-32-EX-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK-32-EX-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK-32-EX-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK-32-EX:       omp.precond.then:
// CHECK-32-EX-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-32-EX-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
// CHECK-32-EX-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-32-EX-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-32-EX-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK-32-EX-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4
// CHECK-32-EX-NEXT:    call void @__kmpc_distribute_static_init_4(ptr @[[GLOB2:[0-9]+]], i32 [[TMP6]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 128)
// CHECK-32-EX-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK-32-EX:       omp.dispatch.cond:
// CHECK-32-EX-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
// CHECK-32-EX-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK-32-EX-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-32-EX:       cond.true:
// CHECK-32-EX-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
// CHECK-32-EX-NEXT:    br label [[COND_END:%.*]]
// CHECK-32-EX:       cond.false:
// CHECK-32-EX-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    br label [[COND_END]]
// CHECK-32-EX:       cond.end:
// CHECK-32-EX-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK-32-EX-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-32-EX-NEXT:    store i32 [[TMP11]], ptr [[DOTOMP_IV]], align 4
// CHECK-32-EX-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-32-EX-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP12]], [[TMP13]]
// CHECK-32-EX-NEXT:    br i1 [[CMP6]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK-32-EX:       omp.dispatch.body:
// CHECK-32-EX-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-32-EX:       omp.inner.for.cond:
// CHECK-32-EX-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12:![0-9]+]]
// CHECK-32-EX-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-32-EX-NEXT:    [[CMP7:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK-32-EX-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-32-EX:       omp.inner.for.body:
// CHECK-32-EX-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-32-EX-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP16]], 1
// CHECK-32-EX-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-32-EX-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-32-EX-NEXT:    [[TMP17:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-32-EX-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], ptr [[TMP0]], i32 0, i32 [[TMP17]]
// CHECK-32-EX-NEXT:    store i32 1, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-32-EX-NEXT:    [[TMP18:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-32-EX-NEXT:    store i32 [[TMP18]], ptr [[L_ADDR]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-32-EX-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-32-EX:       omp.body.continue:
// CHECK-32-EX-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-32-EX:       omp.inner.for.inc:
// CHECK-32-EX-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-32-EX-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK-32-EX-NEXT:    store i32 [[ADD8]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]]
// CHECK-32-EX-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP13:![0-9]+]]
// CHECK-32-EX:       omp.inner.for.end:
// CHECK-32-EX-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK-32-EX:       omp.dispatch.inc:
// CHECK-32-EX-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-32-EX-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-32-EX-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK-32-EX-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_LB]], align 4
// CHECK-32-EX-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-32-EX-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK-32-EX-NEXT:    store i32 [[ADD10]], ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK-32-EX:       omp.dispatch.end:
// CHECK-32-EX-NEXT:    [[TMP24:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK-32-EX-NEXT:    [[TMP25:%.*]] = load i32, ptr [[TMP24]], align 4
// CHECK-32-EX-NEXT:    call void @__kmpc_distribute_static_fini(ptr @[[GLOB2]], i32 [[TMP25]])
// CHECK-32-EX-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-32-EX-NEXT:    [[TMP27:%.*]] = icmp ne i32 [[TMP26]], 0
// CHECK-32-EX-NEXT:    br i1 [[TMP27]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK-32-EX:       .omp.final.then:
// CHECK-32-EX-NEXT:    [[TMP28:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK-32-EX-NEXT:    [[SUB11:%.*]] = sub nsw i32 [[TMP28]], 0
// CHECK-32-EX-NEXT:    [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1
// CHECK-32-EX-NEXT:    [[MUL13:%.*]] = mul nsw i32 [[DIV12]], 1
// CHECK-32-EX-NEXT:    [[ADD14:%.*]] = add nsw i32 0, [[MUL13]]
// CHECK-32-EX-NEXT:    store i32 [[ADD14]], ptr [[I4]], align 4
// CHECK-32-EX-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK-32-EX:       .omp.final.done:
// CHECK-32-EX-NEXT:    [[TMP29:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-32-EX-NEXT:    [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
// CHECK-32-EX-NEXT:    br i1 [[TMP30]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK-32-EX:       .omp.lastprivate.then:
// CHECK-32-EX-NEXT:    [[TMP31:%.*]] = load i32, ptr [[L_ADDR]], align 4
// CHECK-32-EX-NEXT:    store i32 [[TMP31]], ptr [[L_ADDR]], align 4
// CHECK-32-EX-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK-32-EX:       .omp.lastprivate.done:
// CHECK-32-EX-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK-32-EX:       omp.precond.end:
// CHECK-32-EX-NEXT:    call void @__kmpc_free_shared(ptr [[L1]], i32 4)
// CHECK-32-EX-NEXT:    ret void
//
//
// CHECK-32-EX-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l40
// CHECK-32-EX-SAME: (i32 [[N:%.*]], ptr nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR5:[0-9]+]] {
// CHECK-32-EX-NEXT:  entry:
// CHECK-32-EX-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[AA_ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-EX-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK-32-EX-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 4
// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
// CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK-32-EX:       user_code.entry:
// CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-32-EX-NEXT:    [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK-32-EX-NEXT:    store i32 [[TMP3]], ptr [[N_CASTED]], align 4
// CHECK-32-EX-NEXT:    [[TMP4:%.*]] = load i32, ptr [[N_CASTED]], align 4
// CHECK-32-EX-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK-32-EX-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK-32-EX-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l40_omp_outlined(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP4]], ptr [[TMP0]]) #[[ATTR3]]
// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK-32-EX-NEXT:    ret void
// CHECK-32-EX:       worker.exit:
// CHECK-32-EX-NEXT:    ret void
//
//
// CHECK-32-EX-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l40_omp_outlined
// CHECK-32-EX-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], ptr nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR1]] {
// CHECK-32-EX-NEXT:  entry:
// CHECK-32-EX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-EX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-EX-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[AA_ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-EX-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK-32-EX-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK-32-EX-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK-32-EX-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 4
// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK-32-EX-NEXT:    store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK-32-EX-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK-32-EX-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK-32-EX-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK-32-EX-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK-32-EX-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK-32-EX-NEXT:    store i32 0, ptr [[I]], align 4
// CHECK-32-EX-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK-32-EX-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK-32-EX-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK-32-EX:       omp.precond.then:
// CHECK-32-EX-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-32-EX-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK-32-EX-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-32-EX-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-32-EX-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-32-EX-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK-32-EX-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4
// CHECK-32-EX-NEXT:    call void @__kmpc_distribute_static_init_4(ptr @[[GLOB2]], i32 [[TMP6]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK-32-EX-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK-32-EX:       omp.dispatch.cond:
// CHECK-32-EX-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK-32-EX-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK-32-EX-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-32-EX:       cond.true:
// CHECK-32-EX-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
// CHECK-32-EX-NEXT:    br label [[COND_END:%.*]]
// CHECK-32-EX:       cond.false:
// CHECK-32-EX-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    br label [[COND_END]]
// CHECK-32-EX:       cond.end:
// CHECK-32-EX-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK-32-EX-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-32-EX-NEXT:    store i32 [[TMP11]], ptr [[DOTOMP_IV]], align 4
// CHECK-32-EX-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-32-EX-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP12]], [[TMP13]]
// CHECK-32-EX-NEXT:    br i1 [[CMP5]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK-32-EX:       omp.dispatch.body:
// CHECK-32-EX-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-32-EX:       omp.inner.for.cond:
// CHECK-32-EX-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16:![0-9]+]]
// CHECK-32-EX-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK-32-EX-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK-32-EX-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-32-EX:       omp.inner.for.body:
// CHECK-32-EX-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK-32-EX-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP16]], 1
// CHECK-32-EX-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-32-EX-NEXT:    store i32 [[ADD]], ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK-32-EX-NEXT:    [[TMP17:%.*]] = load i32, ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK-32-EX-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i16], ptr [[TMP0]], i32 0, i32 [[TMP17]]
// CHECK-32-EX-NEXT:    [[TMP18:%.*]] = load i16, ptr [[ARRAYIDX]], align 2, !llvm.access.group [[ACC_GRP16]]
// CHECK-32-EX-NEXT:    [[CONV:%.*]] = sext i16 [[TMP18]] to i32
// CHECK-32-EX-NEXT:    [[ADD7:%.*]] = add nsw i32 [[CONV]], 1
// CHECK-32-EX-NEXT:    [[CONV8:%.*]] = trunc i32 [[ADD7]] to i16
// CHECK-32-EX-NEXT:    store i16 [[CONV8]], ptr [[ARRAYIDX]], align 2, !llvm.access.group [[ACC_GRP16]]
// CHECK-32-EX-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-32-EX:       omp.body.continue:
// CHECK-32-EX-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-32-EX:       omp.inner.for.inc:
// CHECK-32-EX-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK-32-EX-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK-32-EX-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16]]
// CHECK-32-EX-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP17:![0-9]+]]
// CHECK-32-EX:       omp.inner.for.end:
// CHECK-32-EX-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK-32-EX:       omp.dispatch.inc:
// CHECK-32-EX-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-32-EX-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-32-EX-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK-32-EX-NEXT:    store i32 [[ADD10]], ptr [[DOTOMP_LB]], align 4
// CHECK-32-EX-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-32-EX-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK-32-EX-NEXT:    store i32 [[ADD11]], ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK-32-EX:       omp.dispatch.end:
// CHECK-32-EX-NEXT:    [[TMP24:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK-32-EX-NEXT:    [[TMP25:%.*]] = load i32, ptr [[TMP24]], align 4
// CHECK-32-EX-NEXT:    call void @__kmpc_distribute_static_fini(ptr @[[GLOB2]], i32 [[TMP25]])
// CHECK-32-EX-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-32-EX-NEXT:    [[TMP27:%.*]] = icmp ne i32 [[TMP26]], 0
// CHECK-32-EX-NEXT:    br i1 [[TMP27]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK-32-EX:       .omp.final.then:
// CHECK-32-EX-NEXT:    [[TMP28:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
// CHECK-32-EX-NEXT:    [[SUB12:%.*]] = sub nsw i32 [[TMP28]], 0
// CHECK-32-EX-NEXT:    [[DIV13:%.*]] = sdiv i32 [[SUB12]], 1
// CHECK-32-EX-NEXT:    [[MUL14:%.*]] = mul nsw i32 [[DIV13]], 1
// CHECK-32-EX-NEXT:    [[ADD15:%.*]] = add nsw i32 0, [[MUL14]]
// CHECK-32-EX-NEXT:    store i32 [[ADD15]], ptr [[I3]], align 4
// CHECK-32-EX-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK-32-EX:       .omp.final.done:
// CHECK-32-EX-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK-32-EX:       omp.precond.end:
// CHECK-32-EX-NEXT:    ret void
//
//
// CHECK-32-EX-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l45
// CHECK-32-EX-SAME: (ptr nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK-32-EX-NEXT:  entry:
// CHECK-32-EX-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
// CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK-32-EX:       user_code.entry:
// CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-32-EX-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK-32-EX-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK-32-EX-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l45_omp_outlined(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]]) #[[ATTR3]]
// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK-32-EX-NEXT:    ret void
// CHECK-32-EX:       worker.exit:
// CHECK-32-EX-NEXT:    ret void
//
//
// CHECK-32-EX-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l45_omp_outlined
// CHECK-32-EX-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], ptr nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR1]] {
// CHECK-32-EX-NEXT:  entry:
// CHECK-32-EX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-EX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-EX-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-EX-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK-32-EX-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK-32-EX-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
// CHECK-32-EX-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-32-EX-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-32-EX-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-32-EX-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK-32-EX-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK-32-EX-NEXT:    call void @__kmpc_distribute_static_init_4(ptr @[[GLOB2]], i32 [[TMP2]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK-32-EX-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK-32-EX:       omp.dispatch.cond:
// CHECK-32-EX-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9
// CHECK-32-EX-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-32-EX:       cond.true:
// CHECK-32-EX-NEXT:    br label [[COND_END:%.*]]
// CHECK-32-EX:       cond.false:
// CHECK-32-EX-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    br label [[COND_END]]
// CHECK-32-EX:       cond.end:
// CHECK-32-EX-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK-32-EX-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-32-EX-NEXT:    store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4
// CHECK-32-EX-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-32-EX-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]]
// CHECK-32-EX-NEXT:    br i1 [[CMP1]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK-32-EX:       omp.dispatch.body:
// CHECK-32-EX-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-32-EX:       omp.inner.for.cond:
// CHECK-32-EX-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19:![0-9]+]]
// CHECK-32-EX-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK-32-EX-NEXT:    [[CMP2:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]]
// CHECK-32-EX-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-32-EX:       omp.inner.for.body:
// CHECK-32-EX-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK-32-EX-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1
// CHECK-32-EX-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-32-EX-NEXT:    store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK-32-EX-NEXT:    [[TMP11:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK-32-EX-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i32 0, i32 [[TMP11]]
// CHECK-32-EX-NEXT:    [[TMP12:%.*]] = load i32, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK-32-EX-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP12]], 1
// CHECK-32-EX-NEXT:    store i32 [[ADD3]], ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK-32-EX-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-32-EX:       omp.body.continue:
// CHECK-32-EX-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-32-EX:       omp.inner.for.inc:
// CHECK-32-EX-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK-32-EX-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK-32-EX-NEXT:    store i32 [[ADD4]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19]]
// CHECK-32-EX-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP20:![0-9]+]]
// CHECK-32-EX:       omp.inner.for.end:
// CHECK-32-EX-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK-32-EX:       omp.dispatch.inc:
// CHECK-32-EX-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-32-EX-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-32-EX-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP14]], [[TMP15]]
// CHECK-32-EX-NEXT:    store i32 [[ADD5]], ptr [[DOTOMP_LB]], align 4
// CHECK-32-EX-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-32-EX-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP16]], [[TMP17]]
// CHECK-32-EX-NEXT:    store i32 [[ADD6]], ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK-32-EX:       omp.dispatch.end:
// CHECK-32-EX-NEXT:    call void @__kmpc_distribute_static_fini(ptr @[[GLOB2]], i32 [[TMP2]])
// CHECK-32-EX-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-32-EX-NEXT:    [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
// CHECK-32-EX-NEXT:    br i1 [[TMP19]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK-32-EX:       .omp.final.then:
// CHECK-32-EX-NEXT:    store i32 10, ptr [[I]], align 4
// CHECK-32-EX-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK-32-EX:       .omp.final.done:
// CHECK-32-EX-NEXT:    ret void
//
//
// CHECK-32-EX-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l50
// CHECK-32-EX-SAME: (ptr nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK-32-EX-NEXT:  entry:
// CHECK-32-EX-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-EX-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[F_CASTED:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
// CHECK-32-EX-NEXT:    store i32 [[F]], ptr [[F_ADDR]], align 4
// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 4
// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
// CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK-32-EX:       user_code.entry:
// CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-32-EX-NEXT:    [[TMP3:%.*]] = load i32, ptr [[F_ADDR]], align 4
// CHECK-32-EX-NEXT:    store i32 [[TMP3]], ptr [[F_CASTED]], align 4
// CHECK-32-EX-NEXT:    [[TMP4:%.*]] = load i32, ptr [[F_CASTED]], align 4
// CHECK-32-EX-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK-32-EX-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK-32-EX-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l50_omp_outlined(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]], i32 [[TMP4]]) #[[ATTR3]]
// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK-32-EX-NEXT:    ret void
// CHECK-32-EX:       worker.exit:
// CHECK-32-EX-NEXT:    ret void
//
//
// CHECK-32-EX-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l50_omp_outlined
// CHECK-32-EX-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], ptr nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR1]] {
// CHECK-32-EX-NEXT:  entry:
// CHECK-32-EX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-EX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-EX-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
// CHECK-32-EX-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK-32-EX-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK-32-EX-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK-32-EX-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
// CHECK-32-EX-NEXT:    store i32 [[F]], ptr [[F_ADDR]], align 4
// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 4
// CHECK-32-EX-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-32-EX-NEXT:    store i32 99, ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-32-EX-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-32-EX-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK-32-EX-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK-32-EX-NEXT:    call void @__kmpc_distribute_static_init_4(ptr @[[GLOB2]], i32 [[TMP2]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK-32-EX-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK-32-EX:       omp.dispatch.cond:
// CHECK-32-EX-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 99
// CHECK-32-EX-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-32-EX:       cond.true:
// CHECK-32-EX-NEXT:    br label [[COND_END:%.*]]
// CHECK-32-EX:       cond.false:
// CHECK-32-EX-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    br label [[COND_END]]
// CHECK-32-EX:       cond.end:
// CHECK-32-EX-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK-32-EX-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-32-EX-NEXT:    store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4
// CHECK-32-EX-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-32-EX-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    [[CMP2:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]]
// CHECK-32-EX-NEXT:    br i1 [[CMP2]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK-32-EX:       omp.dispatch.body:
// CHECK-32-EX-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-32-EX:       omp.inner.for.cond:
// CHECK-32-EX-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22:![0-9]+]]
// CHECK-32-EX-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-EX-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]]
// CHECK-32-EX-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-32-EX:       omp.inner.for.body:
// CHECK-32-EX-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-EX-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP10]], 10
// CHECK-32-EX-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK-32-EX-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-32-EX-NEXT:    store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-EX-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-EX-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-EX-NEXT:    [[DIV4:%.*]] = sdiv i32 [[TMP12]], 10
// CHECK-32-EX-NEXT:    [[MUL5:%.*]] = mul nsw i32 [[DIV4]], 10
// CHECK-32-EX-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP11]], [[MUL5]]
// CHECK-32-EX-NEXT:    [[MUL6:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK-32-EX-NEXT:    [[ADD7:%.*]] = add nsw i32 0, [[MUL6]]
// CHECK-32-EX-NEXT:    store i32 [[ADD7]], ptr [[J]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-EX-NEXT:    store i32 10, ptr [[K]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-EX-NEXT:    [[TMP13:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-EX-NEXT:    [[TMP14:%.*]] = load i32, ptr [[J]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-EX-NEXT:    [[TMP15:%.*]] = load i32, ptr [[F_ADDR]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-EX-NEXT:    [[MUL8:%.*]] = mul nsw i32 [[TMP14]], [[TMP15]]
// CHECK-32-EX-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP13]], [[MUL8]]
// CHECK-32-EX-NEXT:    [[TMP16:%.*]] = load i32, ptr [[K]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-EX-NEXT:    [[ADD10:%.*]] = add nsw i32 [[ADD9]], [[TMP16]]
// CHECK-32-EX-NEXT:    [[TMP17:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-EX-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[TMP0]], i32 0, i32 [[TMP17]]
// CHECK-32-EX-NEXT:    [[TMP18:%.*]] = load i32, ptr [[J]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-EX-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX]], i32 0, i32 [[TMP18]]
// CHECK-32-EX-NEXT:    store i32 [[ADD10]], ptr [[ARRAYIDX11]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-EX-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-32-EX:       omp.body.continue:
// CHECK-32-EX-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-32-EX:       omp.inner.for.inc:
// CHECK-32-EX-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-EX-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK-32-EX-NEXT:    store i32 [[ADD12]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
// CHECK-32-EX-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP23:![0-9]+]]
// CHECK-32-EX:       omp.inner.for.end:
// CHECK-32-EX-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK-32-EX:       omp.dispatch.inc:
// CHECK-32-EX-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-32-EX-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-32-EX-NEXT:    [[ADD13:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK-32-EX-NEXT:    store i32 [[ADD13]], ptr [[DOTOMP_LB]], align 4
// CHECK-32-EX-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-32-EX-NEXT:    [[ADD14:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK-32-EX-NEXT:    store i32 [[ADD14]], ptr [[DOTOMP_UB]], align 4
// CHECK-32-EX-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK-32-EX:       omp.dispatch.end:
// CHECK-32-EX-NEXT:    call void @__kmpc_distribute_static_fini(ptr @[[GLOB2]], i32 [[TMP2]])
// CHECK-32-EX-NEXT:    [[TMP24:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-32-EX-NEXT:    [[TMP25:%.*]] = icmp ne i32 [[TMP24]], 0
// CHECK-32-EX-NEXT:    br i1 [[TMP25]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK-32-EX:       .omp.final.then:
// CHECK-32-EX-NEXT:    store i32 10, ptr [[I]], align 4
// CHECK-32-EX-NEXT:    store i32 10, ptr [[J]], align 4
// CHECK-32-EX-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK-32-EX:       .omp.final.done:
// CHECK-32-EX-NEXT:    ret void
//