clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_codegen.cpp

// 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 -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 -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=CHECK1
// RUN: %clang_cc1 -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 -fopenmp-optimistic-collapse -o - | FileCheck %s --check-prefix=CHECK2
// RUN: %clang_cc1 -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 -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=CHECK3
// RUN: %clang_cc1 -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=CHECK4

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

#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;
  tx *v;

#pragma omp target teams distribute parallel for lastprivate(l) dist_schedule(static,128) schedule(static,32)
  for(int i = 0; i < n; i++) {
    a[i] = 1;
    l = i;
  }

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

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

#pragma omp target teams distribute parallel for 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;
    }
  }

#pragma omp target teams distribute parallel for collapse(2)
  for(int i = 0; i < n; i++) {
    for(int j = 0; j < n; j++) {
      c[i][j] = i + j;
    }
  }

#pragma omp target teams distribute parallel for map(a, v[:N])
  for(int i = 0; i < n; i++)
    a[i] = v[i];
  return a[0];
}

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

  a += ftemplate<int>(n);

  return a;
}

#endif
// CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l43
// CHECK5-SAME: (i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK5-NEXT:  entry:
// CHECK5-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK5-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[L_CASTED:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK5-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK5-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK5-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK5-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK5:       .execute:
// CHECK5-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3:[0-9]+]])
// CHECK5-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK5-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK5-NEXT:    [[TMP4:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP4]], i32* [[L_CASTED]], align 4
// CHECK5-NEXT:    [[TMP5:%.*]] = load i32, i32* [[L_CASTED]], align 4
// CHECK5-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK5-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [1000 x i32]* [[TMP0]], i32 [[TMP5]]) #[[ATTR3:[0-9]+]]
// CHECK5-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK5:       .omp.deinit:
// CHECK5-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK5-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK5:       .exit:
// CHECK5-NEXT:    ret void
// CHECK5-LABEL: define {{[^@]+}}@__omp_outlined__
// CHECK5-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0]] {
// CHECK5-NEXT:  entry:
// CHECK5-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK5-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[I4:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[L_CASTED:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 4
// CHECK5-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK5-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK5-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK5-NEXT:    [[TMP1:%.*]] = load i16, i16* @"_openmp_static_kernel$is_shared", align 2
// CHECK5-NEXT:    [[TMP2:%.*]] = load i32, i32* @"_openmp_static_kernel$size", align 4
// CHECK5-NEXT:    call void @__kmpc_get_team_static_memory(i16 1, i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds (%"union._shared_openmp_static_memory_type_$_", %"union._shared_openmp_static_memory_type_$_" addrspace(3)* @"_openmp_shared_static_glob_rd_$_", i32 0, i32 0, i32 0) to i8*), i32 [[TMP2]], i16 [[TMP1]], i8** addrspacecast (i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr" to i8**))
// CHECK5-NEXT:    [[TMP3:%.*]] = load i8*, i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr", align 4
// CHECK5-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i8, i8* [[TMP3]], i32 0
// CHECK5-NEXT:    [[TMP5:%.*]] = bitcast i8* [[TMP4]] to %struct._globalized_locals_ty*
// CHECK5-NEXT:    [[L1:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY:%.*]], %struct._globalized_locals_ty* [[TMP5]], i32 0, i32 0
// CHECK5-NEXT:    [[TMP6:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP6]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], 0
// CHECK5-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK5-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK5-NEXT:    store i32 [[SUB3]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK5-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK5-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP8]]
// CHECK5-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK5:       omp.precond.then:
// CHECK5-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK5-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK5-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK5-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1:[0-9]+]], i32 [[TMP11]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 128)
// CHECK5-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK5-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
// CHECK5-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK5:       cond.true:
// CHECK5-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK5-NEXT:    br label [[COND_END:%.*]]
// CHECK5:       cond.false:
// CHECK5-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    br label [[COND_END]]
// CHECK5:       cond.end:
// CHECK5-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
// CHECK5-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    store i32 [[TMP16]], i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK5:       omp.inner.for.cond:
// CHECK5-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK5-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP18]], 1
// CHECK5-NEXT:    [[CMP6:%.*]] = icmp slt i32 [[TMP17]], [[ADD]]
// CHECK5-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK5:       omp.inner.for.body:
// CHECK5-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP21:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP21]], i32* [[N_CASTED]], align 4
// CHECK5-NEXT:    [[TMP22:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK5-NEXT:    [[TMP23:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP23]], i32* [[L_CASTED]], align 4
// CHECK5-NEXT:    [[TMP24:%.*]] = load i32, i32* [[L_CASTED]], align 4
// CHECK5-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK5-NEXT:    [[TMP26:%.*]] = inttoptr i32 [[TMP19]] to i8*
// CHECK5-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 4
// CHECK5-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK5-NEXT:    [[TMP28:%.*]] = inttoptr i32 [[TMP20]] to i8*
// CHECK5-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 4
// CHECK5-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK5-NEXT:    [[TMP30:%.*]] = inttoptr i32 [[TMP22]] to i8*
// CHECK5-NEXT:    store i8* [[TMP30]], i8** [[TMP29]], align 4
// CHECK5-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK5-NEXT:    [[TMP32:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK5-NEXT:    store i8* [[TMP32]], i8** [[TMP31]], align 4
// CHECK5-NEXT:    [[TMP33:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 4
// CHECK5-NEXT:    [[TMP34:%.*]] = inttoptr i32 [[TMP24]] to i8*
// CHECK5-NEXT:    store i8* [[TMP34]], i8** [[TMP33]], align 4
// CHECK5-NEXT:    [[TMP35:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    [[TMP36:%.*]] = load i32, i32* [[TMP35]], align 4
// CHECK5-NEXT:    [[TMP37:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK5-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP36]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i32]*, i32)* @__omp_outlined__1 to i8*), i8* null, i8** [[TMP37]], i32 5)
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK5:       omp.inner.for.inc:
// CHECK5-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP38]], [[TMP39]]
// CHECK5-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP40]], [[TMP41]]
// CHECK5-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP43:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP42]], [[TMP43]]
// CHECK5-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP44:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP45:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK5-NEXT:    [[CMP10:%.*]] = icmp sgt i32 [[TMP44]], [[TMP45]]
// CHECK5-NEXT:    br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]]
// CHECK5:       cond.true11:
// CHECK5-NEXT:    [[TMP46:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK5-NEXT:    br label [[COND_END13:%.*]]
// CHECK5:       cond.false12:
// CHECK5-NEXT:    [[TMP47:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    br label [[COND_END13]]
// CHECK5:       cond.end13:
// CHECK5-NEXT:    [[COND14:%.*]] = phi i32 [ [[TMP46]], [[COND_TRUE11]] ], [ [[TMP47]], [[COND_FALSE12]] ]
// CHECK5-NEXT:    store i32 [[COND14]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP48:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    store i32 [[TMP48]], i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK5:       omp.inner.for.end:
// CHECK5-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK5:       omp.loop.exit:
// CHECK5-NEXT:    [[TMP49:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    [[TMP50:%.*]] = load i32, i32* [[TMP49]], align 4
// CHECK5-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP50]])
// CHECK5-NEXT:    [[TMP51:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK5-NEXT:    [[TMP52:%.*]] = icmp ne i32 [[TMP51]], 0
// CHECK5-NEXT:    br i1 [[TMP52]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK5:       .omp.lastprivate.then:
// CHECK5-NEXT:    [[TMP53:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP53]], i32* [[L_ADDR]], align 4
// CHECK5-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK5:       .omp.lastprivate.done:
// CHECK5-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK5:       omp.precond.end:
// CHECK5-NEXT:    [[TMP54:%.*]] = load i16, i16* @"_openmp_static_kernel$is_shared", align 2
// CHECK5-NEXT:    call void @__kmpc_restore_team_static_memory(i16 1, i16 [[TMP54]])
// CHECK5-NEXT:    ret void
// CHECK5-LABEL: define {{[^@]+}}@__omp_outlined__1
// CHECK5-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0]] {
// CHECK5-NEXT:  entry:
// CHECK5-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK5-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK5-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK5-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK5-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK5-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK5-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK5-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK5-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK5-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK5-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK5-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK5-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK5:       omp.precond.then:
// CHECK5-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK5-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK5-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK5-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK5-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK5-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK5-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK5-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK5-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 32)
// CHECK5-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK5:       omp.dispatch.cond:
// CHECK5-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK5-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK5-NEXT:    [[CMP4:%.*]] = icmp ugt i32 [[TMP9]], [[TMP10]]
// CHECK5-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK5:       cond.true:
// CHECK5-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK5-NEXT:    br label [[COND_END:%.*]]
// CHECK5:       cond.false:
// CHECK5-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK5-NEXT:    br label [[COND_END]]
// CHECK5:       cond.end:
// CHECK5-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP11]], [[COND_TRUE]] ], [ [[TMP12]], [[COND_FALSE]] ]
// CHECK5-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
// CHECK5-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK5-NEXT:    store i32 [[TMP13]], i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK5-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK5-NEXT:    br i1 [[CMP5]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK5:       omp.dispatch.body:
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK5:       omp.inner.for.cond:
// CHECK5-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK5-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP16]], [[TMP17]]
// CHECK5-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK5:       omp.inner.for.body:
// CHECK5-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP18]], 1
// CHECK5-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK5-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK5-NEXT:    [[TMP19:%.*]] = load i32, i32* [[I3]], align 4
// CHECK5-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i32 0, i32 [[TMP19]]
// CHECK5-NEXT:    store i32 1, i32* [[ARRAYIDX]], align 4
// CHECK5-NEXT:    [[TMP20:%.*]] = load i32, i32* [[I3]], align 4
// CHECK5-NEXT:    store i32 [[TMP20]], i32* [[L_ADDR]], align 4
// CHECK5-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK5:       omp.body.continue:
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK5:       omp.inner.for.inc:
// CHECK5-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP21]], 1
// CHECK5-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK5:       omp.inner.for.end:
// CHECK5-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK5:       omp.dispatch.inc:
// CHECK5-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK5-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK5-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_LB]], align 4
// CHECK5-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK5-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK5-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_UB]], align 4
// CHECK5-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK5:       omp.dispatch.end:
// CHECK5-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK5-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP27]])
// CHECK5-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK5-NEXT:    [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
// CHECK5-NEXT:    br i1 [[TMP29]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK5:       .omp.lastprivate.then:
// CHECK5-NEXT:    [[TMP30:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP30]], i32* [[L_ADDR]], align 4
// CHECK5-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK5:       .omp.lastprivate.done:
// CHECK5-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK5:       omp.precond.end:
// CHECK5-NEXT:    ret void
// CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l49
// CHECK5-SAME: (i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK5-NEXT:  entry:
// CHECK5-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK5-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK5-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK5-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK5-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK5-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK5:       .execute:
// CHECK5-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK5-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK5-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK5-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK5-NEXT:    call void @__omp_outlined__2(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [1000 x i16]* [[TMP0]]) #[[ATTR3]]
// CHECK5-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK5:       .omp.deinit:
// CHECK5-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK5-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK5:       .exit:
// CHECK5-NEXT:    ret void
// CHECK5-LABEL: define {{[^@]+}}@__omp_outlined__2
// CHECK5-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK5-NEXT:  entry:
// CHECK5-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK5-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK5-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK5-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK5-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK5-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK5-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK5-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK5-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK5-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK5-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK5-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK5:       omp.precond.then:
// CHECK5-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK5-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK5-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK5-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK5-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK5-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK5-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK5-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK5:       cond.true:
// CHECK5-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK5-NEXT:    br label [[COND_END:%.*]]
// CHECK5:       cond.false:
// CHECK5-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    br label [[COND_END]]
// CHECK5:       cond.end:
// CHECK5-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK5-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK5:       omp.inner.for.cond:
// CHECK5-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK5-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK5-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK5-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK5:       omp.inner.for.body:
// CHECK5-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP16:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP16]], i32* [[N_CASTED]], align 4
// CHECK5-NEXT:    [[TMP17:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK5-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK5-NEXT:    [[TMP19:%.*]] = inttoptr i32 [[TMP14]] to i8*
// CHECK5-NEXT:    store i8* [[TMP19]], i8** [[TMP18]], align 4
// CHECK5-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK5-NEXT:    [[TMP21:%.*]] = inttoptr i32 [[TMP15]] to i8*
// CHECK5-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 4
// CHECK5-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK5-NEXT:    [[TMP23:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK5-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 4
// CHECK5-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK5-NEXT:    [[TMP25:%.*]] = bitcast [1000 x i16]* [[TMP0]] to i8*
// CHECK5-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 4
// CHECK5-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK5-NEXT:    [[TMP28:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK5-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP27]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i16]*)* @__omp_outlined__3 to i8*), i8* null, i8** [[TMP28]], i32 4)
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK5:       omp.inner.for.inc:
// CHECK5-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP29]], [[TMP30]]
// CHECK5-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP31]], [[TMP32]]
// CHECK5-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP33]], [[TMP34]]
// CHECK5-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK5-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP35]], [[TMP36]]
// CHECK5-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
// CHECK5:       cond.true10:
// CHECK5-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK5-NEXT:    br label [[COND_END12:%.*]]
// CHECK5:       cond.false11:
// CHECK5-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    br label [[COND_END12]]
// CHECK5:       cond.end12:
// CHECK5-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP37]], [[COND_TRUE10]] ], [ [[TMP38]], [[COND_FALSE11]] ]
// CHECK5-NEXT:    store i32 [[COND13]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    store i32 [[TMP39]], i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK5:       omp.inner.for.end:
// CHECK5-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK5:       omp.loop.exit:
// CHECK5-NEXT:    [[TMP40:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    [[TMP41:%.*]] = load i32, i32* [[TMP40]], align 4
// CHECK5-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP41]])
// CHECK5-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK5:       omp.precond.end:
// CHECK5-NEXT:    ret void
// CHECK5-LABEL: define {{[^@]+}}@__omp_outlined__3
// CHECK5-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK5-NEXT:  entry:
// CHECK5-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK5-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK5-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK5-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK5-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK5-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK5-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK5-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK5-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK5-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK5-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK5-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK5-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK5:       omp.precond.then:
// CHECK5-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK5-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK5-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK5-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK5-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK5-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK5-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK5-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK5-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK5-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK5-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK5:       omp.inner.for.cond:
// CHECK5-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK5-NEXT:    [[CMP4:%.*]] = icmp ule i32 [[TMP10]], [[TMP11]]
// CHECK5-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK5:       omp.inner.for.body:
// CHECK5-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK5-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK5-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK5-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I3]], align 4
// CHECK5-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i16], [1000 x i16]* [[TMP0]], i32 0, i32 [[TMP13]]
// CHECK5-NEXT:    [[TMP14:%.*]] = load i16, i16* [[ARRAYIDX]], align 2
// CHECK5-NEXT:    [[CONV:%.*]] = sext i16 [[TMP14]] to i32
// CHECK5-NEXT:    [[ADD5:%.*]] = add nsw i32 [[CONV]], 1
// CHECK5-NEXT:    [[CONV6:%.*]] = trunc i32 [[ADD5]] to i16
// CHECK5-NEXT:    store i16 [[CONV6]], i16* [[ARRAYIDX]], align 2
// CHECK5-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK5:       omp.body.continue:
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK5:       omp.inner.for.inc:
// CHECK5-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP15]], [[TMP16]]
// CHECK5-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK5:       omp.inner.for.end:
// CHECK5-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK5:       omp.loop.exit:
// CHECK5-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
// CHECK5-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP18]])
// CHECK5-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK5:       omp.precond.end:
// CHECK5-NEXT:    ret void
// CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l54
// CHECK5-SAME: ([10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK5-NEXT:  entry:
// CHECK5-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK5-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK5-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK5-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK5-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK5-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK5-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK5:       .execute:
// CHECK5-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK5-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK5-NEXT:    call void @__omp_outlined__4(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x i32]* [[TMP0]]) #[[ATTR3]]
// CHECK5-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK5:       .omp.deinit:
// CHECK5-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK5-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK5:       .exit:
// CHECK5-NEXT:    ret void
// CHECK5-LABEL: define {{[^@]+}}@__omp_outlined__4
// CHECK5-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK5-NEXT:  entry:
// CHECK5-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK5-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x i8*], align 4
// CHECK5-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK5-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK5-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK5-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    store i32 9, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK5-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK5-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK5-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK5-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9
// CHECK5-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK5:       cond.true:
// CHECK5-NEXT:    br label [[COND_END:%.*]]
// CHECK5:       cond.false:
// CHECK5-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    br label [[COND_END]]
// CHECK5:       cond.end:
// CHECK5-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK5-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK5:       omp.inner.for.cond:
// CHECK5-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[CMP1:%.*]] = icmp slt i32 [[TMP6]], 10
// CHECK5-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK5:       omp.inner.for.body:
// CHECK5-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK5-NEXT:    [[TMP10:%.*]] = inttoptr i32 [[TMP7]] to i8*
// CHECK5-NEXT:    store i8* [[TMP10]], i8** [[TMP9]], align 4
// CHECK5-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK5-NEXT:    [[TMP12:%.*]] = inttoptr i32 [[TMP8]] to i8*
// CHECK5-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 4
// CHECK5-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK5-NEXT:    [[TMP14:%.*]] = bitcast [10 x i32]* [[TMP0]] to i8*
// CHECK5-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 4
// CHECK5-NEXT:    [[TMP15:%.*]] = bitcast [3 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK5-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, [10 x i32]*)* @__omp_outlined__5 to i8*), i8* null, i8** [[TMP15]], i32 3)
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK5:       omp.inner.for.inc:
// CHECK5-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP16]], [[TMP17]]
// CHECK5-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP18]], [[TMP19]]
// CHECK5-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK5-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP22]], 9
// CHECK5-NEXT:    br i1 [[CMP4]], label [[COND_TRUE5:%.*]], label [[COND_FALSE6:%.*]]
// CHECK5:       cond.true5:
// CHECK5-NEXT:    br label [[COND_END7:%.*]]
// CHECK5:       cond.false6:
// CHECK5-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    br label [[COND_END7]]
// CHECK5:       cond.end7:
// CHECK5-NEXT:    [[COND8:%.*]] = phi i32 [ 9, [[COND_TRUE5]] ], [ [[TMP23]], [[COND_FALSE6]] ]
// CHECK5-NEXT:    store i32 [[COND8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    store i32 [[TMP24]], i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK5:       omp.inner.for.end:
// CHECK5-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK5:       omp.loop.exit:
// CHECK5-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]])
// CHECK5-NEXT:    ret void
// CHECK5-LABEL: define {{[^@]+}}@__omp_outlined__5
// CHECK5-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK5-NEXT:  entry:
// CHECK5-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK5-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK5-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK5-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK5-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK5-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK5-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK5-NEXT:    store i32 9, i32* [[DOTOMP_UB]], align 4
// CHECK5-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK5-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP1]], i32* [[DOTOMP_LB]], align 4
// CHECK5-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_UB]], align 4
// CHECK5-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK5-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK5-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK5-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK5-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK5:       omp.inner.for.cond:
// CHECK5-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK5-NEXT:    [[CMP:%.*]] = icmp ule i32 [[TMP6]], [[TMP7]]
// CHECK5-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK5:       omp.inner.for.body:
// CHECK5-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP8]], 1
// CHECK5-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK5-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK5-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4
// CHECK5-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[TMP0]], i32 0, i32 [[TMP9]]
// CHECK5-NEXT:    [[TMP10:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK5-NEXT:    [[ADD1:%.*]] = add nsw i32 [[TMP10]], 1
// CHECK5-NEXT:    store i32 [[ADD1]], i32* [[ARRAYIDX]], align 4
// CHECK5-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK5:       omp.body.continue:
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK5:       omp.inner.for.inc:
// CHECK5-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
// CHECK5-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK5:       omp.inner.for.end:
// CHECK5-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK5:       omp.loop.exit:
// CHECK5-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK5-NEXT:    ret void
// CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l59
// CHECK5-SAME: ([10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK5-NEXT:  entry:
// CHECK5-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK5-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[F_CASTED:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK5-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK5-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK5-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK5-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK5-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK5:       .execute:
// CHECK5-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK5-NEXT:    [[TMP2:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP2]], i32* [[F_CASTED]], align 4
// CHECK5-NEXT:    [[TMP3:%.*]] = load i32, i32* [[F_CASTED]], align 4
// CHECK5-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK5-NEXT:    call void @__omp_outlined__6(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x [10 x i32]]* [[TMP0]], i32 [[TMP3]]) #[[ATTR3]]
// CHECK5-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK5:       .omp.deinit:
// CHECK5-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK5-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK5:       .exit:
// CHECK5-NEXT:    ret void
// CHECK5-LABEL: define {{[^@]+}}@__omp_outlined__6
// CHECK5-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK5-NEXT:  entry:
// CHECK5-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK5-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[F_CASTED:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK5-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK5-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK5-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK5-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    store i32 99, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK5-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK5-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK5-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK5-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 99
// CHECK5-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK5:       cond.true:
// CHECK5-NEXT:    br label [[COND_END:%.*]]
// CHECK5:       cond.false:
// CHECK5-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    br label [[COND_END]]
// CHECK5:       cond.end:
// CHECK5-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK5-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK5:       omp.inner.for.cond:
// CHECK5-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[TMP6]], 100
// CHECK5-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK5:       omp.inner.for.body:
// CHECK5-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP9:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP9]], i32* [[F_CASTED]], align 4
// CHECK5-NEXT:    [[TMP10:%.*]] = load i32, i32* [[F_CASTED]], align 4
// CHECK5-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK5-NEXT:    [[TMP12:%.*]] = inttoptr i32 [[TMP7]] to i8*
// CHECK5-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 4
// CHECK5-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK5-NEXT:    [[TMP14:%.*]] = inttoptr i32 [[TMP8]] to i8*
// CHECK5-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 4
// CHECK5-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK5-NEXT:    [[TMP16:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK5-NEXT:    store i8* [[TMP16]], i8** [[TMP15]], align 4
// CHECK5-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK5-NEXT:    [[TMP18:%.*]] = inttoptr i32 [[TMP10]] to i8*
// CHECK5-NEXT:    store i8* [[TMP18]], i8** [[TMP17]], align 4
// CHECK5-NEXT:    [[TMP19:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK5-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, [10 x [10 x i32]]*, i32)* @__omp_outlined__7 to i8*), i8* null, i8** [[TMP19]], i32 4)
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK5:       omp.inner.for.inc:
// CHECK5-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK5-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK5-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK5-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP26]], 99
// CHECK5-NEXT:    br i1 [[CMP5]], label [[COND_TRUE6:%.*]], label [[COND_FALSE7:%.*]]
// CHECK5:       cond.true6:
// CHECK5-NEXT:    br label [[COND_END8:%.*]]
// CHECK5:       cond.false7:
// CHECK5-NEXT:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    br label [[COND_END8]]
// CHECK5:       cond.end8:
// CHECK5-NEXT:    [[COND9:%.*]] = phi i32 [ 99, [[COND_TRUE6]] ], [ [[TMP27]], [[COND_FALSE7]] ]
// CHECK5-NEXT:    store i32 [[COND9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    store i32 [[TMP28]], i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK5:       omp.inner.for.end:
// CHECK5-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK5:       omp.loop.exit:
// CHECK5-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]])
// CHECK5-NEXT:    ret void
// CHECK5-LABEL: define {{[^@]+}}@__omp_outlined__7
// CHECK5-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK5-NEXT:  entry:
// CHECK5-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK5-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK5-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK5-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK5-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK5-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK5-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK5-NEXT:    store i32 99, i32* [[DOTOMP_UB]], align 4
// CHECK5-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK5-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP1]], i32* [[DOTOMP_LB]], align 4
// CHECK5-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_UB]], align 4
// CHECK5-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK5-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK5-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK5-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK5-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK5:       omp.inner.for.cond:
// CHECK5-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK5-NEXT:    [[CMP:%.*]] = icmp ule i32 [[TMP6]], [[TMP7]]
// CHECK5-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK5:       omp.inner.for.body:
// CHECK5-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP8]], 10
// CHECK5-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK5-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK5-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK5-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[DIV2:%.*]] = sdiv i32 [[TMP10]], 10
// CHECK5-NEXT:    [[MUL3:%.*]] = mul nsw i32 [[DIV2]], 10
// CHECK5-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP9]], [[MUL3]]
// CHECK5-NEXT:    [[MUL4:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK5-NEXT:    [[ADD5:%.*]] = add nsw i32 0, [[MUL4]]
// CHECK5-NEXT:    store i32 [[ADD5]], i32* [[J]], align 4
// CHECK5-NEXT:    store i32 10, i32* [[K]], align 4
// CHECK5-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
// CHECK5-NEXT:    [[TMP12:%.*]] = load i32, i32* [[J]], align 4
// CHECK5-NEXT:    [[TMP13:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK5-NEXT:    [[MUL6:%.*]] = mul nsw i32 [[TMP12]], [[TMP13]]
// CHECK5-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP11]], [[MUL6]]
// CHECK5-NEXT:    [[TMP14:%.*]] = load i32, i32* [[K]], align 4
// CHECK5-NEXT:    [[ADD8:%.*]] = add nsw i32 [[ADD7]], [[TMP14]]
// CHECK5-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
// CHECK5-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i32 0, i32 [[TMP15]]
// CHECK5-NEXT:    [[TMP16:%.*]] = load i32, i32* [[J]], align 4
// CHECK5-NEXT:    [[ARRAYIDX9:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i32 0, i32 [[TMP16]]
// CHECK5-NEXT:    store i32 [[ADD8]], i32* [[ARRAYIDX9]], align 4
// CHECK5-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK5:       omp.body.continue:
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK5:       omp.inner.for.inc:
// CHECK5-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK5-NEXT:    store i32 [[ADD10]], i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK5:       omp.inner.for.end:
// CHECK5-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK5:       omp.loop.exit:
// CHECK5-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK5-NEXT:    ret void
// CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l67
// CHECK5-SAME: (i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK5-NEXT:  entry:
// CHECK5-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK5-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK5-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK5-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK5-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK5-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK5:       .execute:
// CHECK5-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK5-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK5-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK5-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK5-NEXT:    call void @__omp_outlined__8(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [10 x [10 x i32]]* [[TMP0]]) #[[ATTR3]]
// CHECK5-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK5:       .omp.deinit:
// CHECK5-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK5-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK5:       .exit:
// CHECK5-NEXT:    ret void
// CHECK5-LABEL: define {{[^@]+}}@__omp_outlined__8
// CHECK5-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK5-NEXT:  entry:
// CHECK5-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK5-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK5-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK5-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i64, align 8
// CHECK5-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i64, align 8
// CHECK5-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK5-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[I9:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[J10:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK5-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK5-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK5-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK5-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK5-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK5-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK5-NEXT:    [[CONV:%.*]] = sext i32 [[DIV]] to i64
// CHECK5-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK5-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK5-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK5-NEXT:    [[CONV6:%.*]] = sext i32 [[DIV5]] to i64
// CHECK5-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV]], [[CONV6]]
// CHECK5-NEXT:    [[SUB7:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK5-NEXT:    store i64 [[SUB7]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK5-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK5-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK5-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK5-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK5:       land.lhs.true:
// CHECK5-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK5-NEXT:    [[CMP8:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK5-NEXT:    br i1 [[CMP8]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK5:       omp.precond.then:
// CHECK5-NEXT:    store i64 0, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK5-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK5-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK5-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK5-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK5-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK5-NEXT:    [[CONV11:%.*]] = zext i32 [[NVPTX_NUM_THREADS]] to i64
// CHECK5-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    [[TMP9:%.*]] = load i32, i32* [[TMP8]], align 4
// CHECK5-NEXT:    call void @__kmpc_for_static_init_8(%struct.ident_t* @[[GLOB1]], i32 [[TMP9]], i32 91, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_COMB_LB]], i64* [[DOTOMP_COMB_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 [[CONV11]])
// CHECK5-NEXT:    [[TMP10:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK5-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK5-NEXT:    [[CMP12:%.*]] = icmp sgt i64 [[TMP10]], [[TMP11]]
// CHECK5-NEXT:    br i1 [[CMP12]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK5:       cond.true:
// CHECK5-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK5-NEXT:    br label [[COND_END:%.*]]
// CHECK5:       cond.false:
// CHECK5-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK5-NEXT:    br label [[COND_END]]
// CHECK5:       cond.end:
// CHECK5-NEXT:    [[COND:%.*]] = phi i64 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
// CHECK5-NEXT:    store i64 [[COND]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK5-NEXT:    [[TMP14:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK5-NEXT:    store i64 [[TMP14]], i64* [[DOTOMP_IV]], align 8
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK5:       omp.inner.for.cond:
// CHECK5-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK5-NEXT:    [[TMP16:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK5-NEXT:    [[ADD:%.*]] = add nsw i64 [[TMP16]], 1
// CHECK5-NEXT:    [[CMP13:%.*]] = icmp slt i64 [[TMP15]], [[ADD]]
// CHECK5-NEXT:    br i1 [[CMP13]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK5:       omp.inner.for.body:
// CHECK5-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK5-NEXT:    [[TMP18:%.*]] = trunc i64 [[TMP17]] to i32
// CHECK5-NEXT:    [[TMP19:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK5-NEXT:    [[TMP20:%.*]] = trunc i64 [[TMP19]] to i32
// CHECK5-NEXT:    [[TMP21:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP21]], i32* [[N_CASTED]], align 4
// CHECK5-NEXT:    [[TMP22:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK5-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK5-NEXT:    [[TMP24:%.*]] = inttoptr i32 [[TMP18]] to i8*
// CHECK5-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 4
// CHECK5-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK5-NEXT:    [[TMP26:%.*]] = inttoptr i32 [[TMP20]] to i8*
// CHECK5-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 4
// CHECK5-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK5-NEXT:    [[TMP28:%.*]] = inttoptr i32 [[TMP22]] to i8*
// CHECK5-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 4
// CHECK5-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK5-NEXT:    [[TMP30:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK5-NEXT:    store i8* [[TMP30]], i8** [[TMP29]], align 4
// CHECK5-NEXT:    [[TMP31:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    [[TMP32:%.*]] = load i32, i32* [[TMP31]], align 4
// CHECK5-NEXT:    [[TMP33:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK5-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP32]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [10 x [10 x i32]]*)* @__omp_outlined__9 to i8*), i8* null, i8** [[TMP33]], i32 4)
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK5:       omp.inner.for.inc:
// CHECK5-NEXT:    [[TMP34:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK5-NEXT:    [[TMP35:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK5-NEXT:    [[ADD14:%.*]] = add nsw i64 [[TMP34]], [[TMP35]]
// CHECK5-NEXT:    store i64 [[ADD14]], i64* [[DOTOMP_IV]], align 8
// CHECK5-NEXT:    [[TMP36:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK5-NEXT:    [[TMP37:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK5-NEXT:    [[ADD15:%.*]] = add nsw i64 [[TMP36]], [[TMP37]]
// CHECK5-NEXT:    store i64 [[ADD15]], i64* [[DOTOMP_COMB_LB]], align 8
// CHECK5-NEXT:    [[TMP38:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK5-NEXT:    [[TMP39:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK5-NEXT:    [[ADD16:%.*]] = add nsw i64 [[TMP38]], [[TMP39]]
// CHECK5-NEXT:    store i64 [[ADD16]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK5-NEXT:    [[TMP40:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK5-NEXT:    [[TMP41:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK5-NEXT:    [[CMP17:%.*]] = icmp sgt i64 [[TMP40]], [[TMP41]]
// CHECK5-NEXT:    br i1 [[CMP17]], label [[COND_TRUE18:%.*]], label [[COND_FALSE19:%.*]]
// CHECK5:       cond.true18:
// CHECK5-NEXT:    [[TMP42:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK5-NEXT:    br label [[COND_END20:%.*]]
// CHECK5:       cond.false19:
// CHECK5-NEXT:    [[TMP43:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK5-NEXT:    br label [[COND_END20]]
// CHECK5:       cond.end20:
// CHECK5-NEXT:    [[COND21:%.*]] = phi i64 [ [[TMP42]], [[COND_TRUE18]] ], [ [[TMP43]], [[COND_FALSE19]] ]
// CHECK5-NEXT:    store i64 [[COND21]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK5-NEXT:    [[TMP44:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK5-NEXT:    store i64 [[TMP44]], i64* [[DOTOMP_IV]], align 8
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK5:       omp.inner.for.end:
// CHECK5-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK5:       omp.loop.exit:
// CHECK5-NEXT:    [[TMP45:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    [[TMP46:%.*]] = load i32, i32* [[TMP45]], align 4
// CHECK5-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP46]])
// CHECK5-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK5:       omp.precond.end:
// CHECK5-NEXT:    ret void
// CHECK5-LABEL: define {{[^@]+}}@__omp_outlined__9
// CHECK5-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK5-NEXT:  entry:
// CHECK5-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK5-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK5-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK5-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_LB:%.*]] = alloca i64, align 8
// CHECK5-NEXT:    [[DOTOMP_UB:%.*]] = alloca i64, align 8
// CHECK5-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK5-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[I11:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[J12:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK5-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK5-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK5-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK5-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK5-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK5-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK5-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK5-NEXT:    [[CONV:%.*]] = sext i32 [[DIV]] to i64
// CHECK5-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK5-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK5-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK5-NEXT:    [[CONV6:%.*]] = sext i32 [[DIV5]] to i64
// CHECK5-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV]], [[CONV6]]
// CHECK5-NEXT:    [[SUB7:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK5-NEXT:    store i64 [[SUB7]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK5-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK5-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK5-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK5-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK5:       land.lhs.true:
// CHECK5-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK5-NEXT:    [[CMP8:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK5-NEXT:    br i1 [[CMP8]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK5:       omp.precond.then:
// CHECK5-NEXT:    store i64 0, i64* [[DOTOMP_LB]], align 8
// CHECK5-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK5-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_UB]], align 8
// CHECK5-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK5-NEXT:    [[CONV9:%.*]] = zext i32 [[TMP8]] to i64
// CHECK5-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK5-NEXT:    [[CONV10:%.*]] = zext i32 [[TMP9]] to i64
// CHECK5-NEXT:    store i64 [[CONV9]], i64* [[DOTOMP_LB]], align 8
// CHECK5-NEXT:    store i64 [[CONV10]], i64* [[DOTOMP_UB]], align 8
// CHECK5-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK5-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK5-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK5-NEXT:    call void @__kmpc_for_static_init_8(%struct.ident_t* @[[GLOB2]], i32 [[TMP11]], i32 33, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_LB]], i64* [[DOTOMP_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 1)
// CHECK5-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTOMP_LB]], align 8
// CHECK5-NEXT:    store i64 [[TMP12]], i64* [[DOTOMP_IV]], align 8
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK5:       omp.inner.for.cond:
// CHECK5-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK5-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK5-NEXT:    [[CONV13:%.*]] = zext i32 [[TMP14]] to i64
// CHECK5-NEXT:    [[CMP14:%.*]] = icmp sle i64 [[TMP13]], [[CONV13]]
// CHECK5-NEXT:    br i1 [[CMP14]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK5:       omp.inner.for.body:
// CHECK5-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK5-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK5-NEXT:    [[SUB15:%.*]] = sub nsw i32 [[TMP16]], 0
// CHECK5-NEXT:    [[DIV16:%.*]] = sdiv i32 [[SUB15]], 1
// CHECK5-NEXT:    [[MUL17:%.*]] = mul nsw i32 1, [[DIV16]]
// CHECK5-NEXT:    [[CONV18:%.*]] = sext i32 [[MUL17]] to i64
// CHECK5-NEXT:    [[DIV19:%.*]] = sdiv i64 [[TMP15]], [[CONV18]]
// CHECK5-NEXT:    [[MUL20:%.*]] = mul nsw i64 [[DIV19]], 1
// CHECK5-NEXT:    [[ADD:%.*]] = add nsw i64 0, [[MUL20]]
// CHECK5-NEXT:    [[CONV21:%.*]] = trunc i64 [[ADD]] to i32
// CHECK5-NEXT:    store i32 [[CONV21]], i32* [[I11]], align 4
// CHECK5-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK5-NEXT:    [[TMP18:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK5-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK5-NEXT:    [[SUB22:%.*]] = sub nsw i32 [[TMP19]], 0
// CHECK5-NEXT:    [[DIV23:%.*]] = sdiv i32 [[SUB22]], 1
// CHECK5-NEXT:    [[MUL24:%.*]] = mul nsw i32 1, [[DIV23]]
// CHECK5-NEXT:    [[CONV25:%.*]] = sext i32 [[MUL24]] to i64
// CHECK5-NEXT:    [[DIV26:%.*]] = sdiv i64 [[TMP18]], [[CONV25]]
// CHECK5-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK5-NEXT:    [[SUB27:%.*]] = sub nsw i32 [[TMP20]], 0
// CHECK5-NEXT:    [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK5-NEXT:    [[MUL29:%.*]] = mul nsw i32 1, [[DIV28]]
// CHECK5-NEXT:    [[CONV30:%.*]] = sext i32 [[MUL29]] to i64
// CHECK5-NEXT:    [[MUL31:%.*]] = mul nsw i64 [[DIV26]], [[CONV30]]
// CHECK5-NEXT:    [[SUB32:%.*]] = sub nsw i64 [[TMP17]], [[MUL31]]
// CHECK5-NEXT:    [[MUL33:%.*]] = mul nsw i64 [[SUB32]], 1
// CHECK5-NEXT:    [[ADD34:%.*]] = add nsw i64 0, [[MUL33]]
// CHECK5-NEXT:    [[CONV35:%.*]] = trunc i64 [[ADD34]] to i32
// CHECK5-NEXT:    store i32 [[CONV35]], i32* [[J12]], align 4
// CHECK5-NEXT:    [[TMP21:%.*]] = load i32, i32* [[I11]], align 4
// CHECK5-NEXT:    [[TMP22:%.*]] = load i32, i32* [[J12]], align 4
// CHECK5-NEXT:    [[ADD36:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
// CHECK5-NEXT:    [[TMP23:%.*]] = load i32, i32* [[I11]], align 4
// CHECK5-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i32 0, i32 [[TMP23]]
// CHECK5-NEXT:    [[TMP24:%.*]] = load i32, i32* [[J12]], align 4
// CHECK5-NEXT:    [[ARRAYIDX37:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i32 0, i32 [[TMP24]]
// CHECK5-NEXT:    store i32 [[ADD36]], i32* [[ARRAYIDX37]], align 4
// CHECK5-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK5:       omp.body.continue:
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK5:       omp.inner.for.inc:
// CHECK5-NEXT:    [[TMP25:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK5-NEXT:    [[TMP26:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK5-NEXT:    [[ADD38:%.*]] = add nsw i64 [[TMP25]], [[TMP26]]
// CHECK5-NEXT:    store i64 [[ADD38]], i64* [[DOTOMP_IV]], align 8
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK5:       omp.inner.for.end:
// CHECK5-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK5:       omp.loop.exit:
// CHECK5-NEXT:    [[TMP27:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    [[TMP28:%.*]] = load i32, i32* [[TMP27]], align 4
// CHECK5-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP28]])
// CHECK5-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK5:       omp.precond.end:
// CHECK5-NEXT:    ret void
// CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l74
// CHECK5-SAME: (i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK5-NEXT:  entry:
// CHECK5-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK5-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK5-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK5-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK5-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK5-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK5-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK5:       .execute:
// CHECK5-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK5-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK5-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK5-NEXT:    [[TMP4:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK5-NEXT:    call void @__omp_outlined__10(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [1000 x i32]* [[TMP0]], i32* [[TMP4]]) #[[ATTR3]]
// CHECK5-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK5:       .omp.deinit:
// CHECK5-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK5-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK5:       .exit:
// CHECK5-NEXT:    ret void
// CHECK5-LABEL: define {{[^@]+}}@__omp_outlined__10
// CHECK5-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK5-NEXT:  entry:
// CHECK5-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK5-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 4
// CHECK5-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK5-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK5-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK5-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK5-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK5-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK5-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK5-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK5-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK5-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK5-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK5:       omp.precond.then:
// CHECK5-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK5-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK5-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK5-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK5-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK5-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK5-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK5-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK5:       cond.true:
// CHECK5-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK5-NEXT:    br label [[COND_END:%.*]]
// CHECK5:       cond.false:
// CHECK5-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    br label [[COND_END]]
// CHECK5:       cond.end:
// CHECK5-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK5-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK5:       omp.inner.for.cond:
// CHECK5-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK5-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK5-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK5-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK5:       omp.inner.for.body:
// CHECK5-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP16:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP16]], i32* [[N_CASTED]], align 4
// CHECK5-NEXT:    [[TMP17:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK5-NEXT:    [[TMP18:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK5-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK5-NEXT:    [[TMP20:%.*]] = inttoptr i32 [[TMP14]] to i8*
// CHECK5-NEXT:    store i8* [[TMP20]], i8** [[TMP19]], align 4
// CHECK5-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK5-NEXT:    [[TMP22:%.*]] = inttoptr i32 [[TMP15]] to i8*
// CHECK5-NEXT:    store i8* [[TMP22]], i8** [[TMP21]], align 4
// CHECK5-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK5-NEXT:    [[TMP24:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK5-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 4
// CHECK5-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK5-NEXT:    [[TMP26:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK5-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 4
// CHECK5-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 4
// CHECK5-NEXT:    [[TMP28:%.*]] = bitcast i32* [[TMP18]] to i8*
// CHECK5-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 4
// CHECK5-NEXT:    [[TMP29:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    [[TMP30:%.*]] = load i32, i32* [[TMP29]], align 4
// CHECK5-NEXT:    [[TMP31:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK5-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP30]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i32]*, i32*)* @__omp_outlined__11 to i8*), i8* null, i8** [[TMP31]], i32 5)
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK5:       omp.inner.for.inc:
// CHECK5-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP32]], [[TMP33]]
// CHECK5-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP34]], [[TMP35]]
// CHECK5-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP36]], [[TMP37]]
// CHECK5-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK5-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP38]], [[TMP39]]
// CHECK5-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
// CHECK5:       cond.true10:
// CHECK5-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK5-NEXT:    br label [[COND_END12:%.*]]
// CHECK5:       cond.false11:
// CHECK5-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    br label [[COND_END12]]
// CHECK5:       cond.end12:
// CHECK5-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP40]], [[COND_TRUE10]] ], [ [[TMP41]], [[COND_FALSE11]] ]
// CHECK5-NEXT:    store i32 [[COND13]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK5-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK5-NEXT:    store i32 [[TMP42]], i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK5:       omp.inner.for.end:
// CHECK5-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK5:       omp.loop.exit:
// CHECK5-NEXT:    [[TMP43:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    [[TMP44:%.*]] = load i32, i32* [[TMP43]], align 4
// CHECK5-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP44]])
// CHECK5-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK5:       omp.precond.end:
// CHECK5-NEXT:    ret void
// CHECK5-LABEL: define {{[^@]+}}@__omp_outlined__11
// CHECK5-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK5-NEXT:  entry:
// CHECK5-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK5-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK5-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK5-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK5-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK5-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK5-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK5-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK5-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK5-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK5-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK5-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK5-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK5-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK5-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK5:       omp.precond.then:
// CHECK5-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK5-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK5-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK5-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK5-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK5-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK5-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK5-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK5-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK5-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK5-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK5:       omp.inner.for.cond:
// CHECK5-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK5-NEXT:    [[CMP4:%.*]] = icmp ule i32 [[TMP10]], [[TMP11]]
// CHECK5-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK5:       omp.inner.for.body:
// CHECK5-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK5-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK5-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK5-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK5-NEXT:    [[TMP14:%.*]] = load i32, i32* [[I3]], align 4
// CHECK5-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[TMP13]], i32 [[TMP14]]
// CHECK5-NEXT:    [[TMP15:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK5-NEXT:    [[TMP16:%.*]] = load i32, i32* [[I3]], align 4
// CHECK5-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i32 0, i32 [[TMP16]]
// CHECK5-NEXT:    store i32 [[TMP15]], i32* [[ARRAYIDX5]], align 4
// CHECK5-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK5:       omp.body.continue:
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK5:       omp.inner.for.inc:
// CHECK5-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK5-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK5-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK5:       omp.inner.for.end:
// CHECK5-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK5:       omp.loop.exit:
// CHECK5-NEXT:    [[TMP19:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK5-NEXT:    [[TMP20:%.*]] = load i32, i32* [[TMP19]], align 4
// CHECK5-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP20]])
// CHECK5-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK5:       omp.precond.end:
// CHECK5-NEXT:    ret void
// CHECK6-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l43
// CHECK6-SAME: (i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK6-NEXT:  entry:
// CHECK6-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK6-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[L_CASTED:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK6-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK6-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK6-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK6-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK6:       .execute:
// CHECK6-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3:[0-9]+]])
// CHECK6-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK6-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK6-NEXT:    [[TMP4:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP4]], i32* [[L_CASTED]], align 4
// CHECK6-NEXT:    [[TMP5:%.*]] = load i32, i32* [[L_CASTED]], align 4
// CHECK6-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK6-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [1000 x i32]* [[TMP0]], i32 [[TMP5]]) #[[ATTR3:[0-9]+]]
// CHECK6-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK6:       .omp.deinit:
// CHECK6-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK6-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK6:       .exit:
// CHECK6-NEXT:    ret void
// CHECK6-LABEL: define {{[^@]+}}@__omp_outlined__
// CHECK6-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0]] {
// CHECK6-NEXT:  entry:
// CHECK6-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK6-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[I4:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[L_CASTED:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 4
// CHECK6-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK6-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK6-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK6-NEXT:    [[TMP1:%.*]] = load i16, i16* @"_openmp_static_kernel$is_shared", align 2
// CHECK6-NEXT:    [[TMP2:%.*]] = load i32, i32* @"_openmp_static_kernel$size", align 4
// CHECK6-NEXT:    call void @__kmpc_get_team_static_memory(i16 1, i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds (%"union._shared_openmp_static_memory_type_$_", %"union._shared_openmp_static_memory_type_$_" addrspace(3)* @"_openmp_shared_static_glob_rd_$_", i32 0, i32 0, i32 0) to i8*), i32 [[TMP2]], i16 [[TMP1]], i8** addrspacecast (i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr" to i8**))
// CHECK6-NEXT:    [[TMP3:%.*]] = load i8*, i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr", align 4
// CHECK6-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i8, i8* [[TMP3]], i32 0
// CHECK6-NEXT:    [[TMP5:%.*]] = bitcast i8* [[TMP4]] to %struct._globalized_locals_ty*
// CHECK6-NEXT:    [[L1:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY:%.*]], %struct._globalized_locals_ty* [[TMP5]], i32 0, i32 0
// CHECK6-NEXT:    [[TMP6:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP6]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], 0
// CHECK6-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK6-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK6-NEXT:    store i32 [[SUB3]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK6-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK6-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP8]]
// CHECK6-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK6:       omp.precond.then:
// CHECK6-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK6-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK6-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK6-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1:[0-9]+]], i32 [[TMP11]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 128)
// CHECK6-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK6-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
// CHECK6-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK6:       cond.true:
// CHECK6-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK6-NEXT:    br label [[COND_END:%.*]]
// CHECK6:       cond.false:
// CHECK6-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    br label [[COND_END]]
// CHECK6:       cond.end:
// CHECK6-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
// CHECK6-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    store i32 [[TMP16]], i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK6:       omp.inner.for.cond:
// CHECK6-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK6-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP18]], 1
// CHECK6-NEXT:    [[CMP6:%.*]] = icmp slt i32 [[TMP17]], [[ADD]]
// CHECK6-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK6:       omp.inner.for.body:
// CHECK6-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP21:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP21]], i32* [[N_CASTED]], align 4
// CHECK6-NEXT:    [[TMP22:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK6-NEXT:    [[TMP23:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP23]], i32* [[L_CASTED]], align 4
// CHECK6-NEXT:    [[TMP24:%.*]] = load i32, i32* [[L_CASTED]], align 4
// CHECK6-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK6-NEXT:    [[TMP26:%.*]] = inttoptr i32 [[TMP19]] to i8*
// CHECK6-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 4
// CHECK6-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK6-NEXT:    [[TMP28:%.*]] = inttoptr i32 [[TMP20]] to i8*
// CHECK6-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 4
// CHECK6-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK6-NEXT:    [[TMP30:%.*]] = inttoptr i32 [[TMP22]] to i8*
// CHECK6-NEXT:    store i8* [[TMP30]], i8** [[TMP29]], align 4
// CHECK6-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK6-NEXT:    [[TMP32:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK6-NEXT:    store i8* [[TMP32]], i8** [[TMP31]], align 4
// CHECK6-NEXT:    [[TMP33:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 4
// CHECK6-NEXT:    [[TMP34:%.*]] = inttoptr i32 [[TMP24]] to i8*
// CHECK6-NEXT:    store i8* [[TMP34]], i8** [[TMP33]], align 4
// CHECK6-NEXT:    [[TMP35:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    [[TMP36:%.*]] = load i32, i32* [[TMP35]], align 4
// CHECK6-NEXT:    [[TMP37:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK6-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP36]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i32]*, i32)* @__omp_outlined__1 to i8*), i8* null, i8** [[TMP37]], i32 5)
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK6:       omp.inner.for.inc:
// CHECK6-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP38]], [[TMP39]]
// CHECK6-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP40]], [[TMP41]]
// CHECK6-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP43:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP42]], [[TMP43]]
// CHECK6-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP44:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP45:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK6-NEXT:    [[CMP10:%.*]] = icmp sgt i32 [[TMP44]], [[TMP45]]
// CHECK6-NEXT:    br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]]
// CHECK6:       cond.true11:
// CHECK6-NEXT:    [[TMP46:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK6-NEXT:    br label [[COND_END13:%.*]]
// CHECK6:       cond.false12:
// CHECK6-NEXT:    [[TMP47:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    br label [[COND_END13]]
// CHECK6:       cond.end13:
// CHECK6-NEXT:    [[COND14:%.*]] = phi i32 [ [[TMP46]], [[COND_TRUE11]] ], [ [[TMP47]], [[COND_FALSE12]] ]
// CHECK6-NEXT:    store i32 [[COND14]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP48:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    store i32 [[TMP48]], i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK6:       omp.inner.for.end:
// CHECK6-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK6:       omp.loop.exit:
// CHECK6-NEXT:    [[TMP49:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    [[TMP50:%.*]] = load i32, i32* [[TMP49]], align 4
// CHECK6-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP50]])
// CHECK6-NEXT:    [[TMP51:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK6-NEXT:    [[TMP52:%.*]] = icmp ne i32 [[TMP51]], 0
// CHECK6-NEXT:    br i1 [[TMP52]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK6:       .omp.lastprivate.then:
// CHECK6-NEXT:    [[TMP53:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP53]], i32* [[L_ADDR]], align 4
// CHECK6-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK6:       .omp.lastprivate.done:
// CHECK6-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK6:       omp.precond.end:
// CHECK6-NEXT:    [[TMP54:%.*]] = load i16, i16* @"_openmp_static_kernel$is_shared", align 2
// CHECK6-NEXT:    call void @__kmpc_restore_team_static_memory(i16 1, i16 [[TMP54]])
// CHECK6-NEXT:    ret void
// CHECK6-LABEL: define {{[^@]+}}@__omp_outlined__1
// CHECK6-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0]] {
// CHECK6-NEXT:  entry:
// CHECK6-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK6-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK6-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK6-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK6-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK6-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK6-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK6-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK6-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK6-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK6-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK6-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK6-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK6:       omp.precond.then:
// CHECK6-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK6-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK6-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK6-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK6-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK6-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK6-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK6-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK6-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 32)
// CHECK6-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK6:       omp.dispatch.cond:
// CHECK6-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK6-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK6-NEXT:    [[CMP4:%.*]] = icmp ugt i32 [[TMP9]], [[TMP10]]
// CHECK6-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK6:       cond.true:
// CHECK6-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK6-NEXT:    br label [[COND_END:%.*]]
// CHECK6:       cond.false:
// CHECK6-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK6-NEXT:    br label [[COND_END]]
// CHECK6:       cond.end:
// CHECK6-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP11]], [[COND_TRUE]] ], [ [[TMP12]], [[COND_FALSE]] ]
// CHECK6-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
// CHECK6-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK6-NEXT:    store i32 [[TMP13]], i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK6-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK6-NEXT:    br i1 [[CMP5]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK6:       omp.dispatch.body:
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK6:       omp.inner.for.cond:
// CHECK6-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK6-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP16]], [[TMP17]]
// CHECK6-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK6:       omp.inner.for.body:
// CHECK6-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP18]], 1
// CHECK6-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK6-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK6-NEXT:    [[TMP19:%.*]] = load i32, i32* [[I3]], align 4
// CHECK6-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i32 0, i32 [[TMP19]]
// CHECK6-NEXT:    store i32 1, i32* [[ARRAYIDX]], align 4
// CHECK6-NEXT:    [[TMP20:%.*]] = load i32, i32* [[I3]], align 4
// CHECK6-NEXT:    store i32 [[TMP20]], i32* [[L_ADDR]], align 4
// CHECK6-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK6:       omp.body.continue:
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK6:       omp.inner.for.inc:
// CHECK6-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP21]], 1
// CHECK6-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK6:       omp.inner.for.end:
// CHECK6-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK6:       omp.dispatch.inc:
// CHECK6-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK6-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK6-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_LB]], align 4
// CHECK6-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK6-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK6-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_UB]], align 4
// CHECK6-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK6:       omp.dispatch.end:
// CHECK6-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK6-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP27]])
// CHECK6-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK6-NEXT:    [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
// CHECK6-NEXT:    br i1 [[TMP29]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK6:       .omp.lastprivate.then:
// CHECK6-NEXT:    [[TMP30:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP30]], i32* [[L_ADDR]], align 4
// CHECK6-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK6:       .omp.lastprivate.done:
// CHECK6-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK6:       omp.precond.end:
// CHECK6-NEXT:    ret void
// CHECK6-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l49
// CHECK6-SAME: (i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK6-NEXT:  entry:
// CHECK6-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK6-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK6-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK6-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK6-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK6-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK6:       .execute:
// CHECK6-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK6-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK6-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK6-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK6-NEXT:    call void @__omp_outlined__2(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [1000 x i16]* [[TMP0]]) #[[ATTR3]]
// CHECK6-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK6:       .omp.deinit:
// CHECK6-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK6-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK6:       .exit:
// CHECK6-NEXT:    ret void
// CHECK6-LABEL: define {{[^@]+}}@__omp_outlined__2
// CHECK6-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK6-NEXT:  entry:
// CHECK6-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK6-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK6-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK6-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK6-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK6-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK6-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK6-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK6-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK6-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK6-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK6-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK6:       omp.precond.then:
// CHECK6-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK6-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK6-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK6-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK6-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK6-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK6-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK6-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK6:       cond.true:
// CHECK6-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK6-NEXT:    br label [[COND_END:%.*]]
// CHECK6:       cond.false:
// CHECK6-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    br label [[COND_END]]
// CHECK6:       cond.end:
// CHECK6-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK6-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK6:       omp.inner.for.cond:
// CHECK6-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK6-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK6-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK6-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK6:       omp.inner.for.body:
// CHECK6-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP16:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP16]], i32* [[N_CASTED]], align 4
// CHECK6-NEXT:    [[TMP17:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK6-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK6-NEXT:    [[TMP19:%.*]] = inttoptr i32 [[TMP14]] to i8*
// CHECK6-NEXT:    store i8* [[TMP19]], i8** [[TMP18]], align 4
// CHECK6-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK6-NEXT:    [[TMP21:%.*]] = inttoptr i32 [[TMP15]] to i8*
// CHECK6-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 4
// CHECK6-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK6-NEXT:    [[TMP23:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK6-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 4
// CHECK6-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK6-NEXT:    [[TMP25:%.*]] = bitcast [1000 x i16]* [[TMP0]] to i8*
// CHECK6-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 4
// CHECK6-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK6-NEXT:    [[TMP28:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK6-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP27]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i16]*)* @__omp_outlined__3 to i8*), i8* null, i8** [[TMP28]], i32 4)
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK6:       omp.inner.for.inc:
// CHECK6-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP29]], [[TMP30]]
// CHECK6-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP31]], [[TMP32]]
// CHECK6-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP33]], [[TMP34]]
// CHECK6-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK6-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP35]], [[TMP36]]
// CHECK6-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
// CHECK6:       cond.true10:
// CHECK6-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK6-NEXT:    br label [[COND_END12:%.*]]
// CHECK6:       cond.false11:
// CHECK6-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    br label [[COND_END12]]
// CHECK6:       cond.end12:
// CHECK6-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP37]], [[COND_TRUE10]] ], [ [[TMP38]], [[COND_FALSE11]] ]
// CHECK6-NEXT:    store i32 [[COND13]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    store i32 [[TMP39]], i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK6:       omp.inner.for.end:
// CHECK6-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK6:       omp.loop.exit:
// CHECK6-NEXT:    [[TMP40:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    [[TMP41:%.*]] = load i32, i32* [[TMP40]], align 4
// CHECK6-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP41]])
// CHECK6-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK6:       omp.precond.end:
// CHECK6-NEXT:    ret void
// CHECK6-LABEL: define {{[^@]+}}@__omp_outlined__3
// CHECK6-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK6-NEXT:  entry:
// CHECK6-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK6-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK6-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK6-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK6-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK6-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK6-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK6-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK6-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK6-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK6-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK6-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK6-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK6:       omp.precond.then:
// CHECK6-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK6-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK6-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK6-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK6-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK6-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK6-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK6-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK6-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK6-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK6-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK6:       omp.inner.for.cond:
// CHECK6-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK6-NEXT:    [[CMP4:%.*]] = icmp ule i32 [[TMP10]], [[TMP11]]
// CHECK6-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK6:       omp.inner.for.body:
// CHECK6-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK6-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK6-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK6-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I3]], align 4
// CHECK6-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i16], [1000 x i16]* [[TMP0]], i32 0, i32 [[TMP13]]
// CHECK6-NEXT:    [[TMP14:%.*]] = load i16, i16* [[ARRAYIDX]], align 2
// CHECK6-NEXT:    [[CONV:%.*]] = sext i16 [[TMP14]] to i32
// CHECK6-NEXT:    [[ADD5:%.*]] = add nsw i32 [[CONV]], 1
// CHECK6-NEXT:    [[CONV6:%.*]] = trunc i32 [[ADD5]] to i16
// CHECK6-NEXT:    store i16 [[CONV6]], i16* [[ARRAYIDX]], align 2
// CHECK6-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK6:       omp.body.continue:
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK6:       omp.inner.for.inc:
// CHECK6-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP15]], [[TMP16]]
// CHECK6-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK6:       omp.inner.for.end:
// CHECK6-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK6:       omp.loop.exit:
// CHECK6-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
// CHECK6-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP18]])
// CHECK6-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK6:       omp.precond.end:
// CHECK6-NEXT:    ret void
// CHECK6-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l54
// CHECK6-SAME: ([10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK6-NEXT:  entry:
// CHECK6-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK6-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK6-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK6-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK6-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK6-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK6-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK6:       .execute:
// CHECK6-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK6-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK6-NEXT:    call void @__omp_outlined__4(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x i32]* [[TMP0]]) #[[ATTR3]]
// CHECK6-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK6:       .omp.deinit:
// CHECK6-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK6-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK6:       .exit:
// CHECK6-NEXT:    ret void
// CHECK6-LABEL: define {{[^@]+}}@__omp_outlined__4
// CHECK6-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK6-NEXT:  entry:
// CHECK6-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK6-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x i8*], align 4
// CHECK6-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK6-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK6-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK6-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    store i32 9, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK6-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK6-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK6-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK6-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9
// CHECK6-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK6:       cond.true:
// CHECK6-NEXT:    br label [[COND_END:%.*]]
// CHECK6:       cond.false:
// CHECK6-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    br label [[COND_END]]
// CHECK6:       cond.end:
// CHECK6-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK6-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK6:       omp.inner.for.cond:
// CHECK6-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[CMP1:%.*]] = icmp slt i32 [[TMP6]], 10
// CHECK6-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK6:       omp.inner.for.body:
// CHECK6-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK6-NEXT:    [[TMP10:%.*]] = inttoptr i32 [[TMP7]] to i8*
// CHECK6-NEXT:    store i8* [[TMP10]], i8** [[TMP9]], align 4
// CHECK6-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK6-NEXT:    [[TMP12:%.*]] = inttoptr i32 [[TMP8]] to i8*
// CHECK6-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 4
// CHECK6-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK6-NEXT:    [[TMP14:%.*]] = bitcast [10 x i32]* [[TMP0]] to i8*
// CHECK6-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 4
// CHECK6-NEXT:    [[TMP15:%.*]] = bitcast [3 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK6-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, [10 x i32]*)* @__omp_outlined__5 to i8*), i8* null, i8** [[TMP15]], i32 3)
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK6:       omp.inner.for.inc:
// CHECK6-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP16]], [[TMP17]]
// CHECK6-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP18]], [[TMP19]]
// CHECK6-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK6-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP22]], 9
// CHECK6-NEXT:    br i1 [[CMP4]], label [[COND_TRUE5:%.*]], label [[COND_FALSE6:%.*]]
// CHECK6:       cond.true5:
// CHECK6-NEXT:    br label [[COND_END7:%.*]]
// CHECK6:       cond.false6:
// CHECK6-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    br label [[COND_END7]]
// CHECK6:       cond.end7:
// CHECK6-NEXT:    [[COND8:%.*]] = phi i32 [ 9, [[COND_TRUE5]] ], [ [[TMP23]], [[COND_FALSE6]] ]
// CHECK6-NEXT:    store i32 [[COND8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    store i32 [[TMP24]], i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK6:       omp.inner.for.end:
// CHECK6-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK6:       omp.loop.exit:
// CHECK6-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]])
// CHECK6-NEXT:    ret void
// CHECK6-LABEL: define {{[^@]+}}@__omp_outlined__5
// CHECK6-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK6-NEXT:  entry:
// CHECK6-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK6-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK6-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK6-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK6-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK6-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK6-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK6-NEXT:    store i32 9, i32* [[DOTOMP_UB]], align 4
// CHECK6-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK6-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP1]], i32* [[DOTOMP_LB]], align 4
// CHECK6-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_UB]], align 4
// CHECK6-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK6-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK6-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK6-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK6-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK6:       omp.inner.for.cond:
// CHECK6-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK6-NEXT:    [[CMP:%.*]] = icmp ule i32 [[TMP6]], [[TMP7]]
// CHECK6-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK6:       omp.inner.for.body:
// CHECK6-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP8]], 1
// CHECK6-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK6-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK6-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4
// CHECK6-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[TMP0]], i32 0, i32 [[TMP9]]
// CHECK6-NEXT:    [[TMP10:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK6-NEXT:    [[ADD1:%.*]] = add nsw i32 [[TMP10]], 1
// CHECK6-NEXT:    store i32 [[ADD1]], i32* [[ARRAYIDX]], align 4
// CHECK6-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK6:       omp.body.continue:
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK6:       omp.inner.for.inc:
// CHECK6-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
// CHECK6-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK6:       omp.inner.for.end:
// CHECK6-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK6:       omp.loop.exit:
// CHECK6-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK6-NEXT:    ret void
// CHECK6-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l59
// CHECK6-SAME: ([10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK6-NEXT:  entry:
// CHECK6-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK6-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[F_CASTED:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK6-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK6-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK6-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK6-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK6-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK6:       .execute:
// CHECK6-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK6-NEXT:    [[TMP2:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP2]], i32* [[F_CASTED]], align 4
// CHECK6-NEXT:    [[TMP3:%.*]] = load i32, i32* [[F_CASTED]], align 4
// CHECK6-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK6-NEXT:    call void @__omp_outlined__6(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x [10 x i32]]* [[TMP0]], i32 [[TMP3]]) #[[ATTR3]]
// CHECK6-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK6:       .omp.deinit:
// CHECK6-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK6-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK6:       .exit:
// CHECK6-NEXT:    ret void
// CHECK6-LABEL: define {{[^@]+}}@__omp_outlined__6
// CHECK6-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK6-NEXT:  entry:
// CHECK6-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK6-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[F_CASTED:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK6-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK6-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK6-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK6-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    store i32 99, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK6-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK6-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK6-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK6-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 99
// CHECK6-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK6:       cond.true:
// CHECK6-NEXT:    br label [[COND_END:%.*]]
// CHECK6:       cond.false:
// CHECK6-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    br label [[COND_END]]
// CHECK6:       cond.end:
// CHECK6-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK6-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK6:       omp.inner.for.cond:
// CHECK6-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[TMP6]], 100
// CHECK6-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK6:       omp.inner.for.body:
// CHECK6-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP9:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP9]], i32* [[F_CASTED]], align 4
// CHECK6-NEXT:    [[TMP10:%.*]] = load i32, i32* [[F_CASTED]], align 4
// CHECK6-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK6-NEXT:    [[TMP12:%.*]] = inttoptr i32 [[TMP7]] to i8*
// CHECK6-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 4
// CHECK6-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK6-NEXT:    [[TMP14:%.*]] = inttoptr i32 [[TMP8]] to i8*
// CHECK6-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 4
// CHECK6-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK6-NEXT:    [[TMP16:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK6-NEXT:    store i8* [[TMP16]], i8** [[TMP15]], align 4
// CHECK6-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK6-NEXT:    [[TMP18:%.*]] = inttoptr i32 [[TMP10]] to i8*
// CHECK6-NEXT:    store i8* [[TMP18]], i8** [[TMP17]], align 4
// CHECK6-NEXT:    [[TMP19:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK6-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, [10 x [10 x i32]]*, i32)* @__omp_outlined__7 to i8*), i8* null, i8** [[TMP19]], i32 4)
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK6:       omp.inner.for.inc:
// CHECK6-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK6-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK6-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK6-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP26]], 99
// CHECK6-NEXT:    br i1 [[CMP5]], label [[COND_TRUE6:%.*]], label [[COND_FALSE7:%.*]]
// CHECK6:       cond.true6:
// CHECK6-NEXT:    br label [[COND_END8:%.*]]
// CHECK6:       cond.false7:
// CHECK6-NEXT:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    br label [[COND_END8]]
// CHECK6:       cond.end8:
// CHECK6-NEXT:    [[COND9:%.*]] = phi i32 [ 99, [[COND_TRUE6]] ], [ [[TMP27]], [[COND_FALSE7]] ]
// CHECK6-NEXT:    store i32 [[COND9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    store i32 [[TMP28]], i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK6:       omp.inner.for.end:
// CHECK6-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK6:       omp.loop.exit:
// CHECK6-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]])
// CHECK6-NEXT:    ret void
// CHECK6-LABEL: define {{[^@]+}}@__omp_outlined__7
// CHECK6-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK6-NEXT:  entry:
// CHECK6-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK6-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK6-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK6-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK6-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK6-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK6-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK6-NEXT:    store i32 99, i32* [[DOTOMP_UB]], align 4
// CHECK6-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK6-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP1]], i32* [[DOTOMP_LB]], align 4
// CHECK6-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_UB]], align 4
// CHECK6-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK6-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK6-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK6-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK6-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK6:       omp.inner.for.cond:
// CHECK6-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK6-NEXT:    [[CMP:%.*]] = icmp ule i32 [[TMP6]], [[TMP7]]
// CHECK6-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK6:       omp.inner.for.body:
// CHECK6-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP8]], 10
// CHECK6-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK6-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK6-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK6-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[DIV2:%.*]] = sdiv i32 [[TMP10]], 10
// CHECK6-NEXT:    [[MUL3:%.*]] = mul nsw i32 [[DIV2]], 10
// CHECK6-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP9]], [[MUL3]]
// CHECK6-NEXT:    [[MUL4:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK6-NEXT:    [[ADD5:%.*]] = add nsw i32 0, [[MUL4]]
// CHECK6-NEXT:    store i32 [[ADD5]], i32* [[J]], align 4
// CHECK6-NEXT:    store i32 10, i32* [[K]], align 4
// CHECK6-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
// CHECK6-NEXT:    [[TMP12:%.*]] = load i32, i32* [[J]], align 4
// CHECK6-NEXT:    [[TMP13:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK6-NEXT:    [[MUL6:%.*]] = mul nsw i32 [[TMP12]], [[TMP13]]
// CHECK6-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP11]], [[MUL6]]
// CHECK6-NEXT:    [[TMP14:%.*]] = load i32, i32* [[K]], align 4
// CHECK6-NEXT:    [[ADD8:%.*]] = add nsw i32 [[ADD7]], [[TMP14]]
// CHECK6-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
// CHECK6-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i32 0, i32 [[TMP15]]
// CHECK6-NEXT:    [[TMP16:%.*]] = load i32, i32* [[J]], align 4
// CHECK6-NEXT:    [[ARRAYIDX9:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i32 0, i32 [[TMP16]]
// CHECK6-NEXT:    store i32 [[ADD8]], i32* [[ARRAYIDX9]], align 4
// CHECK6-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK6:       omp.body.continue:
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK6:       omp.inner.for.inc:
// CHECK6-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK6-NEXT:    store i32 [[ADD10]], i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK6:       omp.inner.for.end:
// CHECK6-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK6:       omp.loop.exit:
// CHECK6-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK6-NEXT:    ret void
// CHECK6-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l67
// CHECK6-SAME: (i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK6-NEXT:  entry:
// CHECK6-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK6-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK6-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK6-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK6-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK6-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK6:       .execute:
// CHECK6-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK6-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK6-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK6-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK6-NEXT:    call void @__omp_outlined__8(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [10 x [10 x i32]]* [[TMP0]]) #[[ATTR3]]
// CHECK6-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK6:       .omp.deinit:
// CHECK6-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK6-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK6:       .exit:
// CHECK6-NEXT:    ret void
// CHECK6-LABEL: define {{[^@]+}}@__omp_outlined__8
// CHECK6-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK6-NEXT:  entry:
// CHECK6-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK6-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK6-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK6-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i64, align 8
// CHECK6-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i64, align 8
// CHECK6-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK6-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[I9:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[J10:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK6-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK6-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK6-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK6-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK6-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK6-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK6-NEXT:    [[CONV:%.*]] = sext i32 [[DIV]] to i64
// CHECK6-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK6-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK6-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK6-NEXT:    [[CONV6:%.*]] = sext i32 [[DIV5]] to i64
// CHECK6-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV]], [[CONV6]]
// CHECK6-NEXT:    [[SUB7:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK6-NEXT:    store i64 [[SUB7]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK6-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK6-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK6-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK6-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK6:       land.lhs.true:
// CHECK6-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK6-NEXT:    [[CMP8:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK6-NEXT:    br i1 [[CMP8]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK6:       omp.precond.then:
// CHECK6-NEXT:    store i64 0, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK6-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK6-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK6-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK6-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK6-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK6-NEXT:    [[CONV11:%.*]] = zext i32 [[NVPTX_NUM_THREADS]] to i64
// CHECK6-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    [[TMP9:%.*]] = load i32, i32* [[TMP8]], align 4
// CHECK6-NEXT:    call void @__kmpc_for_static_init_8(%struct.ident_t* @[[GLOB1]], i32 [[TMP9]], i32 91, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_COMB_LB]], i64* [[DOTOMP_COMB_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 [[CONV11]])
// CHECK6-NEXT:    [[TMP10:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK6-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK6-NEXT:    [[CMP12:%.*]] = icmp sgt i64 [[TMP10]], [[TMP11]]
// CHECK6-NEXT:    br i1 [[CMP12]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK6:       cond.true:
// CHECK6-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK6-NEXT:    br label [[COND_END:%.*]]
// CHECK6:       cond.false:
// CHECK6-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK6-NEXT:    br label [[COND_END]]
// CHECK6:       cond.end:
// CHECK6-NEXT:    [[COND:%.*]] = phi i64 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
// CHECK6-NEXT:    store i64 [[COND]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK6-NEXT:    [[TMP14:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK6-NEXT:    store i64 [[TMP14]], i64* [[DOTOMP_IV]], align 8
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK6:       omp.inner.for.cond:
// CHECK6-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK6-NEXT:    [[TMP16:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK6-NEXT:    [[ADD:%.*]] = add nsw i64 [[TMP16]], 1
// CHECK6-NEXT:    [[CMP13:%.*]] = icmp slt i64 [[TMP15]], [[ADD]]
// CHECK6-NEXT:    br i1 [[CMP13]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK6:       omp.inner.for.body:
// CHECK6-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK6-NEXT:    [[TMP18:%.*]] = trunc i64 [[TMP17]] to i32
// CHECK6-NEXT:    [[TMP19:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK6-NEXT:    [[TMP20:%.*]] = trunc i64 [[TMP19]] to i32
// CHECK6-NEXT:    [[TMP21:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP21]], i32* [[N_CASTED]], align 4
// CHECK6-NEXT:    [[TMP22:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK6-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK6-NEXT:    [[TMP24:%.*]] = inttoptr i32 [[TMP18]] to i8*
// CHECK6-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 4
// CHECK6-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK6-NEXT:    [[TMP26:%.*]] = inttoptr i32 [[TMP20]] to i8*
// CHECK6-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 4
// CHECK6-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK6-NEXT:    [[TMP28:%.*]] = inttoptr i32 [[TMP22]] to i8*
// CHECK6-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 4
// CHECK6-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK6-NEXT:    [[TMP30:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK6-NEXT:    store i8* [[TMP30]], i8** [[TMP29]], align 4
// CHECK6-NEXT:    [[TMP31:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    [[TMP32:%.*]] = load i32, i32* [[TMP31]], align 4
// CHECK6-NEXT:    [[TMP33:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK6-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP32]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [10 x [10 x i32]]*)* @__omp_outlined__9 to i8*), i8* null, i8** [[TMP33]], i32 4)
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK6:       omp.inner.for.inc:
// CHECK6-NEXT:    [[TMP34:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK6-NEXT:    [[TMP35:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK6-NEXT:    [[ADD14:%.*]] = add nsw i64 [[TMP34]], [[TMP35]]
// CHECK6-NEXT:    store i64 [[ADD14]], i64* [[DOTOMP_IV]], align 8
// CHECK6-NEXT:    [[TMP36:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK6-NEXT:    [[TMP37:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK6-NEXT:    [[ADD15:%.*]] = add nsw i64 [[TMP36]], [[TMP37]]
// CHECK6-NEXT:    store i64 [[ADD15]], i64* [[DOTOMP_COMB_LB]], align 8
// CHECK6-NEXT:    [[TMP38:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK6-NEXT:    [[TMP39:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK6-NEXT:    [[ADD16:%.*]] = add nsw i64 [[TMP38]], [[TMP39]]
// CHECK6-NEXT:    store i64 [[ADD16]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK6-NEXT:    [[TMP40:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK6-NEXT:    [[TMP41:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK6-NEXT:    [[CMP17:%.*]] = icmp sgt i64 [[TMP40]], [[TMP41]]
// CHECK6-NEXT:    br i1 [[CMP17]], label [[COND_TRUE18:%.*]], label [[COND_FALSE19:%.*]]
// CHECK6:       cond.true18:
// CHECK6-NEXT:    [[TMP42:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK6-NEXT:    br label [[COND_END20:%.*]]
// CHECK6:       cond.false19:
// CHECK6-NEXT:    [[TMP43:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK6-NEXT:    br label [[COND_END20]]
// CHECK6:       cond.end20:
// CHECK6-NEXT:    [[COND21:%.*]] = phi i64 [ [[TMP42]], [[COND_TRUE18]] ], [ [[TMP43]], [[COND_FALSE19]] ]
// CHECK6-NEXT:    store i64 [[COND21]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK6-NEXT:    [[TMP44:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK6-NEXT:    store i64 [[TMP44]], i64* [[DOTOMP_IV]], align 8
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK6:       omp.inner.for.end:
// CHECK6-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK6:       omp.loop.exit:
// CHECK6-NEXT:    [[TMP45:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    [[TMP46:%.*]] = load i32, i32* [[TMP45]], align 4
// CHECK6-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP46]])
// CHECK6-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK6:       omp.precond.end:
// CHECK6-NEXT:    ret void
// CHECK6-LABEL: define {{[^@]+}}@__omp_outlined__9
// CHECK6-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK6-NEXT:  entry:
// CHECK6-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK6-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK6-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK6-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_LB:%.*]] = alloca i64, align 8
// CHECK6-NEXT:    [[DOTOMP_UB:%.*]] = alloca i64, align 8
// CHECK6-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK6-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[I11:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[J12:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK6-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK6-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK6-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK6-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK6-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK6-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK6-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK6-NEXT:    [[CONV:%.*]] = sext i32 [[DIV]] to i64
// CHECK6-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK6-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK6-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK6-NEXT:    [[CONV6:%.*]] = sext i32 [[DIV5]] to i64
// CHECK6-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV]], [[CONV6]]
// CHECK6-NEXT:    [[SUB7:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK6-NEXT:    store i64 [[SUB7]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK6-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK6-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK6-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK6-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK6:       land.lhs.true:
// CHECK6-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK6-NEXT:    [[CMP8:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK6-NEXT:    br i1 [[CMP8]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK6:       omp.precond.then:
// CHECK6-NEXT:    store i64 0, i64* [[DOTOMP_LB]], align 8
// CHECK6-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK6-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_UB]], align 8
// CHECK6-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK6-NEXT:    [[CONV9:%.*]] = zext i32 [[TMP8]] to i64
// CHECK6-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK6-NEXT:    [[CONV10:%.*]] = zext i32 [[TMP9]] to i64
// CHECK6-NEXT:    store i64 [[CONV9]], i64* [[DOTOMP_LB]], align 8
// CHECK6-NEXT:    store i64 [[CONV10]], i64* [[DOTOMP_UB]], align 8
// CHECK6-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK6-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK6-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK6-NEXT:    call void @__kmpc_for_static_init_8(%struct.ident_t* @[[GLOB2]], i32 [[TMP11]], i32 33, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_LB]], i64* [[DOTOMP_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 1)
// CHECK6-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTOMP_LB]], align 8
// CHECK6-NEXT:    store i64 [[TMP12]], i64* [[DOTOMP_IV]], align 8
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK6:       omp.inner.for.cond:
// CHECK6-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK6-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK6-NEXT:    [[CONV13:%.*]] = zext i32 [[TMP14]] to i64
// CHECK6-NEXT:    [[CMP14:%.*]] = icmp sle i64 [[TMP13]], [[CONV13]]
// CHECK6-NEXT:    br i1 [[CMP14]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK6:       omp.inner.for.body:
// CHECK6-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK6-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK6-NEXT:    [[SUB15:%.*]] = sub nsw i32 [[TMP16]], 0
// CHECK6-NEXT:    [[DIV16:%.*]] = sdiv i32 [[SUB15]], 1
// CHECK6-NEXT:    [[MUL17:%.*]] = mul nsw i32 1, [[DIV16]]
// CHECK6-NEXT:    [[CONV18:%.*]] = sext i32 [[MUL17]] to i64
// CHECK6-NEXT:    [[DIV19:%.*]] = sdiv i64 [[TMP15]], [[CONV18]]
// CHECK6-NEXT:    [[MUL20:%.*]] = mul nsw i64 [[DIV19]], 1
// CHECK6-NEXT:    [[ADD:%.*]] = add nsw i64 0, [[MUL20]]
// CHECK6-NEXT:    [[CONV21:%.*]] = trunc i64 [[ADD]] to i32
// CHECK6-NEXT:    store i32 [[CONV21]], i32* [[I11]], align 4
// CHECK6-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK6-NEXT:    [[TMP18:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK6-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK6-NEXT:    [[SUB22:%.*]] = sub nsw i32 [[TMP19]], 0
// CHECK6-NEXT:    [[DIV23:%.*]] = sdiv i32 [[SUB22]], 1
// CHECK6-NEXT:    [[MUL24:%.*]] = mul nsw i32 1, [[DIV23]]
// CHECK6-NEXT:    [[CONV25:%.*]] = sext i32 [[MUL24]] to i64
// CHECK6-NEXT:    [[DIV26:%.*]] = sdiv i64 [[TMP18]], [[CONV25]]
// CHECK6-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK6-NEXT:    [[SUB27:%.*]] = sub nsw i32 [[TMP20]], 0
// CHECK6-NEXT:    [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK6-NEXT:    [[MUL29:%.*]] = mul nsw i32 1, [[DIV28]]
// CHECK6-NEXT:    [[CONV30:%.*]] = sext i32 [[MUL29]] to i64
// CHECK6-NEXT:    [[MUL31:%.*]] = mul nsw i64 [[DIV26]], [[CONV30]]
// CHECK6-NEXT:    [[SUB32:%.*]] = sub nsw i64 [[TMP17]], [[MUL31]]
// CHECK6-NEXT:    [[MUL33:%.*]] = mul nsw i64 [[SUB32]], 1
// CHECK6-NEXT:    [[ADD34:%.*]] = add nsw i64 0, [[MUL33]]
// CHECK6-NEXT:    [[CONV35:%.*]] = trunc i64 [[ADD34]] to i32
// CHECK6-NEXT:    store i32 [[CONV35]], i32* [[J12]], align 4
// CHECK6-NEXT:    [[TMP21:%.*]] = load i32, i32* [[I11]], align 4
// CHECK6-NEXT:    [[TMP22:%.*]] = load i32, i32* [[J12]], align 4
// CHECK6-NEXT:    [[ADD36:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
// CHECK6-NEXT:    [[TMP23:%.*]] = load i32, i32* [[I11]], align 4
// CHECK6-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i32 0, i32 [[TMP23]]
// CHECK6-NEXT:    [[TMP24:%.*]] = load i32, i32* [[J12]], align 4
// CHECK6-NEXT:    [[ARRAYIDX37:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i32 0, i32 [[TMP24]]
// CHECK6-NEXT:    store i32 [[ADD36]], i32* [[ARRAYIDX37]], align 4
// CHECK6-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK6:       omp.body.continue:
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK6:       omp.inner.for.inc:
// CHECK6-NEXT:    [[TMP25:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK6-NEXT:    [[TMP26:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK6-NEXT:    [[ADD38:%.*]] = add nsw i64 [[TMP25]], [[TMP26]]
// CHECK6-NEXT:    store i64 [[ADD38]], i64* [[DOTOMP_IV]], align 8
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK6:       omp.inner.for.end:
// CHECK6-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK6:       omp.loop.exit:
// CHECK6-NEXT:    [[TMP27:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    [[TMP28:%.*]] = load i32, i32* [[TMP27]], align 4
// CHECK6-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP28]])
// CHECK6-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK6:       omp.precond.end:
// CHECK6-NEXT:    ret void
// CHECK6-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l74
// CHECK6-SAME: (i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK6-NEXT:  entry:
// CHECK6-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK6-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK6-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK6-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK6-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK6-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK6-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK6:       .execute:
// CHECK6-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK6-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK6-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK6-NEXT:    [[TMP4:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK6-NEXT:    call void @__omp_outlined__10(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [1000 x i32]* [[TMP0]], i32* [[TMP4]]) #[[ATTR3]]
// CHECK6-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK6:       .omp.deinit:
// CHECK6-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK6-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK6:       .exit:
// CHECK6-NEXT:    ret void
// CHECK6-LABEL: define {{[^@]+}}@__omp_outlined__10
// CHECK6-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK6-NEXT:  entry:
// CHECK6-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK6-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 4
// CHECK6-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK6-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK6-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK6-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK6-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK6-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK6-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK6-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK6-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK6-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK6-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK6:       omp.precond.then:
// CHECK6-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK6-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK6-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK6-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK6-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK6-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK6-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK6-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK6:       cond.true:
// CHECK6-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK6-NEXT:    br label [[COND_END:%.*]]
// CHECK6:       cond.false:
// CHECK6-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    br label [[COND_END]]
// CHECK6:       cond.end:
// CHECK6-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK6-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK6:       omp.inner.for.cond:
// CHECK6-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK6-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK6-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK6-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK6:       omp.inner.for.body:
// CHECK6-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP16:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP16]], i32* [[N_CASTED]], align 4
// CHECK6-NEXT:    [[TMP17:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK6-NEXT:    [[TMP18:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK6-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK6-NEXT:    [[TMP20:%.*]] = inttoptr i32 [[TMP14]] to i8*
// CHECK6-NEXT:    store i8* [[TMP20]], i8** [[TMP19]], align 4
// CHECK6-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK6-NEXT:    [[TMP22:%.*]] = inttoptr i32 [[TMP15]] to i8*
// CHECK6-NEXT:    store i8* [[TMP22]], i8** [[TMP21]], align 4
// CHECK6-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK6-NEXT:    [[TMP24:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK6-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 4
// CHECK6-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK6-NEXT:    [[TMP26:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK6-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 4
// CHECK6-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 4
// CHECK6-NEXT:    [[TMP28:%.*]] = bitcast i32* [[TMP18]] to i8*
// CHECK6-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 4
// CHECK6-NEXT:    [[TMP29:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    [[TMP30:%.*]] = load i32, i32* [[TMP29]], align 4
// CHECK6-NEXT:    [[TMP31:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK6-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP30]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i32]*, i32*)* @__omp_outlined__11 to i8*), i8* null, i8** [[TMP31]], i32 5)
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK6:       omp.inner.for.inc:
// CHECK6-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP32]], [[TMP33]]
// CHECK6-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP34]], [[TMP35]]
// CHECK6-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP36]], [[TMP37]]
// CHECK6-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK6-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP38]], [[TMP39]]
// CHECK6-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
// CHECK6:       cond.true10:
// CHECK6-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK6-NEXT:    br label [[COND_END12:%.*]]
// CHECK6:       cond.false11:
// CHECK6-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    br label [[COND_END12]]
// CHECK6:       cond.end12:
// CHECK6-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP40]], [[COND_TRUE10]] ], [ [[TMP41]], [[COND_FALSE11]] ]
// CHECK6-NEXT:    store i32 [[COND13]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK6-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK6-NEXT:    store i32 [[TMP42]], i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK6:       omp.inner.for.end:
// CHECK6-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK6:       omp.loop.exit:
// CHECK6-NEXT:    [[TMP43:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    [[TMP44:%.*]] = load i32, i32* [[TMP43]], align 4
// CHECK6-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP44]])
// CHECK6-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK6:       omp.precond.end:
// CHECK6-NEXT:    ret void
// CHECK6-LABEL: define {{[^@]+}}@__omp_outlined__11
// CHECK6-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK6-NEXT:  entry:
// CHECK6-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK6-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK6-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK6-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK6-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK6-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK6-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK6-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK6-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK6-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK6-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK6-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK6-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK6-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK6-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK6:       omp.precond.then:
// CHECK6-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK6-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK6-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK6-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK6-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK6-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK6-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK6-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK6-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK6-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK6-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK6:       omp.inner.for.cond:
// CHECK6-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK6-NEXT:    [[CMP4:%.*]] = icmp ule i32 [[TMP10]], [[TMP11]]
// CHECK6-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK6:       omp.inner.for.body:
// CHECK6-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK6-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK6-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK6-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK6-NEXT:    [[TMP14:%.*]] = load i32, i32* [[I3]], align 4
// CHECK6-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[TMP13]], i32 [[TMP14]]
// CHECK6-NEXT:    [[TMP15:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK6-NEXT:    [[TMP16:%.*]] = load i32, i32* [[I3]], align 4
// CHECK6-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i32 0, i32 [[TMP16]]
// CHECK6-NEXT:    store i32 [[TMP15]], i32* [[ARRAYIDX5]], align 4
// CHECK6-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK6:       omp.body.continue:
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK6:       omp.inner.for.inc:
// CHECK6-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK6-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK6-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK6:       omp.inner.for.end:
// CHECK6-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK6:       omp.loop.exit:
// CHECK6-NEXT:    [[TMP19:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK6-NEXT:    [[TMP20:%.*]] = load i32, i32* [[TMP19]], align 4
// CHECK6-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP20]])
// CHECK6-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK6:       omp.precond.end:
// CHECK6-NEXT:    ret void
// CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l43
// CHECK7-SAME: (i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK7-NEXT:  entry:
// CHECK7-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK7-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[L_CASTED:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK7-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK7-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK7-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK7-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK7:       .execute:
// CHECK7-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3:[0-9]+]])
// CHECK7-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK7-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK7-NEXT:    [[TMP4:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP4]], i32* [[L_CASTED]], align 4
// CHECK7-NEXT:    [[TMP5:%.*]] = load i32, i32* [[L_CASTED]], align 4
// CHECK7-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK7-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [1000 x i32]* [[TMP0]], i32 [[TMP5]]) #[[ATTR3:[0-9]+]]
// CHECK7-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK7:       .omp.deinit:
// CHECK7-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK7-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK7:       .exit:
// CHECK7-NEXT:    ret void
// CHECK7-LABEL: define {{[^@]+}}@__omp_outlined__
// CHECK7-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0]] {
// CHECK7-NEXT:  entry:
// CHECK7-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK7-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[I4:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[L_CASTED:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 4
// CHECK7-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK7-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK7-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK7-NEXT:    [[TMP1:%.*]] = call i8* @__kmpc_data_sharing_push_stack(i32 4, i16 1)
// CHECK7-NEXT:    [[TMP2:%.*]] = bitcast i8* [[TMP1]] to %struct._globalized_locals_ty*
// CHECK7-NEXT:    [[L1:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY:%.*]], %struct._globalized_locals_ty* [[TMP2]], i32 0, i32 0
// CHECK7-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP3]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK7-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK7-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK7-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK7-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK7-NEXT:    store i32 [[SUB3]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK7-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK7-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK7-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK7-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK7:       omp.precond.then:
// CHECK7-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK7-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK7-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK7-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1:[0-9]+]], i32 [[TMP8]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 128)
// CHECK7-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK7-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
// CHECK7-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK7:       cond.true:
// CHECK7-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK7-NEXT:    br label [[COND_END:%.*]]
// CHECK7:       cond.false:
// CHECK7-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    br label [[COND_END]]
// CHECK7:       cond.end:
// CHECK7-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP11]], [[COND_TRUE]] ], [ [[TMP12]], [[COND_FALSE]] ]
// CHECK7-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    store i32 [[TMP13]], i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK7:       omp.inner.for.cond:
// CHECK7-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK7-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP15]], 1
// CHECK7-NEXT:    [[CMP6:%.*]] = icmp slt i32 [[TMP14]], [[ADD]]
// CHECK7-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK7:       omp.inner.for.body:
// CHECK7-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP18:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP18]], i32* [[N_CASTED]], align 4
// CHECK7-NEXT:    [[TMP19:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK7-NEXT:    [[TMP20:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP20]], i32* [[L_CASTED]], align 4
// CHECK7-NEXT:    [[TMP21:%.*]] = load i32, i32* [[L_CASTED]], align 4
// CHECK7-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK7-NEXT:    [[TMP23:%.*]] = inttoptr i32 [[TMP16]] to i8*
// CHECK7-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 4
// CHECK7-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK7-NEXT:    [[TMP25:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK7-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 4
// CHECK7-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK7-NEXT:    [[TMP27:%.*]] = inttoptr i32 [[TMP19]] to i8*
// CHECK7-NEXT:    store i8* [[TMP27]], i8** [[TMP26]], align 4
// CHECK7-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK7-NEXT:    [[TMP29:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK7-NEXT:    store i8* [[TMP29]], i8** [[TMP28]], align 4
// CHECK7-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 4
// CHECK7-NEXT:    [[TMP31:%.*]] = inttoptr i32 [[TMP21]] to i8*
// CHECK7-NEXT:    store i8* [[TMP31]], i8** [[TMP30]], align 4
// CHECK7-NEXT:    [[TMP32:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    [[TMP33:%.*]] = load i32, i32* [[TMP32]], align 4
// CHECK7-NEXT:    [[TMP34:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK7-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP33]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i32]*, i32)* @__omp_outlined__1 to i8*), i8* null, i8** [[TMP34]], i32 5)
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK7:       omp.inner.for.inc:
// CHECK7-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP35]], [[TMP36]]
// CHECK7-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP37]], [[TMP38]]
// CHECK7-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP39]], [[TMP40]]
// CHECK7-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK7-NEXT:    [[CMP10:%.*]] = icmp sgt i32 [[TMP41]], [[TMP42]]
// CHECK7-NEXT:    br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]]
// CHECK7:       cond.true11:
// CHECK7-NEXT:    [[TMP43:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK7-NEXT:    br label [[COND_END13:%.*]]
// CHECK7:       cond.false12:
// CHECK7-NEXT:    [[TMP44:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    br label [[COND_END13]]
// CHECK7:       cond.end13:
// CHECK7-NEXT:    [[COND14:%.*]] = phi i32 [ [[TMP43]], [[COND_TRUE11]] ], [ [[TMP44]], [[COND_FALSE12]] ]
// CHECK7-NEXT:    store i32 [[COND14]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP45:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    store i32 [[TMP45]], i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK7:       omp.inner.for.end:
// CHECK7-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK7:       omp.loop.exit:
// CHECK7-NEXT:    [[TMP46:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    [[TMP47:%.*]] = load i32, i32* [[TMP46]], align 4
// CHECK7-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP47]])
// CHECK7-NEXT:    [[TMP48:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK7-NEXT:    [[TMP49:%.*]] = icmp ne i32 [[TMP48]], 0
// CHECK7-NEXT:    br i1 [[TMP49]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK7:       .omp.lastprivate.then:
// CHECK7-NEXT:    [[TMP50:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP50]], i32* [[L_ADDR]], align 4
// CHECK7-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK7:       .omp.lastprivate.done:
// CHECK7-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK7:       omp.precond.end:
// CHECK7-NEXT:    call void @__kmpc_data_sharing_pop_stack(i8* [[TMP1]])
// CHECK7-NEXT:    ret void
// CHECK7-LABEL: define {{[^@]+}}@__omp_outlined__1
// CHECK7-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0]] {
// CHECK7-NEXT:  entry:
// CHECK7-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK7-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK7-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK7-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK7-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK7-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK7-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK7-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK7-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK7-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK7-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK7-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK7-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK7-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK7-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK7-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK7:       omp.precond.then:
// CHECK7-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK7-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK7-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK7-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK7-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK7-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK7-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK7-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK7-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 32)
// CHECK7-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK7:       omp.dispatch.cond:
// CHECK7-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK7-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK7-NEXT:    [[CMP4:%.*]] = icmp ugt i32 [[TMP9]], [[TMP10]]
// CHECK7-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK7:       cond.true:
// CHECK7-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK7-NEXT:    br label [[COND_END:%.*]]
// CHECK7:       cond.false:
// CHECK7-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK7-NEXT:    br label [[COND_END]]
// CHECK7:       cond.end:
// CHECK7-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP11]], [[COND_TRUE]] ], [ [[TMP12]], [[COND_FALSE]] ]
// CHECK7-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
// CHECK7-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK7-NEXT:    store i32 [[TMP13]], i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK7-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK7-NEXT:    br i1 [[CMP5]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK7:       omp.dispatch.body:
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK7:       omp.inner.for.cond:
// CHECK7-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK7-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP16]], [[TMP17]]
// CHECK7-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK7:       omp.inner.for.body:
// CHECK7-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP18]], 1
// CHECK7-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK7-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK7-NEXT:    [[TMP19:%.*]] = load i32, i32* [[I3]], align 4
// CHECK7-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i32 0, i32 [[TMP19]]
// CHECK7-NEXT:    store i32 1, i32* [[ARRAYIDX]], align 4
// CHECK7-NEXT:    [[TMP20:%.*]] = load i32, i32* [[I3]], align 4
// CHECK7-NEXT:    store i32 [[TMP20]], i32* [[L_ADDR]], align 4
// CHECK7-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK7:       omp.body.continue:
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK7:       omp.inner.for.inc:
// CHECK7-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP21]], 1
// CHECK7-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK7:       omp.inner.for.end:
// CHECK7-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK7:       omp.dispatch.inc:
// CHECK7-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK7-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK7-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_LB]], align 4
// CHECK7-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK7-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK7-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_UB]], align 4
// CHECK7-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK7:       omp.dispatch.end:
// CHECK7-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK7-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP27]])
// CHECK7-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK7-NEXT:    [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
// CHECK7-NEXT:    br i1 [[TMP29]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK7:       .omp.lastprivate.then:
// CHECK7-NEXT:    [[TMP30:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP30]], i32* [[L_ADDR]], align 4
// CHECK7-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK7:       .omp.lastprivate.done:
// CHECK7-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK7:       omp.precond.end:
// CHECK7-NEXT:    ret void
// CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l49
// CHECK7-SAME: (i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK7-NEXT:  entry:
// CHECK7-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK7-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK7-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK7-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK7-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK7-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK7:       .execute:
// CHECK7-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK7-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK7-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK7-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK7-NEXT:    call void @__omp_outlined__2(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [1000 x i16]* [[TMP0]]) #[[ATTR3]]
// CHECK7-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK7:       .omp.deinit:
// CHECK7-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK7-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK7:       .exit:
// CHECK7-NEXT:    ret void
// CHECK7-LABEL: define {{[^@]+}}@__omp_outlined__2
// CHECK7-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK7-NEXT:  entry:
// CHECK7-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK7-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK7-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK7-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK7-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK7-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK7-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK7-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK7-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK7-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK7-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK7-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK7-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK7-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK7-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK7:       omp.precond.then:
// CHECK7-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK7-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK7-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK7-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK7-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK7-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK7-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK7-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK7:       cond.true:
// CHECK7-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK7-NEXT:    br label [[COND_END:%.*]]
// CHECK7:       cond.false:
// CHECK7-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    br label [[COND_END]]
// CHECK7:       cond.end:
// CHECK7-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK7-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK7:       omp.inner.for.cond:
// CHECK7-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK7-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK7-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK7-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK7:       omp.inner.for.body:
// CHECK7-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP16:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP16]], i32* [[N_CASTED]], align 4
// CHECK7-NEXT:    [[TMP17:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK7-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK7-NEXT:    [[TMP19:%.*]] = inttoptr i32 [[TMP14]] to i8*
// CHECK7-NEXT:    store i8* [[TMP19]], i8** [[TMP18]], align 4
// CHECK7-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK7-NEXT:    [[TMP21:%.*]] = inttoptr i32 [[TMP15]] to i8*
// CHECK7-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 4
// CHECK7-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK7-NEXT:    [[TMP23:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK7-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 4
// CHECK7-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK7-NEXT:    [[TMP25:%.*]] = bitcast [1000 x i16]* [[TMP0]] to i8*
// CHECK7-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 4
// CHECK7-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK7-NEXT:    [[TMP28:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK7-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP27]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i16]*)* @__omp_outlined__3 to i8*), i8* null, i8** [[TMP28]], i32 4)
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK7:       omp.inner.for.inc:
// CHECK7-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP29]], [[TMP30]]
// CHECK7-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP31]], [[TMP32]]
// CHECK7-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP33]], [[TMP34]]
// CHECK7-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK7-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP35]], [[TMP36]]
// CHECK7-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
// CHECK7:       cond.true10:
// CHECK7-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK7-NEXT:    br label [[COND_END12:%.*]]
// CHECK7:       cond.false11:
// CHECK7-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    br label [[COND_END12]]
// CHECK7:       cond.end12:
// CHECK7-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP37]], [[COND_TRUE10]] ], [ [[TMP38]], [[COND_FALSE11]] ]
// CHECK7-NEXT:    store i32 [[COND13]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    store i32 [[TMP39]], i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK7:       omp.inner.for.end:
// CHECK7-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK7:       omp.loop.exit:
// CHECK7-NEXT:    [[TMP40:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    [[TMP41:%.*]] = load i32, i32* [[TMP40]], align 4
// CHECK7-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP41]])
// CHECK7-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK7:       omp.precond.end:
// CHECK7-NEXT:    ret void
// CHECK7-LABEL: define {{[^@]+}}@__omp_outlined__3
// CHECK7-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK7-NEXT:  entry:
// CHECK7-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK7-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK7-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK7-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK7-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK7-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK7-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK7-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK7-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK7-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK7-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK7-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK7-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK7-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK7-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK7-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK7:       omp.precond.then:
// CHECK7-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK7-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK7-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK7-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK7-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK7-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK7-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK7-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK7-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK7-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK7-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK7:       omp.inner.for.cond:
// CHECK7-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK7-NEXT:    [[CMP4:%.*]] = icmp ule i32 [[TMP10]], [[TMP11]]
// CHECK7-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK7:       omp.inner.for.body:
// CHECK7-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK7-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK7-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK7-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I3]], align 4
// CHECK7-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i16], [1000 x i16]* [[TMP0]], i32 0, i32 [[TMP13]]
// CHECK7-NEXT:    [[TMP14:%.*]] = load i16, i16* [[ARRAYIDX]], align 2
// CHECK7-NEXT:    [[CONV:%.*]] = sext i16 [[TMP14]] to i32
// CHECK7-NEXT:    [[ADD5:%.*]] = add nsw i32 [[CONV]], 1
// CHECK7-NEXT:    [[CONV6:%.*]] = trunc i32 [[ADD5]] to i16
// CHECK7-NEXT:    store i16 [[CONV6]], i16* [[ARRAYIDX]], align 2
// CHECK7-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK7:       omp.body.continue:
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK7:       omp.inner.for.inc:
// CHECK7-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP15]], [[TMP16]]
// CHECK7-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK7:       omp.inner.for.end:
// CHECK7-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK7:       omp.loop.exit:
// CHECK7-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
// CHECK7-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP18]])
// CHECK7-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK7:       omp.precond.end:
// CHECK7-NEXT:    ret void
// CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l54
// CHECK7-SAME: ([10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK7-NEXT:  entry:
// CHECK7-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK7-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK7-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK7-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK7-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK7-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK7-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK7:       .execute:
// CHECK7-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK7-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK7-NEXT:    call void @__omp_outlined__4(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x i32]* [[TMP0]]) #[[ATTR3]]
// CHECK7-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK7:       .omp.deinit:
// CHECK7-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK7-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK7:       .exit:
// CHECK7-NEXT:    ret void
// CHECK7-LABEL: define {{[^@]+}}@__omp_outlined__4
// CHECK7-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK7-NEXT:  entry:
// CHECK7-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK7-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x i8*], align 4
// CHECK7-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK7-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK7-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK7-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    store i32 9, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK7-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK7-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK7-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK7-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9
// CHECK7-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK7:       cond.true:
// CHECK7-NEXT:    br label [[COND_END:%.*]]
// CHECK7:       cond.false:
// CHECK7-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    br label [[COND_END]]
// CHECK7:       cond.end:
// CHECK7-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK7-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK7:       omp.inner.for.cond:
// CHECK7-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[CMP1:%.*]] = icmp slt i32 [[TMP6]], 10
// CHECK7-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK7:       omp.inner.for.body:
// CHECK7-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK7-NEXT:    [[TMP10:%.*]] = inttoptr i32 [[TMP7]] to i8*
// CHECK7-NEXT:    store i8* [[TMP10]], i8** [[TMP9]], align 4
// CHECK7-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK7-NEXT:    [[TMP12:%.*]] = inttoptr i32 [[TMP8]] to i8*
// CHECK7-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 4
// CHECK7-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK7-NEXT:    [[TMP14:%.*]] = bitcast [10 x i32]* [[TMP0]] to i8*
// CHECK7-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 4
// CHECK7-NEXT:    [[TMP15:%.*]] = bitcast [3 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK7-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, [10 x i32]*)* @__omp_outlined__5 to i8*), i8* null, i8** [[TMP15]], i32 3)
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK7:       omp.inner.for.inc:
// CHECK7-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP16]], [[TMP17]]
// CHECK7-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP18]], [[TMP19]]
// CHECK7-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK7-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP22]], 9
// CHECK7-NEXT:    br i1 [[CMP4]], label [[COND_TRUE5:%.*]], label [[COND_FALSE6:%.*]]
// CHECK7:       cond.true5:
// CHECK7-NEXT:    br label [[COND_END7:%.*]]
// CHECK7:       cond.false6:
// CHECK7-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    br label [[COND_END7]]
// CHECK7:       cond.end7:
// CHECK7-NEXT:    [[COND8:%.*]] = phi i32 [ 9, [[COND_TRUE5]] ], [ [[TMP23]], [[COND_FALSE6]] ]
// CHECK7-NEXT:    store i32 [[COND8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    store i32 [[TMP24]], i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK7:       omp.inner.for.end:
// CHECK7-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK7:       omp.loop.exit:
// CHECK7-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]])
// CHECK7-NEXT:    ret void
// CHECK7-LABEL: define {{[^@]+}}@__omp_outlined__5
// CHECK7-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK7-NEXT:  entry:
// CHECK7-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK7-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK7-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK7-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK7-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK7-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK7-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK7-NEXT:    store i32 9, i32* [[DOTOMP_UB]], align 4
// CHECK7-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK7-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP1]], i32* [[DOTOMP_LB]], align 4
// CHECK7-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_UB]], align 4
// CHECK7-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK7-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK7-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK7-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK7-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK7:       omp.inner.for.cond:
// CHECK7-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK7-NEXT:    [[CMP:%.*]] = icmp ule i32 [[TMP6]], [[TMP7]]
// CHECK7-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK7:       omp.inner.for.body:
// CHECK7-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP8]], 1
// CHECK7-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK7-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK7-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4
// CHECK7-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[TMP0]], i32 0, i32 [[TMP9]]
// CHECK7-NEXT:    [[TMP10:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK7-NEXT:    [[ADD1:%.*]] = add nsw i32 [[TMP10]], 1
// CHECK7-NEXT:    store i32 [[ADD1]], i32* [[ARRAYIDX]], align 4
// CHECK7-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK7:       omp.body.continue:
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK7:       omp.inner.for.inc:
// CHECK7-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
// CHECK7-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK7:       omp.inner.for.end:
// CHECK7-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK7:       omp.loop.exit:
// CHECK7-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK7-NEXT:    ret void
// CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l59
// CHECK7-SAME: ([10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK7-NEXT:  entry:
// CHECK7-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK7-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[F_CASTED:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK7-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK7-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK7-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK7-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK7-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK7:       .execute:
// CHECK7-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK7-NEXT:    [[TMP2:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP2]], i32* [[F_CASTED]], align 4
// CHECK7-NEXT:    [[TMP3:%.*]] = load i32, i32* [[F_CASTED]], align 4
// CHECK7-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK7-NEXT:    call void @__omp_outlined__6(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x [10 x i32]]* [[TMP0]], i32 [[TMP3]]) #[[ATTR3]]
// CHECK7-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK7:       .omp.deinit:
// CHECK7-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK7-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK7:       .exit:
// CHECK7-NEXT:    ret void
// CHECK7-LABEL: define {{[^@]+}}@__omp_outlined__6
// CHECK7-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK7-NEXT:  entry:
// CHECK7-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK7-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[F_CASTED:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK7-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK7-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK7-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK7-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    store i32 99, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK7-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK7-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK7-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK7-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 99
// CHECK7-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK7:       cond.true:
// CHECK7-NEXT:    br label [[COND_END:%.*]]
// CHECK7:       cond.false:
// CHECK7-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    br label [[COND_END]]
// CHECK7:       cond.end:
// CHECK7-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK7-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK7:       omp.inner.for.cond:
// CHECK7-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[TMP6]], 100
// CHECK7-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK7:       omp.inner.for.body:
// CHECK7-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP9:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP9]], i32* [[F_CASTED]], align 4
// CHECK7-NEXT:    [[TMP10:%.*]] = load i32, i32* [[F_CASTED]], align 4
// CHECK7-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK7-NEXT:    [[TMP12:%.*]] = inttoptr i32 [[TMP7]] to i8*
// CHECK7-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 4
// CHECK7-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK7-NEXT:    [[TMP14:%.*]] = inttoptr i32 [[TMP8]] to i8*
// CHECK7-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 4
// CHECK7-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK7-NEXT:    [[TMP16:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK7-NEXT:    store i8* [[TMP16]], i8** [[TMP15]], align 4
// CHECK7-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK7-NEXT:    [[TMP18:%.*]] = inttoptr i32 [[TMP10]] to i8*
// CHECK7-NEXT:    store i8* [[TMP18]], i8** [[TMP17]], align 4
// CHECK7-NEXT:    [[TMP19:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK7-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, [10 x [10 x i32]]*, i32)* @__omp_outlined__7 to i8*), i8* null, i8** [[TMP19]], i32 4)
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK7:       omp.inner.for.inc:
// CHECK7-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK7-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK7-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK7-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP26]], 99
// CHECK7-NEXT:    br i1 [[CMP5]], label [[COND_TRUE6:%.*]], label [[COND_FALSE7:%.*]]
// CHECK7:       cond.true6:
// CHECK7-NEXT:    br label [[COND_END8:%.*]]
// CHECK7:       cond.false7:
// CHECK7-NEXT:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    br label [[COND_END8]]
// CHECK7:       cond.end8:
// CHECK7-NEXT:    [[COND9:%.*]] = phi i32 [ 99, [[COND_TRUE6]] ], [ [[TMP27]], [[COND_FALSE7]] ]
// CHECK7-NEXT:    store i32 [[COND9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    store i32 [[TMP28]], i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK7:       omp.inner.for.end:
// CHECK7-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK7:       omp.loop.exit:
// CHECK7-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]])
// CHECK7-NEXT:    ret void
// CHECK7-LABEL: define {{[^@]+}}@__omp_outlined__7
// CHECK7-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK7-NEXT:  entry:
// CHECK7-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK7-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK7-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK7-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK7-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK7-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK7-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK7-NEXT:    store i32 99, i32* [[DOTOMP_UB]], align 4
// CHECK7-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK7-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP1]], i32* [[DOTOMP_LB]], align 4
// CHECK7-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_UB]], align 4
// CHECK7-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK7-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK7-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK7-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK7-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK7:       omp.inner.for.cond:
// CHECK7-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK7-NEXT:    [[CMP:%.*]] = icmp ule i32 [[TMP6]], [[TMP7]]
// CHECK7-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK7:       omp.inner.for.body:
// CHECK7-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP8]], 10
// CHECK7-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK7-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK7-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK7-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[DIV2:%.*]] = sdiv i32 [[TMP10]], 10
// CHECK7-NEXT:    [[MUL3:%.*]] = mul nsw i32 [[DIV2]], 10
// CHECK7-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP9]], [[MUL3]]
// CHECK7-NEXT:    [[MUL4:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK7-NEXT:    [[ADD5:%.*]] = add nsw i32 0, [[MUL4]]
// CHECK7-NEXT:    store i32 [[ADD5]], i32* [[J]], align 4
// CHECK7-NEXT:    store i32 10, i32* [[K]], align 4
// CHECK7-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
// CHECK7-NEXT:    [[TMP12:%.*]] = load i32, i32* [[J]], align 4
// CHECK7-NEXT:    [[TMP13:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK7-NEXT:    [[MUL6:%.*]] = mul nsw i32 [[TMP12]], [[TMP13]]
// CHECK7-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP11]], [[MUL6]]
// CHECK7-NEXT:    [[TMP14:%.*]] = load i32, i32* [[K]], align 4
// CHECK7-NEXT:    [[ADD8:%.*]] = add nsw i32 [[ADD7]], [[TMP14]]
// CHECK7-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
// CHECK7-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i32 0, i32 [[TMP15]]
// CHECK7-NEXT:    [[TMP16:%.*]] = load i32, i32* [[J]], align 4
// CHECK7-NEXT:    [[ARRAYIDX9:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i32 0, i32 [[TMP16]]
// CHECK7-NEXT:    store i32 [[ADD8]], i32* [[ARRAYIDX9]], align 4
// CHECK7-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK7:       omp.body.continue:
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK7:       omp.inner.for.inc:
// CHECK7-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK7-NEXT:    store i32 [[ADD10]], i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK7:       omp.inner.for.end:
// CHECK7-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK7:       omp.loop.exit:
// CHECK7-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK7-NEXT:    ret void
// CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l67
// CHECK7-SAME: (i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK7-NEXT:  entry:
// CHECK7-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK7-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK7-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK7-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK7-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK7-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK7:       .execute:
// CHECK7-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK7-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK7-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK7-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK7-NEXT:    call void @__omp_outlined__8(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [10 x [10 x i32]]* [[TMP0]]) #[[ATTR3]]
// CHECK7-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK7:       .omp.deinit:
// CHECK7-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK7-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK7:       .exit:
// CHECK7-NEXT:    ret void
// CHECK7-LABEL: define {{[^@]+}}@__omp_outlined__8
// CHECK7-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK7-NEXT:  entry:
// CHECK7-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK7-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK7-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK7-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i64, align 8
// CHECK7-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i64, align 8
// CHECK7-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK7-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[I9:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[J10:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK7-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK7-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK7-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK7-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK7-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK7-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK7-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK7-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK7-NEXT:    [[CONV:%.*]] = sext i32 [[DIV]] to i64
// CHECK7-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK7-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK7-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK7-NEXT:    [[CONV6:%.*]] = sext i32 [[DIV5]] to i64
// CHECK7-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV]], [[CONV6]]
// CHECK7-NEXT:    [[SUB7:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK7-NEXT:    store i64 [[SUB7]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK7-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK7-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK7-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK7-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK7-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK7:       land.lhs.true:
// CHECK7-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK7-NEXT:    [[CMP8:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK7-NEXT:    br i1 [[CMP8]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK7:       omp.precond.then:
// CHECK7-NEXT:    store i64 0, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK7-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK7-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK7-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK7-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK7-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK7-NEXT:    [[CONV11:%.*]] = zext i32 [[NVPTX_NUM_THREADS]] to i64
// CHECK7-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    [[TMP9:%.*]] = load i32, i32* [[TMP8]], align 4
// CHECK7-NEXT:    call void @__kmpc_for_static_init_8(%struct.ident_t* @[[GLOB1]], i32 [[TMP9]], i32 91, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_COMB_LB]], i64* [[DOTOMP_COMB_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 [[CONV11]])
// CHECK7-NEXT:    [[TMP10:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK7-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK7-NEXT:    [[CMP12:%.*]] = icmp sgt i64 [[TMP10]], [[TMP11]]
// CHECK7-NEXT:    br i1 [[CMP12]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK7:       cond.true:
// CHECK7-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK7-NEXT:    br label [[COND_END:%.*]]
// CHECK7:       cond.false:
// CHECK7-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK7-NEXT:    br label [[COND_END]]
// CHECK7:       cond.end:
// CHECK7-NEXT:    [[COND:%.*]] = phi i64 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
// CHECK7-NEXT:    store i64 [[COND]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK7-NEXT:    [[TMP14:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK7-NEXT:    store i64 [[TMP14]], i64* [[DOTOMP_IV]], align 8
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK7:       omp.inner.for.cond:
// CHECK7-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK7-NEXT:    [[TMP16:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK7-NEXT:    [[ADD:%.*]] = add nsw i64 [[TMP16]], 1
// CHECK7-NEXT:    [[CMP13:%.*]] = icmp slt i64 [[TMP15]], [[ADD]]
// CHECK7-NEXT:    br i1 [[CMP13]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK7:       omp.inner.for.body:
// CHECK7-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK7-NEXT:    [[TMP18:%.*]] = trunc i64 [[TMP17]] to i32
// CHECK7-NEXT:    [[TMP19:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK7-NEXT:    [[TMP20:%.*]] = trunc i64 [[TMP19]] to i32
// CHECK7-NEXT:    [[TMP21:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP21]], i32* [[N_CASTED]], align 4
// CHECK7-NEXT:    [[TMP22:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK7-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK7-NEXT:    [[TMP24:%.*]] = inttoptr i32 [[TMP18]] to i8*
// CHECK7-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 4
// CHECK7-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK7-NEXT:    [[TMP26:%.*]] = inttoptr i32 [[TMP20]] to i8*
// CHECK7-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 4
// CHECK7-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK7-NEXT:    [[TMP28:%.*]] = inttoptr i32 [[TMP22]] to i8*
// CHECK7-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 4
// CHECK7-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK7-NEXT:    [[TMP30:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK7-NEXT:    store i8* [[TMP30]], i8** [[TMP29]], align 4
// CHECK7-NEXT:    [[TMP31:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    [[TMP32:%.*]] = load i32, i32* [[TMP31]], align 4
// CHECK7-NEXT:    [[TMP33:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK7-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP32]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [10 x [10 x i32]]*)* @__omp_outlined__9 to i8*), i8* null, i8** [[TMP33]], i32 4)
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK7:       omp.inner.for.inc:
// CHECK7-NEXT:    [[TMP34:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK7-NEXT:    [[TMP35:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK7-NEXT:    [[ADD14:%.*]] = add nsw i64 [[TMP34]], [[TMP35]]
// CHECK7-NEXT:    store i64 [[ADD14]], i64* [[DOTOMP_IV]], align 8
// CHECK7-NEXT:    [[TMP36:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK7-NEXT:    [[TMP37:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK7-NEXT:    [[ADD15:%.*]] = add nsw i64 [[TMP36]], [[TMP37]]
// CHECK7-NEXT:    store i64 [[ADD15]], i64* [[DOTOMP_COMB_LB]], align 8
// CHECK7-NEXT:    [[TMP38:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK7-NEXT:    [[TMP39:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK7-NEXT:    [[ADD16:%.*]] = add nsw i64 [[TMP38]], [[TMP39]]
// CHECK7-NEXT:    store i64 [[ADD16]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK7-NEXT:    [[TMP40:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK7-NEXT:    [[TMP41:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK7-NEXT:    [[CMP17:%.*]] = icmp sgt i64 [[TMP40]], [[TMP41]]
// CHECK7-NEXT:    br i1 [[CMP17]], label [[COND_TRUE18:%.*]], label [[COND_FALSE19:%.*]]
// CHECK7:       cond.true18:
// CHECK7-NEXT:    [[TMP42:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK7-NEXT:    br label [[COND_END20:%.*]]
// CHECK7:       cond.false19:
// CHECK7-NEXT:    [[TMP43:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK7-NEXT:    br label [[COND_END20]]
// CHECK7:       cond.end20:
// CHECK7-NEXT:    [[COND21:%.*]] = phi i64 [ [[TMP42]], [[COND_TRUE18]] ], [ [[TMP43]], [[COND_FALSE19]] ]
// CHECK7-NEXT:    store i64 [[COND21]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK7-NEXT:    [[TMP44:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK7-NEXT:    store i64 [[TMP44]], i64* [[DOTOMP_IV]], align 8
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK7:       omp.inner.for.end:
// CHECK7-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK7:       omp.loop.exit:
// CHECK7-NEXT:    [[TMP45:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    [[TMP46:%.*]] = load i32, i32* [[TMP45]], align 4
// CHECK7-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP46]])
// CHECK7-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK7:       omp.precond.end:
// CHECK7-NEXT:    ret void
// CHECK7-LABEL: define {{[^@]+}}@__omp_outlined__9
// CHECK7-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK7-NEXT:  entry:
// CHECK7-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK7-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK7-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK7-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_LB:%.*]] = alloca i64, align 8
// CHECK7-NEXT:    [[DOTOMP_UB:%.*]] = alloca i64, align 8
// CHECK7-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK7-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[I11:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[J12:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK7-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK7-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK7-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK7-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK7-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK7-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK7-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK7-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK7-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK7-NEXT:    [[CONV:%.*]] = sext i32 [[DIV]] to i64
// CHECK7-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK7-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK7-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK7-NEXT:    [[CONV6:%.*]] = sext i32 [[DIV5]] to i64
// CHECK7-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV]], [[CONV6]]
// CHECK7-NEXT:    [[SUB7:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK7-NEXT:    store i64 [[SUB7]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK7-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK7-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK7-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK7-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK7-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK7:       land.lhs.true:
// CHECK7-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK7-NEXT:    [[CMP8:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK7-NEXT:    br i1 [[CMP8]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK7:       omp.precond.then:
// CHECK7-NEXT:    store i64 0, i64* [[DOTOMP_LB]], align 8
// CHECK7-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK7-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_UB]], align 8
// CHECK7-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK7-NEXT:    [[CONV9:%.*]] = zext i32 [[TMP8]] to i64
// CHECK7-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK7-NEXT:    [[CONV10:%.*]] = zext i32 [[TMP9]] to i64
// CHECK7-NEXT:    store i64 [[CONV9]], i64* [[DOTOMP_LB]], align 8
// CHECK7-NEXT:    store i64 [[CONV10]], i64* [[DOTOMP_UB]], align 8
// CHECK7-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK7-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK7-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK7-NEXT:    call void @__kmpc_for_static_init_8(%struct.ident_t* @[[GLOB2]], i32 [[TMP11]], i32 33, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_LB]], i64* [[DOTOMP_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 1)
// CHECK7-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTOMP_LB]], align 8
// CHECK7-NEXT:    store i64 [[TMP12]], i64* [[DOTOMP_IV]], align 8
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK7:       omp.inner.for.cond:
// CHECK7-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK7-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK7-NEXT:    [[CONV13:%.*]] = zext i32 [[TMP14]] to i64
// CHECK7-NEXT:    [[CMP14:%.*]] = icmp sle i64 [[TMP13]], [[CONV13]]
// CHECK7-NEXT:    br i1 [[CMP14]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK7:       omp.inner.for.body:
// CHECK7-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK7-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK7-NEXT:    [[SUB15:%.*]] = sub nsw i32 [[TMP16]], 0
// CHECK7-NEXT:    [[DIV16:%.*]] = sdiv i32 [[SUB15]], 1
// CHECK7-NEXT:    [[MUL17:%.*]] = mul nsw i32 1, [[DIV16]]
// CHECK7-NEXT:    [[CONV18:%.*]] = sext i32 [[MUL17]] to i64
// CHECK7-NEXT:    [[DIV19:%.*]] = sdiv i64 [[TMP15]], [[CONV18]]
// CHECK7-NEXT:    [[MUL20:%.*]] = mul nsw i64 [[DIV19]], 1
// CHECK7-NEXT:    [[ADD:%.*]] = add nsw i64 0, [[MUL20]]
// CHECK7-NEXT:    [[CONV21:%.*]] = trunc i64 [[ADD]] to i32
// CHECK7-NEXT:    store i32 [[CONV21]], i32* [[I11]], align 4
// CHECK7-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK7-NEXT:    [[TMP18:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK7-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK7-NEXT:    [[SUB22:%.*]] = sub nsw i32 [[TMP19]], 0
// CHECK7-NEXT:    [[DIV23:%.*]] = sdiv i32 [[SUB22]], 1
// CHECK7-NEXT:    [[MUL24:%.*]] = mul nsw i32 1, [[DIV23]]
// CHECK7-NEXT:    [[CONV25:%.*]] = sext i32 [[MUL24]] to i64
// CHECK7-NEXT:    [[DIV26:%.*]] = sdiv i64 [[TMP18]], [[CONV25]]
// CHECK7-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK7-NEXT:    [[SUB27:%.*]] = sub nsw i32 [[TMP20]], 0
// CHECK7-NEXT:    [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK7-NEXT:    [[MUL29:%.*]] = mul nsw i32 1, [[DIV28]]
// CHECK7-NEXT:    [[CONV30:%.*]] = sext i32 [[MUL29]] to i64
// CHECK7-NEXT:    [[MUL31:%.*]] = mul nsw i64 [[DIV26]], [[CONV30]]
// CHECK7-NEXT:    [[SUB32:%.*]] = sub nsw i64 [[TMP17]], [[MUL31]]
// CHECK7-NEXT:    [[MUL33:%.*]] = mul nsw i64 [[SUB32]], 1
// CHECK7-NEXT:    [[ADD34:%.*]] = add nsw i64 0, [[MUL33]]
// CHECK7-NEXT:    [[CONV35:%.*]] = trunc i64 [[ADD34]] to i32
// CHECK7-NEXT:    store i32 [[CONV35]], i32* [[J12]], align 4
// CHECK7-NEXT:    [[TMP21:%.*]] = load i32, i32* [[I11]], align 4
// CHECK7-NEXT:    [[TMP22:%.*]] = load i32, i32* [[J12]], align 4
// CHECK7-NEXT:    [[ADD36:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
// CHECK7-NEXT:    [[TMP23:%.*]] = load i32, i32* [[I11]], align 4
// CHECK7-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i32 0, i32 [[TMP23]]
// CHECK7-NEXT:    [[TMP24:%.*]] = load i32, i32* [[J12]], align 4
// CHECK7-NEXT:    [[ARRAYIDX37:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i32 0, i32 [[TMP24]]
// CHECK7-NEXT:    store i32 [[ADD36]], i32* [[ARRAYIDX37]], align 4
// CHECK7-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK7:       omp.body.continue:
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK7:       omp.inner.for.inc:
// CHECK7-NEXT:    [[TMP25:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK7-NEXT:    [[TMP26:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK7-NEXT:    [[ADD38:%.*]] = add nsw i64 [[TMP25]], [[TMP26]]
// CHECK7-NEXT:    store i64 [[ADD38]], i64* [[DOTOMP_IV]], align 8
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK7:       omp.inner.for.end:
// CHECK7-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK7:       omp.loop.exit:
// CHECK7-NEXT:    [[TMP27:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    [[TMP28:%.*]] = load i32, i32* [[TMP27]], align 4
// CHECK7-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP28]])
// CHECK7-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK7:       omp.precond.end:
// CHECK7-NEXT:    ret void
// CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l74
// CHECK7-SAME: (i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK7-NEXT:  entry:
// CHECK7-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK7-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK7-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK7-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK7-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK7-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK7-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK7:       .execute:
// CHECK7-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK7-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK7-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK7-NEXT:    [[TMP4:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK7-NEXT:    call void @__omp_outlined__10(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [1000 x i32]* [[TMP0]], i32* [[TMP4]]) #[[ATTR3]]
// CHECK7-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK7:       .omp.deinit:
// CHECK7-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK7-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK7:       .exit:
// CHECK7-NEXT:    ret void
// CHECK7-LABEL: define {{[^@]+}}@__omp_outlined__10
// CHECK7-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK7-NEXT:  entry:
// CHECK7-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK7-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 4
// CHECK7-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK7-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK7-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK7-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK7-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK7-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK7-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK7-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK7-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK7-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK7-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK7-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK7-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK7-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK7:       omp.precond.then:
// CHECK7-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK7-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK7-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK7-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK7-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK7-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK7-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK7-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK7:       cond.true:
// CHECK7-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK7-NEXT:    br label [[COND_END:%.*]]
// CHECK7:       cond.false:
// CHECK7-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    br label [[COND_END]]
// CHECK7:       cond.end:
// CHECK7-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK7-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK7:       omp.inner.for.cond:
// CHECK7-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK7-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK7-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK7-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK7:       omp.inner.for.body:
// CHECK7-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP16:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP16]], i32* [[N_CASTED]], align 4
// CHECK7-NEXT:    [[TMP17:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK7-NEXT:    [[TMP18:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK7-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK7-NEXT:    [[TMP20:%.*]] = inttoptr i32 [[TMP14]] to i8*
// CHECK7-NEXT:    store i8* [[TMP20]], i8** [[TMP19]], align 4
// CHECK7-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK7-NEXT:    [[TMP22:%.*]] = inttoptr i32 [[TMP15]] to i8*
// CHECK7-NEXT:    store i8* [[TMP22]], i8** [[TMP21]], align 4
// CHECK7-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK7-NEXT:    [[TMP24:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK7-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 4
// CHECK7-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK7-NEXT:    [[TMP26:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK7-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 4
// CHECK7-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 4
// CHECK7-NEXT:    [[TMP28:%.*]] = bitcast i32* [[TMP18]] to i8*
// CHECK7-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 4
// CHECK7-NEXT:    [[TMP29:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    [[TMP30:%.*]] = load i32, i32* [[TMP29]], align 4
// CHECK7-NEXT:    [[TMP31:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK7-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP30]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i32]*, i32*)* @__omp_outlined__11 to i8*), i8* null, i8** [[TMP31]], i32 5)
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK7:       omp.inner.for.inc:
// CHECK7-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP32]], [[TMP33]]
// CHECK7-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP34]], [[TMP35]]
// CHECK7-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP36]], [[TMP37]]
// CHECK7-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK7-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP38]], [[TMP39]]
// CHECK7-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
// CHECK7:       cond.true10:
// CHECK7-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK7-NEXT:    br label [[COND_END12:%.*]]
// CHECK7:       cond.false11:
// CHECK7-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    br label [[COND_END12]]
// CHECK7:       cond.end12:
// CHECK7-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP40]], [[COND_TRUE10]] ], [ [[TMP41]], [[COND_FALSE11]] ]
// CHECK7-NEXT:    store i32 [[COND13]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK7-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK7-NEXT:    store i32 [[TMP42]], i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK7:       omp.inner.for.end:
// CHECK7-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK7:       omp.loop.exit:
// CHECK7-NEXT:    [[TMP43:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    [[TMP44:%.*]] = load i32, i32* [[TMP43]], align 4
// CHECK7-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP44]])
// CHECK7-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK7:       omp.precond.end:
// CHECK7-NEXT:    ret void
// CHECK7-LABEL: define {{[^@]+}}@__omp_outlined__11
// CHECK7-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK7-NEXT:  entry:
// CHECK7-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK7-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK7-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK7-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK7-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK7-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK7-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK7-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK7-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK7-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK7-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK7-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK7-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK7-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK7-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK7-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK7-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK7-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK7-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK7:       omp.precond.then:
// CHECK7-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK7-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK7-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK7-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK7-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK7-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK7-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK7-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK7-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK7-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK7-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK7-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK7:       omp.inner.for.cond:
// CHECK7-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK7-NEXT:    [[CMP4:%.*]] = icmp ule i32 [[TMP10]], [[TMP11]]
// CHECK7-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK7:       omp.inner.for.body:
// CHECK7-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK7-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK7-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK7-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK7-NEXT:    [[TMP14:%.*]] = load i32, i32* [[I3]], align 4
// CHECK7-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[TMP13]], i32 [[TMP14]]
// CHECK7-NEXT:    [[TMP15:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK7-NEXT:    [[TMP16:%.*]] = load i32, i32* [[I3]], align 4
// CHECK7-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i32 0, i32 [[TMP16]]
// CHECK7-NEXT:    store i32 [[TMP15]], i32* [[ARRAYIDX5]], align 4
// CHECK7-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK7:       omp.body.continue:
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK7:       omp.inner.for.inc:
// CHECK7-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK7-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK7-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK7-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK7:       omp.inner.for.end:
// CHECK7-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK7:       omp.loop.exit:
// CHECK7-NEXT:    [[TMP19:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK7-NEXT:    [[TMP20:%.*]] = load i32, i32* [[TMP19]], align 4
// CHECK7-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP20]])
// CHECK7-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK7:       omp.precond.end:
// CHECK7-NEXT:    ret void
// CHECK8-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l43
// CHECK8-SAME: (i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK8-NEXT:  entry:
// CHECK8-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK8-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[L_CASTED:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK8-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK8-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK8-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK8-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK8:       .execute:
// CHECK8-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3:[0-9]+]])
// CHECK8-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK8-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK8-NEXT:    [[TMP4:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP4]], i32* [[L_CASTED]], align 4
// CHECK8-NEXT:    [[TMP5:%.*]] = load i32, i32* [[L_CASTED]], align 4
// CHECK8-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK8-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [1000 x i32]* [[TMP0]], i32 [[TMP5]]) #[[ATTR3:[0-9]+]]
// CHECK8-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK8:       .omp.deinit:
// CHECK8-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK8-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK8:       .exit:
// CHECK8-NEXT:    ret void
// CHECK8-LABEL: define {{[^@]+}}@__omp_outlined__
// CHECK8-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0]] {
// CHECK8-NEXT:  entry:
// CHECK8-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK8-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[I4:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[L_CASTED:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 4
// CHECK8-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK8-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK8-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK8-NEXT:    [[TMP1:%.*]] = call i8* @__kmpc_data_sharing_push_stack(i32 4, i16 1)
// CHECK8-NEXT:    [[TMP2:%.*]] = bitcast i8* [[TMP1]] to %struct._globalized_locals_ty*
// CHECK8-NEXT:    [[L1:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY:%.*]], %struct._globalized_locals_ty* [[TMP2]], i32 0, i32 0
// CHECK8-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP3]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK8-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK8-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK8-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK8-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK8-NEXT:    store i32 [[SUB3]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK8-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK8-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK8-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK8-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK8:       omp.precond.then:
// CHECK8-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK8-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK8-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK8-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1:[0-9]+]], i32 [[TMP8]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 128)
// CHECK8-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK8-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
// CHECK8-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK8:       cond.true:
// CHECK8-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK8-NEXT:    br label [[COND_END:%.*]]
// CHECK8:       cond.false:
// CHECK8-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    br label [[COND_END]]
// CHECK8:       cond.end:
// CHECK8-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP11]], [[COND_TRUE]] ], [ [[TMP12]], [[COND_FALSE]] ]
// CHECK8-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    store i32 [[TMP13]], i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK8:       omp.inner.for.cond:
// CHECK8-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK8-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP15]], 1
// CHECK8-NEXT:    [[CMP6:%.*]] = icmp slt i32 [[TMP14]], [[ADD]]
// CHECK8-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK8:       omp.inner.for.body:
// CHECK8-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP18:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP18]], i32* [[N_CASTED]], align 4
// CHECK8-NEXT:    [[TMP19:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK8-NEXT:    [[TMP20:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP20]], i32* [[L_CASTED]], align 4
// CHECK8-NEXT:    [[TMP21:%.*]] = load i32, i32* [[L_CASTED]], align 4
// CHECK8-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK8-NEXT:    [[TMP23:%.*]] = inttoptr i32 [[TMP16]] to i8*
// CHECK8-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 4
// CHECK8-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK8-NEXT:    [[TMP25:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK8-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 4
// CHECK8-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK8-NEXT:    [[TMP27:%.*]] = inttoptr i32 [[TMP19]] to i8*
// CHECK8-NEXT:    store i8* [[TMP27]], i8** [[TMP26]], align 4
// CHECK8-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK8-NEXT:    [[TMP29:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK8-NEXT:    store i8* [[TMP29]], i8** [[TMP28]], align 4
// CHECK8-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 4
// CHECK8-NEXT:    [[TMP31:%.*]] = inttoptr i32 [[TMP21]] to i8*
// CHECK8-NEXT:    store i8* [[TMP31]], i8** [[TMP30]], align 4
// CHECK8-NEXT:    [[TMP32:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    [[TMP33:%.*]] = load i32, i32* [[TMP32]], align 4
// CHECK8-NEXT:    [[TMP34:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK8-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP33]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i32]*, i32)* @__omp_outlined__1 to i8*), i8* null, i8** [[TMP34]], i32 5)
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK8:       omp.inner.for.inc:
// CHECK8-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP35]], [[TMP36]]
// CHECK8-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP37]], [[TMP38]]
// CHECK8-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP39]], [[TMP40]]
// CHECK8-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK8-NEXT:    [[CMP10:%.*]] = icmp sgt i32 [[TMP41]], [[TMP42]]
// CHECK8-NEXT:    br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]]
// CHECK8:       cond.true11:
// CHECK8-NEXT:    [[TMP43:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK8-NEXT:    br label [[COND_END13:%.*]]
// CHECK8:       cond.false12:
// CHECK8-NEXT:    [[TMP44:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    br label [[COND_END13]]
// CHECK8:       cond.end13:
// CHECK8-NEXT:    [[COND14:%.*]] = phi i32 [ [[TMP43]], [[COND_TRUE11]] ], [ [[TMP44]], [[COND_FALSE12]] ]
// CHECK8-NEXT:    store i32 [[COND14]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP45:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    store i32 [[TMP45]], i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK8:       omp.inner.for.end:
// CHECK8-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK8:       omp.loop.exit:
// CHECK8-NEXT:    [[TMP46:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    [[TMP47:%.*]] = load i32, i32* [[TMP46]], align 4
// CHECK8-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP47]])
// CHECK8-NEXT:    [[TMP48:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK8-NEXT:    [[TMP49:%.*]] = icmp ne i32 [[TMP48]], 0
// CHECK8-NEXT:    br i1 [[TMP49]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK8:       .omp.lastprivate.then:
// CHECK8-NEXT:    [[TMP50:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP50]], i32* [[L_ADDR]], align 4
// CHECK8-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK8:       .omp.lastprivate.done:
// CHECK8-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK8:       omp.precond.end:
// CHECK8-NEXT:    call void @__kmpc_data_sharing_pop_stack(i8* [[TMP1]])
// CHECK8-NEXT:    ret void
// CHECK8-LABEL: define {{[^@]+}}@__omp_outlined__1
// CHECK8-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0]] {
// CHECK8-NEXT:  entry:
// CHECK8-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK8-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK8-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK8-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK8-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK8-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK8-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK8-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK8-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK8-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK8-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK8-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK8-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK8-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK8-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK8-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK8:       omp.precond.then:
// CHECK8-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK8-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK8-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK8-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK8-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK8-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK8-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK8-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK8-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 32)
// CHECK8-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK8:       omp.dispatch.cond:
// CHECK8-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK8-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK8-NEXT:    [[CMP4:%.*]] = icmp ugt i32 [[TMP9]], [[TMP10]]
// CHECK8-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK8:       cond.true:
// CHECK8-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK8-NEXT:    br label [[COND_END:%.*]]
// CHECK8:       cond.false:
// CHECK8-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK8-NEXT:    br label [[COND_END]]
// CHECK8:       cond.end:
// CHECK8-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP11]], [[COND_TRUE]] ], [ [[TMP12]], [[COND_FALSE]] ]
// CHECK8-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
// CHECK8-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK8-NEXT:    store i32 [[TMP13]], i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK8-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK8-NEXT:    br i1 [[CMP5]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK8:       omp.dispatch.body:
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK8:       omp.inner.for.cond:
// CHECK8-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK8-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP16]], [[TMP17]]
// CHECK8-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK8:       omp.inner.for.body:
// CHECK8-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP18]], 1
// CHECK8-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK8-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK8-NEXT:    [[TMP19:%.*]] = load i32, i32* [[I3]], align 4
// CHECK8-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i32 0, i32 [[TMP19]]
// CHECK8-NEXT:    store i32 1, i32* [[ARRAYIDX]], align 4
// CHECK8-NEXT:    [[TMP20:%.*]] = load i32, i32* [[I3]], align 4
// CHECK8-NEXT:    store i32 [[TMP20]], i32* [[L_ADDR]], align 4
// CHECK8-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK8:       omp.body.continue:
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK8:       omp.inner.for.inc:
// CHECK8-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP21]], 1
// CHECK8-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK8:       omp.inner.for.end:
// CHECK8-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK8:       omp.dispatch.inc:
// CHECK8-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK8-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK8-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_LB]], align 4
// CHECK8-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK8-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK8-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_UB]], align 4
// CHECK8-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK8:       omp.dispatch.end:
// CHECK8-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK8-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP27]])
// CHECK8-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK8-NEXT:    [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
// CHECK8-NEXT:    br i1 [[TMP29]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK8:       .omp.lastprivate.then:
// CHECK8-NEXT:    [[TMP30:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP30]], i32* [[L_ADDR]], align 4
// CHECK8-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK8:       .omp.lastprivate.done:
// CHECK8-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK8:       omp.precond.end:
// CHECK8-NEXT:    ret void
// CHECK8-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l49
// CHECK8-SAME: (i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK8-NEXT:  entry:
// CHECK8-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK8-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK8-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK8-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK8-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK8-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK8:       .execute:
// CHECK8-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK8-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK8-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK8-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK8-NEXT:    call void @__omp_outlined__2(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [1000 x i16]* [[TMP0]]) #[[ATTR3]]
// CHECK8-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK8:       .omp.deinit:
// CHECK8-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK8-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK8:       .exit:
// CHECK8-NEXT:    ret void
// CHECK8-LABEL: define {{[^@]+}}@__omp_outlined__2
// CHECK8-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK8-NEXT:  entry:
// CHECK8-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK8-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK8-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK8-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK8-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK8-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK8-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK8-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK8-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK8-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK8-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK8-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK8-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK8-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK8-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK8:       omp.precond.then:
// CHECK8-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK8-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK8-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK8-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK8-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK8-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK8-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK8-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK8:       cond.true:
// CHECK8-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK8-NEXT:    br label [[COND_END:%.*]]
// CHECK8:       cond.false:
// CHECK8-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    br label [[COND_END]]
// CHECK8:       cond.end:
// CHECK8-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK8-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK8:       omp.inner.for.cond:
// CHECK8-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK8-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK8-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK8-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK8:       omp.inner.for.body:
// CHECK8-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP16:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP16]], i32* [[N_CASTED]], align 4
// CHECK8-NEXT:    [[TMP17:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK8-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK8-NEXT:    [[TMP19:%.*]] = inttoptr i32 [[TMP14]] to i8*
// CHECK8-NEXT:    store i8* [[TMP19]], i8** [[TMP18]], align 4
// CHECK8-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK8-NEXT:    [[TMP21:%.*]] = inttoptr i32 [[TMP15]] to i8*
// CHECK8-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 4
// CHECK8-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK8-NEXT:    [[TMP23:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK8-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 4
// CHECK8-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK8-NEXT:    [[TMP25:%.*]] = bitcast [1000 x i16]* [[TMP0]] to i8*
// CHECK8-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 4
// CHECK8-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK8-NEXT:    [[TMP28:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK8-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP27]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i16]*)* @__omp_outlined__3 to i8*), i8* null, i8** [[TMP28]], i32 4)
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK8:       omp.inner.for.inc:
// CHECK8-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP29]], [[TMP30]]
// CHECK8-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP31]], [[TMP32]]
// CHECK8-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP33]], [[TMP34]]
// CHECK8-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK8-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP35]], [[TMP36]]
// CHECK8-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
// CHECK8:       cond.true10:
// CHECK8-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK8-NEXT:    br label [[COND_END12:%.*]]
// CHECK8:       cond.false11:
// CHECK8-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    br label [[COND_END12]]
// CHECK8:       cond.end12:
// CHECK8-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP37]], [[COND_TRUE10]] ], [ [[TMP38]], [[COND_FALSE11]] ]
// CHECK8-NEXT:    store i32 [[COND13]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    store i32 [[TMP39]], i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK8:       omp.inner.for.end:
// CHECK8-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK8:       omp.loop.exit:
// CHECK8-NEXT:    [[TMP40:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    [[TMP41:%.*]] = load i32, i32* [[TMP40]], align 4
// CHECK8-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP41]])
// CHECK8-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK8:       omp.precond.end:
// CHECK8-NEXT:    ret void
// CHECK8-LABEL: define {{[^@]+}}@__omp_outlined__3
// CHECK8-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK8-NEXT:  entry:
// CHECK8-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK8-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK8-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK8-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK8-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK8-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK8-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK8-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK8-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK8-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK8-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK8-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK8-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK8-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK8-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK8-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK8:       omp.precond.then:
// CHECK8-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK8-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK8-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK8-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK8-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK8-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK8-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK8-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK8-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK8-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK8-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK8:       omp.inner.for.cond:
// CHECK8-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK8-NEXT:    [[CMP4:%.*]] = icmp ule i32 [[TMP10]], [[TMP11]]
// CHECK8-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK8:       omp.inner.for.body:
// CHECK8-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK8-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK8-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK8-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I3]], align 4
// CHECK8-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i16], [1000 x i16]* [[TMP0]], i32 0, i32 [[TMP13]]
// CHECK8-NEXT:    [[TMP14:%.*]] = load i16, i16* [[ARRAYIDX]], align 2
// CHECK8-NEXT:    [[CONV:%.*]] = sext i16 [[TMP14]] to i32
// CHECK8-NEXT:    [[ADD5:%.*]] = add nsw i32 [[CONV]], 1
// CHECK8-NEXT:    [[CONV6:%.*]] = trunc i32 [[ADD5]] to i16
// CHECK8-NEXT:    store i16 [[CONV6]], i16* [[ARRAYIDX]], align 2
// CHECK8-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK8:       omp.body.continue:
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK8:       omp.inner.for.inc:
// CHECK8-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP15]], [[TMP16]]
// CHECK8-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK8:       omp.inner.for.end:
// CHECK8-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK8:       omp.loop.exit:
// CHECK8-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
// CHECK8-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP18]])
// CHECK8-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK8:       omp.precond.end:
// CHECK8-NEXT:    ret void
// CHECK8-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l54
// CHECK8-SAME: ([10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK8-NEXT:  entry:
// CHECK8-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK8-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK8-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK8-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK8-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK8-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK8-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK8:       .execute:
// CHECK8-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK8-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK8-NEXT:    call void @__omp_outlined__4(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x i32]* [[TMP0]]) #[[ATTR3]]
// CHECK8-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK8:       .omp.deinit:
// CHECK8-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK8-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK8:       .exit:
// CHECK8-NEXT:    ret void
// CHECK8-LABEL: define {{[^@]+}}@__omp_outlined__4
// CHECK8-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK8-NEXT:  entry:
// CHECK8-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK8-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x i8*], align 4
// CHECK8-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK8-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK8-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK8-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    store i32 9, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK8-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK8-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK8-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK8-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9
// CHECK8-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK8:       cond.true:
// CHECK8-NEXT:    br label [[COND_END:%.*]]
// CHECK8:       cond.false:
// CHECK8-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    br label [[COND_END]]
// CHECK8:       cond.end:
// CHECK8-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK8-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK8:       omp.inner.for.cond:
// CHECK8-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[CMP1:%.*]] = icmp slt i32 [[TMP6]], 10
// CHECK8-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK8:       omp.inner.for.body:
// CHECK8-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK8-NEXT:    [[TMP10:%.*]] = inttoptr i32 [[TMP7]] to i8*
// CHECK8-NEXT:    store i8* [[TMP10]], i8** [[TMP9]], align 4
// CHECK8-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK8-NEXT:    [[TMP12:%.*]] = inttoptr i32 [[TMP8]] to i8*
// CHECK8-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 4
// CHECK8-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK8-NEXT:    [[TMP14:%.*]] = bitcast [10 x i32]* [[TMP0]] to i8*
// CHECK8-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 4
// CHECK8-NEXT:    [[TMP15:%.*]] = bitcast [3 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK8-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, [10 x i32]*)* @__omp_outlined__5 to i8*), i8* null, i8** [[TMP15]], i32 3)
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK8:       omp.inner.for.inc:
// CHECK8-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP16]], [[TMP17]]
// CHECK8-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP18]], [[TMP19]]
// CHECK8-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK8-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP22]], 9
// CHECK8-NEXT:    br i1 [[CMP4]], label [[COND_TRUE5:%.*]], label [[COND_FALSE6:%.*]]
// CHECK8:       cond.true5:
// CHECK8-NEXT:    br label [[COND_END7:%.*]]
// CHECK8:       cond.false6:
// CHECK8-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    br label [[COND_END7]]
// CHECK8:       cond.end7:
// CHECK8-NEXT:    [[COND8:%.*]] = phi i32 [ 9, [[COND_TRUE5]] ], [ [[TMP23]], [[COND_FALSE6]] ]
// CHECK8-NEXT:    store i32 [[COND8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    store i32 [[TMP24]], i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK8:       omp.inner.for.end:
// CHECK8-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK8:       omp.loop.exit:
// CHECK8-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]])
// CHECK8-NEXT:    ret void
// CHECK8-LABEL: define {{[^@]+}}@__omp_outlined__5
// CHECK8-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK8-NEXT:  entry:
// CHECK8-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK8-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK8-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK8-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK8-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK8-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK8-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK8-NEXT:    store i32 9, i32* [[DOTOMP_UB]], align 4
// CHECK8-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK8-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP1]], i32* [[DOTOMP_LB]], align 4
// CHECK8-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_UB]], align 4
// CHECK8-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK8-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK8-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK8-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK8-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK8:       omp.inner.for.cond:
// CHECK8-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK8-NEXT:    [[CMP:%.*]] = icmp ule i32 [[TMP6]], [[TMP7]]
// CHECK8-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK8:       omp.inner.for.body:
// CHECK8-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP8]], 1
// CHECK8-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK8-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK8-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4
// CHECK8-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[TMP0]], i32 0, i32 [[TMP9]]
// CHECK8-NEXT:    [[TMP10:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK8-NEXT:    [[ADD1:%.*]] = add nsw i32 [[TMP10]], 1
// CHECK8-NEXT:    store i32 [[ADD1]], i32* [[ARRAYIDX]], align 4
// CHECK8-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK8:       omp.body.continue:
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK8:       omp.inner.for.inc:
// CHECK8-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
// CHECK8-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK8:       omp.inner.for.end:
// CHECK8-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK8:       omp.loop.exit:
// CHECK8-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK8-NEXT:    ret void
// CHECK8-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l59
// CHECK8-SAME: ([10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK8-NEXT:  entry:
// CHECK8-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK8-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[F_CASTED:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK8-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK8-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK8-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK8-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK8-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK8:       .execute:
// CHECK8-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK8-NEXT:    [[TMP2:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP2]], i32* [[F_CASTED]], align 4
// CHECK8-NEXT:    [[TMP3:%.*]] = load i32, i32* [[F_CASTED]], align 4
// CHECK8-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK8-NEXT:    call void @__omp_outlined__6(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x [10 x i32]]* [[TMP0]], i32 [[TMP3]]) #[[ATTR3]]
// CHECK8-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK8:       .omp.deinit:
// CHECK8-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK8-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK8:       .exit:
// CHECK8-NEXT:    ret void
// CHECK8-LABEL: define {{[^@]+}}@__omp_outlined__6
// CHECK8-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK8-NEXT:  entry:
// CHECK8-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK8-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[F_CASTED:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK8-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK8-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK8-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK8-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    store i32 99, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK8-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK8-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK8-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK8-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 99
// CHECK8-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK8:       cond.true:
// CHECK8-NEXT:    br label [[COND_END:%.*]]
// CHECK8:       cond.false:
// CHECK8-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    br label [[COND_END]]
// CHECK8:       cond.end:
// CHECK8-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK8-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK8:       omp.inner.for.cond:
// CHECK8-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[TMP6]], 100
// CHECK8-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK8:       omp.inner.for.body:
// CHECK8-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP9:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP9]], i32* [[F_CASTED]], align 4
// CHECK8-NEXT:    [[TMP10:%.*]] = load i32, i32* [[F_CASTED]], align 4
// CHECK8-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK8-NEXT:    [[TMP12:%.*]] = inttoptr i32 [[TMP7]] to i8*
// CHECK8-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 4
// CHECK8-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK8-NEXT:    [[TMP14:%.*]] = inttoptr i32 [[TMP8]] to i8*
// CHECK8-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 4
// CHECK8-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK8-NEXT:    [[TMP16:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK8-NEXT:    store i8* [[TMP16]], i8** [[TMP15]], align 4
// CHECK8-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK8-NEXT:    [[TMP18:%.*]] = inttoptr i32 [[TMP10]] to i8*
// CHECK8-NEXT:    store i8* [[TMP18]], i8** [[TMP17]], align 4
// CHECK8-NEXT:    [[TMP19:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK8-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, [10 x [10 x i32]]*, i32)* @__omp_outlined__7 to i8*), i8* null, i8** [[TMP19]], i32 4)
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK8:       omp.inner.for.inc:
// CHECK8-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK8-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK8-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK8-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP26]], 99
// CHECK8-NEXT:    br i1 [[CMP5]], label [[COND_TRUE6:%.*]], label [[COND_FALSE7:%.*]]
// CHECK8:       cond.true6:
// CHECK8-NEXT:    br label [[COND_END8:%.*]]
// CHECK8:       cond.false7:
// CHECK8-NEXT:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    br label [[COND_END8]]
// CHECK8:       cond.end8:
// CHECK8-NEXT:    [[COND9:%.*]] = phi i32 [ 99, [[COND_TRUE6]] ], [ [[TMP27]], [[COND_FALSE7]] ]
// CHECK8-NEXT:    store i32 [[COND9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    store i32 [[TMP28]], i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK8:       omp.inner.for.end:
// CHECK8-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK8:       omp.loop.exit:
// CHECK8-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]])
// CHECK8-NEXT:    ret void
// CHECK8-LABEL: define {{[^@]+}}@__omp_outlined__7
// CHECK8-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK8-NEXT:  entry:
// CHECK8-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK8-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK8-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK8-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK8-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK8-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK8-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK8-NEXT:    store i32 99, i32* [[DOTOMP_UB]], align 4
// CHECK8-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK8-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP1]], i32* [[DOTOMP_LB]], align 4
// CHECK8-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_UB]], align 4
// CHECK8-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK8-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK8-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK8-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK8-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK8:       omp.inner.for.cond:
// CHECK8-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK8-NEXT:    [[CMP:%.*]] = icmp ule i32 [[TMP6]], [[TMP7]]
// CHECK8-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK8:       omp.inner.for.body:
// CHECK8-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP8]], 10
// CHECK8-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK8-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK8-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK8-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[DIV2:%.*]] = sdiv i32 [[TMP10]], 10
// CHECK8-NEXT:    [[MUL3:%.*]] = mul nsw i32 [[DIV2]], 10
// CHECK8-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP9]], [[MUL3]]
// CHECK8-NEXT:    [[MUL4:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK8-NEXT:    [[ADD5:%.*]] = add nsw i32 0, [[MUL4]]
// CHECK8-NEXT:    store i32 [[ADD5]], i32* [[J]], align 4
// CHECK8-NEXT:    store i32 10, i32* [[K]], align 4
// CHECK8-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
// CHECK8-NEXT:    [[TMP12:%.*]] = load i32, i32* [[J]], align 4
// CHECK8-NEXT:    [[TMP13:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK8-NEXT:    [[MUL6:%.*]] = mul nsw i32 [[TMP12]], [[TMP13]]
// CHECK8-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP11]], [[MUL6]]
// CHECK8-NEXT:    [[TMP14:%.*]] = load i32, i32* [[K]], align 4
// CHECK8-NEXT:    [[ADD8:%.*]] = add nsw i32 [[ADD7]], [[TMP14]]
// CHECK8-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
// CHECK8-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i32 0, i32 [[TMP15]]
// CHECK8-NEXT:    [[TMP16:%.*]] = load i32, i32* [[J]], align 4
// CHECK8-NEXT:    [[ARRAYIDX9:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i32 0, i32 [[TMP16]]
// CHECK8-NEXT:    store i32 [[ADD8]], i32* [[ARRAYIDX9]], align 4
// CHECK8-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK8:       omp.body.continue:
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK8:       omp.inner.for.inc:
// CHECK8-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK8-NEXT:    store i32 [[ADD10]], i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK8:       omp.inner.for.end:
// CHECK8-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK8:       omp.loop.exit:
// CHECK8-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK8-NEXT:    ret void
// CHECK8-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l67
// CHECK8-SAME: (i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK8-NEXT:  entry:
// CHECK8-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK8-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK8-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK8-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK8-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK8-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK8:       .execute:
// CHECK8-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK8-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK8-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK8-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK8-NEXT:    call void @__omp_outlined__8(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [10 x [10 x i32]]* [[TMP0]]) #[[ATTR3]]
// CHECK8-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK8:       .omp.deinit:
// CHECK8-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK8-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK8:       .exit:
// CHECK8-NEXT:    ret void
// CHECK8-LABEL: define {{[^@]+}}@__omp_outlined__8
// CHECK8-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK8-NEXT:  entry:
// CHECK8-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK8-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK8-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK8-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i64, align 8
// CHECK8-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i64, align 8
// CHECK8-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK8-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[I9:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[J10:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK8-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK8-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK8-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK8-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK8-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK8-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK8-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK8-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK8-NEXT:    [[CONV:%.*]] = sext i32 [[DIV]] to i64
// CHECK8-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK8-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK8-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK8-NEXT:    [[CONV6:%.*]] = sext i32 [[DIV5]] to i64
// CHECK8-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV]], [[CONV6]]
// CHECK8-NEXT:    [[SUB7:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK8-NEXT:    store i64 [[SUB7]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK8-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK8-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK8-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK8-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK8-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK8:       land.lhs.true:
// CHECK8-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK8-NEXT:    [[CMP8:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK8-NEXT:    br i1 [[CMP8]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK8:       omp.precond.then:
// CHECK8-NEXT:    store i64 0, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK8-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK8-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK8-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK8-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK8-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK8-NEXT:    [[CONV11:%.*]] = zext i32 [[NVPTX_NUM_THREADS]] to i64
// CHECK8-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    [[TMP9:%.*]] = load i32, i32* [[TMP8]], align 4
// CHECK8-NEXT:    call void @__kmpc_for_static_init_8(%struct.ident_t* @[[GLOB1]], i32 [[TMP9]], i32 91, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_COMB_LB]], i64* [[DOTOMP_COMB_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 [[CONV11]])
// CHECK8-NEXT:    [[TMP10:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK8-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK8-NEXT:    [[CMP12:%.*]] = icmp sgt i64 [[TMP10]], [[TMP11]]
// CHECK8-NEXT:    br i1 [[CMP12]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK8:       cond.true:
// CHECK8-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK8-NEXT:    br label [[COND_END:%.*]]
// CHECK8:       cond.false:
// CHECK8-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK8-NEXT:    br label [[COND_END]]
// CHECK8:       cond.end:
// CHECK8-NEXT:    [[COND:%.*]] = phi i64 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
// CHECK8-NEXT:    store i64 [[COND]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK8-NEXT:    [[TMP14:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK8-NEXT:    store i64 [[TMP14]], i64* [[DOTOMP_IV]], align 8
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK8:       omp.inner.for.cond:
// CHECK8-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK8-NEXT:    [[TMP16:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK8-NEXT:    [[ADD:%.*]] = add nsw i64 [[TMP16]], 1
// CHECK8-NEXT:    [[CMP13:%.*]] = icmp slt i64 [[TMP15]], [[ADD]]
// CHECK8-NEXT:    br i1 [[CMP13]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK8:       omp.inner.for.body:
// CHECK8-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK8-NEXT:    [[TMP18:%.*]] = trunc i64 [[TMP17]] to i32
// CHECK8-NEXT:    [[TMP19:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK8-NEXT:    [[TMP20:%.*]] = trunc i64 [[TMP19]] to i32
// CHECK8-NEXT:    [[TMP21:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP21]], i32* [[N_CASTED]], align 4
// CHECK8-NEXT:    [[TMP22:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK8-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK8-NEXT:    [[TMP24:%.*]] = inttoptr i32 [[TMP18]] to i8*
// CHECK8-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 4
// CHECK8-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK8-NEXT:    [[TMP26:%.*]] = inttoptr i32 [[TMP20]] to i8*
// CHECK8-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 4
// CHECK8-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK8-NEXT:    [[TMP28:%.*]] = inttoptr i32 [[TMP22]] to i8*
// CHECK8-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 4
// CHECK8-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK8-NEXT:    [[TMP30:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK8-NEXT:    store i8* [[TMP30]], i8** [[TMP29]], align 4
// CHECK8-NEXT:    [[TMP31:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    [[TMP32:%.*]] = load i32, i32* [[TMP31]], align 4
// CHECK8-NEXT:    [[TMP33:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK8-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP32]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [10 x [10 x i32]]*)* @__omp_outlined__9 to i8*), i8* null, i8** [[TMP33]], i32 4)
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK8:       omp.inner.for.inc:
// CHECK8-NEXT:    [[TMP34:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK8-NEXT:    [[TMP35:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK8-NEXT:    [[ADD14:%.*]] = add nsw i64 [[TMP34]], [[TMP35]]
// CHECK8-NEXT:    store i64 [[ADD14]], i64* [[DOTOMP_IV]], align 8
// CHECK8-NEXT:    [[TMP36:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK8-NEXT:    [[TMP37:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK8-NEXT:    [[ADD15:%.*]] = add nsw i64 [[TMP36]], [[TMP37]]
// CHECK8-NEXT:    store i64 [[ADD15]], i64* [[DOTOMP_COMB_LB]], align 8
// CHECK8-NEXT:    [[TMP38:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK8-NEXT:    [[TMP39:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK8-NEXT:    [[ADD16:%.*]] = add nsw i64 [[TMP38]], [[TMP39]]
// CHECK8-NEXT:    store i64 [[ADD16]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK8-NEXT:    [[TMP40:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK8-NEXT:    [[TMP41:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK8-NEXT:    [[CMP17:%.*]] = icmp sgt i64 [[TMP40]], [[TMP41]]
// CHECK8-NEXT:    br i1 [[CMP17]], label [[COND_TRUE18:%.*]], label [[COND_FALSE19:%.*]]
// CHECK8:       cond.true18:
// CHECK8-NEXT:    [[TMP42:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK8-NEXT:    br label [[COND_END20:%.*]]
// CHECK8:       cond.false19:
// CHECK8-NEXT:    [[TMP43:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK8-NEXT:    br label [[COND_END20]]
// CHECK8:       cond.end20:
// CHECK8-NEXT:    [[COND21:%.*]] = phi i64 [ [[TMP42]], [[COND_TRUE18]] ], [ [[TMP43]], [[COND_FALSE19]] ]
// CHECK8-NEXT:    store i64 [[COND21]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK8-NEXT:    [[TMP44:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK8-NEXT:    store i64 [[TMP44]], i64* [[DOTOMP_IV]], align 8
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK8:       omp.inner.for.end:
// CHECK8-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK8:       omp.loop.exit:
// CHECK8-NEXT:    [[TMP45:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    [[TMP46:%.*]] = load i32, i32* [[TMP45]], align 4
// CHECK8-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP46]])
// CHECK8-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK8:       omp.precond.end:
// CHECK8-NEXT:    ret void
// CHECK8-LABEL: define {{[^@]+}}@__omp_outlined__9
// CHECK8-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK8-NEXT:  entry:
// CHECK8-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK8-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK8-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK8-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_LB:%.*]] = alloca i64, align 8
// CHECK8-NEXT:    [[DOTOMP_UB:%.*]] = alloca i64, align 8
// CHECK8-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK8-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[I11:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[J12:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK8-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK8-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK8-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK8-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK8-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK8-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK8-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK8-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK8-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK8-NEXT:    [[CONV:%.*]] = sext i32 [[DIV]] to i64
// CHECK8-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK8-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK8-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK8-NEXT:    [[CONV6:%.*]] = sext i32 [[DIV5]] to i64
// CHECK8-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV]], [[CONV6]]
// CHECK8-NEXT:    [[SUB7:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK8-NEXT:    store i64 [[SUB7]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK8-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK8-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK8-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK8-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK8-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK8:       land.lhs.true:
// CHECK8-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK8-NEXT:    [[CMP8:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK8-NEXT:    br i1 [[CMP8]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK8:       omp.precond.then:
// CHECK8-NEXT:    store i64 0, i64* [[DOTOMP_LB]], align 8
// CHECK8-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK8-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_UB]], align 8
// CHECK8-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK8-NEXT:    [[CONV9:%.*]] = zext i32 [[TMP8]] to i64
// CHECK8-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK8-NEXT:    [[CONV10:%.*]] = zext i32 [[TMP9]] to i64
// CHECK8-NEXT:    store i64 [[CONV9]], i64* [[DOTOMP_LB]], align 8
// CHECK8-NEXT:    store i64 [[CONV10]], i64* [[DOTOMP_UB]], align 8
// CHECK8-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK8-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK8-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK8-NEXT:    call void @__kmpc_for_static_init_8(%struct.ident_t* @[[GLOB2]], i32 [[TMP11]], i32 33, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_LB]], i64* [[DOTOMP_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 1)
// CHECK8-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTOMP_LB]], align 8
// CHECK8-NEXT:    store i64 [[TMP12]], i64* [[DOTOMP_IV]], align 8
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK8:       omp.inner.for.cond:
// CHECK8-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK8-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK8-NEXT:    [[CONV13:%.*]] = zext i32 [[TMP14]] to i64
// CHECK8-NEXT:    [[CMP14:%.*]] = icmp sle i64 [[TMP13]], [[CONV13]]
// CHECK8-NEXT:    br i1 [[CMP14]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK8:       omp.inner.for.body:
// CHECK8-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK8-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK8-NEXT:    [[SUB15:%.*]] = sub nsw i32 [[TMP16]], 0
// CHECK8-NEXT:    [[DIV16:%.*]] = sdiv i32 [[SUB15]], 1
// CHECK8-NEXT:    [[MUL17:%.*]] = mul nsw i32 1, [[DIV16]]
// CHECK8-NEXT:    [[CONV18:%.*]] = sext i32 [[MUL17]] to i64
// CHECK8-NEXT:    [[DIV19:%.*]] = sdiv i64 [[TMP15]], [[CONV18]]
// CHECK8-NEXT:    [[MUL20:%.*]] = mul nsw i64 [[DIV19]], 1
// CHECK8-NEXT:    [[ADD:%.*]] = add nsw i64 0, [[MUL20]]
// CHECK8-NEXT:    [[CONV21:%.*]] = trunc i64 [[ADD]] to i32
// CHECK8-NEXT:    store i32 [[CONV21]], i32* [[I11]], align 4
// CHECK8-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK8-NEXT:    [[TMP18:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK8-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK8-NEXT:    [[SUB22:%.*]] = sub nsw i32 [[TMP19]], 0
// CHECK8-NEXT:    [[DIV23:%.*]] = sdiv i32 [[SUB22]], 1
// CHECK8-NEXT:    [[MUL24:%.*]] = mul nsw i32 1, [[DIV23]]
// CHECK8-NEXT:    [[CONV25:%.*]] = sext i32 [[MUL24]] to i64
// CHECK8-NEXT:    [[DIV26:%.*]] = sdiv i64 [[TMP18]], [[CONV25]]
// CHECK8-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK8-NEXT:    [[SUB27:%.*]] = sub nsw i32 [[TMP20]], 0
// CHECK8-NEXT:    [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK8-NEXT:    [[MUL29:%.*]] = mul nsw i32 1, [[DIV28]]
// CHECK8-NEXT:    [[CONV30:%.*]] = sext i32 [[MUL29]] to i64
// CHECK8-NEXT:    [[MUL31:%.*]] = mul nsw i64 [[DIV26]], [[CONV30]]
// CHECK8-NEXT:    [[SUB32:%.*]] = sub nsw i64 [[TMP17]], [[MUL31]]
// CHECK8-NEXT:    [[MUL33:%.*]] = mul nsw i64 [[SUB32]], 1
// CHECK8-NEXT:    [[ADD34:%.*]] = add nsw i64 0, [[MUL33]]
// CHECK8-NEXT:    [[CONV35:%.*]] = trunc i64 [[ADD34]] to i32
// CHECK8-NEXT:    store i32 [[CONV35]], i32* [[J12]], align 4
// CHECK8-NEXT:    [[TMP21:%.*]] = load i32, i32* [[I11]], align 4
// CHECK8-NEXT:    [[TMP22:%.*]] = load i32, i32* [[J12]], align 4
// CHECK8-NEXT:    [[ADD36:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
// CHECK8-NEXT:    [[TMP23:%.*]] = load i32, i32* [[I11]], align 4
// CHECK8-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i32 0, i32 [[TMP23]]
// CHECK8-NEXT:    [[TMP24:%.*]] = load i32, i32* [[J12]], align 4
// CHECK8-NEXT:    [[ARRAYIDX37:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i32 0, i32 [[TMP24]]
// CHECK8-NEXT:    store i32 [[ADD36]], i32* [[ARRAYIDX37]], align 4
// CHECK8-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK8:       omp.body.continue:
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK8:       omp.inner.for.inc:
// CHECK8-NEXT:    [[TMP25:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK8-NEXT:    [[TMP26:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK8-NEXT:    [[ADD38:%.*]] = add nsw i64 [[TMP25]], [[TMP26]]
// CHECK8-NEXT:    store i64 [[ADD38]], i64* [[DOTOMP_IV]], align 8
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK8:       omp.inner.for.end:
// CHECK8-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK8:       omp.loop.exit:
// CHECK8-NEXT:    [[TMP27:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    [[TMP28:%.*]] = load i32, i32* [[TMP27]], align 4
// CHECK8-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP28]])
// CHECK8-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK8:       omp.precond.end:
// CHECK8-NEXT:    ret void
// CHECK8-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l74
// CHECK8-SAME: (i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK8-NEXT:  entry:
// CHECK8-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK8-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK8-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK8-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK8-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK8-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK8-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK8:       .execute:
// CHECK8-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK8-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK8-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK8-NEXT:    [[TMP4:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK8-NEXT:    call void @__omp_outlined__10(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [1000 x i32]* [[TMP0]], i32* [[TMP4]]) #[[ATTR3]]
// CHECK8-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK8:       .omp.deinit:
// CHECK8-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK8-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK8:       .exit:
// CHECK8-NEXT:    ret void
// CHECK8-LABEL: define {{[^@]+}}@__omp_outlined__10
// CHECK8-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK8-NEXT:  entry:
// CHECK8-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK8-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 4
// CHECK8-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK8-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK8-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK8-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK8-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK8-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK8-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK8-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK8-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK8-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK8-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK8-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK8-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK8-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK8:       omp.precond.then:
// CHECK8-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK8-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK8-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK8-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK8-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK8-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK8-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK8-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK8:       cond.true:
// CHECK8-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK8-NEXT:    br label [[COND_END:%.*]]
// CHECK8:       cond.false:
// CHECK8-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    br label [[COND_END]]
// CHECK8:       cond.end:
// CHECK8-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK8-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK8:       omp.inner.for.cond:
// CHECK8-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK8-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK8-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK8-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK8:       omp.inner.for.body:
// CHECK8-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP16:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP16]], i32* [[N_CASTED]], align 4
// CHECK8-NEXT:    [[TMP17:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK8-NEXT:    [[TMP18:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK8-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK8-NEXT:    [[TMP20:%.*]] = inttoptr i32 [[TMP14]] to i8*
// CHECK8-NEXT:    store i8* [[TMP20]], i8** [[TMP19]], align 4
// CHECK8-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK8-NEXT:    [[TMP22:%.*]] = inttoptr i32 [[TMP15]] to i8*
// CHECK8-NEXT:    store i8* [[TMP22]], i8** [[TMP21]], align 4
// CHECK8-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK8-NEXT:    [[TMP24:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK8-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 4
// CHECK8-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK8-NEXT:    [[TMP26:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK8-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 4
// CHECK8-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 4
// CHECK8-NEXT:    [[TMP28:%.*]] = bitcast i32* [[TMP18]] to i8*
// CHECK8-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 4
// CHECK8-NEXT:    [[TMP29:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    [[TMP30:%.*]] = load i32, i32* [[TMP29]], align 4
// CHECK8-NEXT:    [[TMP31:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK8-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP30]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i32]*, i32*)* @__omp_outlined__11 to i8*), i8* null, i8** [[TMP31]], i32 5)
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK8:       omp.inner.for.inc:
// CHECK8-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP32]], [[TMP33]]
// CHECK8-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP34]], [[TMP35]]
// CHECK8-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP36]], [[TMP37]]
// CHECK8-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK8-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP38]], [[TMP39]]
// CHECK8-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
// CHECK8:       cond.true10:
// CHECK8-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK8-NEXT:    br label [[COND_END12:%.*]]
// CHECK8:       cond.false11:
// CHECK8-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    br label [[COND_END12]]
// CHECK8:       cond.end12:
// CHECK8-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP40]], [[COND_TRUE10]] ], [ [[TMP41]], [[COND_FALSE11]] ]
// CHECK8-NEXT:    store i32 [[COND13]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK8-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK8-NEXT:    store i32 [[TMP42]], i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK8:       omp.inner.for.end:
// CHECK8-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK8:       omp.loop.exit:
// CHECK8-NEXT:    [[TMP43:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    [[TMP44:%.*]] = load i32, i32* [[TMP43]], align 4
// CHECK8-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP44]])
// CHECK8-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK8:       omp.precond.end:
// CHECK8-NEXT:    ret void
// CHECK8-LABEL: define {{[^@]+}}@__omp_outlined__11
// CHECK8-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK8-NEXT:  entry:
// CHECK8-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK8-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK8-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK8-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK8-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK8-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK8-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK8-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK8-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK8-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK8-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK8-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK8-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK8-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK8-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK8-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK8-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK8-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK8-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK8:       omp.precond.then:
// CHECK8-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK8-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK8-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK8-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK8-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK8-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK8-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK8-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK8-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK8-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK8-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK8-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK8:       omp.inner.for.cond:
// CHECK8-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK8-NEXT:    [[CMP4:%.*]] = icmp ule i32 [[TMP10]], [[TMP11]]
// CHECK8-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK8:       omp.inner.for.body:
// CHECK8-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK8-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK8-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK8-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK8-NEXT:    [[TMP14:%.*]] = load i32, i32* [[I3]], align 4
// CHECK8-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[TMP13]], i32 [[TMP14]]
// CHECK8-NEXT:    [[TMP15:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK8-NEXT:    [[TMP16:%.*]] = load i32, i32* [[I3]], align 4
// CHECK8-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i32 0, i32 [[TMP16]]
// CHECK8-NEXT:    store i32 [[TMP15]], i32* [[ARRAYIDX5]], align 4
// CHECK8-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK8:       omp.body.continue:
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK8:       omp.inner.for.inc:
// CHECK8-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK8-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK8-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK8-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK8:       omp.inner.for.end:
// CHECK8-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK8:       omp.loop.exit:
// CHECK8-NEXT:    [[TMP19:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK8-NEXT:    [[TMP20:%.*]] = load i32, i32* [[TMP19]], align 4
// CHECK8-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP20]])
// CHECK8-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK8:       omp.precond.end:
// CHECK8-NEXT:    ret void
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l43
// CHECK9-SAME: (i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i64 [[L:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK9-NEXT:  entry:
// CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK9-NEXT:    [[L_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[L_CASTED:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK9-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK9-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK9-NEXT:    store i64 [[L]], i64* [[L_ADDR]], align 8
// CHECK9-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK9-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK9-NEXT:    [[CONV1:%.*]] = bitcast i64* [[L_ADDR]] to i32*
// CHECK9-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK9-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK9-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK9:       .execute:
// CHECK9-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3:[0-9]+]])
// CHECK9-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK9-NEXT:    [[CONV2:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK9-NEXT:    store i32 [[TMP2]], i32* [[CONV2]], align 4
// CHECK9-NEXT:    [[TMP3:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK9-NEXT:    [[TMP4:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK9-NEXT:    [[CONV3:%.*]] = bitcast i64* [[L_CASTED]] to i32*
// CHECK9-NEXT:    store i32 [[TMP4]], i32* [[CONV3]], align 4
// CHECK9-NEXT:    [[TMP5:%.*]] = load i64, i64* [[L_CASTED]], align 8
// CHECK9-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK9-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP3]], [1000 x i32]* [[TMP0]], i64 [[TMP5]]) #[[ATTR3:[0-9]+]]
// CHECK9-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK9:       .omp.deinit:
// CHECK9-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK9-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK9:       .exit:
// CHECK9-NEXT:    ret void
// CHECK9-LABEL: define {{[^@]+}}@__omp_outlined__
// CHECK9-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i64 [[L:%.*]]) #[[ATTR0]] {
// CHECK9-NEXT:  entry:
// CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK9-NEXT:    [[L_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[I5:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[L_CASTED:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 8
// CHECK9-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK9-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK9-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK9-NEXT:    store i64 [[L]], i64* [[L_ADDR]], align 8
// CHECK9-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK9-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK9-NEXT:    [[CONV1:%.*]] = bitcast i64* [[L_ADDR]] to i32*
// CHECK9-NEXT:    [[TMP1:%.*]] = load i16, i16* @"_openmp_static_kernel$is_shared", align 2
// CHECK9-NEXT:    [[TMP2:%.*]] = load i64, i64* @"_openmp_static_kernel$size", align 8
// CHECK9-NEXT:    call void @__kmpc_get_team_static_memory(i16 1, i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds (%"union._shared_openmp_static_memory_type_$_", %"union._shared_openmp_static_memory_type_$_" addrspace(3)* @"_openmp_shared_static_glob_rd_$_", i32 0, i32 0, i32 0) to i8*), i64 [[TMP2]], i16 [[TMP1]], i8** addrspacecast (i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr" to i8**))
// CHECK9-NEXT:    [[TMP3:%.*]] = load i8*, i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr", align 8
// CHECK9-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i8, i8* [[TMP3]], i64 0
// CHECK9-NEXT:    [[TMP5:%.*]] = bitcast i8* [[TMP4]] to %struct._globalized_locals_ty*
// CHECK9-NEXT:    [[L2:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY:%.*]], %struct._globalized_locals_ty* [[TMP5]], i32 0, i32 0
// CHECK9-NEXT:    [[TMP6:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK9-NEXT:    store i32 [[TMP6]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK9-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], 0
// CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK9-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK9-NEXT:    store i32 [[SUB4]], i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK9-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK9-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP8]]
// CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK9:       omp.precond.then:
// CHECK9-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK9-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK9-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK9-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1:[0-9]+]], i32 [[TMP11]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 128)
// CHECK9-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK9-NEXT:    [[CMP6:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
// CHECK9-NEXT:    br i1 [[CMP6]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK9:       cond.true:
// CHECK9-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK9-NEXT:    br label [[COND_END:%.*]]
// CHECK9:       cond.false:
// CHECK9-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    br label [[COND_END]]
// CHECK9:       cond.end:
// CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
// CHECK9-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    store i32 [[TMP16]], i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK9:       omp.inner.for.cond:
// CHECK9-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP18]], 1
// CHECK9-NEXT:    [[CMP7:%.*]] = icmp slt i32 [[TMP17]], [[ADD]]
// CHECK9-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK9:       omp.inner.for.body:
// CHECK9-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
// CHECK9-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP22:%.*]] = zext i32 [[TMP21]] to i64
// CHECK9-NEXT:    [[TMP23:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK9-NEXT:    [[CONV8:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK9-NEXT:    store i32 [[TMP23]], i32* [[CONV8]], align 4
// CHECK9-NEXT:    [[TMP24:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK9-NEXT:    [[TMP25:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK9-NEXT:    [[CONV9:%.*]] = bitcast i64* [[L_CASTED]] to i32*
// CHECK9-NEXT:    store i32 [[TMP25]], i32* [[CONV9]], align 4
// CHECK9-NEXT:    [[TMP26:%.*]] = load i64, i64* [[L_CASTED]], align 8
// CHECK9-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK9-NEXT:    [[TMP28:%.*]] = inttoptr i64 [[TMP20]] to i8*
// CHECK9-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 8
// CHECK9-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK9-NEXT:    [[TMP30:%.*]] = inttoptr i64 [[TMP22]] to i8*
// CHECK9-NEXT:    store i8* [[TMP30]], i8** [[TMP29]], align 8
// CHECK9-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK9-NEXT:    [[TMP32:%.*]] = inttoptr i64 [[TMP24]] to i8*
// CHECK9-NEXT:    store i8* [[TMP32]], i8** [[TMP31]], align 8
// CHECK9-NEXT:    [[TMP33:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK9-NEXT:    [[TMP34:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK9-NEXT:    store i8* [[TMP34]], i8** [[TMP33]], align 8
// CHECK9-NEXT:    [[TMP35:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 4
// CHECK9-NEXT:    [[TMP36:%.*]] = inttoptr i64 [[TMP26]] to i8*
// CHECK9-NEXT:    store i8* [[TMP36]], i8** [[TMP35]], align 8
// CHECK9-NEXT:    [[TMP37:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    [[TMP38:%.*]] = load i32, i32* [[TMP37]], align 4
// CHECK9-NEXT:    [[TMP39:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK9-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP38]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, i64, [1000 x i32]*, i64)* @__omp_outlined__1 to i8*), i8* null, i8** [[TMP39]], i64 5)
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK9:       omp.inner.for.inc:
// CHECK9-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP40]], [[TMP41]]
// CHECK9-NEXT:    store i32 [[ADD10]], i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    [[TMP43:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP42]], [[TMP43]]
// CHECK9-NEXT:    store i32 [[ADD11]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    [[TMP44:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP45:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP44]], [[TMP45]]
// CHECK9-NEXT:    store i32 [[ADD12]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP46:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP47:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK9-NEXT:    [[CMP13:%.*]] = icmp sgt i32 [[TMP46]], [[TMP47]]
// CHECK9-NEXT:    br i1 [[CMP13]], label [[COND_TRUE14:%.*]], label [[COND_FALSE15:%.*]]
// CHECK9:       cond.true14:
// CHECK9-NEXT:    [[TMP48:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK9-NEXT:    br label [[COND_END16:%.*]]
// CHECK9:       cond.false15:
// CHECK9-NEXT:    [[TMP49:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    br label [[COND_END16]]
// CHECK9:       cond.end16:
// CHECK9-NEXT:    [[COND17:%.*]] = phi i32 [ [[TMP48]], [[COND_TRUE14]] ], [ [[TMP49]], [[COND_FALSE15]] ]
// CHECK9-NEXT:    store i32 [[COND17]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP50:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    store i32 [[TMP50]], i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK9:       omp.inner.for.end:
// CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK9:       omp.loop.exit:
// CHECK9-NEXT:    [[TMP51:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    [[TMP52:%.*]] = load i32, i32* [[TMP51]], align 4
// CHECK9-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP52]])
// CHECK9-NEXT:    [[TMP53:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK9-NEXT:    [[TMP54:%.*]] = icmp ne i32 [[TMP53]], 0
// CHECK9-NEXT:    br i1 [[TMP54]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK9:       .omp.lastprivate.then:
// CHECK9-NEXT:    [[TMP55:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK9-NEXT:    store i32 [[TMP55]], i32* [[CONV1]], align 8
// CHECK9-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK9:       .omp.lastprivate.done:
// CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK9:       omp.precond.end:
// CHECK9-NEXT:    [[TMP56:%.*]] = load i16, i16* @"_openmp_static_kernel$is_shared", align 2
// CHECK9-NEXT:    call void @__kmpc_restore_team_static_memory(i16 1, i16 [[TMP56]])
// CHECK9-NEXT:    ret void
// CHECK9-LABEL: define {{[^@]+}}@__omp_outlined__1
// CHECK9-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i64 [[L:%.*]]) #[[ATTR0]] {
// CHECK9-NEXT:  entry:
// CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK9-NEXT:    [[L_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[I6:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK9-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK9-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK9-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK9-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK9-NEXT:    store i64 [[L]], i64* [[L_ADDR]], align 8
// CHECK9-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK9-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK9-NEXT:    [[CONV1:%.*]] = bitcast i64* [[L_ADDR]] to i32*
// CHECK9-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK9-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK9-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK9-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK9-NEXT:    store i32 [[SUB3]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK9-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK9-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK9:       omp.precond.then:
// CHECK9-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK9-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK9-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK9-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK9-NEXT:    [[CONV4:%.*]] = trunc i64 [[TMP5]] to i32
// CHECK9-NEXT:    [[TMP6:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK9-NEXT:    [[CONV5:%.*]] = trunc i64 [[TMP6]] to i32
// CHECK9-NEXT:    store i32 [[CONV4]], i32* [[DOTOMP_LB]], align 4
// CHECK9-NEXT:    store i32 [[CONV5]], i32* [[DOTOMP_UB]], align 4
// CHECK9-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK9-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK9-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 32)
// CHECK9-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK9:       omp.dispatch.cond:
// CHECK9-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK9-NEXT:    [[CONV7:%.*]] = sext i32 [[TMP9]] to i64
// CHECK9-NEXT:    [[TMP10:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK9-NEXT:    [[CMP8:%.*]] = icmp ugt i64 [[CONV7]], [[TMP10]]
// CHECK9-NEXT:    br i1 [[CMP8]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK9:       cond.true:
// CHECK9-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK9-NEXT:    br label [[COND_END:%.*]]
// CHECK9:       cond.false:
// CHECK9-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK9-NEXT:    [[CONV9:%.*]] = sext i32 [[TMP12]] to i64
// CHECK9-NEXT:    br label [[COND_END]]
// CHECK9:       cond.end:
// CHECK9-NEXT:    [[COND:%.*]] = phi i64 [ [[TMP11]], [[COND_TRUE]] ], [ [[CONV9]], [[COND_FALSE]] ]
// CHECK9-NEXT:    [[CONV10:%.*]] = trunc i64 [[COND]] to i32
// CHECK9-NEXT:    store i32 [[CONV10]], i32* [[DOTOMP_UB]], align 4
// CHECK9-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK9-NEXT:    store i32 [[TMP13]], i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK9-NEXT:    [[CMP11:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK9-NEXT:    br i1 [[CMP11]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK9:       omp.dispatch.body:
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK9:       omp.inner.for.cond:
// CHECK9-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK9-NEXT:    [[CMP12:%.*]] = icmp sle i32 [[TMP16]], [[TMP17]]
// CHECK9-NEXT:    br i1 [[CMP12]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK9:       omp.inner.for.body:
// CHECK9-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP18]], 1
// CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK9-NEXT:    store i32 [[ADD]], i32* [[I6]], align 4
// CHECK9-NEXT:    [[TMP19:%.*]] = load i32, i32* [[I6]], align 4
// CHECK9-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP19]] to i64
// CHECK9-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK9-NEXT:    store i32 1, i32* [[ARRAYIDX]], align 4
// CHECK9-NEXT:    [[TMP20:%.*]] = load i32, i32* [[I6]], align 4
// CHECK9-NEXT:    store i32 [[TMP20]], i32* [[CONV1]], align 8
// CHECK9-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK9:       omp.body.continue:
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK9:       omp.inner.for.inc:
// CHECK9-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[ADD13:%.*]] = add nsw i32 [[TMP21]], 1
// CHECK9-NEXT:    store i32 [[ADD13]], i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK9:       omp.inner.for.end:
// CHECK9-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK9:       omp.dispatch.inc:
// CHECK9-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK9-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    [[ADD14:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK9-NEXT:    store i32 [[ADD14]], i32* [[DOTOMP_LB]], align 4
// CHECK9-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK9-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    [[ADD15:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK9-NEXT:    store i32 [[ADD15]], i32* [[DOTOMP_UB]], align 4
// CHECK9-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK9:       omp.dispatch.end:
// CHECK9-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK9-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP27]])
// CHECK9-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK9-NEXT:    [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
// CHECK9-NEXT:    br i1 [[TMP29]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK9:       .omp.lastprivate.then:
// CHECK9-NEXT:    [[TMP30:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK9-NEXT:    store i32 [[TMP30]], i32* [[CONV1]], align 8
// CHECK9-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK9:       .omp.lastprivate.done:
// CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK9:       omp.precond.end:
// CHECK9-NEXT:    ret void
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l49
// CHECK9-SAME: (i64 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK9-NEXT:  entry:
// CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 8
// CHECK9-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK9-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK9-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 8
// CHECK9-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK9-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 8
// CHECK9-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK9-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK9-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK9:       .execute:
// CHECK9-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK9-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK9-NEXT:    [[CONV1:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK9-NEXT:    store i32 [[TMP2]], i32* [[CONV1]], align 4
// CHECK9-NEXT:    [[TMP3:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK9-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK9-NEXT:    call void @__omp_outlined__2(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP3]], [1000 x i16]* [[TMP0]]) #[[ATTR3]]
// CHECK9-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK9:       .omp.deinit:
// CHECK9-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK9-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK9:       .exit:
// CHECK9-NEXT:    ret void
// CHECK9-LABEL: define {{[^@]+}}@__omp_outlined__2
// CHECK9-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK9-NEXT:  entry:
// CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 8
// CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 8
// CHECK9-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK9-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK9-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 8
// CHECK9-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK9-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 8
// CHECK9-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK9-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK9-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK9-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK9-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK9-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK9:       omp.precond.then:
// CHECK9-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK9-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK9-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK9-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK9-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK9-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK9-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK9-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK9:       cond.true:
// CHECK9-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK9-NEXT:    br label [[COND_END:%.*]]
// CHECK9:       cond.false:
// CHECK9-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    br label [[COND_END]]
// CHECK9:       cond.end:
// CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK9-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK9:       omp.inner.for.cond:
// CHECK9-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK9-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK9-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK9:       omp.inner.for.body:
// CHECK9-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    [[TMP15:%.*]] = zext i32 [[TMP14]] to i64
// CHECK9-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP17:%.*]] = zext i32 [[TMP16]] to i64
// CHECK9-NEXT:    [[TMP18:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK9-NEXT:    [[CONV6:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK9-NEXT:    store i32 [[TMP18]], i32* [[CONV6]], align 4
// CHECK9-NEXT:    [[TMP19:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK9-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK9-NEXT:    [[TMP21:%.*]] = inttoptr i64 [[TMP15]] to i8*
// CHECK9-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 8
// CHECK9-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK9-NEXT:    [[TMP23:%.*]] = inttoptr i64 [[TMP17]] to i8*
// CHECK9-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 8
// CHECK9-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK9-NEXT:    [[TMP25:%.*]] = inttoptr i64 [[TMP19]] to i8*
// CHECK9-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 8
// CHECK9-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK9-NEXT:    [[TMP27:%.*]] = bitcast [1000 x i16]* [[TMP0]] to i8*
// CHECK9-NEXT:    store i8* [[TMP27]], i8** [[TMP26]], align 8
// CHECK9-NEXT:    [[TMP28:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    [[TMP29:%.*]] = load i32, i32* [[TMP28]], align 4
// CHECK9-NEXT:    [[TMP30:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK9-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP29]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, i64, [1000 x i16]*)* @__omp_outlined__3 to i8*), i8* null, i8** [[TMP30]], i64 4)
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK9:       omp.inner.for.inc:
// CHECK9-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP31]], [[TMP32]]
// CHECK9-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP33]], [[TMP34]]
// CHECK9-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP35]], [[TMP36]]
// CHECK9-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK9-NEXT:    [[CMP10:%.*]] = icmp sgt i32 [[TMP37]], [[TMP38]]
// CHECK9-NEXT:    br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]]
// CHECK9:       cond.true11:
// CHECK9-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK9-NEXT:    br label [[COND_END13:%.*]]
// CHECK9:       cond.false12:
// CHECK9-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    br label [[COND_END13]]
// CHECK9:       cond.end13:
// CHECK9-NEXT:    [[COND14:%.*]] = phi i32 [ [[TMP39]], [[COND_TRUE11]] ], [ [[TMP40]], [[COND_FALSE12]] ]
// CHECK9-NEXT:    store i32 [[COND14]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    store i32 [[TMP41]], i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK9:       omp.inner.for.end:
// CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK9:       omp.loop.exit:
// CHECK9-NEXT:    [[TMP42:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    [[TMP43:%.*]] = load i32, i32* [[TMP42]], align 4
// CHECK9-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP43]])
// CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK9:       omp.precond.end:
// CHECK9-NEXT:    ret void
// CHECK9-LABEL: define {{[^@]+}}@__omp_outlined__3
// CHECK9-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], i64 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK9-NEXT:  entry:
// CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 8
// CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[I5:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK9-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK9-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK9-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK9-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 8
// CHECK9-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK9-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 8
// CHECK9-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK9-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK9-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK9-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK9-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK9-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK9:       omp.precond.then:
// CHECK9-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK9-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK9-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK9-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK9-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP5]] to i32
// CHECK9-NEXT:    [[TMP6:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK9-NEXT:    [[CONV4:%.*]] = trunc i64 [[TMP6]] to i32
// CHECK9-NEXT:    store i32 [[CONV3]], i32* [[DOTOMP_LB]], align 4
// CHECK9-NEXT:    store i32 [[CONV4]], i32* [[DOTOMP_UB]], align 4
// CHECK9-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK9-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK9-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK9-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK9-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK9:       omp.inner.for.cond:
// CHECK9-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[CONV6:%.*]] = sext i32 [[TMP10]] to i64
// CHECK9-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK9-NEXT:    [[CMP7:%.*]] = icmp ule i64 [[CONV6]], [[TMP11]]
// CHECK9-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK9:       omp.inner.for.body:
// CHECK9-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK9-NEXT:    store i32 [[ADD]], i32* [[I5]], align 4
// CHECK9-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I5]], align 4
// CHECK9-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP13]] to i64
// CHECK9-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i16], [1000 x i16]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK9-NEXT:    [[TMP14:%.*]] = load i16, i16* [[ARRAYIDX]], align 2
// CHECK9-NEXT:    [[CONV8:%.*]] = sext i16 [[TMP14]] to i32
// CHECK9-NEXT:    [[ADD9:%.*]] = add nsw i32 [[CONV8]], 1
// CHECK9-NEXT:    [[CONV10:%.*]] = trunc i32 [[ADD9]] to i16
// CHECK9-NEXT:    store i16 [[CONV10]], i16* [[ARRAYIDX]], align 2
// CHECK9-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK9:       omp.body.continue:
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK9:       omp.inner.for.inc:
// CHECK9-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP15]], [[TMP16]]
// CHECK9-NEXT:    store i32 [[ADD11]], i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK9:       omp.inner.for.end:
// CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK9:       omp.loop.exit:
// CHECK9-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
// CHECK9-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP18]])
// CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK9:       omp.precond.end:
// CHECK9-NEXT:    ret void
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l54
// CHECK9-SAME: ([10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK9-NEXT:  entry:
// CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 8
// CHECK9-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK9-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 8
// CHECK9-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 8
// CHECK9-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK9-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK9-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK9:       .execute:
// CHECK9-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK9-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK9-NEXT:    call void @__omp_outlined__4(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x i32]* [[TMP0]]) #[[ATTR3]]
// CHECK9-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK9:       .omp.deinit:
// CHECK9-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK9-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK9:       .exit:
// CHECK9-NEXT:    ret void
// CHECK9-LABEL: define {{[^@]+}}@__omp_outlined__4
// CHECK9-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK9-NEXT:  entry:
// CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 8
// CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x i8*], align 8
// CHECK9-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK9-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 8
// CHECK9-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 8
// CHECK9-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    store i32 9, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK9-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK9-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK9-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK9-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9
// CHECK9-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK9:       cond.true:
// CHECK9-NEXT:    br label [[COND_END:%.*]]
// CHECK9:       cond.false:
// CHECK9-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    br label [[COND_END]]
// CHECK9:       cond.end:
// CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK9-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK9:       omp.inner.for.cond:
// CHECK9-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[CMP1:%.*]] = icmp slt i32 [[TMP6]], 10
// CHECK9-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK9:       omp.inner.for.body:
// CHECK9-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
// CHECK9-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
// CHECK9-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK9-NEXT:    [[TMP12:%.*]] = inttoptr i64 [[TMP8]] to i8*
// CHECK9-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 8
// CHECK9-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK9-NEXT:    [[TMP14:%.*]] = inttoptr i64 [[TMP10]] to i8*
// CHECK9-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 8
// CHECK9-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK9-NEXT:    [[TMP16:%.*]] = bitcast [10 x i32]* [[TMP0]] to i8*
// CHECK9-NEXT:    store i8* [[TMP16]], i8** [[TMP15]], align 8
// CHECK9-NEXT:    [[TMP17:%.*]] = bitcast [3 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK9-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, [10 x i32]*)* @__omp_outlined__5 to i8*), i8* null, i8** [[TMP17]], i64 3)
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK9:       omp.inner.for.inc:
// CHECK9-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP18]], [[TMP19]]
// CHECK9-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK9-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK9-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP24]], 9
// CHECK9-NEXT:    br i1 [[CMP4]], label [[COND_TRUE5:%.*]], label [[COND_FALSE6:%.*]]
// CHECK9:       cond.true5:
// CHECK9-NEXT:    br label [[COND_END7:%.*]]
// CHECK9:       cond.false6:
// CHECK9-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    br label [[COND_END7]]
// CHECK9:       cond.end7:
// CHECK9-NEXT:    [[COND8:%.*]] = phi i32 [ 9, [[COND_TRUE5]] ], [ [[TMP25]], [[COND_FALSE6]] ]
// CHECK9-NEXT:    store i32 [[COND8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    store i32 [[TMP26]], i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK9:       omp.inner.for.end:
// CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK9:       omp.loop.exit:
// CHECK9-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]])
// CHECK9-NEXT:    ret void
// CHECK9-LABEL: define {{[^@]+}}@__omp_outlined__5
// CHECK9-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK9-NEXT:  entry:
// CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 8
// CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK9-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK9-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK9-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 8
// CHECK9-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 8
// CHECK9-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK9-NEXT:    store i32 9, i32* [[DOTOMP_UB]], align 4
// CHECK9-NEXT:    [[TMP1:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK9-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP1]] to i32
// CHECK9-NEXT:    [[TMP2:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK9-NEXT:    [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32
// CHECK9-NEXT:    store i32 [[CONV]], i32* [[DOTOMP_LB]], align 4
// CHECK9-NEXT:    store i32 [[CONV1]], i32* [[DOTOMP_UB]], align 4
// CHECK9-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK9-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK9-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK9-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK9-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK9:       omp.inner.for.cond:
// CHECK9-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[CONV2:%.*]] = sext i32 [[TMP6]] to i64
// CHECK9-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK9-NEXT:    [[CMP:%.*]] = icmp ule i64 [[CONV2]], [[TMP7]]
// CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK9:       omp.inner.for.body:
// CHECK9-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP8]], 1
// CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK9-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK9-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4
// CHECK9-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP9]] to i64
// CHECK9-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK9-NEXT:    [[TMP10:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK9-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP10]], 1
// CHECK9-NEXT:    store i32 [[ADD3]], i32* [[ARRAYIDX]], align 4
// CHECK9-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK9:       omp.body.continue:
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK9:       omp.inner.for.inc:
// CHECK9-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
// CHECK9-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK9:       omp.inner.for.end:
// CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK9:       omp.loop.exit:
// CHECK9-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK9-NEXT:    ret void
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l59
// CHECK9-SAME: ([10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i64 [[F:%.*]]) #[[ATTR0]] {
// CHECK9-NEXT:  entry:
// CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK9-NEXT:    [[F_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[F_CASTED:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK9-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK9-NEXT:    store i64 [[F]], i64* [[F_ADDR]], align 8
// CHECK9-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK9-NEXT:    [[CONV:%.*]] = bitcast i64* [[F_ADDR]] to i32*
// CHECK9-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK9-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK9-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK9:       .execute:
// CHECK9-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK9-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK9-NEXT:    [[CONV1:%.*]] = bitcast i64* [[F_CASTED]] to i32*
// CHECK9-NEXT:    store i32 [[TMP2]], i32* [[CONV1]], align 4
// CHECK9-NEXT:    [[TMP3:%.*]] = load i64, i64* [[F_CASTED]], align 8
// CHECK9-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK9-NEXT:    call void @__omp_outlined__6(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x [10 x i32]]* [[TMP0]], i64 [[TMP3]]) #[[ATTR3]]
// CHECK9-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK9:       .omp.deinit:
// CHECK9-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK9-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK9:       .exit:
// CHECK9-NEXT:    ret void
// CHECK9-LABEL: define {{[^@]+}}@__omp_outlined__6
// CHECK9-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i64 [[F:%.*]]) #[[ATTR0]] {
// CHECK9-NEXT:  entry:
// CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK9-NEXT:    [[F_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[F_CASTED:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 8
// CHECK9-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK9-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK9-NEXT:    store i64 [[F]], i64* [[F_ADDR]], align 8
// CHECK9-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK9-NEXT:    [[CONV:%.*]] = bitcast i64* [[F_ADDR]] to i32*
// CHECK9-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    store i32 99, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK9-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK9-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK9-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK9-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 99
// CHECK9-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK9:       cond.true:
// CHECK9-NEXT:    br label [[COND_END:%.*]]
// CHECK9:       cond.false:
// CHECK9-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    br label [[COND_END]]
// CHECK9:       cond.end:
// CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK9-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK9:       omp.inner.for.cond:
// CHECK9-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[TMP6]], 100
// CHECK9-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK9:       omp.inner.for.body:
// CHECK9-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
// CHECK9-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
// CHECK9-NEXT:    [[TMP11:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK9-NEXT:    [[CONV3:%.*]] = bitcast i64* [[F_CASTED]] to i32*
// CHECK9-NEXT:    store i32 [[TMP11]], i32* [[CONV3]], align 4
// CHECK9-NEXT:    [[TMP12:%.*]] = load i64, i64* [[F_CASTED]], align 8
// CHECK9-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK9-NEXT:    [[TMP14:%.*]] = inttoptr i64 [[TMP8]] to i8*
// CHECK9-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 8
// CHECK9-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK9-NEXT:    [[TMP16:%.*]] = inttoptr i64 [[TMP10]] to i8*
// CHECK9-NEXT:    store i8* [[TMP16]], i8** [[TMP15]], align 8
// CHECK9-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK9-NEXT:    [[TMP18:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK9-NEXT:    store i8* [[TMP18]], i8** [[TMP17]], align 8
// CHECK9-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK9-NEXT:    [[TMP20:%.*]] = inttoptr i64 [[TMP12]] to i8*
// CHECK9-NEXT:    store i8* [[TMP20]], i8** [[TMP19]], align 8
// CHECK9-NEXT:    [[TMP21:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK9-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, [10 x [10 x i32]]*, i64)* @__omp_outlined__7 to i8*), i8* null, i8** [[TMP21]], i64 4)
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK9:       omp.inner.for.inc:
// CHECK9-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK9-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK9-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP26]], [[TMP27]]
// CHECK9-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[CMP6:%.*]] = icmp sgt i32 [[TMP28]], 99
// CHECK9-NEXT:    br i1 [[CMP6]], label [[COND_TRUE7:%.*]], label [[COND_FALSE8:%.*]]
// CHECK9:       cond.true7:
// CHECK9-NEXT:    br label [[COND_END9:%.*]]
// CHECK9:       cond.false8:
// CHECK9-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    br label [[COND_END9]]
// CHECK9:       cond.end9:
// CHECK9-NEXT:    [[COND10:%.*]] = phi i32 [ 99, [[COND_TRUE7]] ], [ [[TMP29]], [[COND_FALSE8]] ]
// CHECK9-NEXT:    store i32 [[COND10]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    store i32 [[TMP30]], i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK9:       omp.inner.for.end:
// CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK9:       omp.loop.exit:
// CHECK9-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]])
// CHECK9-NEXT:    ret void
// CHECK9-LABEL: define {{[^@]+}}@__omp_outlined__7
// CHECK9-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i64 [[F:%.*]]) #[[ATTR0]] {
// CHECK9-NEXT:  entry:
// CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK9-NEXT:    [[F_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK9-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK9-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK9-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK9-NEXT:    store i64 [[F]], i64* [[F_ADDR]], align 8
// CHECK9-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK9-NEXT:    [[CONV:%.*]] = bitcast i64* [[F_ADDR]] to i32*
// CHECK9-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK9-NEXT:    store i32 99, i32* [[DOTOMP_UB]], align 4
// CHECK9-NEXT:    [[TMP1:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK9-NEXT:    [[CONV2:%.*]] = trunc i64 [[TMP1]] to i32
// CHECK9-NEXT:    [[TMP2:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK9-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP2]] to i32
// CHECK9-NEXT:    store i32 [[CONV2]], i32* [[DOTOMP_LB]], align 4
// CHECK9-NEXT:    store i32 [[CONV3]], i32* [[DOTOMP_UB]], align 4
// CHECK9-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK9-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK9-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK9-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK9-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK9:       omp.inner.for.cond:
// CHECK9-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[CONV4:%.*]] = sext i32 [[TMP6]] to i64
// CHECK9-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK9-NEXT:    [[CMP:%.*]] = icmp ule i64 [[CONV4]], [[TMP7]]
// CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK9:       omp.inner.for.body:
// CHECK9-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP8]], 10
// CHECK9-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK9-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK9-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[DIV5:%.*]] = sdiv i32 [[TMP10]], 10
// CHECK9-NEXT:    [[MUL6:%.*]] = mul nsw i32 [[DIV5]], 10
// CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP9]], [[MUL6]]
// CHECK9-NEXT:    [[MUL7:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK9-NEXT:    [[ADD8:%.*]] = add nsw i32 0, [[MUL7]]
// CHECK9-NEXT:    store i32 [[ADD8]], i32* [[J]], align 4
// CHECK9-NEXT:    store i32 10, i32* [[K]], align 4
// CHECK9-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
// CHECK9-NEXT:    [[TMP12:%.*]] = load i32, i32* [[J]], align 4
// CHECK9-NEXT:    [[TMP13:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK9-NEXT:    [[MUL9:%.*]] = mul nsw i32 [[TMP12]], [[TMP13]]
// CHECK9-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP11]], [[MUL9]]
// CHECK9-NEXT:    [[TMP14:%.*]] = load i32, i32* [[K]], align 4
// CHECK9-NEXT:    [[ADD11:%.*]] = add nsw i32 [[ADD10]], [[TMP14]]
// CHECK9-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
// CHECK9-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP15]] to i64
// CHECK9-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK9-NEXT:    [[TMP16:%.*]] = load i32, i32* [[J]], align 4
// CHECK9-NEXT:    [[IDXPROM12:%.*]] = sext i32 [[TMP16]] to i64
// CHECK9-NEXT:    [[ARRAYIDX13:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i64 0, i64 [[IDXPROM12]]
// CHECK9-NEXT:    store i32 [[ADD11]], i32* [[ARRAYIDX13]], align 4
// CHECK9-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK9:       omp.body.continue:
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK9:       omp.inner.for.inc:
// CHECK9-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    [[ADD14:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK9-NEXT:    store i32 [[ADD14]], i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK9:       omp.inner.for.end:
// CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK9:       omp.loop.exit:
// CHECK9-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK9-NEXT:    ret void
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l67
// CHECK9-SAME: (i64 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK9-NEXT:  entry:
// CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK9-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK9-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK9-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK9-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK9-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK9-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK9-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK9-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK9:       .execute:
// CHECK9-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK9-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK9-NEXT:    [[CONV1:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK9-NEXT:    store i32 [[TMP2]], i32* [[CONV1]], align 4
// CHECK9-NEXT:    [[TMP3:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK9-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK9-NEXT:    call void @__omp_outlined__8(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP3]], [10 x [10 x i32]]* [[TMP0]]) #[[ATTR3]]
// CHECK9-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK9:       .omp.deinit:
// CHECK9-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK9-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK9:       .exit:
// CHECK9-NEXT:    ret void
// CHECK9-LABEL: define {{[^@]+}}@__omp_outlined__8
// CHECK9-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK9-NEXT:  entry:
// CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[I10:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[J11:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 8
// CHECK9-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK9-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK9-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK9-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK9-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK9-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK9-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK9-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK9-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK9-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK9-NEXT:    [[CONV4:%.*]] = sext i32 [[DIV]] to i64
// CHECK9-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK9-NEXT:    [[SUB5:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK9-NEXT:    [[DIV6:%.*]] = sdiv i32 [[SUB5]], 1
// CHECK9-NEXT:    [[CONV7:%.*]] = sext i32 [[DIV6]] to i64
// CHECK9-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV4]], [[CONV7]]
// CHECK9-NEXT:    [[SUB8:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK9-NEXT:    store i64 [[SUB8]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK9-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK9-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK9-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK9-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK9:       land.lhs.true:
// CHECK9-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK9-NEXT:    [[CMP9:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK9-NEXT:    br i1 [[CMP9]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK9:       omp.precond.then:
// CHECK9-NEXT:    store i64 0, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK9-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK9-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK9-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK9-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK9-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK9-NEXT:    [[CONV12:%.*]] = zext i32 [[NVPTX_NUM_THREADS]] to i64
// CHECK9-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    [[TMP9:%.*]] = load i32, i32* [[TMP8]], align 4
// CHECK9-NEXT:    call void @__kmpc_for_static_init_8(%struct.ident_t* @[[GLOB1]], i32 [[TMP9]], i32 91, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_COMB_LB]], i64* [[DOTOMP_COMB_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 [[CONV12]])
// CHECK9-NEXT:    [[TMP10:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK9-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK9-NEXT:    [[CMP13:%.*]] = icmp sgt i64 [[TMP10]], [[TMP11]]
// CHECK9-NEXT:    br i1 [[CMP13]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK9:       cond.true:
// CHECK9-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK9-NEXT:    br label [[COND_END:%.*]]
// CHECK9:       cond.false:
// CHECK9-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK9-NEXT:    br label [[COND_END]]
// CHECK9:       cond.end:
// CHECK9-NEXT:    [[COND:%.*]] = phi i64 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
// CHECK9-NEXT:    store i64 [[COND]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK9-NEXT:    [[TMP14:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK9-NEXT:    store i64 [[TMP14]], i64* [[DOTOMP_IV]], align 8
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK9:       omp.inner.for.cond:
// CHECK9-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK9-NEXT:    [[TMP16:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK9-NEXT:    [[ADD:%.*]] = add nsw i64 [[TMP16]], 1
// CHECK9-NEXT:    [[CMP14:%.*]] = icmp slt i64 [[TMP15]], [[ADD]]
// CHECK9-NEXT:    br i1 [[CMP14]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK9:       omp.inner.for.body:
// CHECK9-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK9-NEXT:    [[TMP18:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK9-NEXT:    [[TMP19:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK9-NEXT:    [[CONV15:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK9-NEXT:    store i32 [[TMP19]], i32* [[CONV15]], align 4
// CHECK9-NEXT:    [[TMP20:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK9-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK9-NEXT:    [[TMP22:%.*]] = inttoptr i64 [[TMP17]] to i8*
// CHECK9-NEXT:    store i8* [[TMP22]], i8** [[TMP21]], align 8
// CHECK9-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK9-NEXT:    [[TMP24:%.*]] = inttoptr i64 [[TMP18]] to i8*
// CHECK9-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 8
// CHECK9-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK9-NEXT:    [[TMP26:%.*]] = inttoptr i64 [[TMP20]] to i8*
// CHECK9-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 8
// CHECK9-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK9-NEXT:    [[TMP28:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK9-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 8
// CHECK9-NEXT:    [[TMP29:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    [[TMP30:%.*]] = load i32, i32* [[TMP29]], align 4
// CHECK9-NEXT:    [[TMP31:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK9-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP30]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, i64, [10 x [10 x i32]]*)* @__omp_outlined__9 to i8*), i8* null, i8** [[TMP31]], i64 4)
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK9:       omp.inner.for.inc:
// CHECK9-NEXT:    [[TMP32:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK9-NEXT:    [[TMP33:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK9-NEXT:    [[ADD16:%.*]] = add nsw i64 [[TMP32]], [[TMP33]]
// CHECK9-NEXT:    store i64 [[ADD16]], i64* [[DOTOMP_IV]], align 8
// CHECK9-NEXT:    [[TMP34:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK9-NEXT:    [[TMP35:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK9-NEXT:    [[ADD17:%.*]] = add nsw i64 [[TMP34]], [[TMP35]]
// CHECK9-NEXT:    store i64 [[ADD17]], i64* [[DOTOMP_COMB_LB]], align 8
// CHECK9-NEXT:    [[TMP36:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK9-NEXT:    [[TMP37:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK9-NEXT:    [[ADD18:%.*]] = add nsw i64 [[TMP36]], [[TMP37]]
// CHECK9-NEXT:    store i64 [[ADD18]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK9-NEXT:    [[TMP38:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK9-NEXT:    [[TMP39:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK9-NEXT:    [[CMP19:%.*]] = icmp sgt i64 [[TMP38]], [[TMP39]]
// CHECK9-NEXT:    br i1 [[CMP19]], label [[COND_TRUE20:%.*]], label [[COND_FALSE21:%.*]]
// CHECK9:       cond.true20:
// CHECK9-NEXT:    [[TMP40:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK9-NEXT:    br label [[COND_END22:%.*]]
// CHECK9:       cond.false21:
// CHECK9-NEXT:    [[TMP41:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK9-NEXT:    br label [[COND_END22]]
// CHECK9:       cond.end22:
// CHECK9-NEXT:    [[COND23:%.*]] = phi i64 [ [[TMP40]], [[COND_TRUE20]] ], [ [[TMP41]], [[COND_FALSE21]] ]
// CHECK9-NEXT:    store i64 [[COND23]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK9-NEXT:    [[TMP42:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK9-NEXT:    store i64 [[TMP42]], i64* [[DOTOMP_IV]], align 8
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK9:       omp.inner.for.end:
// CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK9:       omp.loop.exit:
// CHECK9-NEXT:    [[TMP43:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    [[TMP44:%.*]] = load i32, i32* [[TMP43]], align 4
// CHECK9-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP44]])
// CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK9:       omp.precond.end:
// CHECK9-NEXT:    ret void
// CHECK9-LABEL: define {{[^@]+}}@__omp_outlined__9
// CHECK9-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], i64 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK9-NEXT:  entry:
// CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_LB:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[DOTOMP_UB:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[I10:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[J11:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK9-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK9-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK9-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK9-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK9-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK9-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK9-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK9-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK9-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK9-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK9-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK9-NEXT:    [[CONV4:%.*]] = sext i32 [[DIV]] to i64
// CHECK9-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK9-NEXT:    [[SUB5:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK9-NEXT:    [[DIV6:%.*]] = sdiv i32 [[SUB5]], 1
// CHECK9-NEXT:    [[CONV7:%.*]] = sext i32 [[DIV6]] to i64
// CHECK9-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV4]], [[CONV7]]
// CHECK9-NEXT:    [[SUB8:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK9-NEXT:    store i64 [[SUB8]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK9-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK9-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK9-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK9-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK9:       land.lhs.true:
// CHECK9-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK9-NEXT:    [[CMP9:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK9-NEXT:    br i1 [[CMP9]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK9:       omp.precond.then:
// CHECK9-NEXT:    store i64 0, i64* [[DOTOMP_LB]], align 8
// CHECK9-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK9-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_UB]], align 8
// CHECK9-NEXT:    [[TMP8:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK9-NEXT:    [[TMP9:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK9-NEXT:    store i64 [[TMP8]], i64* [[DOTOMP_LB]], align 8
// CHECK9-NEXT:    store i64 [[TMP9]], i64* [[DOTOMP_UB]], align 8
// CHECK9-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK9-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK9-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK9-NEXT:    call void @__kmpc_for_static_init_8(%struct.ident_t* @[[GLOB2]], i32 [[TMP11]], i32 33, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_LB]], i64* [[DOTOMP_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 1)
// CHECK9-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTOMP_LB]], align 8
// CHECK9-NEXT:    store i64 [[TMP12]], i64* [[DOTOMP_IV]], align 8
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK9:       omp.inner.for.cond:
// CHECK9-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK9-NEXT:    [[TMP14:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK9-NEXT:    [[CMP12:%.*]] = icmp ule i64 [[TMP13]], [[TMP14]]
// CHECK9-NEXT:    br i1 [[CMP12]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK9:       omp.inner.for.body:
// CHECK9-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK9-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK9-NEXT:    [[SUB13:%.*]] = sub nsw i32 [[TMP16]], 0
// CHECK9-NEXT:    [[DIV14:%.*]] = sdiv i32 [[SUB13]], 1
// CHECK9-NEXT:    [[MUL15:%.*]] = mul nsw i32 1, [[DIV14]]
// CHECK9-NEXT:    [[CONV16:%.*]] = sext i32 [[MUL15]] to i64
// CHECK9-NEXT:    [[DIV17:%.*]] = sdiv i64 [[TMP15]], [[CONV16]]
// CHECK9-NEXT:    [[MUL18:%.*]] = mul nsw i64 [[DIV17]], 1
// CHECK9-NEXT:    [[ADD:%.*]] = add nsw i64 0, [[MUL18]]
// CHECK9-NEXT:    [[CONV19:%.*]] = trunc i64 [[ADD]] to i32
// CHECK9-NEXT:    store i32 [[CONV19]], i32* [[I10]], align 4
// CHECK9-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK9-NEXT:    [[TMP18:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK9-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK9-NEXT:    [[SUB20:%.*]] = sub nsw i32 [[TMP19]], 0
// CHECK9-NEXT:    [[DIV21:%.*]] = sdiv i32 [[SUB20]], 1
// CHECK9-NEXT:    [[MUL22:%.*]] = mul nsw i32 1, [[DIV21]]
// CHECK9-NEXT:    [[CONV23:%.*]] = sext i32 [[MUL22]] to i64
// CHECK9-NEXT:    [[DIV24:%.*]] = sdiv i64 [[TMP18]], [[CONV23]]
// CHECK9-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK9-NEXT:    [[SUB25:%.*]] = sub nsw i32 [[TMP20]], 0
// CHECK9-NEXT:    [[DIV26:%.*]] = sdiv i32 [[SUB25]], 1
// CHECK9-NEXT:    [[MUL27:%.*]] = mul nsw i32 1, [[DIV26]]
// CHECK9-NEXT:    [[CONV28:%.*]] = sext i32 [[MUL27]] to i64
// CHECK9-NEXT:    [[MUL29:%.*]] = mul nsw i64 [[DIV24]], [[CONV28]]
// CHECK9-NEXT:    [[SUB30:%.*]] = sub nsw i64 [[TMP17]], [[MUL29]]
// CHECK9-NEXT:    [[MUL31:%.*]] = mul nsw i64 [[SUB30]], 1
// CHECK9-NEXT:    [[ADD32:%.*]] = add nsw i64 0, [[MUL31]]
// CHECK9-NEXT:    [[CONV33:%.*]] = trunc i64 [[ADD32]] to i32
// CHECK9-NEXT:    store i32 [[CONV33]], i32* [[J11]], align 4
// CHECK9-NEXT:    [[TMP21:%.*]] = load i32, i32* [[I10]], align 4
// CHECK9-NEXT:    [[TMP22:%.*]] = load i32, i32* [[J11]], align 4
// CHECK9-NEXT:    [[ADD34:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
// CHECK9-NEXT:    [[TMP23:%.*]] = load i32, i32* [[I10]], align 4
// CHECK9-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP23]] to i64
// CHECK9-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK9-NEXT:    [[TMP24:%.*]] = load i32, i32* [[J11]], align 4
// CHECK9-NEXT:    [[IDXPROM35:%.*]] = sext i32 [[TMP24]] to i64
// CHECK9-NEXT:    [[ARRAYIDX36:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i64 0, i64 [[IDXPROM35]]
// CHECK9-NEXT:    store i32 [[ADD34]], i32* [[ARRAYIDX36]], align 4
// CHECK9-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK9:       omp.body.continue:
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK9:       omp.inner.for.inc:
// CHECK9-NEXT:    [[TMP25:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK9-NEXT:    [[TMP26:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK9-NEXT:    [[ADD37:%.*]] = add nsw i64 [[TMP25]], [[TMP26]]
// CHECK9-NEXT:    store i64 [[ADD37]], i64* [[DOTOMP_IV]], align 8
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK9:       omp.inner.for.end:
// CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK9:       omp.loop.exit:
// CHECK9-NEXT:    [[TMP27:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    [[TMP28:%.*]] = load i32, i32* [[TMP27]], align 4
// CHECK9-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP28]])
// CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK9:       omp.precond.end:
// CHECK9-NEXT:    ret void
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l74
// CHECK9-SAME: (i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK9-NEXT:  entry:
// CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK9-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK9-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK9-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK9-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 8
// CHECK9-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK9-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK9-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK9-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK9-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK9:       .execute:
// CHECK9-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK9-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK9-NEXT:    [[CONV1:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK9-NEXT:    store i32 [[TMP2]], i32* [[CONV1]], align 4
// CHECK9-NEXT:    [[TMP3:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK9-NEXT:    [[TMP4:%.*]] = load i32*, i32** [[V_ADDR]], align 8
// CHECK9-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK9-NEXT:    call void @__omp_outlined__10(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP3]], [1000 x i32]* [[TMP0]], i32* [[TMP4]]) #[[ATTR3]]
// CHECK9-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK9:       .omp.deinit:
// CHECK9-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK9-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK9:       .exit:
// CHECK9-NEXT:    ret void
// CHECK9-LABEL: define {{[^@]+}}@__omp_outlined__10
// CHECK9-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK9-NEXT:  entry:
// CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK9-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 8
// CHECK9-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK9-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK9-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK9-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 8
// CHECK9-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK9-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK9-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK9-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK9-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK9-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK9-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK9-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK9:       omp.precond.then:
// CHECK9-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK9-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK9-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK9-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK9-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK9-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK9-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK9-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK9:       cond.true:
// CHECK9-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK9-NEXT:    br label [[COND_END:%.*]]
// CHECK9:       cond.false:
// CHECK9-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    br label [[COND_END]]
// CHECK9:       cond.end:
// CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK9-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK9:       omp.inner.for.cond:
// CHECK9-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK9-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK9-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK9:       omp.inner.for.body:
// CHECK9-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    [[TMP15:%.*]] = zext i32 [[TMP14]] to i64
// CHECK9-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP17:%.*]] = zext i32 [[TMP16]] to i64
// CHECK9-NEXT:    [[TMP18:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK9-NEXT:    [[CONV6:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK9-NEXT:    store i32 [[TMP18]], i32* [[CONV6]], align 4
// CHECK9-NEXT:    [[TMP19:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK9-NEXT:    [[TMP20:%.*]] = load i32*, i32** [[V_ADDR]], align 8
// CHECK9-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK9-NEXT:    [[TMP22:%.*]] = inttoptr i64 [[TMP15]] to i8*
// CHECK9-NEXT:    store i8* [[TMP22]], i8** [[TMP21]], align 8
// CHECK9-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK9-NEXT:    [[TMP24:%.*]] = inttoptr i64 [[TMP17]] to i8*
// CHECK9-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 8
// CHECK9-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK9-NEXT:    [[TMP26:%.*]] = inttoptr i64 [[TMP19]] to i8*
// CHECK9-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 8
// CHECK9-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK9-NEXT:    [[TMP28:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK9-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 8
// CHECK9-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 4
// CHECK9-NEXT:    [[TMP30:%.*]] = bitcast i32* [[TMP20]] to i8*
// CHECK9-NEXT:    store i8* [[TMP30]], i8** [[TMP29]], align 8
// CHECK9-NEXT:    [[TMP31:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    [[TMP32:%.*]] = load i32, i32* [[TMP31]], align 4
// CHECK9-NEXT:    [[TMP33:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK9-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP32]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, i64, [1000 x i32]*, i32*)* @__omp_outlined__11 to i8*), i8* null, i8** [[TMP33]], i64 5)
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK9:       omp.inner.for.inc:
// CHECK9-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP34]], [[TMP35]]
// CHECK9-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP36]], [[TMP37]]
// CHECK9-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP38]], [[TMP39]]
// CHECK9-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK9-NEXT:    [[CMP10:%.*]] = icmp sgt i32 [[TMP40]], [[TMP41]]
// CHECK9-NEXT:    br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]]
// CHECK9:       cond.true11:
// CHECK9-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK9-NEXT:    br label [[COND_END13:%.*]]
// CHECK9:       cond.false12:
// CHECK9-NEXT:    [[TMP43:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    br label [[COND_END13]]
// CHECK9:       cond.end13:
// CHECK9-NEXT:    [[COND14:%.*]] = phi i32 [ [[TMP42]], [[COND_TRUE11]] ], [ [[TMP43]], [[COND_FALSE12]] ]
// CHECK9-NEXT:    store i32 [[COND14]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK9-NEXT:    [[TMP44:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK9-NEXT:    store i32 [[TMP44]], i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK9:       omp.inner.for.end:
// CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK9:       omp.loop.exit:
// CHECK9-NEXT:    [[TMP45:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    [[TMP46:%.*]] = load i32, i32* [[TMP45]], align 4
// CHECK9-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP46]])
// CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK9:       omp.precond.end:
// CHECK9-NEXT:    ret void
// CHECK9-LABEL: define {{[^@]+}}@__omp_outlined__11
// CHECK9-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK9-NEXT:  entry:
// CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK9-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 8
// CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    [[I5:%.*]] = alloca i32, align 4
// CHECK9-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK9-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK9-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK9-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK9-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK9-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 8
// CHECK9-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK9-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK9-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK9-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK9-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK9-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK9-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK9-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK9:       omp.precond.then:
// CHECK9-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK9-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK9-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK9-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK9-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP5]] to i32
// CHECK9-NEXT:    [[TMP6:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK9-NEXT:    [[CONV4:%.*]] = trunc i64 [[TMP6]] to i32
// CHECK9-NEXT:    store i32 [[CONV3]], i32* [[DOTOMP_LB]], align 4
// CHECK9-NEXT:    store i32 [[CONV4]], i32* [[DOTOMP_UB]], align 4
// CHECK9-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK9-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK9-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK9-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK9-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK9:       omp.inner.for.cond:
// CHECK9-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[CONV6:%.*]] = sext i32 [[TMP10]] to i64
// CHECK9-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK9-NEXT:    [[CMP7:%.*]] = icmp ule i64 [[CONV6]], [[TMP11]]
// CHECK9-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK9:       omp.inner.for.body:
// CHECK9-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK9-NEXT:    store i32 [[ADD]], i32* [[I5]], align 4
// CHECK9-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[V_ADDR]], align 8
// CHECK9-NEXT:    [[TMP14:%.*]] = load i32, i32* [[I5]], align 4
// CHECK9-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP14]] to i64
// CHECK9-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[TMP13]], i64 [[IDXPROM]]
// CHECK9-NEXT:    [[TMP15:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK9-NEXT:    [[TMP16:%.*]] = load i32, i32* [[I5]], align 4
// CHECK9-NEXT:    [[IDXPROM8:%.*]] = sext i32 [[TMP16]] to i64
// CHECK9-NEXT:    [[ARRAYIDX9:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i64 0, i64 [[IDXPROM8]]
// CHECK9-NEXT:    store i32 [[TMP15]], i32* [[ARRAYIDX9]], align 4
// CHECK9-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK9:       omp.body.continue:
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK9:       omp.inner.for.inc:
// CHECK9-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK9-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK9-NEXT:    store i32 [[ADD10]], i32* [[DOTOMP_IV]], align 4
// CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK9:       omp.inner.for.end:
// CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK9:       omp.loop.exit:
// CHECK9-NEXT:    [[TMP19:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT:    [[TMP20:%.*]] = load i32, i32* [[TMP19]], align 4
// CHECK9-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP20]])
// CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK9:       omp.precond.end:
// CHECK9-NEXT:    ret void
// CHECK10-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l43
// CHECK10-SAME: (i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i64 [[L:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK10-NEXT:  entry:
// CHECK10-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK10-NEXT:    [[L_ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[L_CASTED:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK10-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK10-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK10-NEXT:    store i64 [[L]], i64* [[L_ADDR]], align 8
// CHECK10-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK10-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK10-NEXT:    [[CONV1:%.*]] = bitcast i64* [[L_ADDR]] to i32*
// CHECK10-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK10-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK10-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK10:       .execute:
// CHECK10-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3:[0-9]+]])
// CHECK10-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK10-NEXT:    [[CONV2:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK10-NEXT:    store i32 [[TMP2]], i32* [[CONV2]], align 4
// CHECK10-NEXT:    [[TMP3:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK10-NEXT:    [[TMP4:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK10-NEXT:    [[CONV3:%.*]] = bitcast i64* [[L_CASTED]] to i32*
// CHECK10-NEXT:    store i32 [[TMP4]], i32* [[CONV3]], align 4
// CHECK10-NEXT:    [[TMP5:%.*]] = load i64, i64* [[L_CASTED]], align 8
// CHECK10-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK10-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP3]], [1000 x i32]* [[TMP0]], i64 [[TMP5]]) #[[ATTR3:[0-9]+]]
// CHECK10-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK10:       .omp.deinit:
// CHECK10-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK10-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK10:       .exit:
// CHECK10-NEXT:    ret void
// CHECK10-LABEL: define {{[^@]+}}@__omp_outlined__
// CHECK10-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i64 [[L:%.*]]) #[[ATTR0]] {
// CHECK10-NEXT:  entry:
// CHECK10-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK10-NEXT:    [[L_ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[I5:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[L_CASTED:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 8
// CHECK10-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK10-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK10-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK10-NEXT:    store i64 [[L]], i64* [[L_ADDR]], align 8
// CHECK10-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK10-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK10-NEXT:    [[CONV1:%.*]] = bitcast i64* [[L_ADDR]] to i32*
// CHECK10-NEXT:    [[TMP1:%.*]] = call i8* @__kmpc_data_sharing_push_stack(i64 4, i16 1)
// CHECK10-NEXT:    [[TMP2:%.*]] = bitcast i8* [[TMP1]] to %struct._globalized_locals_ty*
// CHECK10-NEXT:    [[L2:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY:%.*]], %struct._globalized_locals_ty* [[TMP2]], i32 0, i32 0
// CHECK10-NEXT:    [[TMP3:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK10-NEXT:    store i32 [[TMP3]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK10-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK10-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK10-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK10-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK10-NEXT:    store i32 [[SUB4]], i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK10-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK10-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK10-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK10-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK10:       omp.precond.then:
// CHECK10-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK10-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK10-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK10-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1:[0-9]+]], i32 [[TMP8]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 128)
// CHECK10-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK10-NEXT:    [[CMP6:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
// CHECK10-NEXT:    br i1 [[CMP6]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK10:       cond.true:
// CHECK10-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK10-NEXT:    br label [[COND_END:%.*]]
// CHECK10:       cond.false:
// CHECK10-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    br label [[COND_END]]
// CHECK10:       cond.end:
// CHECK10-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP11]], [[COND_TRUE]] ], [ [[TMP12]], [[COND_FALSE]] ]
// CHECK10-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    store i32 [[TMP13]], i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK10:       omp.inner.for.cond:
// CHECK10-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK10-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP15]], 1
// CHECK10-NEXT:    [[CMP7:%.*]] = icmp slt i32 [[TMP14]], [[ADD]]
// CHECK10-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK10:       omp.inner.for.body:
// CHECK10-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    [[TMP17:%.*]] = zext i32 [[TMP16]] to i64
// CHECK10-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP19:%.*]] = zext i32 [[TMP18]] to i64
// CHECK10-NEXT:    [[TMP20:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK10-NEXT:    [[CONV8:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK10-NEXT:    store i32 [[TMP20]], i32* [[CONV8]], align 4
// CHECK10-NEXT:    [[TMP21:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK10-NEXT:    [[TMP22:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK10-NEXT:    [[CONV9:%.*]] = bitcast i64* [[L_CASTED]] to i32*
// CHECK10-NEXT:    store i32 [[TMP22]], i32* [[CONV9]], align 4
// CHECK10-NEXT:    [[TMP23:%.*]] = load i64, i64* [[L_CASTED]], align 8
// CHECK10-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK10-NEXT:    [[TMP25:%.*]] = inttoptr i64 [[TMP17]] to i8*
// CHECK10-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 8
// CHECK10-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK10-NEXT:    [[TMP27:%.*]] = inttoptr i64 [[TMP19]] to i8*
// CHECK10-NEXT:    store i8* [[TMP27]], i8** [[TMP26]], align 8
// CHECK10-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK10-NEXT:    [[TMP29:%.*]] = inttoptr i64 [[TMP21]] to i8*
// CHECK10-NEXT:    store i8* [[TMP29]], i8** [[TMP28]], align 8
// CHECK10-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK10-NEXT:    [[TMP31:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK10-NEXT:    store i8* [[TMP31]], i8** [[TMP30]], align 8
// CHECK10-NEXT:    [[TMP32:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 4
// CHECK10-NEXT:    [[TMP33:%.*]] = inttoptr i64 [[TMP23]] to i8*
// CHECK10-NEXT:    store i8* [[TMP33]], i8** [[TMP32]], align 8
// CHECK10-NEXT:    [[TMP34:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    [[TMP35:%.*]] = load i32, i32* [[TMP34]], align 4
// CHECK10-NEXT:    [[TMP36:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK10-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP35]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, i64, [1000 x i32]*, i64)* @__omp_outlined__1 to i8*), i8* null, i8** [[TMP36]], i64 5)
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK10:       omp.inner.for.inc:
// CHECK10-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP37]], [[TMP38]]
// CHECK10-NEXT:    store i32 [[ADD10]], i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP39]], [[TMP40]]
// CHECK10-NEXT:    store i32 [[ADD11]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP41]], [[TMP42]]
// CHECK10-NEXT:    store i32 [[ADD12]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP43:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP44:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK10-NEXT:    [[CMP13:%.*]] = icmp sgt i32 [[TMP43]], [[TMP44]]
// CHECK10-NEXT:    br i1 [[CMP13]], label [[COND_TRUE14:%.*]], label [[COND_FALSE15:%.*]]
// CHECK10:       cond.true14:
// CHECK10-NEXT:    [[TMP45:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK10-NEXT:    br label [[COND_END16:%.*]]
// CHECK10:       cond.false15:
// CHECK10-NEXT:    [[TMP46:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    br label [[COND_END16]]
// CHECK10:       cond.end16:
// CHECK10-NEXT:    [[COND17:%.*]] = phi i32 [ [[TMP45]], [[COND_TRUE14]] ], [ [[TMP46]], [[COND_FALSE15]] ]
// CHECK10-NEXT:    store i32 [[COND17]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP47:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    store i32 [[TMP47]], i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK10:       omp.inner.for.end:
// CHECK10-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK10:       omp.loop.exit:
// CHECK10-NEXT:    [[TMP48:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    [[TMP49:%.*]] = load i32, i32* [[TMP48]], align 4
// CHECK10-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP49]])
// CHECK10-NEXT:    [[TMP50:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK10-NEXT:    [[TMP51:%.*]] = icmp ne i32 [[TMP50]], 0
// CHECK10-NEXT:    br i1 [[TMP51]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK10:       .omp.lastprivate.then:
// CHECK10-NEXT:    [[TMP52:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK10-NEXT:    store i32 [[TMP52]], i32* [[CONV1]], align 8
// CHECK10-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK10:       .omp.lastprivate.done:
// CHECK10-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK10:       omp.precond.end:
// CHECK10-NEXT:    call void @__kmpc_data_sharing_pop_stack(i8* [[TMP1]])
// CHECK10-NEXT:    ret void
// CHECK10-LABEL: define {{[^@]+}}@__omp_outlined__1
// CHECK10-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i64 [[L:%.*]]) #[[ATTR0]] {
// CHECK10-NEXT:  entry:
// CHECK10-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK10-NEXT:    [[L_ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[I6:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK10-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK10-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK10-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK10-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK10-NEXT:    store i64 [[L]], i64* [[L_ADDR]], align 8
// CHECK10-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK10-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK10-NEXT:    [[CONV1:%.*]] = bitcast i64* [[L_ADDR]] to i32*
// CHECK10-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK10-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK10-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK10-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK10-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK10-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK10-NEXT:    store i32 [[SUB3]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK10-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK10-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK10-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK10-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK10:       omp.precond.then:
// CHECK10-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK10-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK10-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK10-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK10-NEXT:    [[CONV4:%.*]] = trunc i64 [[TMP5]] to i32
// CHECK10-NEXT:    [[TMP6:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK10-NEXT:    [[CONV5:%.*]] = trunc i64 [[TMP6]] to i32
// CHECK10-NEXT:    store i32 [[CONV4]], i32* [[DOTOMP_LB]], align 4
// CHECK10-NEXT:    store i32 [[CONV5]], i32* [[DOTOMP_UB]], align 4
// CHECK10-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK10-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK10-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 32)
// CHECK10-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK10:       omp.dispatch.cond:
// CHECK10-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK10-NEXT:    [[CONV7:%.*]] = sext i32 [[TMP9]] to i64
// CHECK10-NEXT:    [[TMP10:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK10-NEXT:    [[CMP8:%.*]] = icmp ugt i64 [[CONV7]], [[TMP10]]
// CHECK10-NEXT:    br i1 [[CMP8]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK10:       cond.true:
// CHECK10-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK10-NEXT:    br label [[COND_END:%.*]]
// CHECK10:       cond.false:
// CHECK10-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK10-NEXT:    [[CONV9:%.*]] = sext i32 [[TMP12]] to i64
// CHECK10-NEXT:    br label [[COND_END]]
// CHECK10:       cond.end:
// CHECK10-NEXT:    [[COND:%.*]] = phi i64 [ [[TMP11]], [[COND_TRUE]] ], [ [[CONV9]], [[COND_FALSE]] ]
// CHECK10-NEXT:    [[CONV10:%.*]] = trunc i64 [[COND]] to i32
// CHECK10-NEXT:    store i32 [[CONV10]], i32* [[DOTOMP_UB]], align 4
// CHECK10-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK10-NEXT:    store i32 [[TMP13]], i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK10-NEXT:    [[CMP11:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK10-NEXT:    br i1 [[CMP11]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK10:       omp.dispatch.body:
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK10:       omp.inner.for.cond:
// CHECK10-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK10-NEXT:    [[CMP12:%.*]] = icmp sle i32 [[TMP16]], [[TMP17]]
// CHECK10-NEXT:    br i1 [[CMP12]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK10:       omp.inner.for.body:
// CHECK10-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP18]], 1
// CHECK10-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK10-NEXT:    store i32 [[ADD]], i32* [[I6]], align 4
// CHECK10-NEXT:    [[TMP19:%.*]] = load i32, i32* [[I6]], align 4
// CHECK10-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP19]] to i64
// CHECK10-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK10-NEXT:    store i32 1, i32* [[ARRAYIDX]], align 4
// CHECK10-NEXT:    [[TMP20:%.*]] = load i32, i32* [[I6]], align 4
// CHECK10-NEXT:    store i32 [[TMP20]], i32* [[CONV1]], align 8
// CHECK10-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK10:       omp.body.continue:
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK10:       omp.inner.for.inc:
// CHECK10-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[ADD13:%.*]] = add nsw i32 [[TMP21]], 1
// CHECK10-NEXT:    store i32 [[ADD13]], i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK10:       omp.inner.for.end:
// CHECK10-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK10:       omp.dispatch.inc:
// CHECK10-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK10-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    [[ADD14:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK10-NEXT:    store i32 [[ADD14]], i32* [[DOTOMP_LB]], align 4
// CHECK10-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK10-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    [[ADD15:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK10-NEXT:    store i32 [[ADD15]], i32* [[DOTOMP_UB]], align 4
// CHECK10-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK10:       omp.dispatch.end:
// CHECK10-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK10-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP27]])
// CHECK10-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK10-NEXT:    [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
// CHECK10-NEXT:    br i1 [[TMP29]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK10:       .omp.lastprivate.then:
// CHECK10-NEXT:    [[TMP30:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK10-NEXT:    store i32 [[TMP30]], i32* [[CONV1]], align 8
// CHECK10-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK10:       .omp.lastprivate.done:
// CHECK10-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK10:       omp.precond.end:
// CHECK10-NEXT:    ret void
// CHECK10-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l49
// CHECK10-SAME: (i64 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK10-NEXT:  entry:
// CHECK10-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 8
// CHECK10-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK10-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK10-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 8
// CHECK10-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK10-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 8
// CHECK10-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK10-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK10-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK10:       .execute:
// CHECK10-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK10-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK10-NEXT:    [[CONV1:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK10-NEXT:    store i32 [[TMP2]], i32* [[CONV1]], align 4
// CHECK10-NEXT:    [[TMP3:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK10-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK10-NEXT:    call void @__omp_outlined__2(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP3]], [1000 x i16]* [[TMP0]]) #[[ATTR3]]
// CHECK10-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK10:       .omp.deinit:
// CHECK10-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK10-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK10:       .exit:
// CHECK10-NEXT:    ret void
// CHECK10-LABEL: define {{[^@]+}}@__omp_outlined__2
// CHECK10-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK10-NEXT:  entry:
// CHECK10-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 8
// CHECK10-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 8
// CHECK10-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK10-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK10-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 8
// CHECK10-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK10-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 8
// CHECK10-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK10-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK10-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK10-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK10-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK10-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK10-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK10-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK10-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK10-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK10-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK10:       omp.precond.then:
// CHECK10-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK10-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK10-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK10-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK10-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK10-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK10-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK10-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK10:       cond.true:
// CHECK10-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK10-NEXT:    br label [[COND_END:%.*]]
// CHECK10:       cond.false:
// CHECK10-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    br label [[COND_END]]
// CHECK10:       cond.end:
// CHECK10-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK10-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK10:       omp.inner.for.cond:
// CHECK10-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK10-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK10-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK10-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK10:       omp.inner.for.body:
// CHECK10-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    [[TMP15:%.*]] = zext i32 [[TMP14]] to i64
// CHECK10-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP17:%.*]] = zext i32 [[TMP16]] to i64
// CHECK10-NEXT:    [[TMP18:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK10-NEXT:    [[CONV6:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK10-NEXT:    store i32 [[TMP18]], i32* [[CONV6]], align 4
// CHECK10-NEXT:    [[TMP19:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK10-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK10-NEXT:    [[TMP21:%.*]] = inttoptr i64 [[TMP15]] to i8*
// CHECK10-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 8
// CHECK10-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK10-NEXT:    [[TMP23:%.*]] = inttoptr i64 [[TMP17]] to i8*
// CHECK10-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 8
// CHECK10-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK10-NEXT:    [[TMP25:%.*]] = inttoptr i64 [[TMP19]] to i8*
// CHECK10-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 8
// CHECK10-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK10-NEXT:    [[TMP27:%.*]] = bitcast [1000 x i16]* [[TMP0]] to i8*
// CHECK10-NEXT:    store i8* [[TMP27]], i8** [[TMP26]], align 8
// CHECK10-NEXT:    [[TMP28:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    [[TMP29:%.*]] = load i32, i32* [[TMP28]], align 4
// CHECK10-NEXT:    [[TMP30:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK10-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP29]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, i64, [1000 x i16]*)* @__omp_outlined__3 to i8*), i8* null, i8** [[TMP30]], i64 4)
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK10:       omp.inner.for.inc:
// CHECK10-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP31]], [[TMP32]]
// CHECK10-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP33]], [[TMP34]]
// CHECK10-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP35]], [[TMP36]]
// CHECK10-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK10-NEXT:    [[CMP10:%.*]] = icmp sgt i32 [[TMP37]], [[TMP38]]
// CHECK10-NEXT:    br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]]
// CHECK10:       cond.true11:
// CHECK10-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK10-NEXT:    br label [[COND_END13:%.*]]
// CHECK10:       cond.false12:
// CHECK10-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    br label [[COND_END13]]
// CHECK10:       cond.end13:
// CHECK10-NEXT:    [[COND14:%.*]] = phi i32 [ [[TMP39]], [[COND_TRUE11]] ], [ [[TMP40]], [[COND_FALSE12]] ]
// CHECK10-NEXT:    store i32 [[COND14]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    store i32 [[TMP41]], i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK10:       omp.inner.for.end:
// CHECK10-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK10:       omp.loop.exit:
// CHECK10-NEXT:    [[TMP42:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    [[TMP43:%.*]] = load i32, i32* [[TMP42]], align 4
// CHECK10-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP43]])
// CHECK10-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK10:       omp.precond.end:
// CHECK10-NEXT:    ret void
// CHECK10-LABEL: define {{[^@]+}}@__omp_outlined__3
// CHECK10-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], i64 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK10-NEXT:  entry:
// CHECK10-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 8
// CHECK10-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[I5:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK10-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK10-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK10-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK10-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 8
// CHECK10-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK10-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 8
// CHECK10-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK10-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK10-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK10-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK10-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK10-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK10-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK10-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK10-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK10-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK10-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK10:       omp.precond.then:
// CHECK10-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK10-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK10-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK10-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK10-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP5]] to i32
// CHECK10-NEXT:    [[TMP6:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK10-NEXT:    [[CONV4:%.*]] = trunc i64 [[TMP6]] to i32
// CHECK10-NEXT:    store i32 [[CONV3]], i32* [[DOTOMP_LB]], align 4
// CHECK10-NEXT:    store i32 [[CONV4]], i32* [[DOTOMP_UB]], align 4
// CHECK10-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK10-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK10-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK10-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK10-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK10:       omp.inner.for.cond:
// CHECK10-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[CONV6:%.*]] = sext i32 [[TMP10]] to i64
// CHECK10-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK10-NEXT:    [[CMP7:%.*]] = icmp ule i64 [[CONV6]], [[TMP11]]
// CHECK10-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK10:       omp.inner.for.body:
// CHECK10-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK10-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK10-NEXT:    store i32 [[ADD]], i32* [[I5]], align 4
// CHECK10-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I5]], align 4
// CHECK10-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP13]] to i64
// CHECK10-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i16], [1000 x i16]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK10-NEXT:    [[TMP14:%.*]] = load i16, i16* [[ARRAYIDX]], align 2
// CHECK10-NEXT:    [[CONV8:%.*]] = sext i16 [[TMP14]] to i32
// CHECK10-NEXT:    [[ADD9:%.*]] = add nsw i32 [[CONV8]], 1
// CHECK10-NEXT:    [[CONV10:%.*]] = trunc i32 [[ADD9]] to i16
// CHECK10-NEXT:    store i16 [[CONV10]], i16* [[ARRAYIDX]], align 2
// CHECK10-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK10:       omp.body.continue:
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK10:       omp.inner.for.inc:
// CHECK10-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP15]], [[TMP16]]
// CHECK10-NEXT:    store i32 [[ADD11]], i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK10:       omp.inner.for.end:
// CHECK10-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK10:       omp.loop.exit:
// CHECK10-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
// CHECK10-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP18]])
// CHECK10-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK10:       omp.precond.end:
// CHECK10-NEXT:    ret void
// CHECK10-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l54
// CHECK10-SAME: ([10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK10-NEXT:  entry:
// CHECK10-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 8
// CHECK10-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK10-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 8
// CHECK10-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 8
// CHECK10-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK10-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK10-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK10:       .execute:
// CHECK10-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK10-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK10-NEXT:    call void @__omp_outlined__4(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x i32]* [[TMP0]]) #[[ATTR3]]
// CHECK10-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK10:       .omp.deinit:
// CHECK10-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK10-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK10:       .exit:
// CHECK10-NEXT:    ret void
// CHECK10-LABEL: define {{[^@]+}}@__omp_outlined__4
// CHECK10-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK10-NEXT:  entry:
// CHECK10-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 8
// CHECK10-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x i8*], align 8
// CHECK10-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK10-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 8
// CHECK10-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 8
// CHECK10-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    store i32 9, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK10-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK10-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK10-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK10-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9
// CHECK10-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK10:       cond.true:
// CHECK10-NEXT:    br label [[COND_END:%.*]]
// CHECK10:       cond.false:
// CHECK10-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    br label [[COND_END]]
// CHECK10:       cond.end:
// CHECK10-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK10-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK10:       omp.inner.for.cond:
// CHECK10-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[CMP1:%.*]] = icmp slt i32 [[TMP6]], 10
// CHECK10-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK10:       omp.inner.for.body:
// CHECK10-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
// CHECK10-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
// CHECK10-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK10-NEXT:    [[TMP12:%.*]] = inttoptr i64 [[TMP8]] to i8*
// CHECK10-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 8
// CHECK10-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK10-NEXT:    [[TMP14:%.*]] = inttoptr i64 [[TMP10]] to i8*
// CHECK10-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 8
// CHECK10-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK10-NEXT:    [[TMP16:%.*]] = bitcast [10 x i32]* [[TMP0]] to i8*
// CHECK10-NEXT:    store i8* [[TMP16]], i8** [[TMP15]], align 8
// CHECK10-NEXT:    [[TMP17:%.*]] = bitcast [3 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK10-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, [10 x i32]*)* @__omp_outlined__5 to i8*), i8* null, i8** [[TMP17]], i64 3)
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK10:       omp.inner.for.inc:
// CHECK10-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP18]], [[TMP19]]
// CHECK10-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK10-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK10-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP24]], 9
// CHECK10-NEXT:    br i1 [[CMP4]], label [[COND_TRUE5:%.*]], label [[COND_FALSE6:%.*]]
// CHECK10:       cond.true5:
// CHECK10-NEXT:    br label [[COND_END7:%.*]]
// CHECK10:       cond.false6:
// CHECK10-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    br label [[COND_END7]]
// CHECK10:       cond.end7:
// CHECK10-NEXT:    [[COND8:%.*]] = phi i32 [ 9, [[COND_TRUE5]] ], [ [[TMP25]], [[COND_FALSE6]] ]
// CHECK10-NEXT:    store i32 [[COND8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    store i32 [[TMP26]], i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK10:       omp.inner.for.end:
// CHECK10-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK10:       omp.loop.exit:
// CHECK10-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]])
// CHECK10-NEXT:    ret void
// CHECK10-LABEL: define {{[^@]+}}@__omp_outlined__5
// CHECK10-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK10-NEXT:  entry:
// CHECK10-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 8
// CHECK10-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK10-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK10-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK10-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 8
// CHECK10-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 8
// CHECK10-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK10-NEXT:    store i32 9, i32* [[DOTOMP_UB]], align 4
// CHECK10-NEXT:    [[TMP1:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK10-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP1]] to i32
// CHECK10-NEXT:    [[TMP2:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK10-NEXT:    [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32
// CHECK10-NEXT:    store i32 [[CONV]], i32* [[DOTOMP_LB]], align 4
// CHECK10-NEXT:    store i32 [[CONV1]], i32* [[DOTOMP_UB]], align 4
// CHECK10-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK10-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK10-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK10-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK10-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK10:       omp.inner.for.cond:
// CHECK10-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[CONV2:%.*]] = sext i32 [[TMP6]] to i64
// CHECK10-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK10-NEXT:    [[CMP:%.*]] = icmp ule i64 [[CONV2]], [[TMP7]]
// CHECK10-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK10:       omp.inner.for.body:
// CHECK10-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP8]], 1
// CHECK10-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK10-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK10-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4
// CHECK10-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP9]] to i64
// CHECK10-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK10-NEXT:    [[TMP10:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK10-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP10]], 1
// CHECK10-NEXT:    store i32 [[ADD3]], i32* [[ARRAYIDX]], align 4
// CHECK10-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK10:       omp.body.continue:
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK10:       omp.inner.for.inc:
// CHECK10-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
// CHECK10-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK10:       omp.inner.for.end:
// CHECK10-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK10:       omp.loop.exit:
// CHECK10-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK10-NEXT:    ret void
// CHECK10-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l59
// CHECK10-SAME: ([10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i64 [[F:%.*]]) #[[ATTR0]] {
// CHECK10-NEXT:  entry:
// CHECK10-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK10-NEXT:    [[F_ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[F_CASTED:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK10-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK10-NEXT:    store i64 [[F]], i64* [[F_ADDR]], align 8
// CHECK10-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK10-NEXT:    [[CONV:%.*]] = bitcast i64* [[F_ADDR]] to i32*
// CHECK10-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK10-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK10-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK10:       .execute:
// CHECK10-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK10-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK10-NEXT:    [[CONV1:%.*]] = bitcast i64* [[F_CASTED]] to i32*
// CHECK10-NEXT:    store i32 [[TMP2]], i32* [[CONV1]], align 4
// CHECK10-NEXT:    [[TMP3:%.*]] = load i64, i64* [[F_CASTED]], align 8
// CHECK10-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK10-NEXT:    call void @__omp_outlined__6(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x [10 x i32]]* [[TMP0]], i64 [[TMP3]]) #[[ATTR3]]
// CHECK10-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK10:       .omp.deinit:
// CHECK10-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK10-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK10:       .exit:
// CHECK10-NEXT:    ret void
// CHECK10-LABEL: define {{[^@]+}}@__omp_outlined__6
// CHECK10-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i64 [[F:%.*]]) #[[ATTR0]] {
// CHECK10-NEXT:  entry:
// CHECK10-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK10-NEXT:    [[F_ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[F_CASTED:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 8
// CHECK10-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK10-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK10-NEXT:    store i64 [[F]], i64* [[F_ADDR]], align 8
// CHECK10-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK10-NEXT:    [[CONV:%.*]] = bitcast i64* [[F_ADDR]] to i32*
// CHECK10-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    store i32 99, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK10-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK10-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK10-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK10-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 99
// CHECK10-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK10:       cond.true:
// CHECK10-NEXT:    br label [[COND_END:%.*]]
// CHECK10:       cond.false:
// CHECK10-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    br label [[COND_END]]
// CHECK10:       cond.end:
// CHECK10-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK10-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK10:       omp.inner.for.cond:
// CHECK10-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[TMP6]], 100
// CHECK10-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK10:       omp.inner.for.body:
// CHECK10-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
// CHECK10-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
// CHECK10-NEXT:    [[TMP11:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK10-NEXT:    [[CONV3:%.*]] = bitcast i64* [[F_CASTED]] to i32*
// CHECK10-NEXT:    store i32 [[TMP11]], i32* [[CONV3]], align 4
// CHECK10-NEXT:    [[TMP12:%.*]] = load i64, i64* [[F_CASTED]], align 8
// CHECK10-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK10-NEXT:    [[TMP14:%.*]] = inttoptr i64 [[TMP8]] to i8*
// CHECK10-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 8
// CHECK10-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK10-NEXT:    [[TMP16:%.*]] = inttoptr i64 [[TMP10]] to i8*
// CHECK10-NEXT:    store i8* [[TMP16]], i8** [[TMP15]], align 8
// CHECK10-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK10-NEXT:    [[TMP18:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK10-NEXT:    store i8* [[TMP18]], i8** [[TMP17]], align 8
// CHECK10-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK10-NEXT:    [[TMP20:%.*]] = inttoptr i64 [[TMP12]] to i8*
// CHECK10-NEXT:    store i8* [[TMP20]], i8** [[TMP19]], align 8
// CHECK10-NEXT:    [[TMP21:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK10-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, [10 x [10 x i32]]*, i64)* @__omp_outlined__7 to i8*), i8* null, i8** [[TMP21]], i64 4)
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK10:       omp.inner.for.inc:
// CHECK10-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK10-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK10-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP26]], [[TMP27]]
// CHECK10-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[CMP6:%.*]] = icmp sgt i32 [[TMP28]], 99
// CHECK10-NEXT:    br i1 [[CMP6]], label [[COND_TRUE7:%.*]], label [[COND_FALSE8:%.*]]
// CHECK10:       cond.true7:
// CHECK10-NEXT:    br label [[COND_END9:%.*]]
// CHECK10:       cond.false8:
// CHECK10-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    br label [[COND_END9]]
// CHECK10:       cond.end9:
// CHECK10-NEXT:    [[COND10:%.*]] = phi i32 [ 99, [[COND_TRUE7]] ], [ [[TMP29]], [[COND_FALSE8]] ]
// CHECK10-NEXT:    store i32 [[COND10]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    store i32 [[TMP30]], i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK10:       omp.inner.for.end:
// CHECK10-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK10:       omp.loop.exit:
// CHECK10-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]])
// CHECK10-NEXT:    ret void
// CHECK10-LABEL: define {{[^@]+}}@__omp_outlined__7
// CHECK10-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i64 [[F:%.*]]) #[[ATTR0]] {
// CHECK10-NEXT:  entry:
// CHECK10-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK10-NEXT:    [[F_ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK10-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK10-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK10-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK10-NEXT:    store i64 [[F]], i64* [[F_ADDR]], align 8
// CHECK10-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK10-NEXT:    [[CONV:%.*]] = bitcast i64* [[F_ADDR]] to i32*
// CHECK10-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK10-NEXT:    store i32 99, i32* [[DOTOMP_UB]], align 4
// CHECK10-NEXT:    [[TMP1:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK10-NEXT:    [[CONV2:%.*]] = trunc i64 [[TMP1]] to i32
// CHECK10-NEXT:    [[TMP2:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK10-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP2]] to i32
// CHECK10-NEXT:    store i32 [[CONV2]], i32* [[DOTOMP_LB]], align 4
// CHECK10-NEXT:    store i32 [[CONV3]], i32* [[DOTOMP_UB]], align 4
// CHECK10-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK10-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK10-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK10-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK10-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK10:       omp.inner.for.cond:
// CHECK10-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[CONV4:%.*]] = sext i32 [[TMP6]] to i64
// CHECK10-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK10-NEXT:    [[CMP:%.*]] = icmp ule i64 [[CONV4]], [[TMP7]]
// CHECK10-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK10:       omp.inner.for.body:
// CHECK10-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP8]], 10
// CHECK10-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK10-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK10-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK10-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[DIV5:%.*]] = sdiv i32 [[TMP10]], 10
// CHECK10-NEXT:    [[MUL6:%.*]] = mul nsw i32 [[DIV5]], 10
// CHECK10-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP9]], [[MUL6]]
// CHECK10-NEXT:    [[MUL7:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK10-NEXT:    [[ADD8:%.*]] = add nsw i32 0, [[MUL7]]
// CHECK10-NEXT:    store i32 [[ADD8]], i32* [[J]], align 4
// CHECK10-NEXT:    store i32 10, i32* [[K]], align 4
// CHECK10-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
// CHECK10-NEXT:    [[TMP12:%.*]] = load i32, i32* [[J]], align 4
// CHECK10-NEXT:    [[TMP13:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK10-NEXT:    [[MUL9:%.*]] = mul nsw i32 [[TMP12]], [[TMP13]]
// CHECK10-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP11]], [[MUL9]]
// CHECK10-NEXT:    [[TMP14:%.*]] = load i32, i32* [[K]], align 4
// CHECK10-NEXT:    [[ADD11:%.*]] = add nsw i32 [[ADD10]], [[TMP14]]
// CHECK10-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
// CHECK10-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP15]] to i64
// CHECK10-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK10-NEXT:    [[TMP16:%.*]] = load i32, i32* [[J]], align 4
// CHECK10-NEXT:    [[IDXPROM12:%.*]] = sext i32 [[TMP16]] to i64
// CHECK10-NEXT:    [[ARRAYIDX13:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i64 0, i64 [[IDXPROM12]]
// CHECK10-NEXT:    store i32 [[ADD11]], i32* [[ARRAYIDX13]], align 4
// CHECK10-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK10:       omp.body.continue:
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK10:       omp.inner.for.inc:
// CHECK10-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    [[ADD14:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK10-NEXT:    store i32 [[ADD14]], i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK10:       omp.inner.for.end:
// CHECK10-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK10:       omp.loop.exit:
// CHECK10-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK10-NEXT:    ret void
// CHECK10-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l67
// CHECK10-SAME: (i64 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK10-NEXT:  entry:
// CHECK10-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK10-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK10-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK10-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK10-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK10-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK10-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK10-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK10-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK10:       .execute:
// CHECK10-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK10-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK10-NEXT:    [[CONV1:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK10-NEXT:    store i32 [[TMP2]], i32* [[CONV1]], align 4
// CHECK10-NEXT:    [[TMP3:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK10-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK10-NEXT:    call void @__omp_outlined__8(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP3]], [10 x [10 x i32]]* [[TMP0]]) #[[ATTR3]]
// CHECK10-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK10:       .omp.deinit:
// CHECK10-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK10-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK10:       .exit:
// CHECK10-NEXT:    ret void
// CHECK10-LABEL: define {{[^@]+}}@__omp_outlined__8
// CHECK10-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK10-NEXT:  entry:
// CHECK10-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK10-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[I10:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[J11:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 8
// CHECK10-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK10-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK10-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK10-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK10-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK10-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK10-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK10-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK10-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK10-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK10-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK10-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK10-NEXT:    [[CONV4:%.*]] = sext i32 [[DIV]] to i64
// CHECK10-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK10-NEXT:    [[SUB5:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK10-NEXT:    [[DIV6:%.*]] = sdiv i32 [[SUB5]], 1
// CHECK10-NEXT:    [[CONV7:%.*]] = sext i32 [[DIV6]] to i64
// CHECK10-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV4]], [[CONV7]]
// CHECK10-NEXT:    [[SUB8:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK10-NEXT:    store i64 [[SUB8]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK10-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK10-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK10-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK10-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK10-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK10:       land.lhs.true:
// CHECK10-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK10-NEXT:    [[CMP9:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK10-NEXT:    br i1 [[CMP9]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK10:       omp.precond.then:
// CHECK10-NEXT:    store i64 0, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK10-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK10-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK10-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK10-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK10-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK10-NEXT:    [[CONV12:%.*]] = zext i32 [[NVPTX_NUM_THREADS]] to i64
// CHECK10-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    [[TMP9:%.*]] = load i32, i32* [[TMP8]], align 4
// CHECK10-NEXT:    call void @__kmpc_for_static_init_8(%struct.ident_t* @[[GLOB1]], i32 [[TMP9]], i32 91, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_COMB_LB]], i64* [[DOTOMP_COMB_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 [[CONV12]])
// CHECK10-NEXT:    [[TMP10:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK10-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK10-NEXT:    [[CMP13:%.*]] = icmp sgt i64 [[TMP10]], [[TMP11]]
// CHECK10-NEXT:    br i1 [[CMP13]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK10:       cond.true:
// CHECK10-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK10-NEXT:    br label [[COND_END:%.*]]
// CHECK10:       cond.false:
// CHECK10-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK10-NEXT:    br label [[COND_END]]
// CHECK10:       cond.end:
// CHECK10-NEXT:    [[COND:%.*]] = phi i64 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
// CHECK10-NEXT:    store i64 [[COND]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK10-NEXT:    [[TMP14:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK10-NEXT:    store i64 [[TMP14]], i64* [[DOTOMP_IV]], align 8
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK10:       omp.inner.for.cond:
// CHECK10-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK10-NEXT:    [[TMP16:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK10-NEXT:    [[ADD:%.*]] = add nsw i64 [[TMP16]], 1
// CHECK10-NEXT:    [[CMP14:%.*]] = icmp slt i64 [[TMP15]], [[ADD]]
// CHECK10-NEXT:    br i1 [[CMP14]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK10:       omp.inner.for.body:
// CHECK10-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK10-NEXT:    [[TMP18:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK10-NEXT:    [[TMP19:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK10-NEXT:    [[CONV15:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK10-NEXT:    store i32 [[TMP19]], i32* [[CONV15]], align 4
// CHECK10-NEXT:    [[TMP20:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK10-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK10-NEXT:    [[TMP22:%.*]] = inttoptr i64 [[TMP17]] to i8*
// CHECK10-NEXT:    store i8* [[TMP22]], i8** [[TMP21]], align 8
// CHECK10-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK10-NEXT:    [[TMP24:%.*]] = inttoptr i64 [[TMP18]] to i8*
// CHECK10-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 8
// CHECK10-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK10-NEXT:    [[TMP26:%.*]] = inttoptr i64 [[TMP20]] to i8*
// CHECK10-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 8
// CHECK10-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK10-NEXT:    [[TMP28:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK10-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 8
// CHECK10-NEXT:    [[TMP29:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    [[TMP30:%.*]] = load i32, i32* [[TMP29]], align 4
// CHECK10-NEXT:    [[TMP31:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK10-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP30]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, i64, [10 x [10 x i32]]*)* @__omp_outlined__9 to i8*), i8* null, i8** [[TMP31]], i64 4)
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK10:       omp.inner.for.inc:
// CHECK10-NEXT:    [[TMP32:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK10-NEXT:    [[TMP33:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK10-NEXT:    [[ADD16:%.*]] = add nsw i64 [[TMP32]], [[TMP33]]
// CHECK10-NEXT:    store i64 [[ADD16]], i64* [[DOTOMP_IV]], align 8
// CHECK10-NEXT:    [[TMP34:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK10-NEXT:    [[TMP35:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK10-NEXT:    [[ADD17:%.*]] = add nsw i64 [[TMP34]], [[TMP35]]
// CHECK10-NEXT:    store i64 [[ADD17]], i64* [[DOTOMP_COMB_LB]], align 8
// CHECK10-NEXT:    [[TMP36:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK10-NEXT:    [[TMP37:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK10-NEXT:    [[ADD18:%.*]] = add nsw i64 [[TMP36]], [[TMP37]]
// CHECK10-NEXT:    store i64 [[ADD18]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK10-NEXT:    [[TMP38:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK10-NEXT:    [[TMP39:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK10-NEXT:    [[CMP19:%.*]] = icmp sgt i64 [[TMP38]], [[TMP39]]
// CHECK10-NEXT:    br i1 [[CMP19]], label [[COND_TRUE20:%.*]], label [[COND_FALSE21:%.*]]
// CHECK10:       cond.true20:
// CHECK10-NEXT:    [[TMP40:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK10-NEXT:    br label [[COND_END22:%.*]]
// CHECK10:       cond.false21:
// CHECK10-NEXT:    [[TMP41:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK10-NEXT:    br label [[COND_END22]]
// CHECK10:       cond.end22:
// CHECK10-NEXT:    [[COND23:%.*]] = phi i64 [ [[TMP40]], [[COND_TRUE20]] ], [ [[TMP41]], [[COND_FALSE21]] ]
// CHECK10-NEXT:    store i64 [[COND23]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK10-NEXT:    [[TMP42:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK10-NEXT:    store i64 [[TMP42]], i64* [[DOTOMP_IV]], align 8
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK10:       omp.inner.for.end:
// CHECK10-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK10:       omp.loop.exit:
// CHECK10-NEXT:    [[TMP43:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    [[TMP44:%.*]] = load i32, i32* [[TMP43]], align 4
// CHECK10-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP44]])
// CHECK10-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK10:       omp.precond.end:
// CHECK10-NEXT:    ret void
// CHECK10-LABEL: define {{[^@]+}}@__omp_outlined__9
// CHECK10-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], i64 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK10-NEXT:  entry:
// CHECK10-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK10-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_LB:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[DOTOMP_UB:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[I10:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[J11:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK10-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK10-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK10-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK10-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK10-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK10-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK10-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK10-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK10-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK10-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK10-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK10-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK10-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK10-NEXT:    [[CONV4:%.*]] = sext i32 [[DIV]] to i64
// CHECK10-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK10-NEXT:    [[SUB5:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK10-NEXT:    [[DIV6:%.*]] = sdiv i32 [[SUB5]], 1
// CHECK10-NEXT:    [[CONV7:%.*]] = sext i32 [[DIV6]] to i64
// CHECK10-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV4]], [[CONV7]]
// CHECK10-NEXT:    [[SUB8:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK10-NEXT:    store i64 [[SUB8]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK10-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK10-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK10-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK10-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK10-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK10:       land.lhs.true:
// CHECK10-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK10-NEXT:    [[CMP9:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK10-NEXT:    br i1 [[CMP9]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK10:       omp.precond.then:
// CHECK10-NEXT:    store i64 0, i64* [[DOTOMP_LB]], align 8
// CHECK10-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK10-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_UB]], align 8
// CHECK10-NEXT:    [[TMP8:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK10-NEXT:    [[TMP9:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK10-NEXT:    store i64 [[TMP8]], i64* [[DOTOMP_LB]], align 8
// CHECK10-NEXT:    store i64 [[TMP9]], i64* [[DOTOMP_UB]], align 8
// CHECK10-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK10-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK10-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK10-NEXT:    call void @__kmpc_for_static_init_8(%struct.ident_t* @[[GLOB2]], i32 [[TMP11]], i32 33, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_LB]], i64* [[DOTOMP_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 1)
// CHECK10-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTOMP_LB]], align 8
// CHECK10-NEXT:    store i64 [[TMP12]], i64* [[DOTOMP_IV]], align 8
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK10:       omp.inner.for.cond:
// CHECK10-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK10-NEXT:    [[TMP14:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK10-NEXT:    [[CMP12:%.*]] = icmp ule i64 [[TMP13]], [[TMP14]]
// CHECK10-NEXT:    br i1 [[CMP12]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK10:       omp.inner.for.body:
// CHECK10-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK10-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK10-NEXT:    [[SUB13:%.*]] = sub nsw i32 [[TMP16]], 0
// CHECK10-NEXT:    [[DIV14:%.*]] = sdiv i32 [[SUB13]], 1
// CHECK10-NEXT:    [[MUL15:%.*]] = mul nsw i32 1, [[DIV14]]
// CHECK10-NEXT:    [[CONV16:%.*]] = sext i32 [[MUL15]] to i64
// CHECK10-NEXT:    [[DIV17:%.*]] = sdiv i64 [[TMP15]], [[CONV16]]
// CHECK10-NEXT:    [[MUL18:%.*]] = mul nsw i64 [[DIV17]], 1
// CHECK10-NEXT:    [[ADD:%.*]] = add nsw i64 0, [[MUL18]]
// CHECK10-NEXT:    [[CONV19:%.*]] = trunc i64 [[ADD]] to i32
// CHECK10-NEXT:    store i32 [[CONV19]], i32* [[I10]], align 4
// CHECK10-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK10-NEXT:    [[TMP18:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK10-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK10-NEXT:    [[SUB20:%.*]] = sub nsw i32 [[TMP19]], 0
// CHECK10-NEXT:    [[DIV21:%.*]] = sdiv i32 [[SUB20]], 1
// CHECK10-NEXT:    [[MUL22:%.*]] = mul nsw i32 1, [[DIV21]]
// CHECK10-NEXT:    [[CONV23:%.*]] = sext i32 [[MUL22]] to i64
// CHECK10-NEXT:    [[DIV24:%.*]] = sdiv i64 [[TMP18]], [[CONV23]]
// CHECK10-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK10-NEXT:    [[SUB25:%.*]] = sub nsw i32 [[TMP20]], 0
// CHECK10-NEXT:    [[DIV26:%.*]] = sdiv i32 [[SUB25]], 1
// CHECK10-NEXT:    [[MUL27:%.*]] = mul nsw i32 1, [[DIV26]]
// CHECK10-NEXT:    [[CONV28:%.*]] = sext i32 [[MUL27]] to i64
// CHECK10-NEXT:    [[MUL29:%.*]] = mul nsw i64 [[DIV24]], [[CONV28]]
// CHECK10-NEXT:    [[SUB30:%.*]] = sub nsw i64 [[TMP17]], [[MUL29]]
// CHECK10-NEXT:    [[MUL31:%.*]] = mul nsw i64 [[SUB30]], 1
// CHECK10-NEXT:    [[ADD32:%.*]] = add nsw i64 0, [[MUL31]]
// CHECK10-NEXT:    [[CONV33:%.*]] = trunc i64 [[ADD32]] to i32
// CHECK10-NEXT:    store i32 [[CONV33]], i32* [[J11]], align 4
// CHECK10-NEXT:    [[TMP21:%.*]] = load i32, i32* [[I10]], align 4
// CHECK10-NEXT:    [[TMP22:%.*]] = load i32, i32* [[J11]], align 4
// CHECK10-NEXT:    [[ADD34:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
// CHECK10-NEXT:    [[TMP23:%.*]] = load i32, i32* [[I10]], align 4
// CHECK10-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP23]] to i64
// CHECK10-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK10-NEXT:    [[TMP24:%.*]] = load i32, i32* [[J11]], align 4
// CHECK10-NEXT:    [[IDXPROM35:%.*]] = sext i32 [[TMP24]] to i64
// CHECK10-NEXT:    [[ARRAYIDX36:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i64 0, i64 [[IDXPROM35]]
// CHECK10-NEXT:    store i32 [[ADD34]], i32* [[ARRAYIDX36]], align 4
// CHECK10-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK10:       omp.body.continue:
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK10:       omp.inner.for.inc:
// CHECK10-NEXT:    [[TMP25:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK10-NEXT:    [[TMP26:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK10-NEXT:    [[ADD37:%.*]] = add nsw i64 [[TMP25]], [[TMP26]]
// CHECK10-NEXT:    store i64 [[ADD37]], i64* [[DOTOMP_IV]], align 8
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK10:       omp.inner.for.end:
// CHECK10-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK10:       omp.loop.exit:
// CHECK10-NEXT:    [[TMP27:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    [[TMP28:%.*]] = load i32, i32* [[TMP27]], align 4
// CHECK10-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP28]])
// CHECK10-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK10:       omp.precond.end:
// CHECK10-NEXT:    ret void
// CHECK10-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l74
// CHECK10-SAME: (i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK10-NEXT:  entry:
// CHECK10-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK10-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK10-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK10-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK10-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 8
// CHECK10-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK10-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK10-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK10-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK10-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK10:       .execute:
// CHECK10-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK10-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK10-NEXT:    [[CONV1:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK10-NEXT:    store i32 [[TMP2]], i32* [[CONV1]], align 4
// CHECK10-NEXT:    [[TMP3:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK10-NEXT:    [[TMP4:%.*]] = load i32*, i32** [[V_ADDR]], align 8
// CHECK10-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK10-NEXT:    call void @__omp_outlined__10(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP3]], [1000 x i32]* [[TMP0]], i32* [[TMP4]]) #[[ATTR3]]
// CHECK10-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK10:       .omp.deinit:
// CHECK10-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK10-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK10:       .exit:
// CHECK10-NEXT:    ret void
// CHECK10-LABEL: define {{[^@]+}}@__omp_outlined__10
// CHECK10-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK10-NEXT:  entry:
// CHECK10-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK10-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 8
// CHECK10-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK10-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK10-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK10-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 8
// CHECK10-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK10-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK10-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK10-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK10-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK10-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK10-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK10-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK10-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK10-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK10-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK10-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK10-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK10:       omp.precond.then:
// CHECK10-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK10-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK10-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK10-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK10-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK10-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK10-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK10-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK10:       cond.true:
// CHECK10-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK10-NEXT:    br label [[COND_END:%.*]]
// CHECK10:       cond.false:
// CHECK10-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    br label [[COND_END]]
// CHECK10:       cond.end:
// CHECK10-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK10-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK10:       omp.inner.for.cond:
// CHECK10-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK10-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK10-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK10-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK10:       omp.inner.for.body:
// CHECK10-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    [[TMP15:%.*]] = zext i32 [[TMP14]] to i64
// CHECK10-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP17:%.*]] = zext i32 [[TMP16]] to i64
// CHECK10-NEXT:    [[TMP18:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK10-NEXT:    [[CONV6:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK10-NEXT:    store i32 [[TMP18]], i32* [[CONV6]], align 4
// CHECK10-NEXT:    [[TMP19:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK10-NEXT:    [[TMP20:%.*]] = load i32*, i32** [[V_ADDR]], align 8
// CHECK10-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK10-NEXT:    [[TMP22:%.*]] = inttoptr i64 [[TMP15]] to i8*
// CHECK10-NEXT:    store i8* [[TMP22]], i8** [[TMP21]], align 8
// CHECK10-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK10-NEXT:    [[TMP24:%.*]] = inttoptr i64 [[TMP17]] to i8*
// CHECK10-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 8
// CHECK10-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK10-NEXT:    [[TMP26:%.*]] = inttoptr i64 [[TMP19]] to i8*
// CHECK10-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 8
// CHECK10-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK10-NEXT:    [[TMP28:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK10-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 8
// CHECK10-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 4
// CHECK10-NEXT:    [[TMP30:%.*]] = bitcast i32* [[TMP20]] to i8*
// CHECK10-NEXT:    store i8* [[TMP30]], i8** [[TMP29]], align 8
// CHECK10-NEXT:    [[TMP31:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    [[TMP32:%.*]] = load i32, i32* [[TMP31]], align 4
// CHECK10-NEXT:    [[TMP33:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK10-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP32]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, i64, [1000 x i32]*, i32*)* @__omp_outlined__11 to i8*), i8* null, i8** [[TMP33]], i64 5)
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK10:       omp.inner.for.inc:
// CHECK10-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP34]], [[TMP35]]
// CHECK10-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP36]], [[TMP37]]
// CHECK10-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP38]], [[TMP39]]
// CHECK10-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK10-NEXT:    [[CMP10:%.*]] = icmp sgt i32 [[TMP40]], [[TMP41]]
// CHECK10-NEXT:    br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]]
// CHECK10:       cond.true11:
// CHECK10-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK10-NEXT:    br label [[COND_END13:%.*]]
// CHECK10:       cond.false12:
// CHECK10-NEXT:    [[TMP43:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    br label [[COND_END13]]
// CHECK10:       cond.end13:
// CHECK10-NEXT:    [[COND14:%.*]] = phi i32 [ [[TMP42]], [[COND_TRUE11]] ], [ [[TMP43]], [[COND_FALSE12]] ]
// CHECK10-NEXT:    store i32 [[COND14]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK10-NEXT:    [[TMP44:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK10-NEXT:    store i32 [[TMP44]], i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK10:       omp.inner.for.end:
// CHECK10-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK10:       omp.loop.exit:
// CHECK10-NEXT:    [[TMP45:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    [[TMP46:%.*]] = load i32, i32* [[TMP45]], align 4
// CHECK10-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP46]])
// CHECK10-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK10:       omp.precond.end:
// CHECK10-NEXT:    ret void
// CHECK10-LABEL: define {{[^@]+}}@__omp_outlined__11
// CHECK10-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK10-NEXT:  entry:
// CHECK10-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK10-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK10-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 8
// CHECK10-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    [[I5:%.*]] = alloca i32, align 4
// CHECK10-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK10-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK10-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK10-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK10-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK10-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 8
// CHECK10-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK10-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK10-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK10-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK10-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK10-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK10-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK10-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK10-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK10-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK10-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK10-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK10-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK10:       omp.precond.then:
// CHECK10-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK10-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK10-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK10-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK10-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP5]] to i32
// CHECK10-NEXT:    [[TMP6:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK10-NEXT:    [[CONV4:%.*]] = trunc i64 [[TMP6]] to i32
// CHECK10-NEXT:    store i32 [[CONV3]], i32* [[DOTOMP_LB]], align 4
// CHECK10-NEXT:    store i32 [[CONV4]], i32* [[DOTOMP_UB]], align 4
// CHECK10-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK10-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK10-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK10-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK10-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK10:       omp.inner.for.cond:
// CHECK10-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[CONV6:%.*]] = sext i32 [[TMP10]] to i64
// CHECK10-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK10-NEXT:    [[CMP7:%.*]] = icmp ule i64 [[CONV6]], [[TMP11]]
// CHECK10-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK10:       omp.inner.for.body:
// CHECK10-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK10-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK10-NEXT:    store i32 [[ADD]], i32* [[I5]], align 4
// CHECK10-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[V_ADDR]], align 8
// CHECK10-NEXT:    [[TMP14:%.*]] = load i32, i32* [[I5]], align 4
// CHECK10-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP14]] to i64
// CHECK10-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[TMP13]], i64 [[IDXPROM]]
// CHECK10-NEXT:    [[TMP15:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK10-NEXT:    [[TMP16:%.*]] = load i32, i32* [[I5]], align 4
// CHECK10-NEXT:    [[IDXPROM8:%.*]] = sext i32 [[TMP16]] to i64
// CHECK10-NEXT:    [[ARRAYIDX9:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i64 0, i64 [[IDXPROM8]]
// CHECK10-NEXT:    store i32 [[TMP15]], i32* [[ARRAYIDX9]], align 4
// CHECK10-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK10:       omp.body.continue:
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK10:       omp.inner.for.inc:
// CHECK10-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK10-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK10-NEXT:    store i32 [[ADD10]], i32* [[DOTOMP_IV]], align 4
// CHECK10-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK10:       omp.inner.for.end:
// CHECK10-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK10:       omp.loop.exit:
// CHECK10-NEXT:    [[TMP19:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK10-NEXT:    [[TMP20:%.*]] = load i32, i32* [[TMP19]], align 4
// CHECK10-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP20]])
// CHECK10-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK10:       omp.precond.end:
// CHECK10-NEXT:    ret void
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l43
// CHECK11-SAME: (i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i64 [[L:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK11-NEXT:  entry:
// CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK11-NEXT:    [[L_ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[L_CASTED:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK11-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK11-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK11-NEXT:    store i64 [[L]], i64* [[L_ADDR]], align 8
// CHECK11-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK11-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK11-NEXT:    [[CONV1:%.*]] = bitcast i64* [[L_ADDR]] to i32*
// CHECK11-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK11-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK11-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK11:       .execute:
// CHECK11-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3:[0-9]+]])
// CHECK11-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK11-NEXT:    [[CONV2:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK11-NEXT:    store i32 [[TMP2]], i32* [[CONV2]], align 4
// CHECK11-NEXT:    [[TMP3:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK11-NEXT:    [[TMP4:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK11-NEXT:    [[CONV3:%.*]] = bitcast i64* [[L_CASTED]] to i32*
// CHECK11-NEXT:    store i32 [[TMP4]], i32* [[CONV3]], align 4
// CHECK11-NEXT:    [[TMP5:%.*]] = load i64, i64* [[L_CASTED]], align 8
// CHECK11-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK11-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP3]], [1000 x i32]* [[TMP0]], i64 [[TMP5]]) #[[ATTR3:[0-9]+]]
// CHECK11-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK11:       .omp.deinit:
// CHECK11-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK11-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK11:       .exit:
// CHECK11-NEXT:    ret void
// CHECK11-LABEL: define {{[^@]+}}@__omp_outlined__
// CHECK11-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i64 [[L:%.*]]) #[[ATTR0]] {
// CHECK11-NEXT:  entry:
// CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK11-NEXT:    [[L_ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[I5:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[L_CASTED:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 8
// CHECK11-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK11-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK11-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK11-NEXT:    store i64 [[L]], i64* [[L_ADDR]], align 8
// CHECK11-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK11-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK11-NEXT:    [[CONV1:%.*]] = bitcast i64* [[L_ADDR]] to i32*
// CHECK11-NEXT:    [[TMP1:%.*]] = load i16, i16* @"_openmp_static_kernel$is_shared", align 2
// CHECK11-NEXT:    [[TMP2:%.*]] = load i64, i64* @"_openmp_static_kernel$size", align 8
// CHECK11-NEXT:    call void @__kmpc_get_team_static_memory(i16 1, i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds (%"union._shared_openmp_static_memory_type_$_", %"union._shared_openmp_static_memory_type_$_" addrspace(3)* @"_openmp_shared_static_glob_rd_$_", i32 0, i32 0, i32 0) to i8*), i64 [[TMP2]], i16 [[TMP1]], i8** addrspacecast (i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr" to i8**))
// CHECK11-NEXT:    [[TMP3:%.*]] = load i8*, i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr", align 8
// CHECK11-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i8, i8* [[TMP3]], i64 0
// CHECK11-NEXT:    [[TMP5:%.*]] = bitcast i8* [[TMP4]] to %struct._globalized_locals_ty*
// CHECK11-NEXT:    [[L2:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY:%.*]], %struct._globalized_locals_ty* [[TMP5]], i32 0, i32 0
// CHECK11-NEXT:    [[TMP6:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK11-NEXT:    store i32 [[TMP6]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK11-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], 0
// CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK11-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK11-NEXT:    store i32 [[SUB4]], i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK11-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK11-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP8]]
// CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK11:       omp.precond.then:
// CHECK11-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK11-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK11-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK11-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1:[0-9]+]], i32 [[TMP11]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 128)
// CHECK11-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK11-NEXT:    [[CMP6:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
// CHECK11-NEXT:    br i1 [[CMP6]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK11:       cond.true:
// CHECK11-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK11-NEXT:    br label [[COND_END:%.*]]
// CHECK11:       cond.false:
// CHECK11-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    br label [[COND_END]]
// CHECK11:       cond.end:
// CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
// CHECK11-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    store i32 [[TMP16]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK11:       omp.inner.for.cond:
// CHECK11-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP18]], 1
// CHECK11-NEXT:    [[CMP7:%.*]] = icmp slt i32 [[TMP17]], [[ADD]]
// CHECK11-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK11:       omp.inner.for.body:
// CHECK11-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
// CHECK11-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP22:%.*]] = zext i32 [[TMP21]] to i64
// CHECK11-NEXT:    [[TMP23:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK11-NEXT:    [[CONV8:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK11-NEXT:    store i32 [[TMP23]], i32* [[CONV8]], align 4
// CHECK11-NEXT:    [[TMP24:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK11-NEXT:    [[TMP25:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK11-NEXT:    [[CONV9:%.*]] = bitcast i64* [[L_CASTED]] to i32*
// CHECK11-NEXT:    store i32 [[TMP25]], i32* [[CONV9]], align 4
// CHECK11-NEXT:    [[TMP26:%.*]] = load i64, i64* [[L_CASTED]], align 8
// CHECK11-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK11-NEXT:    [[TMP28:%.*]] = inttoptr i64 [[TMP20]] to i8*
// CHECK11-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 8
// CHECK11-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK11-NEXT:    [[TMP30:%.*]] = inttoptr i64 [[TMP22]] to i8*
// CHECK11-NEXT:    store i8* [[TMP30]], i8** [[TMP29]], align 8
// CHECK11-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK11-NEXT:    [[TMP32:%.*]] = inttoptr i64 [[TMP24]] to i8*
// CHECK11-NEXT:    store i8* [[TMP32]], i8** [[TMP31]], align 8
// CHECK11-NEXT:    [[TMP33:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK11-NEXT:    [[TMP34:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK11-NEXT:    store i8* [[TMP34]], i8** [[TMP33]], align 8
// CHECK11-NEXT:    [[TMP35:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 4
// CHECK11-NEXT:    [[TMP36:%.*]] = inttoptr i64 [[TMP26]] to i8*
// CHECK11-NEXT:    store i8* [[TMP36]], i8** [[TMP35]], align 8
// CHECK11-NEXT:    [[TMP37:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    [[TMP38:%.*]] = load i32, i32* [[TMP37]], align 4
// CHECK11-NEXT:    [[TMP39:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK11-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP38]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, i64, [1000 x i32]*, i64)* @__omp_outlined__1 to i8*), i8* null, i8** [[TMP39]], i64 5)
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK11:       omp.inner.for.inc:
// CHECK11-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP40]], [[TMP41]]
// CHECK11-NEXT:    store i32 [[ADD10]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    [[TMP43:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP42]], [[TMP43]]
// CHECK11-NEXT:    store i32 [[ADD11]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    [[TMP44:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP45:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP44]], [[TMP45]]
// CHECK11-NEXT:    store i32 [[ADD12]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP46:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP47:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK11-NEXT:    [[CMP13:%.*]] = icmp sgt i32 [[TMP46]], [[TMP47]]
// CHECK11-NEXT:    br i1 [[CMP13]], label [[COND_TRUE14:%.*]], label [[COND_FALSE15:%.*]]
// CHECK11:       cond.true14:
// CHECK11-NEXT:    [[TMP48:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK11-NEXT:    br label [[COND_END16:%.*]]
// CHECK11:       cond.false15:
// CHECK11-NEXT:    [[TMP49:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    br label [[COND_END16]]
// CHECK11:       cond.end16:
// CHECK11-NEXT:    [[COND17:%.*]] = phi i32 [ [[TMP48]], [[COND_TRUE14]] ], [ [[TMP49]], [[COND_FALSE15]] ]
// CHECK11-NEXT:    store i32 [[COND17]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP50:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    store i32 [[TMP50]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK11:       omp.inner.for.end:
// CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK11:       omp.loop.exit:
// CHECK11-NEXT:    [[TMP51:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    [[TMP52:%.*]] = load i32, i32* [[TMP51]], align 4
// CHECK11-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP52]])
// CHECK11-NEXT:    [[TMP53:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK11-NEXT:    [[TMP54:%.*]] = icmp ne i32 [[TMP53]], 0
// CHECK11-NEXT:    br i1 [[TMP54]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK11:       .omp.lastprivate.then:
// CHECK11-NEXT:    [[TMP55:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK11-NEXT:    store i32 [[TMP55]], i32* [[CONV1]], align 8
// CHECK11-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK11:       .omp.lastprivate.done:
// CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK11:       omp.precond.end:
// CHECK11-NEXT:    [[TMP56:%.*]] = load i16, i16* @"_openmp_static_kernel$is_shared", align 2
// CHECK11-NEXT:    call void @__kmpc_restore_team_static_memory(i16 1, i16 [[TMP56]])
// CHECK11-NEXT:    ret void
// CHECK11-LABEL: define {{[^@]+}}@__omp_outlined__1
// CHECK11-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i64 [[L:%.*]]) #[[ATTR0]] {
// CHECK11-NEXT:  entry:
// CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK11-NEXT:    [[L_ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[I6:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK11-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK11-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK11-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK11-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK11-NEXT:    store i64 [[L]], i64* [[L_ADDR]], align 8
// CHECK11-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK11-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK11-NEXT:    [[CONV1:%.*]] = bitcast i64* [[L_ADDR]] to i32*
// CHECK11-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK11-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK11-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK11-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK11-NEXT:    store i32 [[SUB3]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK11-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK11-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK11:       omp.precond.then:
// CHECK11-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK11-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK11-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK11-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK11-NEXT:    [[CONV4:%.*]] = trunc i64 [[TMP5]] to i32
// CHECK11-NEXT:    [[TMP6:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK11-NEXT:    [[CONV5:%.*]] = trunc i64 [[TMP6]] to i32
// CHECK11-NEXT:    store i32 [[CONV4]], i32* [[DOTOMP_LB]], align 4
// CHECK11-NEXT:    store i32 [[CONV5]], i32* [[DOTOMP_UB]], align 4
// CHECK11-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK11-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK11-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 32)
// CHECK11-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK11:       omp.dispatch.cond:
// CHECK11-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK11-NEXT:    [[CONV7:%.*]] = sext i32 [[TMP9]] to i64
// CHECK11-NEXT:    [[TMP10:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK11-NEXT:    [[CMP8:%.*]] = icmp ugt i64 [[CONV7]], [[TMP10]]
// CHECK11-NEXT:    br i1 [[CMP8]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK11:       cond.true:
// CHECK11-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK11-NEXT:    br label [[COND_END:%.*]]
// CHECK11:       cond.false:
// CHECK11-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK11-NEXT:    [[CONV9:%.*]] = sext i32 [[TMP12]] to i64
// CHECK11-NEXT:    br label [[COND_END]]
// CHECK11:       cond.end:
// CHECK11-NEXT:    [[COND:%.*]] = phi i64 [ [[TMP11]], [[COND_TRUE]] ], [ [[CONV9]], [[COND_FALSE]] ]
// CHECK11-NEXT:    [[CONV10:%.*]] = trunc i64 [[COND]] to i32
// CHECK11-NEXT:    store i32 [[CONV10]], i32* [[DOTOMP_UB]], align 4
// CHECK11-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK11-NEXT:    store i32 [[TMP13]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK11-NEXT:    [[CMP11:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK11-NEXT:    br i1 [[CMP11]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK11:       omp.dispatch.body:
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK11:       omp.inner.for.cond:
// CHECK11-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK11-NEXT:    [[CMP12:%.*]] = icmp sle i32 [[TMP16]], [[TMP17]]
// CHECK11-NEXT:    br i1 [[CMP12]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK11:       omp.inner.for.body:
// CHECK11-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP18]], 1
// CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK11-NEXT:    store i32 [[ADD]], i32* [[I6]], align 4
// CHECK11-NEXT:    [[TMP19:%.*]] = load i32, i32* [[I6]], align 4
// CHECK11-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP19]] to i64
// CHECK11-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK11-NEXT:    store i32 1, i32* [[ARRAYIDX]], align 4
// CHECK11-NEXT:    [[TMP20:%.*]] = load i32, i32* [[I6]], align 4
// CHECK11-NEXT:    store i32 [[TMP20]], i32* [[CONV1]], align 8
// CHECK11-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK11:       omp.body.continue:
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK11:       omp.inner.for.inc:
// CHECK11-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[ADD13:%.*]] = add nsw i32 [[TMP21]], 1
// CHECK11-NEXT:    store i32 [[ADD13]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK11:       omp.inner.for.end:
// CHECK11-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK11:       omp.dispatch.inc:
// CHECK11-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK11-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    [[ADD14:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK11-NEXT:    store i32 [[ADD14]], i32* [[DOTOMP_LB]], align 4
// CHECK11-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK11-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    [[ADD15:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK11-NEXT:    store i32 [[ADD15]], i32* [[DOTOMP_UB]], align 4
// CHECK11-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK11:       omp.dispatch.end:
// CHECK11-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK11-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP27]])
// CHECK11-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK11-NEXT:    [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
// CHECK11-NEXT:    br i1 [[TMP29]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK11:       .omp.lastprivate.then:
// CHECK11-NEXT:    [[TMP30:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK11-NEXT:    store i32 [[TMP30]], i32* [[CONV1]], align 8
// CHECK11-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK11:       .omp.lastprivate.done:
// CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK11:       omp.precond.end:
// CHECK11-NEXT:    ret void
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l49
// CHECK11-SAME: (i64 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK11-NEXT:  entry:
// CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 8
// CHECK11-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK11-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK11-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 8
// CHECK11-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK11-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 8
// CHECK11-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK11-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK11-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK11:       .execute:
// CHECK11-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK11-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK11-NEXT:    [[CONV1:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK11-NEXT:    store i32 [[TMP2]], i32* [[CONV1]], align 4
// CHECK11-NEXT:    [[TMP3:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK11-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK11-NEXT:    call void @__omp_outlined__2(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP3]], [1000 x i16]* [[TMP0]]) #[[ATTR3]]
// CHECK11-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK11:       .omp.deinit:
// CHECK11-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK11-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK11:       .exit:
// CHECK11-NEXT:    ret void
// CHECK11-LABEL: define {{[^@]+}}@__omp_outlined__2
// CHECK11-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK11-NEXT:  entry:
// CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 8
// CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 8
// CHECK11-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK11-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK11-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 8
// CHECK11-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK11-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 8
// CHECK11-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK11-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK11-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK11-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK11-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK11-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK11:       omp.precond.then:
// CHECK11-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK11-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK11-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK11-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK11-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK11-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK11-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK11-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK11:       cond.true:
// CHECK11-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK11-NEXT:    br label [[COND_END:%.*]]
// CHECK11:       cond.false:
// CHECK11-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    br label [[COND_END]]
// CHECK11:       cond.end:
// CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK11-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK11:       omp.inner.for.cond:
// CHECK11-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK11-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK11-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK11:       omp.inner.for.body:
// CHECK11-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    [[TMP15:%.*]] = zext i32 [[TMP14]] to i64
// CHECK11-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP17:%.*]] = zext i32 [[TMP16]] to i64
// CHECK11-NEXT:    [[TMP18:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK11-NEXT:    [[CONV6:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK11-NEXT:    store i32 [[TMP18]], i32* [[CONV6]], align 4
// CHECK11-NEXT:    [[TMP19:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK11-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK11-NEXT:    [[TMP21:%.*]] = inttoptr i64 [[TMP15]] to i8*
// CHECK11-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 8
// CHECK11-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK11-NEXT:    [[TMP23:%.*]] = inttoptr i64 [[TMP17]] to i8*
// CHECK11-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 8
// CHECK11-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK11-NEXT:    [[TMP25:%.*]] = inttoptr i64 [[TMP19]] to i8*
// CHECK11-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 8
// CHECK11-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK11-NEXT:    [[TMP27:%.*]] = bitcast [1000 x i16]* [[TMP0]] to i8*
// CHECK11-NEXT:    store i8* [[TMP27]], i8** [[TMP26]], align 8
// CHECK11-NEXT:    [[TMP28:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    [[TMP29:%.*]] = load i32, i32* [[TMP28]], align 4
// CHECK11-NEXT:    [[TMP30:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK11-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP29]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, i64, [1000 x i16]*)* @__omp_outlined__3 to i8*), i8* null, i8** [[TMP30]], i64 4)
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK11:       omp.inner.for.inc:
// CHECK11-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP31]], [[TMP32]]
// CHECK11-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP33]], [[TMP34]]
// CHECK11-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP35]], [[TMP36]]
// CHECK11-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK11-NEXT:    [[CMP10:%.*]] = icmp sgt i32 [[TMP37]], [[TMP38]]
// CHECK11-NEXT:    br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]]
// CHECK11:       cond.true11:
// CHECK11-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK11-NEXT:    br label [[COND_END13:%.*]]
// CHECK11:       cond.false12:
// CHECK11-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    br label [[COND_END13]]
// CHECK11:       cond.end13:
// CHECK11-NEXT:    [[COND14:%.*]] = phi i32 [ [[TMP39]], [[COND_TRUE11]] ], [ [[TMP40]], [[COND_FALSE12]] ]
// CHECK11-NEXT:    store i32 [[COND14]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    store i32 [[TMP41]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK11:       omp.inner.for.end:
// CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK11:       omp.loop.exit:
// CHECK11-NEXT:    [[TMP42:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    [[TMP43:%.*]] = load i32, i32* [[TMP42]], align 4
// CHECK11-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP43]])
// CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK11:       omp.precond.end:
// CHECK11-NEXT:    ret void
// CHECK11-LABEL: define {{[^@]+}}@__omp_outlined__3
// CHECK11-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], i64 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK11-NEXT:  entry:
// CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 8
// CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[I5:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK11-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK11-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK11-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK11-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 8
// CHECK11-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK11-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 8
// CHECK11-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK11-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK11-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK11-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK11-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK11-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK11:       omp.precond.then:
// CHECK11-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK11-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK11-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK11-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK11-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP5]] to i32
// CHECK11-NEXT:    [[TMP6:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK11-NEXT:    [[CONV4:%.*]] = trunc i64 [[TMP6]] to i32
// CHECK11-NEXT:    store i32 [[CONV3]], i32* [[DOTOMP_LB]], align 4
// CHECK11-NEXT:    store i32 [[CONV4]], i32* [[DOTOMP_UB]], align 4
// CHECK11-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK11-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK11-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK11-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK11-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK11:       omp.inner.for.cond:
// CHECK11-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[CONV6:%.*]] = sext i32 [[TMP10]] to i64
// CHECK11-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK11-NEXT:    [[CMP7:%.*]] = icmp ule i64 [[CONV6]], [[TMP11]]
// CHECK11-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK11:       omp.inner.for.body:
// CHECK11-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK11-NEXT:    store i32 [[ADD]], i32* [[I5]], align 4
// CHECK11-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I5]], align 4
// CHECK11-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP13]] to i64
// CHECK11-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i16], [1000 x i16]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK11-NEXT:    [[TMP14:%.*]] = load i16, i16* [[ARRAYIDX]], align 2
// CHECK11-NEXT:    [[CONV8:%.*]] = sext i16 [[TMP14]] to i32
// CHECK11-NEXT:    [[ADD9:%.*]] = add nsw i32 [[CONV8]], 1
// CHECK11-NEXT:    [[CONV10:%.*]] = trunc i32 [[ADD9]] to i16
// CHECK11-NEXT:    store i16 [[CONV10]], i16* [[ARRAYIDX]], align 2
// CHECK11-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK11:       omp.body.continue:
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK11:       omp.inner.for.inc:
// CHECK11-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP15]], [[TMP16]]
// CHECK11-NEXT:    store i32 [[ADD11]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK11:       omp.inner.for.end:
// CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK11:       omp.loop.exit:
// CHECK11-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
// CHECK11-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP18]])
// CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK11:       omp.precond.end:
// CHECK11-NEXT:    ret void
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l54
// CHECK11-SAME: ([10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK11-NEXT:  entry:
// CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 8
// CHECK11-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK11-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 8
// CHECK11-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 8
// CHECK11-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK11-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK11-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK11:       .execute:
// CHECK11-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK11-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK11-NEXT:    call void @__omp_outlined__4(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x i32]* [[TMP0]]) #[[ATTR3]]
// CHECK11-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK11:       .omp.deinit:
// CHECK11-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK11-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK11:       .exit:
// CHECK11-NEXT:    ret void
// CHECK11-LABEL: define {{[^@]+}}@__omp_outlined__4
// CHECK11-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK11-NEXT:  entry:
// CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 8
// CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x i8*], align 8
// CHECK11-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK11-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 8
// CHECK11-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 8
// CHECK11-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    store i32 9, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK11-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK11-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK11-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK11-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9
// CHECK11-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK11:       cond.true:
// CHECK11-NEXT:    br label [[COND_END:%.*]]
// CHECK11:       cond.false:
// CHECK11-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    br label [[COND_END]]
// CHECK11:       cond.end:
// CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK11-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK11:       omp.inner.for.cond:
// CHECK11-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[CMP1:%.*]] = icmp slt i32 [[TMP6]], 10
// CHECK11-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK11:       omp.inner.for.body:
// CHECK11-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
// CHECK11-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
// CHECK11-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK11-NEXT:    [[TMP12:%.*]] = inttoptr i64 [[TMP8]] to i8*
// CHECK11-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 8
// CHECK11-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK11-NEXT:    [[TMP14:%.*]] = inttoptr i64 [[TMP10]] to i8*
// CHECK11-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 8
// CHECK11-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK11-NEXT:    [[TMP16:%.*]] = bitcast [10 x i32]* [[TMP0]] to i8*
// CHECK11-NEXT:    store i8* [[TMP16]], i8** [[TMP15]], align 8
// CHECK11-NEXT:    [[TMP17:%.*]] = bitcast [3 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK11-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, [10 x i32]*)* @__omp_outlined__5 to i8*), i8* null, i8** [[TMP17]], i64 3)
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK11:       omp.inner.for.inc:
// CHECK11-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP18]], [[TMP19]]
// CHECK11-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK11-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK11-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP24]], 9
// CHECK11-NEXT:    br i1 [[CMP4]], label [[COND_TRUE5:%.*]], label [[COND_FALSE6:%.*]]
// CHECK11:       cond.true5:
// CHECK11-NEXT:    br label [[COND_END7:%.*]]
// CHECK11:       cond.false6:
// CHECK11-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    br label [[COND_END7]]
// CHECK11:       cond.end7:
// CHECK11-NEXT:    [[COND8:%.*]] = phi i32 [ 9, [[COND_TRUE5]] ], [ [[TMP25]], [[COND_FALSE6]] ]
// CHECK11-NEXT:    store i32 [[COND8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    store i32 [[TMP26]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK11:       omp.inner.for.end:
// CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK11:       omp.loop.exit:
// CHECK11-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]])
// CHECK11-NEXT:    ret void
// CHECK11-LABEL: define {{[^@]+}}@__omp_outlined__5
// CHECK11-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK11-NEXT:  entry:
// CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 8
// CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK11-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK11-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK11-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 8
// CHECK11-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 8
// CHECK11-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK11-NEXT:    store i32 9, i32* [[DOTOMP_UB]], align 4
// CHECK11-NEXT:    [[TMP1:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK11-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP1]] to i32
// CHECK11-NEXT:    [[TMP2:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK11-NEXT:    [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32
// CHECK11-NEXT:    store i32 [[CONV]], i32* [[DOTOMP_LB]], align 4
// CHECK11-NEXT:    store i32 [[CONV1]], i32* [[DOTOMP_UB]], align 4
// CHECK11-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK11-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK11-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK11-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK11-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK11:       omp.inner.for.cond:
// CHECK11-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[CONV2:%.*]] = sext i32 [[TMP6]] to i64
// CHECK11-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK11-NEXT:    [[CMP:%.*]] = icmp ule i64 [[CONV2]], [[TMP7]]
// CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK11:       omp.inner.for.body:
// CHECK11-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP8]], 1
// CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK11-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK11-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4
// CHECK11-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP9]] to i64
// CHECK11-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK11-NEXT:    [[TMP10:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK11-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP10]], 1
// CHECK11-NEXT:    store i32 [[ADD3]], i32* [[ARRAYIDX]], align 4
// CHECK11-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK11:       omp.body.continue:
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK11:       omp.inner.for.inc:
// CHECK11-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
// CHECK11-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK11:       omp.inner.for.end:
// CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK11:       omp.loop.exit:
// CHECK11-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK11-NEXT:    ret void
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l59
// CHECK11-SAME: ([10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i64 [[F:%.*]]) #[[ATTR0]] {
// CHECK11-NEXT:  entry:
// CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK11-NEXT:    [[F_ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[F_CASTED:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK11-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK11-NEXT:    store i64 [[F]], i64* [[F_ADDR]], align 8
// CHECK11-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK11-NEXT:    [[CONV:%.*]] = bitcast i64* [[F_ADDR]] to i32*
// CHECK11-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK11-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK11-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK11:       .execute:
// CHECK11-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK11-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK11-NEXT:    [[CONV1:%.*]] = bitcast i64* [[F_CASTED]] to i32*
// CHECK11-NEXT:    store i32 [[TMP2]], i32* [[CONV1]], align 4
// CHECK11-NEXT:    [[TMP3:%.*]] = load i64, i64* [[F_CASTED]], align 8
// CHECK11-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK11-NEXT:    call void @__omp_outlined__6(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x [10 x i32]]* [[TMP0]], i64 [[TMP3]]) #[[ATTR3]]
// CHECK11-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK11:       .omp.deinit:
// CHECK11-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK11-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK11:       .exit:
// CHECK11-NEXT:    ret void
// CHECK11-LABEL: define {{[^@]+}}@__omp_outlined__6
// CHECK11-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i64 [[F:%.*]]) #[[ATTR0]] {
// CHECK11-NEXT:  entry:
// CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK11-NEXT:    [[F_ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[F_CASTED:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 8
// CHECK11-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK11-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK11-NEXT:    store i64 [[F]], i64* [[F_ADDR]], align 8
// CHECK11-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK11-NEXT:    [[CONV:%.*]] = bitcast i64* [[F_ADDR]] to i32*
// CHECK11-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    store i32 99, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK11-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK11-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK11-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK11-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 99
// CHECK11-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK11:       cond.true:
// CHECK11-NEXT:    br label [[COND_END:%.*]]
// CHECK11:       cond.false:
// CHECK11-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    br label [[COND_END]]
// CHECK11:       cond.end:
// CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK11-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK11:       omp.inner.for.cond:
// CHECK11-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[TMP6]], 100
// CHECK11-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK11:       omp.inner.for.body:
// CHECK11-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
// CHECK11-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
// CHECK11-NEXT:    [[TMP11:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK11-NEXT:    [[CONV3:%.*]] = bitcast i64* [[F_CASTED]] to i32*
// CHECK11-NEXT:    store i32 [[TMP11]], i32* [[CONV3]], align 4
// CHECK11-NEXT:    [[TMP12:%.*]] = load i64, i64* [[F_CASTED]], align 8
// CHECK11-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK11-NEXT:    [[TMP14:%.*]] = inttoptr i64 [[TMP8]] to i8*
// CHECK11-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 8
// CHECK11-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK11-NEXT:    [[TMP16:%.*]] = inttoptr i64 [[TMP10]] to i8*
// CHECK11-NEXT:    store i8* [[TMP16]], i8** [[TMP15]], align 8
// CHECK11-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK11-NEXT:    [[TMP18:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK11-NEXT:    store i8* [[TMP18]], i8** [[TMP17]], align 8
// CHECK11-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK11-NEXT:    [[TMP20:%.*]] = inttoptr i64 [[TMP12]] to i8*
// CHECK11-NEXT:    store i8* [[TMP20]], i8** [[TMP19]], align 8
// CHECK11-NEXT:    [[TMP21:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK11-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, [10 x [10 x i32]]*, i64)* @__omp_outlined__7 to i8*), i8* null, i8** [[TMP21]], i64 4)
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK11:       omp.inner.for.inc:
// CHECK11-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK11-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK11-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP26]], [[TMP27]]
// CHECK11-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[CMP6:%.*]] = icmp sgt i32 [[TMP28]], 99
// CHECK11-NEXT:    br i1 [[CMP6]], label [[COND_TRUE7:%.*]], label [[COND_FALSE8:%.*]]
// CHECK11:       cond.true7:
// CHECK11-NEXT:    br label [[COND_END9:%.*]]
// CHECK11:       cond.false8:
// CHECK11-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    br label [[COND_END9]]
// CHECK11:       cond.end9:
// CHECK11-NEXT:    [[COND10:%.*]] = phi i32 [ 99, [[COND_TRUE7]] ], [ [[TMP29]], [[COND_FALSE8]] ]
// CHECK11-NEXT:    store i32 [[COND10]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    store i32 [[TMP30]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK11:       omp.inner.for.end:
// CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK11:       omp.loop.exit:
// CHECK11-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]])
// CHECK11-NEXT:    ret void
// CHECK11-LABEL: define {{[^@]+}}@__omp_outlined__7
// CHECK11-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i64 [[F:%.*]]) #[[ATTR0]] {
// CHECK11-NEXT:  entry:
// CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK11-NEXT:    [[F_ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK11-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK11-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK11-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK11-NEXT:    store i64 [[F]], i64* [[F_ADDR]], align 8
// CHECK11-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK11-NEXT:    [[CONV:%.*]] = bitcast i64* [[F_ADDR]] to i32*
// CHECK11-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK11-NEXT:    store i32 99, i32* [[DOTOMP_UB]], align 4
// CHECK11-NEXT:    [[TMP1:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK11-NEXT:    [[CONV2:%.*]] = trunc i64 [[TMP1]] to i32
// CHECK11-NEXT:    [[TMP2:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK11-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP2]] to i32
// CHECK11-NEXT:    store i32 [[CONV2]], i32* [[DOTOMP_LB]], align 4
// CHECK11-NEXT:    store i32 [[CONV3]], i32* [[DOTOMP_UB]], align 4
// CHECK11-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK11-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK11-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK11-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK11-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK11:       omp.inner.for.cond:
// CHECK11-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[CONV4:%.*]] = sext i32 [[TMP6]] to i64
// CHECK11-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK11-NEXT:    [[CMP:%.*]] = icmp ule i64 [[CONV4]], [[TMP7]]
// CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK11:       omp.inner.for.body:
// CHECK11-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP8]], 10
// CHECK11-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK11-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK11-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[DIV5:%.*]] = sdiv i32 [[TMP10]], 10
// CHECK11-NEXT:    [[MUL6:%.*]] = mul nsw i32 [[DIV5]], 10
// CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP9]], [[MUL6]]
// CHECK11-NEXT:    [[MUL7:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK11-NEXT:    [[ADD8:%.*]] = add nsw i32 0, [[MUL7]]
// CHECK11-NEXT:    store i32 [[ADD8]], i32* [[J]], align 4
// CHECK11-NEXT:    store i32 10, i32* [[K]], align 4
// CHECK11-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
// CHECK11-NEXT:    [[TMP12:%.*]] = load i32, i32* [[J]], align 4
// CHECK11-NEXT:    [[TMP13:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK11-NEXT:    [[MUL9:%.*]] = mul nsw i32 [[TMP12]], [[TMP13]]
// CHECK11-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP11]], [[MUL9]]
// CHECK11-NEXT:    [[TMP14:%.*]] = load i32, i32* [[K]], align 4
// CHECK11-NEXT:    [[ADD11:%.*]] = add nsw i32 [[ADD10]], [[TMP14]]
// CHECK11-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
// CHECK11-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP15]] to i64
// CHECK11-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK11-NEXT:    [[TMP16:%.*]] = load i32, i32* [[J]], align 4
// CHECK11-NEXT:    [[IDXPROM12:%.*]] = sext i32 [[TMP16]] to i64
// CHECK11-NEXT:    [[ARRAYIDX13:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i64 0, i64 [[IDXPROM12]]
// CHECK11-NEXT:    store i32 [[ADD11]], i32* [[ARRAYIDX13]], align 4
// CHECK11-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK11:       omp.body.continue:
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK11:       omp.inner.for.inc:
// CHECK11-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    [[ADD14:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK11-NEXT:    store i32 [[ADD14]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK11:       omp.inner.for.end:
// CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK11:       omp.loop.exit:
// CHECK11-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK11-NEXT:    ret void
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l67
// CHECK11-SAME: (i64 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK11-NEXT:  entry:
// CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK11-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK11-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK11-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK11-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK11-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK11-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK11-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK11-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK11:       .execute:
// CHECK11-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK11-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK11-NEXT:    [[CONV1:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK11-NEXT:    store i32 [[TMP2]], i32* [[CONV1]], align 4
// CHECK11-NEXT:    [[TMP3:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK11-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK11-NEXT:    call void @__omp_outlined__8(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP3]], [10 x [10 x i32]]* [[TMP0]]) #[[ATTR3]]
// CHECK11-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK11:       .omp.deinit:
// CHECK11-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK11-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK11:       .exit:
// CHECK11-NEXT:    ret void
// CHECK11-LABEL: define {{[^@]+}}@__omp_outlined__8
// CHECK11-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK11-NEXT:  entry:
// CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[I8:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[J9:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 8
// CHECK11-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK11-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK11-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK11-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK11-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK11-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK11-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK11-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK11-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK11-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK11-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK11-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK11-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK11-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], [[DIV5]]
// CHECK11-NEXT:    [[SUB6:%.*]] = sub nsw i32 [[MUL]], 1
// CHECK11-NEXT:    store i32 [[SUB6]], i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK11-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK11-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK11-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK11-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK11:       land.lhs.true:
// CHECK11-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK11-NEXT:    [[CMP7:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK11-NEXT:    br i1 [[CMP7]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK11:       omp.precond.then:
// CHECK11-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK11-NEXT:    store i32 [[TMP7]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK11-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK11-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    [[TMP9:%.*]] = load i32, i32* [[TMP8]], align 4
// CHECK11-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP9]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK11-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK11-NEXT:    [[CMP10:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
// CHECK11-NEXT:    br i1 [[CMP10]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK11:       cond.true:
// CHECK11-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK11-NEXT:    br label [[COND_END:%.*]]
// CHECK11:       cond.false:
// CHECK11-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    br label [[COND_END]]
// CHECK11:       cond.end:
// CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
// CHECK11-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    store i32 [[TMP14]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK11:       omp.inner.for.cond:
// CHECK11-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP16]], 1
// CHECK11-NEXT:    [[CMP11:%.*]] = icmp slt i32 [[TMP15]], [[ADD]]
// CHECK11-NEXT:    br i1 [[CMP11]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK11:       omp.inner.for.body:
// CHECK11-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    [[TMP18:%.*]] = zext i32 [[TMP17]] to i64
// CHECK11-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
// CHECK11-NEXT:    [[TMP21:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK11-NEXT:    [[CONV12:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK11-NEXT:    store i32 [[TMP21]], i32* [[CONV12]], align 4
// CHECK11-NEXT:    [[TMP22:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK11-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK11-NEXT:    [[TMP24:%.*]] = inttoptr i64 [[TMP18]] to i8*
// CHECK11-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 8
// CHECK11-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK11-NEXT:    [[TMP26:%.*]] = inttoptr i64 [[TMP20]] to i8*
// CHECK11-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 8
// CHECK11-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK11-NEXT:    [[TMP28:%.*]] = inttoptr i64 [[TMP22]] to i8*
// CHECK11-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 8
// CHECK11-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK11-NEXT:    [[TMP30:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK11-NEXT:    store i8* [[TMP30]], i8** [[TMP29]], align 8
// CHECK11-NEXT:    [[TMP31:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    [[TMP32:%.*]] = load i32, i32* [[TMP31]], align 4
// CHECK11-NEXT:    [[TMP33:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK11-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP32]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, i64, [10 x [10 x i32]]*)* @__omp_outlined__9 to i8*), i8* null, i8** [[TMP33]], i64 4)
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK11:       omp.inner.for.inc:
// CHECK11-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    [[ADD13:%.*]] = add nsw i32 [[TMP34]], [[TMP35]]
// CHECK11-NEXT:    store i32 [[ADD13]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    [[ADD14:%.*]] = add nsw i32 [[TMP36]], [[TMP37]]
// CHECK11-NEXT:    store i32 [[ADD14]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    [[ADD15:%.*]] = add nsw i32 [[TMP38]], [[TMP39]]
// CHECK11-NEXT:    store i32 [[ADD15]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK11-NEXT:    [[CMP16:%.*]] = icmp sgt i32 [[TMP40]], [[TMP41]]
// CHECK11-NEXT:    br i1 [[CMP16]], label [[COND_TRUE17:%.*]], label [[COND_FALSE18:%.*]]
// CHECK11:       cond.true17:
// CHECK11-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK11-NEXT:    br label [[COND_END19:%.*]]
// CHECK11:       cond.false18:
// CHECK11-NEXT:    [[TMP43:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    br label [[COND_END19]]
// CHECK11:       cond.end19:
// CHECK11-NEXT:    [[COND20:%.*]] = phi i32 [ [[TMP42]], [[COND_TRUE17]] ], [ [[TMP43]], [[COND_FALSE18]] ]
// CHECK11-NEXT:    store i32 [[COND20]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP44:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    store i32 [[TMP44]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK11:       omp.inner.for.end:
// CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK11:       omp.loop.exit:
// CHECK11-NEXT:    [[TMP45:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    [[TMP46:%.*]] = load i32, i32* [[TMP45]], align 4
// CHECK11-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP46]])
// CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK11:       omp.precond.end:
// CHECK11-NEXT:    ret void
// CHECK11-LABEL: define {{[^@]+}}@__omp_outlined__9
// CHECK11-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], i64 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK11-NEXT:  entry:
// CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[I10:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[J11:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK11-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK11-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK11-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK11-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK11-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK11-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK11-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK11-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK11-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK11-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK11-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK11-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK11-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK11-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK11-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], [[DIV5]]
// CHECK11-NEXT:    [[SUB6:%.*]] = sub nsw i32 [[MUL]], 1
// CHECK11-NEXT:    store i32 [[SUB6]], i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK11-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK11-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK11-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK11-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK11:       land.lhs.true:
// CHECK11-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK11-NEXT:    [[CMP7:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK11-NEXT:    br i1 [[CMP7]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK11:       omp.precond.then:
// CHECK11-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK11-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK11-NEXT:    store i32 [[TMP7]], i32* [[DOTOMP_UB]], align 4
// CHECK11-NEXT:    [[TMP8:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK11-NEXT:    [[CONV8:%.*]] = trunc i64 [[TMP8]] to i32
// CHECK11-NEXT:    [[TMP9:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK11-NEXT:    [[CONV9:%.*]] = trunc i64 [[TMP9]] to i32
// CHECK11-NEXT:    store i32 [[CONV8]], i32* [[DOTOMP_LB]], align 4
// CHECK11-NEXT:    store i32 [[CONV9]], i32* [[DOTOMP_UB]], align 4
// CHECK11-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK11-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK11-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP11]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK11-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK11-NEXT:    store i32 [[TMP12]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK11:       omp.inner.for.cond:
// CHECK11-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[CONV12:%.*]] = sext i32 [[TMP13]] to i64
// CHECK11-NEXT:    [[TMP14:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK11-NEXT:    [[CMP13:%.*]] = icmp ule i64 [[CONV12]], [[TMP14]]
// CHECK11-NEXT:    br i1 [[CMP13]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK11:       omp.inner.for.body:
// CHECK11-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK11-NEXT:    [[SUB14:%.*]] = sub nsw i32 [[TMP16]], 0
// CHECK11-NEXT:    [[DIV15:%.*]] = sdiv i32 [[SUB14]], 1
// CHECK11-NEXT:    [[MUL16:%.*]] = mul nsw i32 1, [[DIV15]]
// CHECK11-NEXT:    [[DIV17:%.*]] = sdiv i32 [[TMP15]], [[MUL16]]
// CHECK11-NEXT:    [[MUL18:%.*]] = mul nsw i32 [[DIV17]], 1
// CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL18]]
// CHECK11-NEXT:    store i32 [[ADD]], i32* [[I10]], align 4
// CHECK11-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK11-NEXT:    [[SUB19:%.*]] = sub nsw i32 [[TMP19]], 0
// CHECK11-NEXT:    [[DIV20:%.*]] = sdiv i32 [[SUB19]], 1
// CHECK11-NEXT:    [[MUL21:%.*]] = mul nsw i32 1, [[DIV20]]
// CHECK11-NEXT:    [[DIV22:%.*]] = sdiv i32 [[TMP18]], [[MUL21]]
// CHECK11-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK11-NEXT:    [[SUB23:%.*]] = sub nsw i32 [[TMP20]], 0
// CHECK11-NEXT:    [[DIV24:%.*]] = sdiv i32 [[SUB23]], 1
// CHECK11-NEXT:    [[MUL25:%.*]] = mul nsw i32 1, [[DIV24]]
// CHECK11-NEXT:    [[MUL26:%.*]] = mul nsw i32 [[DIV22]], [[MUL25]]
// CHECK11-NEXT:    [[SUB27:%.*]] = sub nsw i32 [[TMP17]], [[MUL26]]
// CHECK11-NEXT:    [[MUL28:%.*]] = mul nsw i32 [[SUB27]], 1
// CHECK11-NEXT:    [[ADD29:%.*]] = add nsw i32 0, [[MUL28]]
// CHECK11-NEXT:    store i32 [[ADD29]], i32* [[J11]], align 4
// CHECK11-NEXT:    [[TMP21:%.*]] = load i32, i32* [[I10]], align 4
// CHECK11-NEXT:    [[TMP22:%.*]] = load i32, i32* [[J11]], align 4
// CHECK11-NEXT:    [[ADD30:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
// CHECK11-NEXT:    [[TMP23:%.*]] = load i32, i32* [[I10]], align 4
// CHECK11-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP23]] to i64
// CHECK11-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK11-NEXT:    [[TMP24:%.*]] = load i32, i32* [[J11]], align 4
// CHECK11-NEXT:    [[IDXPROM31:%.*]] = sext i32 [[TMP24]] to i64
// CHECK11-NEXT:    [[ARRAYIDX32:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i64 0, i64 [[IDXPROM31]]
// CHECK11-NEXT:    store i32 [[ADD30]], i32* [[ARRAYIDX32]], align 4
// CHECK11-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK11:       omp.body.continue:
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK11:       omp.inner.for.inc:
// CHECK11-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    [[ADD33:%.*]] = add nsw i32 [[TMP25]], [[TMP26]]
// CHECK11-NEXT:    store i32 [[ADD33]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK11:       omp.inner.for.end:
// CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK11:       omp.loop.exit:
// CHECK11-NEXT:    [[TMP27:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    [[TMP28:%.*]] = load i32, i32* [[TMP27]], align 4
// CHECK11-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP28]])
// CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK11:       omp.precond.end:
// CHECK11-NEXT:    ret void
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l74
// CHECK11-SAME: (i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK11-NEXT:  entry:
// CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK11-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK11-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK11-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK11-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 8
// CHECK11-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK11-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK11-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK11-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK11-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK11:       .execute:
// CHECK11-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK11-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK11-NEXT:    [[CONV1:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK11-NEXT:    store i32 [[TMP2]], i32* [[CONV1]], align 4
// CHECK11-NEXT:    [[TMP3:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK11-NEXT:    [[TMP4:%.*]] = load i32*, i32** [[V_ADDR]], align 8
// CHECK11-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK11-NEXT:    call void @__omp_outlined__10(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP3]], [1000 x i32]* [[TMP0]], i32* [[TMP4]]) #[[ATTR3]]
// CHECK11-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK11:       .omp.deinit:
// CHECK11-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK11-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK11:       .exit:
// CHECK11-NEXT:    ret void
// CHECK11-LABEL: define {{[^@]+}}@__omp_outlined__10
// CHECK11-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK11-NEXT:  entry:
// CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK11-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 8
// CHECK11-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK11-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK11-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK11-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 8
// CHECK11-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK11-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK11-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK11-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK11-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK11-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK11-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK11-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK11:       omp.precond.then:
// CHECK11-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK11-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK11-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK11-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK11-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK11-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK11-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK11-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK11:       cond.true:
// CHECK11-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK11-NEXT:    br label [[COND_END:%.*]]
// CHECK11:       cond.false:
// CHECK11-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    br label [[COND_END]]
// CHECK11:       cond.end:
// CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK11-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK11:       omp.inner.for.cond:
// CHECK11-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK11-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK11-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK11:       omp.inner.for.body:
// CHECK11-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    [[TMP15:%.*]] = zext i32 [[TMP14]] to i64
// CHECK11-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP17:%.*]] = zext i32 [[TMP16]] to i64
// CHECK11-NEXT:    [[TMP18:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK11-NEXT:    [[CONV6:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK11-NEXT:    store i32 [[TMP18]], i32* [[CONV6]], align 4
// CHECK11-NEXT:    [[TMP19:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK11-NEXT:    [[TMP20:%.*]] = load i32*, i32** [[V_ADDR]], align 8
// CHECK11-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK11-NEXT:    [[TMP22:%.*]] = inttoptr i64 [[TMP15]] to i8*
// CHECK11-NEXT:    store i8* [[TMP22]], i8** [[TMP21]], align 8
// CHECK11-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK11-NEXT:    [[TMP24:%.*]] = inttoptr i64 [[TMP17]] to i8*
// CHECK11-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 8
// CHECK11-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK11-NEXT:    [[TMP26:%.*]] = inttoptr i64 [[TMP19]] to i8*
// CHECK11-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 8
// CHECK11-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK11-NEXT:    [[TMP28:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK11-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 8
// CHECK11-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 4
// CHECK11-NEXT:    [[TMP30:%.*]] = bitcast i32* [[TMP20]] to i8*
// CHECK11-NEXT:    store i8* [[TMP30]], i8** [[TMP29]], align 8
// CHECK11-NEXT:    [[TMP31:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    [[TMP32:%.*]] = load i32, i32* [[TMP31]], align 4
// CHECK11-NEXT:    [[TMP33:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK11-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP32]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, i64, [1000 x i32]*, i32*)* @__omp_outlined__11 to i8*), i8* null, i8** [[TMP33]], i64 5)
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK11:       omp.inner.for.inc:
// CHECK11-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP34]], [[TMP35]]
// CHECK11-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP36]], [[TMP37]]
// CHECK11-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP38]], [[TMP39]]
// CHECK11-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK11-NEXT:    [[CMP10:%.*]] = icmp sgt i32 [[TMP40]], [[TMP41]]
// CHECK11-NEXT:    br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]]
// CHECK11:       cond.true11:
// CHECK11-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK11-NEXT:    br label [[COND_END13:%.*]]
// CHECK11:       cond.false12:
// CHECK11-NEXT:    [[TMP43:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    br label [[COND_END13]]
// CHECK11:       cond.end13:
// CHECK11-NEXT:    [[COND14:%.*]] = phi i32 [ [[TMP42]], [[COND_TRUE11]] ], [ [[TMP43]], [[COND_FALSE12]] ]
// CHECK11-NEXT:    store i32 [[COND14]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK11-NEXT:    [[TMP44:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK11-NEXT:    store i32 [[TMP44]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK11:       omp.inner.for.end:
// CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK11:       omp.loop.exit:
// CHECK11-NEXT:    [[TMP45:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    [[TMP46:%.*]] = load i32, i32* [[TMP45]], align 4
// CHECK11-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP46]])
// CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK11:       omp.precond.end:
// CHECK11-NEXT:    ret void
// CHECK11-LABEL: define {{[^@]+}}@__omp_outlined__11
// CHECK11-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK11-NEXT:  entry:
// CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK11-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 8
// CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    [[I5:%.*]] = alloca i32, align 4
// CHECK11-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK11-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK11-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK11-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK11-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK11-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 8
// CHECK11-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK11-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK11-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK11-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK11-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK11-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK11-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK11-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK11:       omp.precond.then:
// CHECK11-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK11-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK11-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK11-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK11-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP5]] to i32
// CHECK11-NEXT:    [[TMP6:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK11-NEXT:    [[CONV4:%.*]] = trunc i64 [[TMP6]] to i32
// CHECK11-NEXT:    store i32 [[CONV3]], i32* [[DOTOMP_LB]], align 4
// CHECK11-NEXT:    store i32 [[CONV4]], i32* [[DOTOMP_UB]], align 4
// CHECK11-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK11-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK11-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK11-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK11-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK11:       omp.inner.for.cond:
// CHECK11-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[CONV6:%.*]] = sext i32 [[TMP10]] to i64
// CHECK11-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK11-NEXT:    [[CMP7:%.*]] = icmp ule i64 [[CONV6]], [[TMP11]]
// CHECK11-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK11:       omp.inner.for.body:
// CHECK11-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK11-NEXT:    store i32 [[ADD]], i32* [[I5]], align 4
// CHECK11-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[V_ADDR]], align 8
// CHECK11-NEXT:    [[TMP14:%.*]] = load i32, i32* [[I5]], align 4
// CHECK11-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP14]] to i64
// CHECK11-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[TMP13]], i64 [[IDXPROM]]
// CHECK11-NEXT:    [[TMP15:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK11-NEXT:    [[TMP16:%.*]] = load i32, i32* [[I5]], align 4
// CHECK11-NEXT:    [[IDXPROM8:%.*]] = sext i32 [[TMP16]] to i64
// CHECK11-NEXT:    [[ARRAYIDX9:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i64 0, i64 [[IDXPROM8]]
// CHECK11-NEXT:    store i32 [[TMP15]], i32* [[ARRAYIDX9]], align 4
// CHECK11-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK11:       omp.body.continue:
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK11:       omp.inner.for.inc:
// CHECK11-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK11-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK11-NEXT:    store i32 [[ADD10]], i32* [[DOTOMP_IV]], align 4
// CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK11:       omp.inner.for.end:
// CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK11:       omp.loop.exit:
// CHECK11-NEXT:    [[TMP19:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK11-NEXT:    [[TMP20:%.*]] = load i32, i32* [[TMP19]], align 4
// CHECK11-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP20]])
// CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK11:       omp.precond.end:
// CHECK11-NEXT:    ret void
// CHECK12-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l43
// CHECK12-SAME: (i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i64 [[L:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK12-NEXT:  entry:
// CHECK12-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK12-NEXT:    [[L_ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[L_CASTED:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK12-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK12-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK12-NEXT:    store i64 [[L]], i64* [[L_ADDR]], align 8
// CHECK12-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK12-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK12-NEXT:    [[CONV1:%.*]] = bitcast i64* [[L_ADDR]] to i32*
// CHECK12-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK12-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK12-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK12:       .execute:
// CHECK12-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3:[0-9]+]])
// CHECK12-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK12-NEXT:    [[CONV2:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK12-NEXT:    store i32 [[TMP2]], i32* [[CONV2]], align 4
// CHECK12-NEXT:    [[TMP3:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK12-NEXT:    [[TMP4:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK12-NEXT:    [[CONV3:%.*]] = bitcast i64* [[L_CASTED]] to i32*
// CHECK12-NEXT:    store i32 [[TMP4]], i32* [[CONV3]], align 4
// CHECK12-NEXT:    [[TMP5:%.*]] = load i64, i64* [[L_CASTED]], align 8
// CHECK12-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK12-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP3]], [1000 x i32]* [[TMP0]], i64 [[TMP5]]) #[[ATTR3:[0-9]+]]
// CHECK12-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK12:       .omp.deinit:
// CHECK12-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK12-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK12:       .exit:
// CHECK12-NEXT:    ret void
// CHECK12-LABEL: define {{[^@]+}}@__omp_outlined__
// CHECK12-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i64 [[L:%.*]]) #[[ATTR0]] {
// CHECK12-NEXT:  entry:
// CHECK12-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK12-NEXT:    [[L_ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[I5:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[L_CASTED:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 8
// CHECK12-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK12-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK12-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK12-NEXT:    store i64 [[L]], i64* [[L_ADDR]], align 8
// CHECK12-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK12-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK12-NEXT:    [[CONV1:%.*]] = bitcast i64* [[L_ADDR]] to i32*
// CHECK12-NEXT:    [[TMP1:%.*]] = call i8* @__kmpc_data_sharing_push_stack(i64 4, i16 1)
// CHECK12-NEXT:    [[TMP2:%.*]] = bitcast i8* [[TMP1]] to %struct._globalized_locals_ty*
// CHECK12-NEXT:    [[L2:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY:%.*]], %struct._globalized_locals_ty* [[TMP2]], i32 0, i32 0
// CHECK12-NEXT:    [[TMP3:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK12-NEXT:    store i32 [[TMP3]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK12-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK12-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK12-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK12-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK12-NEXT:    store i32 [[SUB4]], i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK12-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK12-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK12-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK12-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK12:       omp.precond.then:
// CHECK12-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK12-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK12-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK12-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1:[0-9]+]], i32 [[TMP8]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 128)
// CHECK12-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK12-NEXT:    [[CMP6:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
// CHECK12-NEXT:    br i1 [[CMP6]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK12:       cond.true:
// CHECK12-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK12-NEXT:    br label [[COND_END:%.*]]
// CHECK12:       cond.false:
// CHECK12-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    br label [[COND_END]]
// CHECK12:       cond.end:
// CHECK12-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP11]], [[COND_TRUE]] ], [ [[TMP12]], [[COND_FALSE]] ]
// CHECK12-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    store i32 [[TMP13]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK12:       omp.inner.for.cond:
// CHECK12-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK12-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP15]], 1
// CHECK12-NEXT:    [[CMP7:%.*]] = icmp slt i32 [[TMP14]], [[ADD]]
// CHECK12-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK12:       omp.inner.for.body:
// CHECK12-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    [[TMP17:%.*]] = zext i32 [[TMP16]] to i64
// CHECK12-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP19:%.*]] = zext i32 [[TMP18]] to i64
// CHECK12-NEXT:    [[TMP20:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK12-NEXT:    [[CONV8:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK12-NEXT:    store i32 [[TMP20]], i32* [[CONV8]], align 4
// CHECK12-NEXT:    [[TMP21:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK12-NEXT:    [[TMP22:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK12-NEXT:    [[CONV9:%.*]] = bitcast i64* [[L_CASTED]] to i32*
// CHECK12-NEXT:    store i32 [[TMP22]], i32* [[CONV9]], align 4
// CHECK12-NEXT:    [[TMP23:%.*]] = load i64, i64* [[L_CASTED]], align 8
// CHECK12-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK12-NEXT:    [[TMP25:%.*]] = inttoptr i64 [[TMP17]] to i8*
// CHECK12-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 8
// CHECK12-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK12-NEXT:    [[TMP27:%.*]] = inttoptr i64 [[TMP19]] to i8*
// CHECK12-NEXT:    store i8* [[TMP27]], i8** [[TMP26]], align 8
// CHECK12-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK12-NEXT:    [[TMP29:%.*]] = inttoptr i64 [[TMP21]] to i8*
// CHECK12-NEXT:    store i8* [[TMP29]], i8** [[TMP28]], align 8
// CHECK12-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK12-NEXT:    [[TMP31:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK12-NEXT:    store i8* [[TMP31]], i8** [[TMP30]], align 8
// CHECK12-NEXT:    [[TMP32:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 4
// CHECK12-NEXT:    [[TMP33:%.*]] = inttoptr i64 [[TMP23]] to i8*
// CHECK12-NEXT:    store i8* [[TMP33]], i8** [[TMP32]], align 8
// CHECK12-NEXT:    [[TMP34:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    [[TMP35:%.*]] = load i32, i32* [[TMP34]], align 4
// CHECK12-NEXT:    [[TMP36:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK12-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP35]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, i64, [1000 x i32]*, i64)* @__omp_outlined__1 to i8*), i8* null, i8** [[TMP36]], i64 5)
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK12:       omp.inner.for.inc:
// CHECK12-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP37]], [[TMP38]]
// CHECK12-NEXT:    store i32 [[ADD10]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP39]], [[TMP40]]
// CHECK12-NEXT:    store i32 [[ADD11]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP41]], [[TMP42]]
// CHECK12-NEXT:    store i32 [[ADD12]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP43:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP44:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK12-NEXT:    [[CMP13:%.*]] = icmp sgt i32 [[TMP43]], [[TMP44]]
// CHECK12-NEXT:    br i1 [[CMP13]], label [[COND_TRUE14:%.*]], label [[COND_FALSE15:%.*]]
// CHECK12:       cond.true14:
// CHECK12-NEXT:    [[TMP45:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK12-NEXT:    br label [[COND_END16:%.*]]
// CHECK12:       cond.false15:
// CHECK12-NEXT:    [[TMP46:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    br label [[COND_END16]]
// CHECK12:       cond.end16:
// CHECK12-NEXT:    [[COND17:%.*]] = phi i32 [ [[TMP45]], [[COND_TRUE14]] ], [ [[TMP46]], [[COND_FALSE15]] ]
// CHECK12-NEXT:    store i32 [[COND17]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP47:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    store i32 [[TMP47]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK12:       omp.inner.for.end:
// CHECK12-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK12:       omp.loop.exit:
// CHECK12-NEXT:    [[TMP48:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    [[TMP49:%.*]] = load i32, i32* [[TMP48]], align 4
// CHECK12-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP49]])
// CHECK12-NEXT:    [[TMP50:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK12-NEXT:    [[TMP51:%.*]] = icmp ne i32 [[TMP50]], 0
// CHECK12-NEXT:    br i1 [[TMP51]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK12:       .omp.lastprivate.then:
// CHECK12-NEXT:    [[TMP52:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK12-NEXT:    store i32 [[TMP52]], i32* [[CONV1]], align 8
// CHECK12-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK12:       .omp.lastprivate.done:
// CHECK12-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK12:       omp.precond.end:
// CHECK12-NEXT:    call void @__kmpc_data_sharing_pop_stack(i8* [[TMP1]])
// CHECK12-NEXT:    ret void
// CHECK12-LABEL: define {{[^@]+}}@__omp_outlined__1
// CHECK12-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i64 [[L:%.*]]) #[[ATTR0]] {
// CHECK12-NEXT:  entry:
// CHECK12-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK12-NEXT:    [[L_ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[I6:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK12-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK12-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK12-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK12-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK12-NEXT:    store i64 [[L]], i64* [[L_ADDR]], align 8
// CHECK12-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK12-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK12-NEXT:    [[CONV1:%.*]] = bitcast i64* [[L_ADDR]] to i32*
// CHECK12-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK12-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK12-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK12-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK12-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK12-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK12-NEXT:    store i32 [[SUB3]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK12-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK12-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK12-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK12-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK12:       omp.precond.then:
// CHECK12-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK12-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK12-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK12-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK12-NEXT:    [[CONV4:%.*]] = trunc i64 [[TMP5]] to i32
// CHECK12-NEXT:    [[TMP6:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK12-NEXT:    [[CONV5:%.*]] = trunc i64 [[TMP6]] to i32
// CHECK12-NEXT:    store i32 [[CONV4]], i32* [[DOTOMP_LB]], align 4
// CHECK12-NEXT:    store i32 [[CONV5]], i32* [[DOTOMP_UB]], align 4
// CHECK12-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK12-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK12-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 32)
// CHECK12-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK12:       omp.dispatch.cond:
// CHECK12-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK12-NEXT:    [[CONV7:%.*]] = sext i32 [[TMP9]] to i64
// CHECK12-NEXT:    [[TMP10:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK12-NEXT:    [[CMP8:%.*]] = icmp ugt i64 [[CONV7]], [[TMP10]]
// CHECK12-NEXT:    br i1 [[CMP8]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK12:       cond.true:
// CHECK12-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK12-NEXT:    br label [[COND_END:%.*]]
// CHECK12:       cond.false:
// CHECK12-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK12-NEXT:    [[CONV9:%.*]] = sext i32 [[TMP12]] to i64
// CHECK12-NEXT:    br label [[COND_END]]
// CHECK12:       cond.end:
// CHECK12-NEXT:    [[COND:%.*]] = phi i64 [ [[TMP11]], [[COND_TRUE]] ], [ [[CONV9]], [[COND_FALSE]] ]
// CHECK12-NEXT:    [[CONV10:%.*]] = trunc i64 [[COND]] to i32
// CHECK12-NEXT:    store i32 [[CONV10]], i32* [[DOTOMP_UB]], align 4
// CHECK12-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK12-NEXT:    store i32 [[TMP13]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK12-NEXT:    [[CMP11:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK12-NEXT:    br i1 [[CMP11]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK12:       omp.dispatch.body:
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK12:       omp.inner.for.cond:
// CHECK12-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK12-NEXT:    [[CMP12:%.*]] = icmp sle i32 [[TMP16]], [[TMP17]]
// CHECK12-NEXT:    br i1 [[CMP12]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK12:       omp.inner.for.body:
// CHECK12-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP18]], 1
// CHECK12-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK12-NEXT:    store i32 [[ADD]], i32* [[I6]], align 4
// CHECK12-NEXT:    [[TMP19:%.*]] = load i32, i32* [[I6]], align 4
// CHECK12-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP19]] to i64
// CHECK12-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK12-NEXT:    store i32 1, i32* [[ARRAYIDX]], align 4
// CHECK12-NEXT:    [[TMP20:%.*]] = load i32, i32* [[I6]], align 4
// CHECK12-NEXT:    store i32 [[TMP20]], i32* [[CONV1]], align 8
// CHECK12-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK12:       omp.body.continue:
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK12:       omp.inner.for.inc:
// CHECK12-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[ADD13:%.*]] = add nsw i32 [[TMP21]], 1
// CHECK12-NEXT:    store i32 [[ADD13]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK12:       omp.inner.for.end:
// CHECK12-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK12:       omp.dispatch.inc:
// CHECK12-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK12-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    [[ADD14:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK12-NEXT:    store i32 [[ADD14]], i32* [[DOTOMP_LB]], align 4
// CHECK12-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK12-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    [[ADD15:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK12-NEXT:    store i32 [[ADD15]], i32* [[DOTOMP_UB]], align 4
// CHECK12-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK12:       omp.dispatch.end:
// CHECK12-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK12-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP27]])
// CHECK12-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK12-NEXT:    [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
// CHECK12-NEXT:    br i1 [[TMP29]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK12:       .omp.lastprivate.then:
// CHECK12-NEXT:    [[TMP30:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK12-NEXT:    store i32 [[TMP30]], i32* [[CONV1]], align 8
// CHECK12-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK12:       .omp.lastprivate.done:
// CHECK12-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK12:       omp.precond.end:
// CHECK12-NEXT:    ret void
// CHECK12-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l49
// CHECK12-SAME: (i64 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK12-NEXT:  entry:
// CHECK12-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 8
// CHECK12-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK12-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK12-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 8
// CHECK12-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK12-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 8
// CHECK12-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK12-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK12-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK12:       .execute:
// CHECK12-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK12-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK12-NEXT:    [[CONV1:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK12-NEXT:    store i32 [[TMP2]], i32* [[CONV1]], align 4
// CHECK12-NEXT:    [[TMP3:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK12-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK12-NEXT:    call void @__omp_outlined__2(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP3]], [1000 x i16]* [[TMP0]]) #[[ATTR3]]
// CHECK12-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK12:       .omp.deinit:
// CHECK12-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK12-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK12:       .exit:
// CHECK12-NEXT:    ret void
// CHECK12-LABEL: define {{[^@]+}}@__omp_outlined__2
// CHECK12-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK12-NEXT:  entry:
// CHECK12-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 8
// CHECK12-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 8
// CHECK12-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK12-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK12-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 8
// CHECK12-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK12-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 8
// CHECK12-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK12-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK12-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK12-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK12-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK12-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK12-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK12-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK12-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK12-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK12-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK12:       omp.precond.then:
// CHECK12-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK12-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK12-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK12-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK12-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK12-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK12-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK12-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK12:       cond.true:
// CHECK12-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK12-NEXT:    br label [[COND_END:%.*]]
// CHECK12:       cond.false:
// CHECK12-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    br label [[COND_END]]
// CHECK12:       cond.end:
// CHECK12-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK12-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK12:       omp.inner.for.cond:
// CHECK12-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK12-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK12-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK12-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK12:       omp.inner.for.body:
// CHECK12-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    [[TMP15:%.*]] = zext i32 [[TMP14]] to i64
// CHECK12-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP17:%.*]] = zext i32 [[TMP16]] to i64
// CHECK12-NEXT:    [[TMP18:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK12-NEXT:    [[CONV6:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK12-NEXT:    store i32 [[TMP18]], i32* [[CONV6]], align 4
// CHECK12-NEXT:    [[TMP19:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK12-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK12-NEXT:    [[TMP21:%.*]] = inttoptr i64 [[TMP15]] to i8*
// CHECK12-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 8
// CHECK12-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK12-NEXT:    [[TMP23:%.*]] = inttoptr i64 [[TMP17]] to i8*
// CHECK12-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 8
// CHECK12-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK12-NEXT:    [[TMP25:%.*]] = inttoptr i64 [[TMP19]] to i8*
// CHECK12-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 8
// CHECK12-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK12-NEXT:    [[TMP27:%.*]] = bitcast [1000 x i16]* [[TMP0]] to i8*
// CHECK12-NEXT:    store i8* [[TMP27]], i8** [[TMP26]], align 8
// CHECK12-NEXT:    [[TMP28:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    [[TMP29:%.*]] = load i32, i32* [[TMP28]], align 4
// CHECK12-NEXT:    [[TMP30:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK12-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP29]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, i64, [1000 x i16]*)* @__omp_outlined__3 to i8*), i8* null, i8** [[TMP30]], i64 4)
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK12:       omp.inner.for.inc:
// CHECK12-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP31]], [[TMP32]]
// CHECK12-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP33]], [[TMP34]]
// CHECK12-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP35]], [[TMP36]]
// CHECK12-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK12-NEXT:    [[CMP10:%.*]] = icmp sgt i32 [[TMP37]], [[TMP38]]
// CHECK12-NEXT:    br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]]
// CHECK12:       cond.true11:
// CHECK12-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK12-NEXT:    br label [[COND_END13:%.*]]
// CHECK12:       cond.false12:
// CHECK12-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    br label [[COND_END13]]
// CHECK12:       cond.end13:
// CHECK12-NEXT:    [[COND14:%.*]] = phi i32 [ [[TMP39]], [[COND_TRUE11]] ], [ [[TMP40]], [[COND_FALSE12]] ]
// CHECK12-NEXT:    store i32 [[COND14]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    store i32 [[TMP41]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK12:       omp.inner.for.end:
// CHECK12-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK12:       omp.loop.exit:
// CHECK12-NEXT:    [[TMP42:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    [[TMP43:%.*]] = load i32, i32* [[TMP42]], align 4
// CHECK12-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP43]])
// CHECK12-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK12:       omp.precond.end:
// CHECK12-NEXT:    ret void
// CHECK12-LABEL: define {{[^@]+}}@__omp_outlined__3
// CHECK12-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], i64 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK12-NEXT:  entry:
// CHECK12-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 8
// CHECK12-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[I5:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK12-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK12-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK12-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK12-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 8
// CHECK12-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK12-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 8
// CHECK12-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK12-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK12-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK12-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK12-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK12-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK12-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK12-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK12-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK12-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK12-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK12:       omp.precond.then:
// CHECK12-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK12-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK12-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK12-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK12-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP5]] to i32
// CHECK12-NEXT:    [[TMP6:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK12-NEXT:    [[CONV4:%.*]] = trunc i64 [[TMP6]] to i32
// CHECK12-NEXT:    store i32 [[CONV3]], i32* [[DOTOMP_LB]], align 4
// CHECK12-NEXT:    store i32 [[CONV4]], i32* [[DOTOMP_UB]], align 4
// CHECK12-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK12-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK12-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK12-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK12-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK12:       omp.inner.for.cond:
// CHECK12-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[CONV6:%.*]] = sext i32 [[TMP10]] to i64
// CHECK12-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK12-NEXT:    [[CMP7:%.*]] = icmp ule i64 [[CONV6]], [[TMP11]]
// CHECK12-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK12:       omp.inner.for.body:
// CHECK12-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK12-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK12-NEXT:    store i32 [[ADD]], i32* [[I5]], align 4
// CHECK12-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I5]], align 4
// CHECK12-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP13]] to i64
// CHECK12-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i16], [1000 x i16]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK12-NEXT:    [[TMP14:%.*]] = load i16, i16* [[ARRAYIDX]], align 2
// CHECK12-NEXT:    [[CONV8:%.*]] = sext i16 [[TMP14]] to i32
// CHECK12-NEXT:    [[ADD9:%.*]] = add nsw i32 [[CONV8]], 1
// CHECK12-NEXT:    [[CONV10:%.*]] = trunc i32 [[ADD9]] to i16
// CHECK12-NEXT:    store i16 [[CONV10]], i16* [[ARRAYIDX]], align 2
// CHECK12-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK12:       omp.body.continue:
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK12:       omp.inner.for.inc:
// CHECK12-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP15]], [[TMP16]]
// CHECK12-NEXT:    store i32 [[ADD11]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK12:       omp.inner.for.end:
// CHECK12-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK12:       omp.loop.exit:
// CHECK12-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
// CHECK12-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP18]])
// CHECK12-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK12:       omp.precond.end:
// CHECK12-NEXT:    ret void
// CHECK12-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l54
// CHECK12-SAME: ([10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK12-NEXT:  entry:
// CHECK12-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 8
// CHECK12-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK12-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 8
// CHECK12-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 8
// CHECK12-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK12-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK12-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK12:       .execute:
// CHECK12-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK12-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK12-NEXT:    call void @__omp_outlined__4(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x i32]* [[TMP0]]) #[[ATTR3]]
// CHECK12-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK12:       .omp.deinit:
// CHECK12-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK12-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK12:       .exit:
// CHECK12-NEXT:    ret void
// CHECK12-LABEL: define {{[^@]+}}@__omp_outlined__4
// CHECK12-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK12-NEXT:  entry:
// CHECK12-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 8
// CHECK12-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x i8*], align 8
// CHECK12-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK12-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 8
// CHECK12-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 8
// CHECK12-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    store i32 9, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK12-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK12-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK12-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK12-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9
// CHECK12-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK12:       cond.true:
// CHECK12-NEXT:    br label [[COND_END:%.*]]
// CHECK12:       cond.false:
// CHECK12-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    br label [[COND_END]]
// CHECK12:       cond.end:
// CHECK12-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK12-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK12:       omp.inner.for.cond:
// CHECK12-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[CMP1:%.*]] = icmp slt i32 [[TMP6]], 10
// CHECK12-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK12:       omp.inner.for.body:
// CHECK12-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
// CHECK12-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
// CHECK12-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK12-NEXT:    [[TMP12:%.*]] = inttoptr i64 [[TMP8]] to i8*
// CHECK12-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 8
// CHECK12-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK12-NEXT:    [[TMP14:%.*]] = inttoptr i64 [[TMP10]] to i8*
// CHECK12-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 8
// CHECK12-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK12-NEXT:    [[TMP16:%.*]] = bitcast [10 x i32]* [[TMP0]] to i8*
// CHECK12-NEXT:    store i8* [[TMP16]], i8** [[TMP15]], align 8
// CHECK12-NEXT:    [[TMP17:%.*]] = bitcast [3 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK12-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, [10 x i32]*)* @__omp_outlined__5 to i8*), i8* null, i8** [[TMP17]], i64 3)
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK12:       omp.inner.for.inc:
// CHECK12-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP18]], [[TMP19]]
// CHECK12-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK12-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK12-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP24]], 9
// CHECK12-NEXT:    br i1 [[CMP4]], label [[COND_TRUE5:%.*]], label [[COND_FALSE6:%.*]]
// CHECK12:       cond.true5:
// CHECK12-NEXT:    br label [[COND_END7:%.*]]
// CHECK12:       cond.false6:
// CHECK12-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    br label [[COND_END7]]
// CHECK12:       cond.end7:
// CHECK12-NEXT:    [[COND8:%.*]] = phi i32 [ 9, [[COND_TRUE5]] ], [ [[TMP25]], [[COND_FALSE6]] ]
// CHECK12-NEXT:    store i32 [[COND8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    store i32 [[TMP26]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK12:       omp.inner.for.end:
// CHECK12-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK12:       omp.loop.exit:
// CHECK12-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]])
// CHECK12-NEXT:    ret void
// CHECK12-LABEL: define {{[^@]+}}@__omp_outlined__5
// CHECK12-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK12-NEXT:  entry:
// CHECK12-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 8
// CHECK12-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK12-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK12-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK12-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 8
// CHECK12-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 8
// CHECK12-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK12-NEXT:    store i32 9, i32* [[DOTOMP_UB]], align 4
// CHECK12-NEXT:    [[TMP1:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK12-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP1]] to i32
// CHECK12-NEXT:    [[TMP2:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK12-NEXT:    [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32
// CHECK12-NEXT:    store i32 [[CONV]], i32* [[DOTOMP_LB]], align 4
// CHECK12-NEXT:    store i32 [[CONV1]], i32* [[DOTOMP_UB]], align 4
// CHECK12-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK12-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK12-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK12-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK12-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK12:       omp.inner.for.cond:
// CHECK12-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[CONV2:%.*]] = sext i32 [[TMP6]] to i64
// CHECK12-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK12-NEXT:    [[CMP:%.*]] = icmp ule i64 [[CONV2]], [[TMP7]]
// CHECK12-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK12:       omp.inner.for.body:
// CHECK12-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP8]], 1
// CHECK12-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK12-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK12-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4
// CHECK12-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP9]] to i64
// CHECK12-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK12-NEXT:    [[TMP10:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK12-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP10]], 1
// CHECK12-NEXT:    store i32 [[ADD3]], i32* [[ARRAYIDX]], align 4
// CHECK12-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK12:       omp.body.continue:
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK12:       omp.inner.for.inc:
// CHECK12-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
// CHECK12-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK12:       omp.inner.for.end:
// CHECK12-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK12:       omp.loop.exit:
// CHECK12-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK12-NEXT:    ret void
// CHECK12-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l59
// CHECK12-SAME: ([10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i64 [[F:%.*]]) #[[ATTR0]] {
// CHECK12-NEXT:  entry:
// CHECK12-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK12-NEXT:    [[F_ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[F_CASTED:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK12-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK12-NEXT:    store i64 [[F]], i64* [[F_ADDR]], align 8
// CHECK12-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK12-NEXT:    [[CONV:%.*]] = bitcast i64* [[F_ADDR]] to i32*
// CHECK12-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK12-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK12-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK12:       .execute:
// CHECK12-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK12-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK12-NEXT:    [[CONV1:%.*]] = bitcast i64* [[F_CASTED]] to i32*
// CHECK12-NEXT:    store i32 [[TMP2]], i32* [[CONV1]], align 4
// CHECK12-NEXT:    [[TMP3:%.*]] = load i64, i64* [[F_CASTED]], align 8
// CHECK12-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK12-NEXT:    call void @__omp_outlined__6(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x [10 x i32]]* [[TMP0]], i64 [[TMP3]]) #[[ATTR3]]
// CHECK12-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK12:       .omp.deinit:
// CHECK12-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK12-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK12:       .exit:
// CHECK12-NEXT:    ret void
// CHECK12-LABEL: define {{[^@]+}}@__omp_outlined__6
// CHECK12-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i64 [[F:%.*]]) #[[ATTR0]] {
// CHECK12-NEXT:  entry:
// CHECK12-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK12-NEXT:    [[F_ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[F_CASTED:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 8
// CHECK12-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK12-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK12-NEXT:    store i64 [[F]], i64* [[F_ADDR]], align 8
// CHECK12-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK12-NEXT:    [[CONV:%.*]] = bitcast i64* [[F_ADDR]] to i32*
// CHECK12-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    store i32 99, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK12-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK12-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK12-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK12-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 99
// CHECK12-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK12:       cond.true:
// CHECK12-NEXT:    br label [[COND_END:%.*]]
// CHECK12:       cond.false:
// CHECK12-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    br label [[COND_END]]
// CHECK12:       cond.end:
// CHECK12-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK12-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK12:       omp.inner.for.cond:
// CHECK12-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[TMP6]], 100
// CHECK12-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK12:       omp.inner.for.body:
// CHECK12-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
// CHECK12-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
// CHECK12-NEXT:    [[TMP11:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK12-NEXT:    [[CONV3:%.*]] = bitcast i64* [[F_CASTED]] to i32*
// CHECK12-NEXT:    store i32 [[TMP11]], i32* [[CONV3]], align 4
// CHECK12-NEXT:    [[TMP12:%.*]] = load i64, i64* [[F_CASTED]], align 8
// CHECK12-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK12-NEXT:    [[TMP14:%.*]] = inttoptr i64 [[TMP8]] to i8*
// CHECK12-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 8
// CHECK12-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK12-NEXT:    [[TMP16:%.*]] = inttoptr i64 [[TMP10]] to i8*
// CHECK12-NEXT:    store i8* [[TMP16]], i8** [[TMP15]], align 8
// CHECK12-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK12-NEXT:    [[TMP18:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK12-NEXT:    store i8* [[TMP18]], i8** [[TMP17]], align 8
// CHECK12-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK12-NEXT:    [[TMP20:%.*]] = inttoptr i64 [[TMP12]] to i8*
// CHECK12-NEXT:    store i8* [[TMP20]], i8** [[TMP19]], align 8
// CHECK12-NEXT:    [[TMP21:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK12-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, [10 x [10 x i32]]*, i64)* @__omp_outlined__7 to i8*), i8* null, i8** [[TMP21]], i64 4)
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK12:       omp.inner.for.inc:
// CHECK12-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK12-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK12-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP26]], [[TMP27]]
// CHECK12-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[CMP6:%.*]] = icmp sgt i32 [[TMP28]], 99
// CHECK12-NEXT:    br i1 [[CMP6]], label [[COND_TRUE7:%.*]], label [[COND_FALSE8:%.*]]
// CHECK12:       cond.true7:
// CHECK12-NEXT:    br label [[COND_END9:%.*]]
// CHECK12:       cond.false8:
// CHECK12-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    br label [[COND_END9]]
// CHECK12:       cond.end9:
// CHECK12-NEXT:    [[COND10:%.*]] = phi i32 [ 99, [[COND_TRUE7]] ], [ [[TMP29]], [[COND_FALSE8]] ]
// CHECK12-NEXT:    store i32 [[COND10]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    store i32 [[TMP30]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK12:       omp.inner.for.end:
// CHECK12-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK12:       omp.loop.exit:
// CHECK12-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]])
// CHECK12-NEXT:    ret void
// CHECK12-LABEL: define {{[^@]+}}@__omp_outlined__7
// CHECK12-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i64 [[F:%.*]]) #[[ATTR0]] {
// CHECK12-NEXT:  entry:
// CHECK12-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK12-NEXT:    [[F_ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK12-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK12-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK12-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK12-NEXT:    store i64 [[F]], i64* [[F_ADDR]], align 8
// CHECK12-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK12-NEXT:    [[CONV:%.*]] = bitcast i64* [[F_ADDR]] to i32*
// CHECK12-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK12-NEXT:    store i32 99, i32* [[DOTOMP_UB]], align 4
// CHECK12-NEXT:    [[TMP1:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK12-NEXT:    [[CONV2:%.*]] = trunc i64 [[TMP1]] to i32
// CHECK12-NEXT:    [[TMP2:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK12-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP2]] to i32
// CHECK12-NEXT:    store i32 [[CONV2]], i32* [[DOTOMP_LB]], align 4
// CHECK12-NEXT:    store i32 [[CONV3]], i32* [[DOTOMP_UB]], align 4
// CHECK12-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK12-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK12-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK12-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK12-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK12:       omp.inner.for.cond:
// CHECK12-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[CONV4:%.*]] = sext i32 [[TMP6]] to i64
// CHECK12-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK12-NEXT:    [[CMP:%.*]] = icmp ule i64 [[CONV4]], [[TMP7]]
// CHECK12-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK12:       omp.inner.for.body:
// CHECK12-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP8]], 10
// CHECK12-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK12-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK12-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK12-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[DIV5:%.*]] = sdiv i32 [[TMP10]], 10
// CHECK12-NEXT:    [[MUL6:%.*]] = mul nsw i32 [[DIV5]], 10
// CHECK12-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP9]], [[MUL6]]
// CHECK12-NEXT:    [[MUL7:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK12-NEXT:    [[ADD8:%.*]] = add nsw i32 0, [[MUL7]]
// CHECK12-NEXT:    store i32 [[ADD8]], i32* [[J]], align 4
// CHECK12-NEXT:    store i32 10, i32* [[K]], align 4
// CHECK12-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
// CHECK12-NEXT:    [[TMP12:%.*]] = load i32, i32* [[J]], align 4
// CHECK12-NEXT:    [[TMP13:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK12-NEXT:    [[MUL9:%.*]] = mul nsw i32 [[TMP12]], [[TMP13]]
// CHECK12-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP11]], [[MUL9]]
// CHECK12-NEXT:    [[TMP14:%.*]] = load i32, i32* [[K]], align 4
// CHECK12-NEXT:    [[ADD11:%.*]] = add nsw i32 [[ADD10]], [[TMP14]]
// CHECK12-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
// CHECK12-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP15]] to i64
// CHECK12-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK12-NEXT:    [[TMP16:%.*]] = load i32, i32* [[J]], align 4
// CHECK12-NEXT:    [[IDXPROM12:%.*]] = sext i32 [[TMP16]] to i64
// CHECK12-NEXT:    [[ARRAYIDX13:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i64 0, i64 [[IDXPROM12]]
// CHECK12-NEXT:    store i32 [[ADD11]], i32* [[ARRAYIDX13]], align 4
// CHECK12-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK12:       omp.body.continue:
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK12:       omp.inner.for.inc:
// CHECK12-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    [[ADD14:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK12-NEXT:    store i32 [[ADD14]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK12:       omp.inner.for.end:
// CHECK12-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK12:       omp.loop.exit:
// CHECK12-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK12-NEXT:    ret void
// CHECK12-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l67
// CHECK12-SAME: (i64 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK12-NEXT:  entry:
// CHECK12-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK12-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK12-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK12-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK12-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK12-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK12-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK12-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK12-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK12:       .execute:
// CHECK12-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK12-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK12-NEXT:    [[CONV1:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK12-NEXT:    store i32 [[TMP2]], i32* [[CONV1]], align 4
// CHECK12-NEXT:    [[TMP3:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK12-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK12-NEXT:    call void @__omp_outlined__8(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP3]], [10 x [10 x i32]]* [[TMP0]]) #[[ATTR3]]
// CHECK12-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK12:       .omp.deinit:
// CHECK12-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK12-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK12:       .exit:
// CHECK12-NEXT:    ret void
// CHECK12-LABEL: define {{[^@]+}}@__omp_outlined__8
// CHECK12-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK12-NEXT:  entry:
// CHECK12-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK12-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[I8:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[J9:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 8
// CHECK12-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK12-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK12-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK12-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK12-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK12-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK12-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK12-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK12-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK12-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK12-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK12-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK12-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK12-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK12-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK12-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], [[DIV5]]
// CHECK12-NEXT:    [[SUB6:%.*]] = sub nsw i32 [[MUL]], 1
// CHECK12-NEXT:    store i32 [[SUB6]], i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK12-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK12-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK12-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK12-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK12-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK12:       land.lhs.true:
// CHECK12-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK12-NEXT:    [[CMP7:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK12-NEXT:    br i1 [[CMP7]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK12:       omp.precond.then:
// CHECK12-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK12-NEXT:    store i32 [[TMP7]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK12-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK12-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    [[TMP9:%.*]] = load i32, i32* [[TMP8]], align 4
// CHECK12-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP9]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK12-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK12-NEXT:    [[CMP10:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
// CHECK12-NEXT:    br i1 [[CMP10]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK12:       cond.true:
// CHECK12-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK12-NEXT:    br label [[COND_END:%.*]]
// CHECK12:       cond.false:
// CHECK12-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    br label [[COND_END]]
// CHECK12:       cond.end:
// CHECK12-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
// CHECK12-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    store i32 [[TMP14]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK12:       omp.inner.for.cond:
// CHECK12-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK12-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP16]], 1
// CHECK12-NEXT:    [[CMP11:%.*]] = icmp slt i32 [[TMP15]], [[ADD]]
// CHECK12-NEXT:    br i1 [[CMP11]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK12:       omp.inner.for.body:
// CHECK12-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    [[TMP18:%.*]] = zext i32 [[TMP17]] to i64
// CHECK12-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
// CHECK12-NEXT:    [[TMP21:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK12-NEXT:    [[CONV12:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK12-NEXT:    store i32 [[TMP21]], i32* [[CONV12]], align 4
// CHECK12-NEXT:    [[TMP22:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK12-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK12-NEXT:    [[TMP24:%.*]] = inttoptr i64 [[TMP18]] to i8*
// CHECK12-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 8
// CHECK12-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK12-NEXT:    [[TMP26:%.*]] = inttoptr i64 [[TMP20]] to i8*
// CHECK12-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 8
// CHECK12-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK12-NEXT:    [[TMP28:%.*]] = inttoptr i64 [[TMP22]] to i8*
// CHECK12-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 8
// CHECK12-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK12-NEXT:    [[TMP30:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK12-NEXT:    store i8* [[TMP30]], i8** [[TMP29]], align 8
// CHECK12-NEXT:    [[TMP31:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    [[TMP32:%.*]] = load i32, i32* [[TMP31]], align 4
// CHECK12-NEXT:    [[TMP33:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK12-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP32]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, i64, [10 x [10 x i32]]*)* @__omp_outlined__9 to i8*), i8* null, i8** [[TMP33]], i64 4)
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK12:       omp.inner.for.inc:
// CHECK12-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    [[ADD13:%.*]] = add nsw i32 [[TMP34]], [[TMP35]]
// CHECK12-NEXT:    store i32 [[ADD13]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    [[ADD14:%.*]] = add nsw i32 [[TMP36]], [[TMP37]]
// CHECK12-NEXT:    store i32 [[ADD14]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    [[ADD15:%.*]] = add nsw i32 [[TMP38]], [[TMP39]]
// CHECK12-NEXT:    store i32 [[ADD15]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK12-NEXT:    [[CMP16:%.*]] = icmp sgt i32 [[TMP40]], [[TMP41]]
// CHECK12-NEXT:    br i1 [[CMP16]], label [[COND_TRUE17:%.*]], label [[COND_FALSE18:%.*]]
// CHECK12:       cond.true17:
// CHECK12-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK12-NEXT:    br label [[COND_END19:%.*]]
// CHECK12:       cond.false18:
// CHECK12-NEXT:    [[TMP43:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    br label [[COND_END19]]
// CHECK12:       cond.end19:
// CHECK12-NEXT:    [[COND20:%.*]] = phi i32 [ [[TMP42]], [[COND_TRUE17]] ], [ [[TMP43]], [[COND_FALSE18]] ]
// CHECK12-NEXT:    store i32 [[COND20]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP44:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    store i32 [[TMP44]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK12:       omp.inner.for.end:
// CHECK12-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK12:       omp.loop.exit:
// CHECK12-NEXT:    [[TMP45:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    [[TMP46:%.*]] = load i32, i32* [[TMP45]], align 4
// CHECK12-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP46]])
// CHECK12-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK12:       omp.precond.end:
// CHECK12-NEXT:    ret void
// CHECK12-LABEL: define {{[^@]+}}@__omp_outlined__9
// CHECK12-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], i64 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK12-NEXT:  entry:
// CHECK12-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK12-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[I10:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[J11:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK12-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK12-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK12-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK12-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK12-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK12-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK12-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK12-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK12-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK12-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK12-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK12-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK12-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK12-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK12-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK12-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK12-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], [[DIV5]]
// CHECK12-NEXT:    [[SUB6:%.*]] = sub nsw i32 [[MUL]], 1
// CHECK12-NEXT:    store i32 [[SUB6]], i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK12-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK12-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK12-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK12-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK12-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK12:       land.lhs.true:
// CHECK12-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK12-NEXT:    [[CMP7:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK12-NEXT:    br i1 [[CMP7]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK12:       omp.precond.then:
// CHECK12-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK12-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK12-NEXT:    store i32 [[TMP7]], i32* [[DOTOMP_UB]], align 4
// CHECK12-NEXT:    [[TMP8:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK12-NEXT:    [[CONV8:%.*]] = trunc i64 [[TMP8]] to i32
// CHECK12-NEXT:    [[TMP9:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK12-NEXT:    [[CONV9:%.*]] = trunc i64 [[TMP9]] to i32
// CHECK12-NEXT:    store i32 [[CONV8]], i32* [[DOTOMP_LB]], align 4
// CHECK12-NEXT:    store i32 [[CONV9]], i32* [[DOTOMP_UB]], align 4
// CHECK12-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK12-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK12-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP11]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK12-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK12-NEXT:    store i32 [[TMP12]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK12:       omp.inner.for.cond:
// CHECK12-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[CONV12:%.*]] = sext i32 [[TMP13]] to i64
// CHECK12-NEXT:    [[TMP14:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK12-NEXT:    [[CMP13:%.*]] = icmp ule i64 [[CONV12]], [[TMP14]]
// CHECK12-NEXT:    br i1 [[CMP13]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK12:       omp.inner.for.body:
// CHECK12-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK12-NEXT:    [[SUB14:%.*]] = sub nsw i32 [[TMP16]], 0
// CHECK12-NEXT:    [[DIV15:%.*]] = sdiv i32 [[SUB14]], 1
// CHECK12-NEXT:    [[MUL16:%.*]] = mul nsw i32 1, [[DIV15]]
// CHECK12-NEXT:    [[DIV17:%.*]] = sdiv i32 [[TMP15]], [[MUL16]]
// CHECK12-NEXT:    [[MUL18:%.*]] = mul nsw i32 [[DIV17]], 1
// CHECK12-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL18]]
// CHECK12-NEXT:    store i32 [[ADD]], i32* [[I10]], align 4
// CHECK12-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK12-NEXT:    [[SUB19:%.*]] = sub nsw i32 [[TMP19]], 0
// CHECK12-NEXT:    [[DIV20:%.*]] = sdiv i32 [[SUB19]], 1
// CHECK12-NEXT:    [[MUL21:%.*]] = mul nsw i32 1, [[DIV20]]
// CHECK12-NEXT:    [[DIV22:%.*]] = sdiv i32 [[TMP18]], [[MUL21]]
// CHECK12-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK12-NEXT:    [[SUB23:%.*]] = sub nsw i32 [[TMP20]], 0
// CHECK12-NEXT:    [[DIV24:%.*]] = sdiv i32 [[SUB23]], 1
// CHECK12-NEXT:    [[MUL25:%.*]] = mul nsw i32 1, [[DIV24]]
// CHECK12-NEXT:    [[MUL26:%.*]] = mul nsw i32 [[DIV22]], [[MUL25]]
// CHECK12-NEXT:    [[SUB27:%.*]] = sub nsw i32 [[TMP17]], [[MUL26]]
// CHECK12-NEXT:    [[MUL28:%.*]] = mul nsw i32 [[SUB27]], 1
// CHECK12-NEXT:    [[ADD29:%.*]] = add nsw i32 0, [[MUL28]]
// CHECK12-NEXT:    store i32 [[ADD29]], i32* [[J11]], align 4
// CHECK12-NEXT:    [[TMP21:%.*]] = load i32, i32* [[I10]], align 4
// CHECK12-NEXT:    [[TMP22:%.*]] = load i32, i32* [[J11]], align 4
// CHECK12-NEXT:    [[ADD30:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
// CHECK12-NEXT:    [[TMP23:%.*]] = load i32, i32* [[I10]], align 4
// CHECK12-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP23]] to i64
// CHECK12-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK12-NEXT:    [[TMP24:%.*]] = load i32, i32* [[J11]], align 4
// CHECK12-NEXT:    [[IDXPROM31:%.*]] = sext i32 [[TMP24]] to i64
// CHECK12-NEXT:    [[ARRAYIDX32:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i64 0, i64 [[IDXPROM31]]
// CHECK12-NEXT:    store i32 [[ADD30]], i32* [[ARRAYIDX32]], align 4
// CHECK12-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK12:       omp.body.continue:
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK12:       omp.inner.for.inc:
// CHECK12-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    [[ADD33:%.*]] = add nsw i32 [[TMP25]], [[TMP26]]
// CHECK12-NEXT:    store i32 [[ADD33]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK12:       omp.inner.for.end:
// CHECK12-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK12:       omp.loop.exit:
// CHECK12-NEXT:    [[TMP27:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    [[TMP28:%.*]] = load i32, i32* [[TMP27]], align 4
// CHECK12-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP28]])
// CHECK12-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK12:       omp.precond.end:
// CHECK12-NEXT:    ret void
// CHECK12-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l74
// CHECK12-SAME: (i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK12-NEXT:  entry:
// CHECK12-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK12-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK12-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK12-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK12-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 8
// CHECK12-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK12-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK12-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK12-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK12-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK12:       .execute:
// CHECK12-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK12-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK12-NEXT:    [[CONV1:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK12-NEXT:    store i32 [[TMP2]], i32* [[CONV1]], align 4
// CHECK12-NEXT:    [[TMP3:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK12-NEXT:    [[TMP4:%.*]] = load i32*, i32** [[V_ADDR]], align 8
// CHECK12-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK12-NEXT:    call void @__omp_outlined__10(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP3]], [1000 x i32]* [[TMP0]], i32* [[TMP4]]) #[[ATTR3]]
// CHECK12-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK12:       .omp.deinit:
// CHECK12-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK12-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK12:       .exit:
// CHECK12-NEXT:    ret void
// CHECK12-LABEL: define {{[^@]+}}@__omp_outlined__10
// CHECK12-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK12-NEXT:  entry:
// CHECK12-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK12-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 8
// CHECK12-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK12-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK12-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK12-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 8
// CHECK12-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK12-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK12-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK12-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK12-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK12-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK12-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK12-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK12-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK12-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK12-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK12-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK12-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK12:       omp.precond.then:
// CHECK12-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK12-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK12-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK12-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK12-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK12-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK12-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK12-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK12:       cond.true:
// CHECK12-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK12-NEXT:    br label [[COND_END:%.*]]
// CHECK12:       cond.false:
// CHECK12-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    br label [[COND_END]]
// CHECK12:       cond.end:
// CHECK12-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK12-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK12:       omp.inner.for.cond:
// CHECK12-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK12-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK12-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK12-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK12:       omp.inner.for.body:
// CHECK12-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    [[TMP15:%.*]] = zext i32 [[TMP14]] to i64
// CHECK12-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP17:%.*]] = zext i32 [[TMP16]] to i64
// CHECK12-NEXT:    [[TMP18:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK12-NEXT:    [[CONV6:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK12-NEXT:    store i32 [[TMP18]], i32* [[CONV6]], align 4
// CHECK12-NEXT:    [[TMP19:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK12-NEXT:    [[TMP20:%.*]] = load i32*, i32** [[V_ADDR]], align 8
// CHECK12-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK12-NEXT:    [[TMP22:%.*]] = inttoptr i64 [[TMP15]] to i8*
// CHECK12-NEXT:    store i8* [[TMP22]], i8** [[TMP21]], align 8
// CHECK12-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK12-NEXT:    [[TMP24:%.*]] = inttoptr i64 [[TMP17]] to i8*
// CHECK12-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 8
// CHECK12-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK12-NEXT:    [[TMP26:%.*]] = inttoptr i64 [[TMP19]] to i8*
// CHECK12-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 8
// CHECK12-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK12-NEXT:    [[TMP28:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK12-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 8
// CHECK12-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 4
// CHECK12-NEXT:    [[TMP30:%.*]] = bitcast i32* [[TMP20]] to i8*
// CHECK12-NEXT:    store i8* [[TMP30]], i8** [[TMP29]], align 8
// CHECK12-NEXT:    [[TMP31:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    [[TMP32:%.*]] = load i32, i32* [[TMP31]], align 4
// CHECK12-NEXT:    [[TMP33:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK12-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP32]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, i64, [1000 x i32]*, i32*)* @__omp_outlined__11 to i8*), i8* null, i8** [[TMP33]], i64 5)
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK12:       omp.inner.for.inc:
// CHECK12-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP34]], [[TMP35]]
// CHECK12-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP36]], [[TMP37]]
// CHECK12-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP38]], [[TMP39]]
// CHECK12-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK12-NEXT:    [[CMP10:%.*]] = icmp sgt i32 [[TMP40]], [[TMP41]]
// CHECK12-NEXT:    br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]]
// CHECK12:       cond.true11:
// CHECK12-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK12-NEXT:    br label [[COND_END13:%.*]]
// CHECK12:       cond.false12:
// CHECK12-NEXT:    [[TMP43:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    br label [[COND_END13]]
// CHECK12:       cond.end13:
// CHECK12-NEXT:    [[COND14:%.*]] = phi i32 [ [[TMP42]], [[COND_TRUE11]] ], [ [[TMP43]], [[COND_FALSE12]] ]
// CHECK12-NEXT:    store i32 [[COND14]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK12-NEXT:    [[TMP44:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK12-NEXT:    store i32 [[TMP44]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK12:       omp.inner.for.end:
// CHECK12-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK12:       omp.loop.exit:
// CHECK12-NEXT:    [[TMP45:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    [[TMP46:%.*]] = load i32, i32* [[TMP45]], align 4
// CHECK12-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP46]])
// CHECK12-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK12:       omp.precond.end:
// CHECK12-NEXT:    ret void
// CHECK12-LABEL: define {{[^@]+}}@__omp_outlined__11
// CHECK12-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK12-NEXT:  entry:
// CHECK12-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK12-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK12-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 8
// CHECK12-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    [[I5:%.*]] = alloca i32, align 4
// CHECK12-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK12-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK12-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK12-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK12-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK12-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 8
// CHECK12-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK12-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK12-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK12-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK12-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK12-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK12-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK12-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK12-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK12-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK12-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK12-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK12-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK12:       omp.precond.then:
// CHECK12-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK12-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK12-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK12-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK12-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP5]] to i32
// CHECK12-NEXT:    [[TMP6:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK12-NEXT:    [[CONV4:%.*]] = trunc i64 [[TMP6]] to i32
// CHECK12-NEXT:    store i32 [[CONV3]], i32* [[DOTOMP_LB]], align 4
// CHECK12-NEXT:    store i32 [[CONV4]], i32* [[DOTOMP_UB]], align 4
// CHECK12-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK12-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK12-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK12-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK12-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK12:       omp.inner.for.cond:
// CHECK12-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[CONV6:%.*]] = sext i32 [[TMP10]] to i64
// CHECK12-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK12-NEXT:    [[CMP7:%.*]] = icmp ule i64 [[CONV6]], [[TMP11]]
// CHECK12-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK12:       omp.inner.for.body:
// CHECK12-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK12-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK12-NEXT:    store i32 [[ADD]], i32* [[I5]], align 4
// CHECK12-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[V_ADDR]], align 8
// CHECK12-NEXT:    [[TMP14:%.*]] = load i32, i32* [[I5]], align 4
// CHECK12-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP14]] to i64
// CHECK12-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[TMP13]], i64 [[IDXPROM]]
// CHECK12-NEXT:    [[TMP15:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK12-NEXT:    [[TMP16:%.*]] = load i32, i32* [[I5]], align 4
// CHECK12-NEXT:    [[IDXPROM8:%.*]] = sext i32 [[TMP16]] to i64
// CHECK12-NEXT:    [[ARRAYIDX9:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i64 0, i64 [[IDXPROM8]]
// CHECK12-NEXT:    store i32 [[TMP15]], i32* [[ARRAYIDX9]], align 4
// CHECK12-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK12:       omp.body.continue:
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK12:       omp.inner.for.inc:
// CHECK12-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK12-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK12-NEXT:    store i32 [[ADD10]], i32* [[DOTOMP_IV]], align 4
// CHECK12-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK12:       omp.inner.for.end:
// CHECK12-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK12:       omp.loop.exit:
// CHECK12-NEXT:    [[TMP19:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK12-NEXT:    [[TMP20:%.*]] = load i32, i32* [[TMP19]], align 4
// CHECK12-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP20]])
// CHECK12-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK12:       omp.precond.end:
// CHECK12-NEXT:    ret void
// CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l43
// CHECK13-SAME: (i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK13-NEXT:  entry:
// CHECK13-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK13-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[L_CASTED:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK13-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK13-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK13-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK13-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK13:       .execute:
// CHECK13-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3:[0-9]+]])
// CHECK13-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK13-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK13-NEXT:    [[TMP4:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP4]], i32* [[L_CASTED]], align 4
// CHECK13-NEXT:    [[TMP5:%.*]] = load i32, i32* [[L_CASTED]], align 4
// CHECK13-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK13-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [1000 x i32]* [[TMP0]], i32 [[TMP5]]) #[[ATTR3:[0-9]+]]
// CHECK13-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK13:       .omp.deinit:
// CHECK13-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK13-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK13:       .exit:
// CHECK13-NEXT:    ret void
// CHECK13-LABEL: define {{[^@]+}}@__omp_outlined__
// CHECK13-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0]] {
// CHECK13-NEXT:  entry:
// CHECK13-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK13-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[I4:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[L_CASTED:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 4
// CHECK13-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK13-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK13-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK13-NEXT:    [[TMP1:%.*]] = load i16, i16* @"_openmp_static_kernel$is_shared", align 2
// CHECK13-NEXT:    [[TMP2:%.*]] = load i32, i32* @"_openmp_static_kernel$size", align 4
// CHECK13-NEXT:    call void @__kmpc_get_team_static_memory(i16 1, i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds (%"union._shared_openmp_static_memory_type_$_", %"union._shared_openmp_static_memory_type_$_" addrspace(3)* @"_openmp_shared_static_glob_rd_$_", i32 0, i32 0, i32 0) to i8*), i32 [[TMP2]], i16 [[TMP1]], i8** addrspacecast (i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr" to i8**))
// CHECK13-NEXT:    [[TMP3:%.*]] = load i8*, i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr", align 4
// CHECK13-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i8, i8* [[TMP3]], i32 0
// CHECK13-NEXT:    [[TMP5:%.*]] = bitcast i8* [[TMP4]] to %struct._globalized_locals_ty*
// CHECK13-NEXT:    [[L1:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY:%.*]], %struct._globalized_locals_ty* [[TMP5]], i32 0, i32 0
// CHECK13-NEXT:    [[TMP6:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP6]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK13-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK13-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], 0
// CHECK13-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK13-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK13-NEXT:    store i32 [[SUB3]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK13-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK13-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK13-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP8]]
// CHECK13-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK13:       omp.precond.then:
// CHECK13-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK13-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK13-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK13-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1:[0-9]+]], i32 [[TMP11]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 128)
// CHECK13-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK13-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
// CHECK13-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK13:       cond.true:
// CHECK13-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK13-NEXT:    br label [[COND_END:%.*]]
// CHECK13:       cond.false:
// CHECK13-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    br label [[COND_END]]
// CHECK13:       cond.end:
// CHECK13-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
// CHECK13-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    store i32 [[TMP16]], i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK13:       omp.inner.for.cond:
// CHECK13-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK13-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP18]], 1
// CHECK13-NEXT:    [[CMP6:%.*]] = icmp slt i32 [[TMP17]], [[ADD]]
// CHECK13-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK13:       omp.inner.for.body:
// CHECK13-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP21:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP21]], i32* [[N_CASTED]], align 4
// CHECK13-NEXT:    [[TMP22:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK13-NEXT:    [[TMP23:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP23]], i32* [[L_CASTED]], align 4
// CHECK13-NEXT:    [[TMP24:%.*]] = load i32, i32* [[L_CASTED]], align 4
// CHECK13-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK13-NEXT:    [[TMP26:%.*]] = inttoptr i32 [[TMP19]] to i8*
// CHECK13-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 4
// CHECK13-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK13-NEXT:    [[TMP28:%.*]] = inttoptr i32 [[TMP20]] to i8*
// CHECK13-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 4
// CHECK13-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK13-NEXT:    [[TMP30:%.*]] = inttoptr i32 [[TMP22]] to i8*
// CHECK13-NEXT:    store i8* [[TMP30]], i8** [[TMP29]], align 4
// CHECK13-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK13-NEXT:    [[TMP32:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK13-NEXT:    store i8* [[TMP32]], i8** [[TMP31]], align 4
// CHECK13-NEXT:    [[TMP33:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 4
// CHECK13-NEXT:    [[TMP34:%.*]] = inttoptr i32 [[TMP24]] to i8*
// CHECK13-NEXT:    store i8* [[TMP34]], i8** [[TMP33]], align 4
// CHECK13-NEXT:    [[TMP35:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    [[TMP36:%.*]] = load i32, i32* [[TMP35]], align 4
// CHECK13-NEXT:    [[TMP37:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK13-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP36]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i32]*, i32)* @__omp_outlined__1 to i8*), i8* null, i8** [[TMP37]], i32 5)
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK13:       omp.inner.for.inc:
// CHECK13-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP38]], [[TMP39]]
// CHECK13-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP40]], [[TMP41]]
// CHECK13-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP43:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP42]], [[TMP43]]
// CHECK13-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP44:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP45:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK13-NEXT:    [[CMP10:%.*]] = icmp sgt i32 [[TMP44]], [[TMP45]]
// CHECK13-NEXT:    br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]]
// CHECK13:       cond.true11:
// CHECK13-NEXT:    [[TMP46:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK13-NEXT:    br label [[COND_END13:%.*]]
// CHECK13:       cond.false12:
// CHECK13-NEXT:    [[TMP47:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    br label [[COND_END13]]
// CHECK13:       cond.end13:
// CHECK13-NEXT:    [[COND14:%.*]] = phi i32 [ [[TMP46]], [[COND_TRUE11]] ], [ [[TMP47]], [[COND_FALSE12]] ]
// CHECK13-NEXT:    store i32 [[COND14]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP48:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    store i32 [[TMP48]], i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK13:       omp.inner.for.end:
// CHECK13-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK13:       omp.loop.exit:
// CHECK13-NEXT:    [[TMP49:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    [[TMP50:%.*]] = load i32, i32* [[TMP49]], align 4
// CHECK13-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP50]])
// CHECK13-NEXT:    [[TMP51:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK13-NEXT:    [[TMP52:%.*]] = icmp ne i32 [[TMP51]], 0
// CHECK13-NEXT:    br i1 [[TMP52]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK13:       .omp.lastprivate.then:
// CHECK13-NEXT:    [[TMP53:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP53]], i32* [[L_ADDR]], align 4
// CHECK13-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK13:       .omp.lastprivate.done:
// CHECK13-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK13:       omp.precond.end:
// CHECK13-NEXT:    [[TMP54:%.*]] = load i16, i16* @"_openmp_static_kernel$is_shared", align 2
// CHECK13-NEXT:    call void @__kmpc_restore_team_static_memory(i16 1, i16 [[TMP54]])
// CHECK13-NEXT:    ret void
// CHECK13-LABEL: define {{[^@]+}}@__omp_outlined__1
// CHECK13-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0]] {
// CHECK13-NEXT:  entry:
// CHECK13-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK13-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK13-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK13-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK13-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK13-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK13-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK13-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK13-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK13-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK13-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK13-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK13-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK13-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK13-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK13-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK13:       omp.precond.then:
// CHECK13-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK13-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK13-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK13-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK13-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK13-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK13-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK13-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK13-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 32)
// CHECK13-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK13:       omp.dispatch.cond:
// CHECK13-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK13-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK13-NEXT:    [[CMP4:%.*]] = icmp ugt i32 [[TMP9]], [[TMP10]]
// CHECK13-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK13:       cond.true:
// CHECK13-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK13-NEXT:    br label [[COND_END:%.*]]
// CHECK13:       cond.false:
// CHECK13-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK13-NEXT:    br label [[COND_END]]
// CHECK13:       cond.end:
// CHECK13-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP11]], [[COND_TRUE]] ], [ [[TMP12]], [[COND_FALSE]] ]
// CHECK13-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
// CHECK13-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK13-NEXT:    store i32 [[TMP13]], i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK13-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK13-NEXT:    br i1 [[CMP5]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK13:       omp.dispatch.body:
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK13:       omp.inner.for.cond:
// CHECK13-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK13-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP16]], [[TMP17]]
// CHECK13-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK13:       omp.inner.for.body:
// CHECK13-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP18]], 1
// CHECK13-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK13-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK13-NEXT:    [[TMP19:%.*]] = load i32, i32* [[I3]], align 4
// CHECK13-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i32 0, i32 [[TMP19]]
// CHECK13-NEXT:    store i32 1, i32* [[ARRAYIDX]], align 4
// CHECK13-NEXT:    [[TMP20:%.*]] = load i32, i32* [[I3]], align 4
// CHECK13-NEXT:    store i32 [[TMP20]], i32* [[L_ADDR]], align 4
// CHECK13-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK13:       omp.body.continue:
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK13:       omp.inner.for.inc:
// CHECK13-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP21]], 1
// CHECK13-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK13:       omp.inner.for.end:
// CHECK13-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK13:       omp.dispatch.inc:
// CHECK13-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK13-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK13-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_LB]], align 4
// CHECK13-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK13-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK13-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_UB]], align 4
// CHECK13-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK13:       omp.dispatch.end:
// CHECK13-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK13-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP27]])
// CHECK13-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK13-NEXT:    [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
// CHECK13-NEXT:    br i1 [[TMP29]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK13:       .omp.lastprivate.then:
// CHECK13-NEXT:    [[TMP30:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP30]], i32* [[L_ADDR]], align 4
// CHECK13-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK13:       .omp.lastprivate.done:
// CHECK13-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK13:       omp.precond.end:
// CHECK13-NEXT:    ret void
// CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l49
// CHECK13-SAME: (i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK13-NEXT:  entry:
// CHECK13-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK13-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK13-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK13-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK13-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK13-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK13:       .execute:
// CHECK13-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK13-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK13-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK13-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK13-NEXT:    call void @__omp_outlined__2(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [1000 x i16]* [[TMP0]]) #[[ATTR3]]
// CHECK13-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK13:       .omp.deinit:
// CHECK13-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK13-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK13:       .exit:
// CHECK13-NEXT:    ret void
// CHECK13-LABEL: define {{[^@]+}}@__omp_outlined__2
// CHECK13-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK13-NEXT:  entry:
// CHECK13-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK13-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK13-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK13-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK13-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK13-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK13-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK13-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK13-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK13-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK13-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK13-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK13-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK13-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK13-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK13:       omp.precond.then:
// CHECK13-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK13-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK13-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK13-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK13-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK13-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK13-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK13-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK13:       cond.true:
// CHECK13-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK13-NEXT:    br label [[COND_END:%.*]]
// CHECK13:       cond.false:
// CHECK13-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    br label [[COND_END]]
// CHECK13:       cond.end:
// CHECK13-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK13-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK13:       omp.inner.for.cond:
// CHECK13-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK13-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK13-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK13-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK13:       omp.inner.for.body:
// CHECK13-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP16:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP16]], i32* [[N_CASTED]], align 4
// CHECK13-NEXT:    [[TMP17:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK13-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK13-NEXT:    [[TMP19:%.*]] = inttoptr i32 [[TMP14]] to i8*
// CHECK13-NEXT:    store i8* [[TMP19]], i8** [[TMP18]], align 4
// CHECK13-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK13-NEXT:    [[TMP21:%.*]] = inttoptr i32 [[TMP15]] to i8*
// CHECK13-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 4
// CHECK13-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK13-NEXT:    [[TMP23:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK13-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 4
// CHECK13-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK13-NEXT:    [[TMP25:%.*]] = bitcast [1000 x i16]* [[TMP0]] to i8*
// CHECK13-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 4
// CHECK13-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK13-NEXT:    [[TMP28:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK13-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP27]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i16]*)* @__omp_outlined__3 to i8*), i8* null, i8** [[TMP28]], i32 4)
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK13:       omp.inner.for.inc:
// CHECK13-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP29]], [[TMP30]]
// CHECK13-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP31]], [[TMP32]]
// CHECK13-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP33]], [[TMP34]]
// CHECK13-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK13-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP35]], [[TMP36]]
// CHECK13-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
// CHECK13:       cond.true10:
// CHECK13-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK13-NEXT:    br label [[COND_END12:%.*]]
// CHECK13:       cond.false11:
// CHECK13-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    br label [[COND_END12]]
// CHECK13:       cond.end12:
// CHECK13-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP37]], [[COND_TRUE10]] ], [ [[TMP38]], [[COND_FALSE11]] ]
// CHECK13-NEXT:    store i32 [[COND13]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    store i32 [[TMP39]], i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK13:       omp.inner.for.end:
// CHECK13-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK13:       omp.loop.exit:
// CHECK13-NEXT:    [[TMP40:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    [[TMP41:%.*]] = load i32, i32* [[TMP40]], align 4
// CHECK13-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP41]])
// CHECK13-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK13:       omp.precond.end:
// CHECK13-NEXT:    ret void
// CHECK13-LABEL: define {{[^@]+}}@__omp_outlined__3
// CHECK13-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK13-NEXT:  entry:
// CHECK13-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK13-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK13-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK13-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK13-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK13-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK13-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK13-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK13-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK13-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK13-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK13-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK13-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK13-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK13-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK13-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK13:       omp.precond.then:
// CHECK13-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK13-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK13-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK13-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK13-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK13-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK13-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK13-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK13-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK13-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK13-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK13:       omp.inner.for.cond:
// CHECK13-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK13-NEXT:    [[CMP4:%.*]] = icmp ule i32 [[TMP10]], [[TMP11]]
// CHECK13-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK13:       omp.inner.for.body:
// CHECK13-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK13-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK13-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK13-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I3]], align 4
// CHECK13-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i16], [1000 x i16]* [[TMP0]], i32 0, i32 [[TMP13]]
// CHECK13-NEXT:    [[TMP14:%.*]] = load i16, i16* [[ARRAYIDX]], align 2
// CHECK13-NEXT:    [[CONV:%.*]] = sext i16 [[TMP14]] to i32
// CHECK13-NEXT:    [[ADD5:%.*]] = add nsw i32 [[CONV]], 1
// CHECK13-NEXT:    [[CONV6:%.*]] = trunc i32 [[ADD5]] to i16
// CHECK13-NEXT:    store i16 [[CONV6]], i16* [[ARRAYIDX]], align 2
// CHECK13-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK13:       omp.body.continue:
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK13:       omp.inner.for.inc:
// CHECK13-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP15]], [[TMP16]]
// CHECK13-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK13:       omp.inner.for.end:
// CHECK13-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK13:       omp.loop.exit:
// CHECK13-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
// CHECK13-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP18]])
// CHECK13-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK13:       omp.precond.end:
// CHECK13-NEXT:    ret void
// CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l54
// CHECK13-SAME: ([10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK13-NEXT:  entry:
// CHECK13-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK13-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK13-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK13-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK13-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK13-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK13-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK13:       .execute:
// CHECK13-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK13-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK13-NEXT:    call void @__omp_outlined__4(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x i32]* [[TMP0]]) #[[ATTR3]]
// CHECK13-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK13:       .omp.deinit:
// CHECK13-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK13-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK13:       .exit:
// CHECK13-NEXT:    ret void
// CHECK13-LABEL: define {{[^@]+}}@__omp_outlined__4
// CHECK13-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK13-NEXT:  entry:
// CHECK13-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK13-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x i8*], align 4
// CHECK13-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK13-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK13-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK13-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    store i32 9, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK13-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK13-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK13-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK13-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9
// CHECK13-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK13:       cond.true:
// CHECK13-NEXT:    br label [[COND_END:%.*]]
// CHECK13:       cond.false:
// CHECK13-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    br label [[COND_END]]
// CHECK13:       cond.end:
// CHECK13-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK13-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK13:       omp.inner.for.cond:
// CHECK13-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[CMP1:%.*]] = icmp slt i32 [[TMP6]], 10
// CHECK13-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK13:       omp.inner.for.body:
// CHECK13-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK13-NEXT:    [[TMP10:%.*]] = inttoptr i32 [[TMP7]] to i8*
// CHECK13-NEXT:    store i8* [[TMP10]], i8** [[TMP9]], align 4
// CHECK13-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK13-NEXT:    [[TMP12:%.*]] = inttoptr i32 [[TMP8]] to i8*
// CHECK13-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 4
// CHECK13-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK13-NEXT:    [[TMP14:%.*]] = bitcast [10 x i32]* [[TMP0]] to i8*
// CHECK13-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 4
// CHECK13-NEXT:    [[TMP15:%.*]] = bitcast [3 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK13-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, [10 x i32]*)* @__omp_outlined__5 to i8*), i8* null, i8** [[TMP15]], i32 3)
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK13:       omp.inner.for.inc:
// CHECK13-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP16]], [[TMP17]]
// CHECK13-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP18]], [[TMP19]]
// CHECK13-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK13-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP22]], 9
// CHECK13-NEXT:    br i1 [[CMP4]], label [[COND_TRUE5:%.*]], label [[COND_FALSE6:%.*]]
// CHECK13:       cond.true5:
// CHECK13-NEXT:    br label [[COND_END7:%.*]]
// CHECK13:       cond.false6:
// CHECK13-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    br label [[COND_END7]]
// CHECK13:       cond.end7:
// CHECK13-NEXT:    [[COND8:%.*]] = phi i32 [ 9, [[COND_TRUE5]] ], [ [[TMP23]], [[COND_FALSE6]] ]
// CHECK13-NEXT:    store i32 [[COND8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    store i32 [[TMP24]], i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK13:       omp.inner.for.end:
// CHECK13-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK13:       omp.loop.exit:
// CHECK13-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]])
// CHECK13-NEXT:    ret void
// CHECK13-LABEL: define {{[^@]+}}@__omp_outlined__5
// CHECK13-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK13-NEXT:  entry:
// CHECK13-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK13-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK13-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK13-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK13-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK13-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK13-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK13-NEXT:    store i32 9, i32* [[DOTOMP_UB]], align 4
// CHECK13-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK13-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP1]], i32* [[DOTOMP_LB]], align 4
// CHECK13-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_UB]], align 4
// CHECK13-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK13-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK13-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK13-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK13-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK13:       omp.inner.for.cond:
// CHECK13-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK13-NEXT:    [[CMP:%.*]] = icmp ule i32 [[TMP6]], [[TMP7]]
// CHECK13-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK13:       omp.inner.for.body:
// CHECK13-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP8]], 1
// CHECK13-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK13-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK13-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4
// CHECK13-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[TMP0]], i32 0, i32 [[TMP9]]
// CHECK13-NEXT:    [[TMP10:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK13-NEXT:    [[ADD1:%.*]] = add nsw i32 [[TMP10]], 1
// CHECK13-NEXT:    store i32 [[ADD1]], i32* [[ARRAYIDX]], align 4
// CHECK13-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK13:       omp.body.continue:
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK13:       omp.inner.for.inc:
// CHECK13-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
// CHECK13-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK13:       omp.inner.for.end:
// CHECK13-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK13:       omp.loop.exit:
// CHECK13-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK13-NEXT:    ret void
// CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l59
// CHECK13-SAME: ([10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK13-NEXT:  entry:
// CHECK13-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK13-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[F_CASTED:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK13-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK13-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK13-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK13-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK13-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK13:       .execute:
// CHECK13-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK13-NEXT:    [[TMP2:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP2]], i32* [[F_CASTED]], align 4
// CHECK13-NEXT:    [[TMP3:%.*]] = load i32, i32* [[F_CASTED]], align 4
// CHECK13-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK13-NEXT:    call void @__omp_outlined__6(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x [10 x i32]]* [[TMP0]], i32 [[TMP3]]) #[[ATTR3]]
// CHECK13-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK13:       .omp.deinit:
// CHECK13-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK13-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK13:       .exit:
// CHECK13-NEXT:    ret void
// CHECK13-LABEL: define {{[^@]+}}@__omp_outlined__6
// CHECK13-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK13-NEXT:  entry:
// CHECK13-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK13-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[F_CASTED:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK13-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK13-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK13-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK13-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    store i32 99, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK13-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK13-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK13-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK13-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 99
// CHECK13-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK13:       cond.true:
// CHECK13-NEXT:    br label [[COND_END:%.*]]
// CHECK13:       cond.false:
// CHECK13-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    br label [[COND_END]]
// CHECK13:       cond.end:
// CHECK13-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK13-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK13:       omp.inner.for.cond:
// CHECK13-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[TMP6]], 100
// CHECK13-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK13:       omp.inner.for.body:
// CHECK13-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP9:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP9]], i32* [[F_CASTED]], align 4
// CHECK13-NEXT:    [[TMP10:%.*]] = load i32, i32* [[F_CASTED]], align 4
// CHECK13-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK13-NEXT:    [[TMP12:%.*]] = inttoptr i32 [[TMP7]] to i8*
// CHECK13-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 4
// CHECK13-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK13-NEXT:    [[TMP14:%.*]] = inttoptr i32 [[TMP8]] to i8*
// CHECK13-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 4
// CHECK13-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK13-NEXT:    [[TMP16:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK13-NEXT:    store i8* [[TMP16]], i8** [[TMP15]], align 4
// CHECK13-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK13-NEXT:    [[TMP18:%.*]] = inttoptr i32 [[TMP10]] to i8*
// CHECK13-NEXT:    store i8* [[TMP18]], i8** [[TMP17]], align 4
// CHECK13-NEXT:    [[TMP19:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK13-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, [10 x [10 x i32]]*, i32)* @__omp_outlined__7 to i8*), i8* null, i8** [[TMP19]], i32 4)
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK13:       omp.inner.for.inc:
// CHECK13-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK13-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK13-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK13-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP26]], 99
// CHECK13-NEXT:    br i1 [[CMP5]], label [[COND_TRUE6:%.*]], label [[COND_FALSE7:%.*]]
// CHECK13:       cond.true6:
// CHECK13-NEXT:    br label [[COND_END8:%.*]]
// CHECK13:       cond.false7:
// CHECK13-NEXT:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    br label [[COND_END8]]
// CHECK13:       cond.end8:
// CHECK13-NEXT:    [[COND9:%.*]] = phi i32 [ 99, [[COND_TRUE6]] ], [ [[TMP27]], [[COND_FALSE7]] ]
// CHECK13-NEXT:    store i32 [[COND9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    store i32 [[TMP28]], i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK13:       omp.inner.for.end:
// CHECK13-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK13:       omp.loop.exit:
// CHECK13-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]])
// CHECK13-NEXT:    ret void
// CHECK13-LABEL: define {{[^@]+}}@__omp_outlined__7
// CHECK13-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK13-NEXT:  entry:
// CHECK13-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK13-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK13-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK13-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK13-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK13-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK13-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK13-NEXT:    store i32 99, i32* [[DOTOMP_UB]], align 4
// CHECK13-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK13-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP1]], i32* [[DOTOMP_LB]], align 4
// CHECK13-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_UB]], align 4
// CHECK13-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK13-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK13-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK13-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK13-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK13:       omp.inner.for.cond:
// CHECK13-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK13-NEXT:    [[CMP:%.*]] = icmp ule i32 [[TMP6]], [[TMP7]]
// CHECK13-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK13:       omp.inner.for.body:
// CHECK13-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP8]], 10
// CHECK13-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK13-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK13-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK13-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[DIV2:%.*]] = sdiv i32 [[TMP10]], 10
// CHECK13-NEXT:    [[MUL3:%.*]] = mul nsw i32 [[DIV2]], 10
// CHECK13-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP9]], [[MUL3]]
// CHECK13-NEXT:    [[MUL4:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK13-NEXT:    [[ADD5:%.*]] = add nsw i32 0, [[MUL4]]
// CHECK13-NEXT:    store i32 [[ADD5]], i32* [[J]], align 4
// CHECK13-NEXT:    store i32 10, i32* [[K]], align 4
// CHECK13-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
// CHECK13-NEXT:    [[TMP12:%.*]] = load i32, i32* [[J]], align 4
// CHECK13-NEXT:    [[TMP13:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK13-NEXT:    [[MUL6:%.*]] = mul nsw i32 [[TMP12]], [[TMP13]]
// CHECK13-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP11]], [[MUL6]]
// CHECK13-NEXT:    [[TMP14:%.*]] = load i32, i32* [[K]], align 4
// CHECK13-NEXT:    [[ADD8:%.*]] = add nsw i32 [[ADD7]], [[TMP14]]
// CHECK13-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
// CHECK13-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i32 0, i32 [[TMP15]]
// CHECK13-NEXT:    [[TMP16:%.*]] = load i32, i32* [[J]], align 4
// CHECK13-NEXT:    [[ARRAYIDX9:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i32 0, i32 [[TMP16]]
// CHECK13-NEXT:    store i32 [[ADD8]], i32* [[ARRAYIDX9]], align 4
// CHECK13-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK13:       omp.body.continue:
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK13:       omp.inner.for.inc:
// CHECK13-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK13-NEXT:    store i32 [[ADD10]], i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK13:       omp.inner.for.end:
// CHECK13-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK13:       omp.loop.exit:
// CHECK13-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK13-NEXT:    ret void
// CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l67
// CHECK13-SAME: (i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK13-NEXT:  entry:
// CHECK13-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK13-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK13-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK13-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK13-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK13-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK13:       .execute:
// CHECK13-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK13-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK13-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK13-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK13-NEXT:    call void @__omp_outlined__8(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [10 x [10 x i32]]* [[TMP0]]) #[[ATTR3]]
// CHECK13-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK13:       .omp.deinit:
// CHECK13-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK13-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK13:       .exit:
// CHECK13-NEXT:    ret void
// CHECK13-LABEL: define {{[^@]+}}@__omp_outlined__8
// CHECK13-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK13-NEXT:  entry:
// CHECK13-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK13-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK13-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK13-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i64, align 8
// CHECK13-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i64, align 8
// CHECK13-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK13-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[I9:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[J10:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK13-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK13-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK13-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK13-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK13-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK13-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK13-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK13-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK13-NEXT:    [[CONV:%.*]] = sext i32 [[DIV]] to i64
// CHECK13-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK13-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK13-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK13-NEXT:    [[CONV6:%.*]] = sext i32 [[DIV5]] to i64
// CHECK13-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV]], [[CONV6]]
// CHECK13-NEXT:    [[SUB7:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK13-NEXT:    store i64 [[SUB7]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK13-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK13-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK13-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK13-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK13-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK13:       land.lhs.true:
// CHECK13-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK13-NEXT:    [[CMP8:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK13-NEXT:    br i1 [[CMP8]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK13:       omp.precond.then:
// CHECK13-NEXT:    store i64 0, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK13-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK13-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK13-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK13-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK13-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK13-NEXT:    [[CONV11:%.*]] = zext i32 [[NVPTX_NUM_THREADS]] to i64
// CHECK13-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    [[TMP9:%.*]] = load i32, i32* [[TMP8]], align 4
// CHECK13-NEXT:    call void @__kmpc_for_static_init_8(%struct.ident_t* @[[GLOB1]], i32 [[TMP9]], i32 91, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_COMB_LB]], i64* [[DOTOMP_COMB_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 [[CONV11]])
// CHECK13-NEXT:    [[TMP10:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK13-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK13-NEXT:    [[CMP12:%.*]] = icmp sgt i64 [[TMP10]], [[TMP11]]
// CHECK13-NEXT:    br i1 [[CMP12]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK13:       cond.true:
// CHECK13-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK13-NEXT:    br label [[COND_END:%.*]]
// CHECK13:       cond.false:
// CHECK13-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK13-NEXT:    br label [[COND_END]]
// CHECK13:       cond.end:
// CHECK13-NEXT:    [[COND:%.*]] = phi i64 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
// CHECK13-NEXT:    store i64 [[COND]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK13-NEXT:    [[TMP14:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK13-NEXT:    store i64 [[TMP14]], i64* [[DOTOMP_IV]], align 8
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK13:       omp.inner.for.cond:
// CHECK13-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK13-NEXT:    [[TMP16:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK13-NEXT:    [[ADD:%.*]] = add nsw i64 [[TMP16]], 1
// CHECK13-NEXT:    [[CMP13:%.*]] = icmp slt i64 [[TMP15]], [[ADD]]
// CHECK13-NEXT:    br i1 [[CMP13]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK13:       omp.inner.for.body:
// CHECK13-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK13-NEXT:    [[TMP18:%.*]] = trunc i64 [[TMP17]] to i32
// CHECK13-NEXT:    [[TMP19:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK13-NEXT:    [[TMP20:%.*]] = trunc i64 [[TMP19]] to i32
// CHECK13-NEXT:    [[TMP21:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP21]], i32* [[N_CASTED]], align 4
// CHECK13-NEXT:    [[TMP22:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK13-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK13-NEXT:    [[TMP24:%.*]] = inttoptr i32 [[TMP18]] to i8*
// CHECK13-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 4
// CHECK13-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK13-NEXT:    [[TMP26:%.*]] = inttoptr i32 [[TMP20]] to i8*
// CHECK13-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 4
// CHECK13-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK13-NEXT:    [[TMP28:%.*]] = inttoptr i32 [[TMP22]] to i8*
// CHECK13-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 4
// CHECK13-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK13-NEXT:    [[TMP30:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK13-NEXT:    store i8* [[TMP30]], i8** [[TMP29]], align 4
// CHECK13-NEXT:    [[TMP31:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    [[TMP32:%.*]] = load i32, i32* [[TMP31]], align 4
// CHECK13-NEXT:    [[TMP33:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK13-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP32]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [10 x [10 x i32]]*)* @__omp_outlined__9 to i8*), i8* null, i8** [[TMP33]], i32 4)
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK13:       omp.inner.for.inc:
// CHECK13-NEXT:    [[TMP34:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK13-NEXT:    [[TMP35:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK13-NEXT:    [[ADD14:%.*]] = add nsw i64 [[TMP34]], [[TMP35]]
// CHECK13-NEXT:    store i64 [[ADD14]], i64* [[DOTOMP_IV]], align 8
// CHECK13-NEXT:    [[TMP36:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK13-NEXT:    [[TMP37:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK13-NEXT:    [[ADD15:%.*]] = add nsw i64 [[TMP36]], [[TMP37]]
// CHECK13-NEXT:    store i64 [[ADD15]], i64* [[DOTOMP_COMB_LB]], align 8
// CHECK13-NEXT:    [[TMP38:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK13-NEXT:    [[TMP39:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK13-NEXT:    [[ADD16:%.*]] = add nsw i64 [[TMP38]], [[TMP39]]
// CHECK13-NEXT:    store i64 [[ADD16]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK13-NEXT:    [[TMP40:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK13-NEXT:    [[TMP41:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK13-NEXT:    [[CMP17:%.*]] = icmp sgt i64 [[TMP40]], [[TMP41]]
// CHECK13-NEXT:    br i1 [[CMP17]], label [[COND_TRUE18:%.*]], label [[COND_FALSE19:%.*]]
// CHECK13:       cond.true18:
// CHECK13-NEXT:    [[TMP42:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK13-NEXT:    br label [[COND_END20:%.*]]
// CHECK13:       cond.false19:
// CHECK13-NEXT:    [[TMP43:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK13-NEXT:    br label [[COND_END20]]
// CHECK13:       cond.end20:
// CHECK13-NEXT:    [[COND21:%.*]] = phi i64 [ [[TMP42]], [[COND_TRUE18]] ], [ [[TMP43]], [[COND_FALSE19]] ]
// CHECK13-NEXT:    store i64 [[COND21]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK13-NEXT:    [[TMP44:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK13-NEXT:    store i64 [[TMP44]], i64* [[DOTOMP_IV]], align 8
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK13:       omp.inner.for.end:
// CHECK13-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK13:       omp.loop.exit:
// CHECK13-NEXT:    [[TMP45:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    [[TMP46:%.*]] = load i32, i32* [[TMP45]], align 4
// CHECK13-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP46]])
// CHECK13-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK13:       omp.precond.end:
// CHECK13-NEXT:    ret void
// CHECK13-LABEL: define {{[^@]+}}@__omp_outlined__9
// CHECK13-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK13-NEXT:  entry:
// CHECK13-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK13-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK13-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK13-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_LB:%.*]] = alloca i64, align 8
// CHECK13-NEXT:    [[DOTOMP_UB:%.*]] = alloca i64, align 8
// CHECK13-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK13-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[I11:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[J12:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK13-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK13-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK13-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK13-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK13-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK13-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK13-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK13-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK13-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK13-NEXT:    [[CONV:%.*]] = sext i32 [[DIV]] to i64
// CHECK13-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK13-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK13-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK13-NEXT:    [[CONV6:%.*]] = sext i32 [[DIV5]] to i64
// CHECK13-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV]], [[CONV6]]
// CHECK13-NEXT:    [[SUB7:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK13-NEXT:    store i64 [[SUB7]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK13-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK13-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK13-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK13-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK13-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK13:       land.lhs.true:
// CHECK13-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK13-NEXT:    [[CMP8:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK13-NEXT:    br i1 [[CMP8]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK13:       omp.precond.then:
// CHECK13-NEXT:    store i64 0, i64* [[DOTOMP_LB]], align 8
// CHECK13-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK13-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_UB]], align 8
// CHECK13-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK13-NEXT:    [[CONV9:%.*]] = zext i32 [[TMP8]] to i64
// CHECK13-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK13-NEXT:    [[CONV10:%.*]] = zext i32 [[TMP9]] to i64
// CHECK13-NEXT:    store i64 [[CONV9]], i64* [[DOTOMP_LB]], align 8
// CHECK13-NEXT:    store i64 [[CONV10]], i64* [[DOTOMP_UB]], align 8
// CHECK13-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK13-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK13-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK13-NEXT:    call void @__kmpc_for_static_init_8(%struct.ident_t* @[[GLOB2]], i32 [[TMP11]], i32 33, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_LB]], i64* [[DOTOMP_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 1)
// CHECK13-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTOMP_LB]], align 8
// CHECK13-NEXT:    store i64 [[TMP12]], i64* [[DOTOMP_IV]], align 8
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK13:       omp.inner.for.cond:
// CHECK13-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK13-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK13-NEXT:    [[CONV13:%.*]] = zext i32 [[TMP14]] to i64
// CHECK13-NEXT:    [[CMP14:%.*]] = icmp sle i64 [[TMP13]], [[CONV13]]
// CHECK13-NEXT:    br i1 [[CMP14]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK13:       omp.inner.for.body:
// CHECK13-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK13-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK13-NEXT:    [[SUB15:%.*]] = sub nsw i32 [[TMP16]], 0
// CHECK13-NEXT:    [[DIV16:%.*]] = sdiv i32 [[SUB15]], 1
// CHECK13-NEXT:    [[MUL17:%.*]] = mul nsw i32 1, [[DIV16]]
// CHECK13-NEXT:    [[CONV18:%.*]] = sext i32 [[MUL17]] to i64
// CHECK13-NEXT:    [[DIV19:%.*]] = sdiv i64 [[TMP15]], [[CONV18]]
// CHECK13-NEXT:    [[MUL20:%.*]] = mul nsw i64 [[DIV19]], 1
// CHECK13-NEXT:    [[ADD:%.*]] = add nsw i64 0, [[MUL20]]
// CHECK13-NEXT:    [[CONV21:%.*]] = trunc i64 [[ADD]] to i32
// CHECK13-NEXT:    store i32 [[CONV21]], i32* [[I11]], align 4
// CHECK13-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK13-NEXT:    [[TMP18:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK13-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK13-NEXT:    [[SUB22:%.*]] = sub nsw i32 [[TMP19]], 0
// CHECK13-NEXT:    [[DIV23:%.*]] = sdiv i32 [[SUB22]], 1
// CHECK13-NEXT:    [[MUL24:%.*]] = mul nsw i32 1, [[DIV23]]
// CHECK13-NEXT:    [[CONV25:%.*]] = sext i32 [[MUL24]] to i64
// CHECK13-NEXT:    [[DIV26:%.*]] = sdiv i64 [[TMP18]], [[CONV25]]
// CHECK13-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK13-NEXT:    [[SUB27:%.*]] = sub nsw i32 [[TMP20]], 0
// CHECK13-NEXT:    [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK13-NEXT:    [[MUL29:%.*]] = mul nsw i32 1, [[DIV28]]
// CHECK13-NEXT:    [[CONV30:%.*]] = sext i32 [[MUL29]] to i64
// CHECK13-NEXT:    [[MUL31:%.*]] = mul nsw i64 [[DIV26]], [[CONV30]]
// CHECK13-NEXT:    [[SUB32:%.*]] = sub nsw i64 [[TMP17]], [[MUL31]]
// CHECK13-NEXT:    [[MUL33:%.*]] = mul nsw i64 [[SUB32]], 1
// CHECK13-NEXT:    [[ADD34:%.*]] = add nsw i64 0, [[MUL33]]
// CHECK13-NEXT:    [[CONV35:%.*]] = trunc i64 [[ADD34]] to i32
// CHECK13-NEXT:    store i32 [[CONV35]], i32* [[J12]], align 4
// CHECK13-NEXT:    [[TMP21:%.*]] = load i32, i32* [[I11]], align 4
// CHECK13-NEXT:    [[TMP22:%.*]] = load i32, i32* [[J12]], align 4
// CHECK13-NEXT:    [[ADD36:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
// CHECK13-NEXT:    [[TMP23:%.*]] = load i32, i32* [[I11]], align 4
// CHECK13-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i32 0, i32 [[TMP23]]
// CHECK13-NEXT:    [[TMP24:%.*]] = load i32, i32* [[J12]], align 4
// CHECK13-NEXT:    [[ARRAYIDX37:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i32 0, i32 [[TMP24]]
// CHECK13-NEXT:    store i32 [[ADD36]], i32* [[ARRAYIDX37]], align 4
// CHECK13-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK13:       omp.body.continue:
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK13:       omp.inner.for.inc:
// CHECK13-NEXT:    [[TMP25:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK13-NEXT:    [[TMP26:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK13-NEXT:    [[ADD38:%.*]] = add nsw i64 [[TMP25]], [[TMP26]]
// CHECK13-NEXT:    store i64 [[ADD38]], i64* [[DOTOMP_IV]], align 8
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK13:       omp.inner.for.end:
// CHECK13-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK13:       omp.loop.exit:
// CHECK13-NEXT:    [[TMP27:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    [[TMP28:%.*]] = load i32, i32* [[TMP27]], align 4
// CHECK13-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP28]])
// CHECK13-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK13:       omp.precond.end:
// CHECK13-NEXT:    ret void
// CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l74
// CHECK13-SAME: (i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK13-NEXT:  entry:
// CHECK13-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK13-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK13-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK13-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK13-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK13-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK13-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK13:       .execute:
// CHECK13-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK13-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK13-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK13-NEXT:    [[TMP4:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK13-NEXT:    call void @__omp_outlined__10(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [1000 x i32]* [[TMP0]], i32* [[TMP4]]) #[[ATTR3]]
// CHECK13-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK13:       .omp.deinit:
// CHECK13-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK13-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK13:       .exit:
// CHECK13-NEXT:    ret void
// CHECK13-LABEL: define {{[^@]+}}@__omp_outlined__10
// CHECK13-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK13-NEXT:  entry:
// CHECK13-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK13-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 4
// CHECK13-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK13-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK13-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK13-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK13-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK13-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK13-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK13-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK13-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK13-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK13-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK13-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK13-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK13-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK13:       omp.precond.then:
// CHECK13-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK13-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK13-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK13-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK13-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK13-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK13-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK13-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK13:       cond.true:
// CHECK13-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK13-NEXT:    br label [[COND_END:%.*]]
// CHECK13:       cond.false:
// CHECK13-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    br label [[COND_END]]
// CHECK13:       cond.end:
// CHECK13-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK13-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK13:       omp.inner.for.cond:
// CHECK13-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK13-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK13-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK13-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK13:       omp.inner.for.body:
// CHECK13-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP16:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP16]], i32* [[N_CASTED]], align 4
// CHECK13-NEXT:    [[TMP17:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK13-NEXT:    [[TMP18:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK13-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK13-NEXT:    [[TMP20:%.*]] = inttoptr i32 [[TMP14]] to i8*
// CHECK13-NEXT:    store i8* [[TMP20]], i8** [[TMP19]], align 4
// CHECK13-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK13-NEXT:    [[TMP22:%.*]] = inttoptr i32 [[TMP15]] to i8*
// CHECK13-NEXT:    store i8* [[TMP22]], i8** [[TMP21]], align 4
// CHECK13-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK13-NEXT:    [[TMP24:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK13-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 4
// CHECK13-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK13-NEXT:    [[TMP26:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK13-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 4
// CHECK13-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 4
// CHECK13-NEXT:    [[TMP28:%.*]] = bitcast i32* [[TMP18]] to i8*
// CHECK13-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 4
// CHECK13-NEXT:    [[TMP29:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    [[TMP30:%.*]] = load i32, i32* [[TMP29]], align 4
// CHECK13-NEXT:    [[TMP31:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK13-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP30]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i32]*, i32*)* @__omp_outlined__11 to i8*), i8* null, i8** [[TMP31]], i32 5)
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK13:       omp.inner.for.inc:
// CHECK13-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP32]], [[TMP33]]
// CHECK13-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP34]], [[TMP35]]
// CHECK13-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP36]], [[TMP37]]
// CHECK13-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK13-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP38]], [[TMP39]]
// CHECK13-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
// CHECK13:       cond.true10:
// CHECK13-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK13-NEXT:    br label [[COND_END12:%.*]]
// CHECK13:       cond.false11:
// CHECK13-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    br label [[COND_END12]]
// CHECK13:       cond.end12:
// CHECK13-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP40]], [[COND_TRUE10]] ], [ [[TMP41]], [[COND_FALSE11]] ]
// CHECK13-NEXT:    store i32 [[COND13]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK13-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK13-NEXT:    store i32 [[TMP42]], i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK13:       omp.inner.for.end:
// CHECK13-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK13:       omp.loop.exit:
// CHECK13-NEXT:    [[TMP43:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    [[TMP44:%.*]] = load i32, i32* [[TMP43]], align 4
// CHECK13-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP44]])
// CHECK13-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK13:       omp.precond.end:
// CHECK13-NEXT:    ret void
// CHECK13-LABEL: define {{[^@]+}}@__omp_outlined__11
// CHECK13-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK13-NEXT:  entry:
// CHECK13-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK13-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK13-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK13-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK13-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK13-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK13-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK13-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK13-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK13-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK13-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK13-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK13-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK13-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK13-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK13-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK13-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK13-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK13-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK13:       omp.precond.then:
// CHECK13-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK13-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK13-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK13-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK13-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK13-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK13-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK13-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK13-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK13-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK13-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK13-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK13:       omp.inner.for.cond:
// CHECK13-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK13-NEXT:    [[CMP4:%.*]] = icmp ule i32 [[TMP10]], [[TMP11]]
// CHECK13-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK13:       omp.inner.for.body:
// CHECK13-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK13-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK13-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK13-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK13-NEXT:    [[TMP14:%.*]] = load i32, i32* [[I3]], align 4
// CHECK13-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[TMP13]], i32 [[TMP14]]
// CHECK13-NEXT:    [[TMP15:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK13-NEXT:    [[TMP16:%.*]] = load i32, i32* [[I3]], align 4
// CHECK13-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i32 0, i32 [[TMP16]]
// CHECK13-NEXT:    store i32 [[TMP15]], i32* [[ARRAYIDX5]], align 4
// CHECK13-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK13:       omp.body.continue:
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK13:       omp.inner.for.inc:
// CHECK13-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK13-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK13-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK13-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK13:       omp.inner.for.end:
// CHECK13-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK13:       omp.loop.exit:
// CHECK13-NEXT:    [[TMP19:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK13-NEXT:    [[TMP20:%.*]] = load i32, i32* [[TMP19]], align 4
// CHECK13-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP20]])
// CHECK13-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK13:       omp.precond.end:
// CHECK13-NEXT:    ret void
// CHECK14-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l43
// CHECK14-SAME: (i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK14-NEXT:  entry:
// CHECK14-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK14-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[L_CASTED:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK14-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK14-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK14-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK14-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK14:       .execute:
// CHECK14-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3:[0-9]+]])
// CHECK14-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK14-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK14-NEXT:    [[TMP4:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP4]], i32* [[L_CASTED]], align 4
// CHECK14-NEXT:    [[TMP5:%.*]] = load i32, i32* [[L_CASTED]], align 4
// CHECK14-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK14-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [1000 x i32]* [[TMP0]], i32 [[TMP5]]) #[[ATTR3:[0-9]+]]
// CHECK14-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK14:       .omp.deinit:
// CHECK14-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK14-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK14:       .exit:
// CHECK14-NEXT:    ret void
// CHECK14-LABEL: define {{[^@]+}}@__omp_outlined__
// CHECK14-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0]] {
// CHECK14-NEXT:  entry:
// CHECK14-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK14-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[I4:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[L_CASTED:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 4
// CHECK14-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK14-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK14-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK14-NEXT:    [[TMP1:%.*]] = load i16, i16* @"_openmp_static_kernel$is_shared", align 2
// CHECK14-NEXT:    [[TMP2:%.*]] = load i32, i32* @"_openmp_static_kernel$size", align 4
// CHECK14-NEXT:    call void @__kmpc_get_team_static_memory(i16 1, i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds (%"union._shared_openmp_static_memory_type_$_", %"union._shared_openmp_static_memory_type_$_" addrspace(3)* @"_openmp_shared_static_glob_rd_$_", i32 0, i32 0, i32 0) to i8*), i32 [[TMP2]], i16 [[TMP1]], i8** addrspacecast (i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr" to i8**))
// CHECK14-NEXT:    [[TMP3:%.*]] = load i8*, i8* addrspace(3)* @"_openmp_kernel_static_glob_rd$ptr", align 4
// CHECK14-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i8, i8* [[TMP3]], i32 0
// CHECK14-NEXT:    [[TMP5:%.*]] = bitcast i8* [[TMP4]] to %struct._globalized_locals_ty*
// CHECK14-NEXT:    [[L1:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY:%.*]], %struct._globalized_locals_ty* [[TMP5]], i32 0, i32 0
// CHECK14-NEXT:    [[TMP6:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP6]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK14-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK14-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], 0
// CHECK14-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK14-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK14-NEXT:    store i32 [[SUB3]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK14-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK14-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK14-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP8]]
// CHECK14-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK14:       omp.precond.then:
// CHECK14-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK14-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK14-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK14-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1:[0-9]+]], i32 [[TMP11]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 128)
// CHECK14-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK14-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
// CHECK14-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK14:       cond.true:
// CHECK14-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK14-NEXT:    br label [[COND_END:%.*]]
// CHECK14:       cond.false:
// CHECK14-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    br label [[COND_END]]
// CHECK14:       cond.end:
// CHECK14-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
// CHECK14-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    store i32 [[TMP16]], i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK14:       omp.inner.for.cond:
// CHECK14-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK14-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP18]], 1
// CHECK14-NEXT:    [[CMP6:%.*]] = icmp slt i32 [[TMP17]], [[ADD]]
// CHECK14-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK14:       omp.inner.for.body:
// CHECK14-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP21:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP21]], i32* [[N_CASTED]], align 4
// CHECK14-NEXT:    [[TMP22:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK14-NEXT:    [[TMP23:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP23]], i32* [[L_CASTED]], align 4
// CHECK14-NEXT:    [[TMP24:%.*]] = load i32, i32* [[L_CASTED]], align 4
// CHECK14-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK14-NEXT:    [[TMP26:%.*]] = inttoptr i32 [[TMP19]] to i8*
// CHECK14-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 4
// CHECK14-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK14-NEXT:    [[TMP28:%.*]] = inttoptr i32 [[TMP20]] to i8*
// CHECK14-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 4
// CHECK14-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK14-NEXT:    [[TMP30:%.*]] = inttoptr i32 [[TMP22]] to i8*
// CHECK14-NEXT:    store i8* [[TMP30]], i8** [[TMP29]], align 4
// CHECK14-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK14-NEXT:    [[TMP32:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK14-NEXT:    store i8* [[TMP32]], i8** [[TMP31]], align 4
// CHECK14-NEXT:    [[TMP33:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 4
// CHECK14-NEXT:    [[TMP34:%.*]] = inttoptr i32 [[TMP24]] to i8*
// CHECK14-NEXT:    store i8* [[TMP34]], i8** [[TMP33]], align 4
// CHECK14-NEXT:    [[TMP35:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    [[TMP36:%.*]] = load i32, i32* [[TMP35]], align 4
// CHECK14-NEXT:    [[TMP37:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK14-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP36]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i32]*, i32)* @__omp_outlined__1 to i8*), i8* null, i8** [[TMP37]], i32 5)
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK14:       omp.inner.for.inc:
// CHECK14-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP38]], [[TMP39]]
// CHECK14-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP40]], [[TMP41]]
// CHECK14-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP43:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP42]], [[TMP43]]
// CHECK14-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP44:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP45:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK14-NEXT:    [[CMP10:%.*]] = icmp sgt i32 [[TMP44]], [[TMP45]]
// CHECK14-NEXT:    br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]]
// CHECK14:       cond.true11:
// CHECK14-NEXT:    [[TMP46:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK14-NEXT:    br label [[COND_END13:%.*]]
// CHECK14:       cond.false12:
// CHECK14-NEXT:    [[TMP47:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    br label [[COND_END13]]
// CHECK14:       cond.end13:
// CHECK14-NEXT:    [[COND14:%.*]] = phi i32 [ [[TMP46]], [[COND_TRUE11]] ], [ [[TMP47]], [[COND_FALSE12]] ]
// CHECK14-NEXT:    store i32 [[COND14]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP48:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    store i32 [[TMP48]], i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK14:       omp.inner.for.end:
// CHECK14-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK14:       omp.loop.exit:
// CHECK14-NEXT:    [[TMP49:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    [[TMP50:%.*]] = load i32, i32* [[TMP49]], align 4
// CHECK14-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP50]])
// CHECK14-NEXT:    [[TMP51:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK14-NEXT:    [[TMP52:%.*]] = icmp ne i32 [[TMP51]], 0
// CHECK14-NEXT:    br i1 [[TMP52]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK14:       .omp.lastprivate.then:
// CHECK14-NEXT:    [[TMP53:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP53]], i32* [[L_ADDR]], align 4
// CHECK14-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK14:       .omp.lastprivate.done:
// CHECK14-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK14:       omp.precond.end:
// CHECK14-NEXT:    [[TMP54:%.*]] = load i16, i16* @"_openmp_static_kernel$is_shared", align 2
// CHECK14-NEXT:    call void @__kmpc_restore_team_static_memory(i16 1, i16 [[TMP54]])
// CHECK14-NEXT:    ret void
// CHECK14-LABEL: define {{[^@]+}}@__omp_outlined__1
// CHECK14-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0]] {
// CHECK14-NEXT:  entry:
// CHECK14-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK14-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK14-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK14-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK14-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK14-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK14-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK14-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK14-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK14-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK14-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK14-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK14-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK14-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK14-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK14-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK14:       omp.precond.then:
// CHECK14-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK14-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK14-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK14-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK14-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK14-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK14-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK14-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK14-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 32)
// CHECK14-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK14:       omp.dispatch.cond:
// CHECK14-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK14-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK14-NEXT:    [[CMP4:%.*]] = icmp ugt i32 [[TMP9]], [[TMP10]]
// CHECK14-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK14:       cond.true:
// CHECK14-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK14-NEXT:    br label [[COND_END:%.*]]
// CHECK14:       cond.false:
// CHECK14-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK14-NEXT:    br label [[COND_END]]
// CHECK14:       cond.end:
// CHECK14-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP11]], [[COND_TRUE]] ], [ [[TMP12]], [[COND_FALSE]] ]
// CHECK14-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
// CHECK14-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK14-NEXT:    store i32 [[TMP13]], i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK14-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK14-NEXT:    br i1 [[CMP5]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK14:       omp.dispatch.body:
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK14:       omp.inner.for.cond:
// CHECK14-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK14-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP16]], [[TMP17]]
// CHECK14-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK14:       omp.inner.for.body:
// CHECK14-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP18]], 1
// CHECK14-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK14-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK14-NEXT:    [[TMP19:%.*]] = load i32, i32* [[I3]], align 4
// CHECK14-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i32 0, i32 [[TMP19]]
// CHECK14-NEXT:    store i32 1, i32* [[ARRAYIDX]], align 4
// CHECK14-NEXT:    [[TMP20:%.*]] = load i32, i32* [[I3]], align 4
// CHECK14-NEXT:    store i32 [[TMP20]], i32* [[L_ADDR]], align 4
// CHECK14-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK14:       omp.body.continue:
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK14:       omp.inner.for.inc:
// CHECK14-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP21]], 1
// CHECK14-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK14:       omp.inner.for.end:
// CHECK14-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK14:       omp.dispatch.inc:
// CHECK14-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK14-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK14-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_LB]], align 4
// CHECK14-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK14-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK14-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_UB]], align 4
// CHECK14-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK14:       omp.dispatch.end:
// CHECK14-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK14-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP27]])
// CHECK14-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK14-NEXT:    [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
// CHECK14-NEXT:    br i1 [[TMP29]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK14:       .omp.lastprivate.then:
// CHECK14-NEXT:    [[TMP30:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP30]], i32* [[L_ADDR]], align 4
// CHECK14-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK14:       .omp.lastprivate.done:
// CHECK14-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK14:       omp.precond.end:
// CHECK14-NEXT:    ret void
// CHECK14-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l49
// CHECK14-SAME: (i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK14-NEXT:  entry:
// CHECK14-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK14-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK14-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK14-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK14-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK14-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK14:       .execute:
// CHECK14-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK14-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK14-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK14-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK14-NEXT:    call void @__omp_outlined__2(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [1000 x i16]* [[TMP0]]) #[[ATTR3]]
// CHECK14-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK14:       .omp.deinit:
// CHECK14-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK14-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK14:       .exit:
// CHECK14-NEXT:    ret void
// CHECK14-LABEL: define {{[^@]+}}@__omp_outlined__2
// CHECK14-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK14-NEXT:  entry:
// CHECK14-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK14-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK14-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK14-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK14-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK14-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK14-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK14-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK14-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK14-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK14-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK14-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK14-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK14-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK14-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK14:       omp.precond.then:
// CHECK14-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK14-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK14-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK14-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK14-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK14-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK14-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK14-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK14:       cond.true:
// CHECK14-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK14-NEXT:    br label [[COND_END:%.*]]
// CHECK14:       cond.false:
// CHECK14-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    br label [[COND_END]]
// CHECK14:       cond.end:
// CHECK14-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK14-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK14:       omp.inner.for.cond:
// CHECK14-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK14-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK14-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK14-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK14:       omp.inner.for.body:
// CHECK14-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP16:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP16]], i32* [[N_CASTED]], align 4
// CHECK14-NEXT:    [[TMP17:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK14-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK14-NEXT:    [[TMP19:%.*]] = inttoptr i32 [[TMP14]] to i8*
// CHECK14-NEXT:    store i8* [[TMP19]], i8** [[TMP18]], align 4
// CHECK14-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK14-NEXT:    [[TMP21:%.*]] = inttoptr i32 [[TMP15]] to i8*
// CHECK14-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 4
// CHECK14-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK14-NEXT:    [[TMP23:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK14-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 4
// CHECK14-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK14-NEXT:    [[TMP25:%.*]] = bitcast [1000 x i16]* [[TMP0]] to i8*
// CHECK14-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 4
// CHECK14-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK14-NEXT:    [[TMP28:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK14-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP27]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i16]*)* @__omp_outlined__3 to i8*), i8* null, i8** [[TMP28]], i32 4)
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK14:       omp.inner.for.inc:
// CHECK14-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP29]], [[TMP30]]
// CHECK14-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP31]], [[TMP32]]
// CHECK14-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP33]], [[TMP34]]
// CHECK14-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK14-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP35]], [[TMP36]]
// CHECK14-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
// CHECK14:       cond.true10:
// CHECK14-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK14-NEXT:    br label [[COND_END12:%.*]]
// CHECK14:       cond.false11:
// CHECK14-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    br label [[COND_END12]]
// CHECK14:       cond.end12:
// CHECK14-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP37]], [[COND_TRUE10]] ], [ [[TMP38]], [[COND_FALSE11]] ]
// CHECK14-NEXT:    store i32 [[COND13]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    store i32 [[TMP39]], i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK14:       omp.inner.for.end:
// CHECK14-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK14:       omp.loop.exit:
// CHECK14-NEXT:    [[TMP40:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    [[TMP41:%.*]] = load i32, i32* [[TMP40]], align 4
// CHECK14-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP41]])
// CHECK14-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK14:       omp.precond.end:
// CHECK14-NEXT:    ret void
// CHECK14-LABEL: define {{[^@]+}}@__omp_outlined__3
// CHECK14-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK14-NEXT:  entry:
// CHECK14-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK14-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK14-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK14-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK14-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK14-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK14-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK14-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK14-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK14-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK14-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK14-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK14-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK14-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK14-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK14-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK14:       omp.precond.then:
// CHECK14-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK14-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK14-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK14-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK14-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK14-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK14-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK14-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK14-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK14-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK14-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK14:       omp.inner.for.cond:
// CHECK14-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK14-NEXT:    [[CMP4:%.*]] = icmp ule i32 [[TMP10]], [[TMP11]]
// CHECK14-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK14:       omp.inner.for.body:
// CHECK14-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK14-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK14-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK14-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I3]], align 4
// CHECK14-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i16], [1000 x i16]* [[TMP0]], i32 0, i32 [[TMP13]]
// CHECK14-NEXT:    [[TMP14:%.*]] = load i16, i16* [[ARRAYIDX]], align 2
// CHECK14-NEXT:    [[CONV:%.*]] = sext i16 [[TMP14]] to i32
// CHECK14-NEXT:    [[ADD5:%.*]] = add nsw i32 [[CONV]], 1
// CHECK14-NEXT:    [[CONV6:%.*]] = trunc i32 [[ADD5]] to i16
// CHECK14-NEXT:    store i16 [[CONV6]], i16* [[ARRAYIDX]], align 2
// CHECK14-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK14:       omp.body.continue:
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK14:       omp.inner.for.inc:
// CHECK14-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP15]], [[TMP16]]
// CHECK14-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK14:       omp.inner.for.end:
// CHECK14-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK14:       omp.loop.exit:
// CHECK14-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
// CHECK14-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP18]])
// CHECK14-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK14:       omp.precond.end:
// CHECK14-NEXT:    ret void
// CHECK14-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l54
// CHECK14-SAME: ([10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK14-NEXT:  entry:
// CHECK14-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK14-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK14-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK14-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK14-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK14-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK14-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK14:       .execute:
// CHECK14-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK14-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK14-NEXT:    call void @__omp_outlined__4(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x i32]* [[TMP0]]) #[[ATTR3]]
// CHECK14-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK14:       .omp.deinit:
// CHECK14-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK14-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK14:       .exit:
// CHECK14-NEXT:    ret void
// CHECK14-LABEL: define {{[^@]+}}@__omp_outlined__4
// CHECK14-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK14-NEXT:  entry:
// CHECK14-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK14-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x i8*], align 4
// CHECK14-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK14-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK14-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK14-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    store i32 9, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK14-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK14-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK14-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK14-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9
// CHECK14-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK14:       cond.true:
// CHECK14-NEXT:    br label [[COND_END:%.*]]
// CHECK14:       cond.false:
// CHECK14-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    br label [[COND_END]]
// CHECK14:       cond.end:
// CHECK14-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK14-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK14:       omp.inner.for.cond:
// CHECK14-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[CMP1:%.*]] = icmp slt i32 [[TMP6]], 10
// CHECK14-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK14:       omp.inner.for.body:
// CHECK14-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK14-NEXT:    [[TMP10:%.*]] = inttoptr i32 [[TMP7]] to i8*
// CHECK14-NEXT:    store i8* [[TMP10]], i8** [[TMP9]], align 4
// CHECK14-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK14-NEXT:    [[TMP12:%.*]] = inttoptr i32 [[TMP8]] to i8*
// CHECK14-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 4
// CHECK14-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK14-NEXT:    [[TMP14:%.*]] = bitcast [10 x i32]* [[TMP0]] to i8*
// CHECK14-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 4
// CHECK14-NEXT:    [[TMP15:%.*]] = bitcast [3 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK14-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, [10 x i32]*)* @__omp_outlined__5 to i8*), i8* null, i8** [[TMP15]], i32 3)
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK14:       omp.inner.for.inc:
// CHECK14-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP16]], [[TMP17]]
// CHECK14-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP18]], [[TMP19]]
// CHECK14-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK14-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP22]], 9
// CHECK14-NEXT:    br i1 [[CMP4]], label [[COND_TRUE5:%.*]], label [[COND_FALSE6:%.*]]
// CHECK14:       cond.true5:
// CHECK14-NEXT:    br label [[COND_END7:%.*]]
// CHECK14:       cond.false6:
// CHECK14-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    br label [[COND_END7]]
// CHECK14:       cond.end7:
// CHECK14-NEXT:    [[COND8:%.*]] = phi i32 [ 9, [[COND_TRUE5]] ], [ [[TMP23]], [[COND_FALSE6]] ]
// CHECK14-NEXT:    store i32 [[COND8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    store i32 [[TMP24]], i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK14:       omp.inner.for.end:
// CHECK14-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK14:       omp.loop.exit:
// CHECK14-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]])
// CHECK14-NEXT:    ret void
// CHECK14-LABEL: define {{[^@]+}}@__omp_outlined__5
// CHECK14-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK14-NEXT:  entry:
// CHECK14-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK14-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK14-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK14-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK14-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK14-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK14-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK14-NEXT:    store i32 9, i32* [[DOTOMP_UB]], align 4
// CHECK14-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK14-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP1]], i32* [[DOTOMP_LB]], align 4
// CHECK14-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_UB]], align 4
// CHECK14-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK14-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK14-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK14-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK14-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK14:       omp.inner.for.cond:
// CHECK14-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK14-NEXT:    [[CMP:%.*]] = icmp ule i32 [[TMP6]], [[TMP7]]
// CHECK14-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK14:       omp.inner.for.body:
// CHECK14-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP8]], 1
// CHECK14-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK14-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK14-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4
// CHECK14-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[TMP0]], i32 0, i32 [[TMP9]]
// CHECK14-NEXT:    [[TMP10:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK14-NEXT:    [[ADD1:%.*]] = add nsw i32 [[TMP10]], 1
// CHECK14-NEXT:    store i32 [[ADD1]], i32* [[ARRAYIDX]], align 4
// CHECK14-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK14:       omp.body.continue:
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK14:       omp.inner.for.inc:
// CHECK14-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
// CHECK14-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK14:       omp.inner.for.end:
// CHECK14-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK14:       omp.loop.exit:
// CHECK14-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK14-NEXT:    ret void
// CHECK14-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l59
// CHECK14-SAME: ([10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK14-NEXT:  entry:
// CHECK14-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK14-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[F_CASTED:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK14-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK14-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK14-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK14-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK14-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK14:       .execute:
// CHECK14-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK14-NEXT:    [[TMP2:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP2]], i32* [[F_CASTED]], align 4
// CHECK14-NEXT:    [[TMP3:%.*]] = load i32, i32* [[F_CASTED]], align 4
// CHECK14-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK14-NEXT:    call void @__omp_outlined__6(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x [10 x i32]]* [[TMP0]], i32 [[TMP3]]) #[[ATTR3]]
// CHECK14-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK14:       .omp.deinit:
// CHECK14-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK14-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK14:       .exit:
// CHECK14-NEXT:    ret void
// CHECK14-LABEL: define {{[^@]+}}@__omp_outlined__6
// CHECK14-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK14-NEXT:  entry:
// CHECK14-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK14-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[F_CASTED:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK14-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK14-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK14-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK14-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    store i32 99, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK14-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK14-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK14-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK14-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 99
// CHECK14-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK14:       cond.true:
// CHECK14-NEXT:    br label [[COND_END:%.*]]
// CHECK14:       cond.false:
// CHECK14-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    br label [[COND_END]]
// CHECK14:       cond.end:
// CHECK14-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK14-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK14:       omp.inner.for.cond:
// CHECK14-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[TMP6]], 100
// CHECK14-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK14:       omp.inner.for.body:
// CHECK14-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP9:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP9]], i32* [[F_CASTED]], align 4
// CHECK14-NEXT:    [[TMP10:%.*]] = load i32, i32* [[F_CASTED]], align 4
// CHECK14-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK14-NEXT:    [[TMP12:%.*]] = inttoptr i32 [[TMP7]] to i8*
// CHECK14-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 4
// CHECK14-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK14-NEXT:    [[TMP14:%.*]] = inttoptr i32 [[TMP8]] to i8*
// CHECK14-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 4
// CHECK14-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK14-NEXT:    [[TMP16:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK14-NEXT:    store i8* [[TMP16]], i8** [[TMP15]], align 4
// CHECK14-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK14-NEXT:    [[TMP18:%.*]] = inttoptr i32 [[TMP10]] to i8*
// CHECK14-NEXT:    store i8* [[TMP18]], i8** [[TMP17]], align 4
// CHECK14-NEXT:    [[TMP19:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK14-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, [10 x [10 x i32]]*, i32)* @__omp_outlined__7 to i8*), i8* null, i8** [[TMP19]], i32 4)
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK14:       omp.inner.for.inc:
// CHECK14-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK14-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK14-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK14-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP26]], 99
// CHECK14-NEXT:    br i1 [[CMP5]], label [[COND_TRUE6:%.*]], label [[COND_FALSE7:%.*]]
// CHECK14:       cond.true6:
// CHECK14-NEXT:    br label [[COND_END8:%.*]]
// CHECK14:       cond.false7:
// CHECK14-NEXT:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    br label [[COND_END8]]
// CHECK14:       cond.end8:
// CHECK14-NEXT:    [[COND9:%.*]] = phi i32 [ 99, [[COND_TRUE6]] ], [ [[TMP27]], [[COND_FALSE7]] ]
// CHECK14-NEXT:    store i32 [[COND9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    store i32 [[TMP28]], i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK14:       omp.inner.for.end:
// CHECK14-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK14:       omp.loop.exit:
// CHECK14-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]])
// CHECK14-NEXT:    ret void
// CHECK14-LABEL: define {{[^@]+}}@__omp_outlined__7
// CHECK14-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK14-NEXT:  entry:
// CHECK14-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK14-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK14-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK14-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK14-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK14-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK14-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK14-NEXT:    store i32 99, i32* [[DOTOMP_UB]], align 4
// CHECK14-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK14-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP1]], i32* [[DOTOMP_LB]], align 4
// CHECK14-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_UB]], align 4
// CHECK14-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK14-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK14-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK14-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK14-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK14:       omp.inner.for.cond:
// CHECK14-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK14-NEXT:    [[CMP:%.*]] = icmp ule i32 [[TMP6]], [[TMP7]]
// CHECK14-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK14:       omp.inner.for.body:
// CHECK14-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP8]], 10
// CHECK14-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK14-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK14-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK14-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[DIV2:%.*]] = sdiv i32 [[TMP10]], 10
// CHECK14-NEXT:    [[MUL3:%.*]] = mul nsw i32 [[DIV2]], 10
// CHECK14-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP9]], [[MUL3]]
// CHECK14-NEXT:    [[MUL4:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK14-NEXT:    [[ADD5:%.*]] = add nsw i32 0, [[MUL4]]
// CHECK14-NEXT:    store i32 [[ADD5]], i32* [[J]], align 4
// CHECK14-NEXT:    store i32 10, i32* [[K]], align 4
// CHECK14-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
// CHECK14-NEXT:    [[TMP12:%.*]] = load i32, i32* [[J]], align 4
// CHECK14-NEXT:    [[TMP13:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK14-NEXT:    [[MUL6:%.*]] = mul nsw i32 [[TMP12]], [[TMP13]]
// CHECK14-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP11]], [[MUL6]]
// CHECK14-NEXT:    [[TMP14:%.*]] = load i32, i32* [[K]], align 4
// CHECK14-NEXT:    [[ADD8:%.*]] = add nsw i32 [[ADD7]], [[TMP14]]
// CHECK14-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
// CHECK14-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i32 0, i32 [[TMP15]]
// CHECK14-NEXT:    [[TMP16:%.*]] = load i32, i32* [[J]], align 4
// CHECK14-NEXT:    [[ARRAYIDX9:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i32 0, i32 [[TMP16]]
// CHECK14-NEXT:    store i32 [[ADD8]], i32* [[ARRAYIDX9]], align 4
// CHECK14-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK14:       omp.body.continue:
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK14:       omp.inner.for.inc:
// CHECK14-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK14-NEXT:    store i32 [[ADD10]], i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK14:       omp.inner.for.end:
// CHECK14-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK14:       omp.loop.exit:
// CHECK14-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK14-NEXT:    ret void
// CHECK14-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l67
// CHECK14-SAME: (i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK14-NEXT:  entry:
// CHECK14-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK14-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK14-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK14-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK14-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK14-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK14:       .execute:
// CHECK14-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK14-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK14-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK14-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK14-NEXT:    call void @__omp_outlined__8(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [10 x [10 x i32]]* [[TMP0]]) #[[ATTR3]]
// CHECK14-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK14:       .omp.deinit:
// CHECK14-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK14-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK14:       .exit:
// CHECK14-NEXT:    ret void
// CHECK14-LABEL: define {{[^@]+}}@__omp_outlined__8
// CHECK14-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK14-NEXT:  entry:
// CHECK14-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK14-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK14-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK14-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i64, align 8
// CHECK14-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i64, align 8
// CHECK14-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK14-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[I9:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[J10:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK14-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK14-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK14-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK14-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK14-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK14-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK14-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK14-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK14-NEXT:    [[CONV:%.*]] = sext i32 [[DIV]] to i64
// CHECK14-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK14-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK14-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK14-NEXT:    [[CONV6:%.*]] = sext i32 [[DIV5]] to i64
// CHECK14-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV]], [[CONV6]]
// CHECK14-NEXT:    [[SUB7:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK14-NEXT:    store i64 [[SUB7]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK14-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK14-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK14-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK14-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK14-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK14:       land.lhs.true:
// CHECK14-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK14-NEXT:    [[CMP8:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK14-NEXT:    br i1 [[CMP8]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK14:       omp.precond.then:
// CHECK14-NEXT:    store i64 0, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK14-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK14-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK14-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK14-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK14-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK14-NEXT:    [[CONV11:%.*]] = zext i32 [[NVPTX_NUM_THREADS]] to i64
// CHECK14-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    [[TMP9:%.*]] = load i32, i32* [[TMP8]], align 4
// CHECK14-NEXT:    call void @__kmpc_for_static_init_8(%struct.ident_t* @[[GLOB1]], i32 [[TMP9]], i32 91, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_COMB_LB]], i64* [[DOTOMP_COMB_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 [[CONV11]])
// CHECK14-NEXT:    [[TMP10:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK14-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK14-NEXT:    [[CMP12:%.*]] = icmp sgt i64 [[TMP10]], [[TMP11]]
// CHECK14-NEXT:    br i1 [[CMP12]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK14:       cond.true:
// CHECK14-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK14-NEXT:    br label [[COND_END:%.*]]
// CHECK14:       cond.false:
// CHECK14-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK14-NEXT:    br label [[COND_END]]
// CHECK14:       cond.end:
// CHECK14-NEXT:    [[COND:%.*]] = phi i64 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
// CHECK14-NEXT:    store i64 [[COND]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK14-NEXT:    [[TMP14:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK14-NEXT:    store i64 [[TMP14]], i64* [[DOTOMP_IV]], align 8
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK14:       omp.inner.for.cond:
// CHECK14-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK14-NEXT:    [[TMP16:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK14-NEXT:    [[ADD:%.*]] = add nsw i64 [[TMP16]], 1
// CHECK14-NEXT:    [[CMP13:%.*]] = icmp slt i64 [[TMP15]], [[ADD]]
// CHECK14-NEXT:    br i1 [[CMP13]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK14:       omp.inner.for.body:
// CHECK14-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK14-NEXT:    [[TMP18:%.*]] = trunc i64 [[TMP17]] to i32
// CHECK14-NEXT:    [[TMP19:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK14-NEXT:    [[TMP20:%.*]] = trunc i64 [[TMP19]] to i32
// CHECK14-NEXT:    [[TMP21:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP21]], i32* [[N_CASTED]], align 4
// CHECK14-NEXT:    [[TMP22:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK14-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK14-NEXT:    [[TMP24:%.*]] = inttoptr i32 [[TMP18]] to i8*
// CHECK14-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 4
// CHECK14-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK14-NEXT:    [[TMP26:%.*]] = inttoptr i32 [[TMP20]] to i8*
// CHECK14-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 4
// CHECK14-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK14-NEXT:    [[TMP28:%.*]] = inttoptr i32 [[TMP22]] to i8*
// CHECK14-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 4
// CHECK14-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK14-NEXT:    [[TMP30:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK14-NEXT:    store i8* [[TMP30]], i8** [[TMP29]], align 4
// CHECK14-NEXT:    [[TMP31:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    [[TMP32:%.*]] = load i32, i32* [[TMP31]], align 4
// CHECK14-NEXT:    [[TMP33:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK14-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP32]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [10 x [10 x i32]]*)* @__omp_outlined__9 to i8*), i8* null, i8** [[TMP33]], i32 4)
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK14:       omp.inner.for.inc:
// CHECK14-NEXT:    [[TMP34:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK14-NEXT:    [[TMP35:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK14-NEXT:    [[ADD14:%.*]] = add nsw i64 [[TMP34]], [[TMP35]]
// CHECK14-NEXT:    store i64 [[ADD14]], i64* [[DOTOMP_IV]], align 8
// CHECK14-NEXT:    [[TMP36:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK14-NEXT:    [[TMP37:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK14-NEXT:    [[ADD15:%.*]] = add nsw i64 [[TMP36]], [[TMP37]]
// CHECK14-NEXT:    store i64 [[ADD15]], i64* [[DOTOMP_COMB_LB]], align 8
// CHECK14-NEXT:    [[TMP38:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK14-NEXT:    [[TMP39:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK14-NEXT:    [[ADD16:%.*]] = add nsw i64 [[TMP38]], [[TMP39]]
// CHECK14-NEXT:    store i64 [[ADD16]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK14-NEXT:    [[TMP40:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK14-NEXT:    [[TMP41:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK14-NEXT:    [[CMP17:%.*]] = icmp sgt i64 [[TMP40]], [[TMP41]]
// CHECK14-NEXT:    br i1 [[CMP17]], label [[COND_TRUE18:%.*]], label [[COND_FALSE19:%.*]]
// CHECK14:       cond.true18:
// CHECK14-NEXT:    [[TMP42:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK14-NEXT:    br label [[COND_END20:%.*]]
// CHECK14:       cond.false19:
// CHECK14-NEXT:    [[TMP43:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK14-NEXT:    br label [[COND_END20]]
// CHECK14:       cond.end20:
// CHECK14-NEXT:    [[COND21:%.*]] = phi i64 [ [[TMP42]], [[COND_TRUE18]] ], [ [[TMP43]], [[COND_FALSE19]] ]
// CHECK14-NEXT:    store i64 [[COND21]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK14-NEXT:    [[TMP44:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK14-NEXT:    store i64 [[TMP44]], i64* [[DOTOMP_IV]], align 8
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK14:       omp.inner.for.end:
// CHECK14-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK14:       omp.loop.exit:
// CHECK14-NEXT:    [[TMP45:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    [[TMP46:%.*]] = load i32, i32* [[TMP45]], align 4
// CHECK14-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP46]])
// CHECK14-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK14:       omp.precond.end:
// CHECK14-NEXT:    ret void
// CHECK14-LABEL: define {{[^@]+}}@__omp_outlined__9
// CHECK14-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK14-NEXT:  entry:
// CHECK14-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK14-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK14-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK14-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_LB:%.*]] = alloca i64, align 8
// CHECK14-NEXT:    [[DOTOMP_UB:%.*]] = alloca i64, align 8
// CHECK14-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK14-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[I11:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[J12:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK14-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK14-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK14-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK14-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK14-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK14-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK14-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK14-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK14-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK14-NEXT:    [[CONV:%.*]] = sext i32 [[DIV]] to i64
// CHECK14-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK14-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK14-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK14-NEXT:    [[CONV6:%.*]] = sext i32 [[DIV5]] to i64
// CHECK14-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV]], [[CONV6]]
// CHECK14-NEXT:    [[SUB7:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK14-NEXT:    store i64 [[SUB7]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK14-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK14-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK14-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK14-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK14-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK14:       land.lhs.true:
// CHECK14-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK14-NEXT:    [[CMP8:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK14-NEXT:    br i1 [[CMP8]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK14:       omp.precond.then:
// CHECK14-NEXT:    store i64 0, i64* [[DOTOMP_LB]], align 8
// CHECK14-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK14-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_UB]], align 8
// CHECK14-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK14-NEXT:    [[CONV9:%.*]] = zext i32 [[TMP8]] to i64
// CHECK14-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK14-NEXT:    [[CONV10:%.*]] = zext i32 [[TMP9]] to i64
// CHECK14-NEXT:    store i64 [[CONV9]], i64* [[DOTOMP_LB]], align 8
// CHECK14-NEXT:    store i64 [[CONV10]], i64* [[DOTOMP_UB]], align 8
// CHECK14-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK14-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK14-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK14-NEXT:    call void @__kmpc_for_static_init_8(%struct.ident_t* @[[GLOB2]], i32 [[TMP11]], i32 33, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_LB]], i64* [[DOTOMP_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 1)
// CHECK14-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTOMP_LB]], align 8
// CHECK14-NEXT:    store i64 [[TMP12]], i64* [[DOTOMP_IV]], align 8
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK14:       omp.inner.for.cond:
// CHECK14-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK14-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK14-NEXT:    [[CONV13:%.*]] = zext i32 [[TMP14]] to i64
// CHECK14-NEXT:    [[CMP14:%.*]] = icmp sle i64 [[TMP13]], [[CONV13]]
// CHECK14-NEXT:    br i1 [[CMP14]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK14:       omp.inner.for.body:
// CHECK14-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK14-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK14-NEXT:    [[SUB15:%.*]] = sub nsw i32 [[TMP16]], 0
// CHECK14-NEXT:    [[DIV16:%.*]] = sdiv i32 [[SUB15]], 1
// CHECK14-NEXT:    [[MUL17:%.*]] = mul nsw i32 1, [[DIV16]]
// CHECK14-NEXT:    [[CONV18:%.*]] = sext i32 [[MUL17]] to i64
// CHECK14-NEXT:    [[DIV19:%.*]] = sdiv i64 [[TMP15]], [[CONV18]]
// CHECK14-NEXT:    [[MUL20:%.*]] = mul nsw i64 [[DIV19]], 1
// CHECK14-NEXT:    [[ADD:%.*]] = add nsw i64 0, [[MUL20]]
// CHECK14-NEXT:    [[CONV21:%.*]] = trunc i64 [[ADD]] to i32
// CHECK14-NEXT:    store i32 [[CONV21]], i32* [[I11]], align 4
// CHECK14-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK14-NEXT:    [[TMP18:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK14-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK14-NEXT:    [[SUB22:%.*]] = sub nsw i32 [[TMP19]], 0
// CHECK14-NEXT:    [[DIV23:%.*]] = sdiv i32 [[SUB22]], 1
// CHECK14-NEXT:    [[MUL24:%.*]] = mul nsw i32 1, [[DIV23]]
// CHECK14-NEXT:    [[CONV25:%.*]] = sext i32 [[MUL24]] to i64
// CHECK14-NEXT:    [[DIV26:%.*]] = sdiv i64 [[TMP18]], [[CONV25]]
// CHECK14-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK14-NEXT:    [[SUB27:%.*]] = sub nsw i32 [[TMP20]], 0
// CHECK14-NEXT:    [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK14-NEXT:    [[MUL29:%.*]] = mul nsw i32 1, [[DIV28]]
// CHECK14-NEXT:    [[CONV30:%.*]] = sext i32 [[MUL29]] to i64
// CHECK14-NEXT:    [[MUL31:%.*]] = mul nsw i64 [[DIV26]], [[CONV30]]
// CHECK14-NEXT:    [[SUB32:%.*]] = sub nsw i64 [[TMP17]], [[MUL31]]
// CHECK14-NEXT:    [[MUL33:%.*]] = mul nsw i64 [[SUB32]], 1
// CHECK14-NEXT:    [[ADD34:%.*]] = add nsw i64 0, [[MUL33]]
// CHECK14-NEXT:    [[CONV35:%.*]] = trunc i64 [[ADD34]] to i32
// CHECK14-NEXT:    store i32 [[CONV35]], i32* [[J12]], align 4
// CHECK14-NEXT:    [[TMP21:%.*]] = load i32, i32* [[I11]], align 4
// CHECK14-NEXT:    [[TMP22:%.*]] = load i32, i32* [[J12]], align 4
// CHECK14-NEXT:    [[ADD36:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
// CHECK14-NEXT:    [[TMP23:%.*]] = load i32, i32* [[I11]], align 4
// CHECK14-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i32 0, i32 [[TMP23]]
// CHECK14-NEXT:    [[TMP24:%.*]] = load i32, i32* [[J12]], align 4
// CHECK14-NEXT:    [[ARRAYIDX37:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i32 0, i32 [[TMP24]]
// CHECK14-NEXT:    store i32 [[ADD36]], i32* [[ARRAYIDX37]], align 4
// CHECK14-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK14:       omp.body.continue:
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK14:       omp.inner.for.inc:
// CHECK14-NEXT:    [[TMP25:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK14-NEXT:    [[TMP26:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK14-NEXT:    [[ADD38:%.*]] = add nsw i64 [[TMP25]], [[TMP26]]
// CHECK14-NEXT:    store i64 [[ADD38]], i64* [[DOTOMP_IV]], align 8
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK14:       omp.inner.for.end:
// CHECK14-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK14:       omp.loop.exit:
// CHECK14-NEXT:    [[TMP27:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    [[TMP28:%.*]] = load i32, i32* [[TMP27]], align 4
// CHECK14-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP28]])
// CHECK14-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK14:       omp.precond.end:
// CHECK14-NEXT:    ret void
// CHECK14-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l74
// CHECK14-SAME: (i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK14-NEXT:  entry:
// CHECK14-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK14-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK14-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK14-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK14-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK14-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK14-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK14:       .execute:
// CHECK14-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK14-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK14-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK14-NEXT:    [[TMP4:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK14-NEXT:    call void @__omp_outlined__10(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [1000 x i32]* [[TMP0]], i32* [[TMP4]]) #[[ATTR3]]
// CHECK14-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK14:       .omp.deinit:
// CHECK14-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK14-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK14:       .exit:
// CHECK14-NEXT:    ret void
// CHECK14-LABEL: define {{[^@]+}}@__omp_outlined__10
// CHECK14-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK14-NEXT:  entry:
// CHECK14-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK14-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 4
// CHECK14-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK14-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK14-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK14-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK14-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK14-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK14-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK14-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK14-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK14-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK14-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK14-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK14-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK14-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK14:       omp.precond.then:
// CHECK14-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK14-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK14-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK14-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK14-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK14-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK14-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK14-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK14:       cond.true:
// CHECK14-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK14-NEXT:    br label [[COND_END:%.*]]
// CHECK14:       cond.false:
// CHECK14-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    br label [[COND_END]]
// CHECK14:       cond.end:
// CHECK14-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK14-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK14:       omp.inner.for.cond:
// CHECK14-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK14-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK14-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK14-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK14:       omp.inner.for.body:
// CHECK14-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP16:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP16]], i32* [[N_CASTED]], align 4
// CHECK14-NEXT:    [[TMP17:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK14-NEXT:    [[TMP18:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK14-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK14-NEXT:    [[TMP20:%.*]] = inttoptr i32 [[TMP14]] to i8*
// CHECK14-NEXT:    store i8* [[TMP20]], i8** [[TMP19]], align 4
// CHECK14-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK14-NEXT:    [[TMP22:%.*]] = inttoptr i32 [[TMP15]] to i8*
// CHECK14-NEXT:    store i8* [[TMP22]], i8** [[TMP21]], align 4
// CHECK14-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK14-NEXT:    [[TMP24:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK14-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 4
// CHECK14-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK14-NEXT:    [[TMP26:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK14-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 4
// CHECK14-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 4
// CHECK14-NEXT:    [[TMP28:%.*]] = bitcast i32* [[TMP18]] to i8*
// CHECK14-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 4
// CHECK14-NEXT:    [[TMP29:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    [[TMP30:%.*]] = load i32, i32* [[TMP29]], align 4
// CHECK14-NEXT:    [[TMP31:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK14-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP30]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i32]*, i32*)* @__omp_outlined__11 to i8*), i8* null, i8** [[TMP31]], i32 5)
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK14:       omp.inner.for.inc:
// CHECK14-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP32]], [[TMP33]]
// CHECK14-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP34]], [[TMP35]]
// CHECK14-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP36]], [[TMP37]]
// CHECK14-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK14-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP38]], [[TMP39]]
// CHECK14-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
// CHECK14:       cond.true10:
// CHECK14-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK14-NEXT:    br label [[COND_END12:%.*]]
// CHECK14:       cond.false11:
// CHECK14-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    br label [[COND_END12]]
// CHECK14:       cond.end12:
// CHECK14-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP40]], [[COND_TRUE10]] ], [ [[TMP41]], [[COND_FALSE11]] ]
// CHECK14-NEXT:    store i32 [[COND13]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK14-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK14-NEXT:    store i32 [[TMP42]], i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK14:       omp.inner.for.end:
// CHECK14-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK14:       omp.loop.exit:
// CHECK14-NEXT:    [[TMP43:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    [[TMP44:%.*]] = load i32, i32* [[TMP43]], align 4
// CHECK14-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP44]])
// CHECK14-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK14:       omp.precond.end:
// CHECK14-NEXT:    ret void
// CHECK14-LABEL: define {{[^@]+}}@__omp_outlined__11
// CHECK14-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK14-NEXT:  entry:
// CHECK14-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK14-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK14-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK14-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK14-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK14-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK14-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK14-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK14-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK14-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK14-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK14-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK14-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK14-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK14-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK14-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK14-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK14-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK14-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK14:       omp.precond.then:
// CHECK14-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK14-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK14-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK14-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK14-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK14-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK14-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK14-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK14-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK14-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK14-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK14-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK14:       omp.inner.for.cond:
// CHECK14-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK14-NEXT:    [[CMP4:%.*]] = icmp ule i32 [[TMP10]], [[TMP11]]
// CHECK14-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK14:       omp.inner.for.body:
// CHECK14-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK14-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK14-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK14-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK14-NEXT:    [[TMP14:%.*]] = load i32, i32* [[I3]], align 4
// CHECK14-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[TMP13]], i32 [[TMP14]]
// CHECK14-NEXT:    [[TMP15:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK14-NEXT:    [[TMP16:%.*]] = load i32, i32* [[I3]], align 4
// CHECK14-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i32 0, i32 [[TMP16]]
// CHECK14-NEXT:    store i32 [[TMP15]], i32* [[ARRAYIDX5]], align 4
// CHECK14-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK14:       omp.body.continue:
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK14:       omp.inner.for.inc:
// CHECK14-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK14-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK14-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK14-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK14:       omp.inner.for.end:
// CHECK14-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK14:       omp.loop.exit:
// CHECK14-NEXT:    [[TMP19:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK14-NEXT:    [[TMP20:%.*]] = load i32, i32* [[TMP19]], align 4
// CHECK14-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP20]])
// CHECK14-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK14:       omp.precond.end:
// CHECK14-NEXT:    ret void
// CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l43
// CHECK15-SAME: (i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK15-NEXT:  entry:
// CHECK15-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK15-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[L_CASTED:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK15-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK15-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK15-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK15-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK15:       .execute:
// CHECK15-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3:[0-9]+]])
// CHECK15-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK15-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK15-NEXT:    [[TMP4:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP4]], i32* [[L_CASTED]], align 4
// CHECK15-NEXT:    [[TMP5:%.*]] = load i32, i32* [[L_CASTED]], align 4
// CHECK15-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK15-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [1000 x i32]* [[TMP0]], i32 [[TMP5]]) #[[ATTR3:[0-9]+]]
// CHECK15-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK15:       .omp.deinit:
// CHECK15-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK15-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK15:       .exit:
// CHECK15-NEXT:    ret void
// CHECK15-LABEL: define {{[^@]+}}@__omp_outlined__
// CHECK15-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0]] {
// CHECK15-NEXT:  entry:
// CHECK15-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK15-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[I4:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[L_CASTED:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 4
// CHECK15-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK15-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK15-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK15-NEXT:    [[TMP1:%.*]] = call i8* @__kmpc_data_sharing_push_stack(i32 4, i16 1)
// CHECK15-NEXT:    [[TMP2:%.*]] = bitcast i8* [[TMP1]] to %struct._globalized_locals_ty*
// CHECK15-NEXT:    [[L1:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY:%.*]], %struct._globalized_locals_ty* [[TMP2]], i32 0, i32 0
// CHECK15-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP3]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK15-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK15-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK15-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK15-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK15-NEXT:    store i32 [[SUB3]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK15-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK15-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK15-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK15-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK15:       omp.precond.then:
// CHECK15-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK15-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK15-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK15-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1:[0-9]+]], i32 [[TMP8]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 128)
// CHECK15-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK15-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
// CHECK15-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK15:       cond.true:
// CHECK15-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK15-NEXT:    br label [[COND_END:%.*]]
// CHECK15:       cond.false:
// CHECK15-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    br label [[COND_END]]
// CHECK15:       cond.end:
// CHECK15-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP11]], [[COND_TRUE]] ], [ [[TMP12]], [[COND_FALSE]] ]
// CHECK15-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    store i32 [[TMP13]], i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK15:       omp.inner.for.cond:
// CHECK15-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK15-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP15]], 1
// CHECK15-NEXT:    [[CMP6:%.*]] = icmp slt i32 [[TMP14]], [[ADD]]
// CHECK15-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK15:       omp.inner.for.body:
// CHECK15-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP18:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP18]], i32* [[N_CASTED]], align 4
// CHECK15-NEXT:    [[TMP19:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK15-NEXT:    [[TMP20:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP20]], i32* [[L_CASTED]], align 4
// CHECK15-NEXT:    [[TMP21:%.*]] = load i32, i32* [[L_CASTED]], align 4
// CHECK15-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK15-NEXT:    [[TMP23:%.*]] = inttoptr i32 [[TMP16]] to i8*
// CHECK15-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 4
// CHECK15-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK15-NEXT:    [[TMP25:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK15-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 4
// CHECK15-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK15-NEXT:    [[TMP27:%.*]] = inttoptr i32 [[TMP19]] to i8*
// CHECK15-NEXT:    store i8* [[TMP27]], i8** [[TMP26]], align 4
// CHECK15-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK15-NEXT:    [[TMP29:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK15-NEXT:    store i8* [[TMP29]], i8** [[TMP28]], align 4
// CHECK15-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 4
// CHECK15-NEXT:    [[TMP31:%.*]] = inttoptr i32 [[TMP21]] to i8*
// CHECK15-NEXT:    store i8* [[TMP31]], i8** [[TMP30]], align 4
// CHECK15-NEXT:    [[TMP32:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    [[TMP33:%.*]] = load i32, i32* [[TMP32]], align 4
// CHECK15-NEXT:    [[TMP34:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK15-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP33]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i32]*, i32)* @__omp_outlined__1 to i8*), i8* null, i8** [[TMP34]], i32 5)
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK15:       omp.inner.for.inc:
// CHECK15-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP35]], [[TMP36]]
// CHECK15-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP37]], [[TMP38]]
// CHECK15-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP39]], [[TMP40]]
// CHECK15-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK15-NEXT:    [[CMP10:%.*]] = icmp sgt i32 [[TMP41]], [[TMP42]]
// CHECK15-NEXT:    br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]]
// CHECK15:       cond.true11:
// CHECK15-NEXT:    [[TMP43:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK15-NEXT:    br label [[COND_END13:%.*]]
// CHECK15:       cond.false12:
// CHECK15-NEXT:    [[TMP44:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    br label [[COND_END13]]
// CHECK15:       cond.end13:
// CHECK15-NEXT:    [[COND14:%.*]] = phi i32 [ [[TMP43]], [[COND_TRUE11]] ], [ [[TMP44]], [[COND_FALSE12]] ]
// CHECK15-NEXT:    store i32 [[COND14]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP45:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    store i32 [[TMP45]], i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK15:       omp.inner.for.end:
// CHECK15-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK15:       omp.loop.exit:
// CHECK15-NEXT:    [[TMP46:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    [[TMP47:%.*]] = load i32, i32* [[TMP46]], align 4
// CHECK15-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP47]])
// CHECK15-NEXT:    [[TMP48:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK15-NEXT:    [[TMP49:%.*]] = icmp ne i32 [[TMP48]], 0
// CHECK15-NEXT:    br i1 [[TMP49]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK15:       .omp.lastprivate.then:
// CHECK15-NEXT:    [[TMP50:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP50]], i32* [[L_ADDR]], align 4
// CHECK15-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK15:       .omp.lastprivate.done:
// CHECK15-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK15:       omp.precond.end:
// CHECK15-NEXT:    call void @__kmpc_data_sharing_pop_stack(i8* [[TMP1]])
// CHECK15-NEXT:    ret void
// CHECK15-LABEL: define {{[^@]+}}@__omp_outlined__1
// CHECK15-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0]] {
// CHECK15-NEXT:  entry:
// CHECK15-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK15-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK15-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK15-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK15-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK15-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK15-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK15-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK15-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK15-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK15-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK15-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK15-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK15-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK15-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK15-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK15:       omp.precond.then:
// CHECK15-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK15-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK15-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK15-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK15-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK15-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK15-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK15-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK15-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 32)
// CHECK15-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK15:       omp.dispatch.cond:
// CHECK15-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK15-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK15-NEXT:    [[CMP4:%.*]] = icmp ugt i32 [[TMP9]], [[TMP10]]
// CHECK15-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK15:       cond.true:
// CHECK15-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK15-NEXT:    br label [[COND_END:%.*]]
// CHECK15:       cond.false:
// CHECK15-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK15-NEXT:    br label [[COND_END]]
// CHECK15:       cond.end:
// CHECK15-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP11]], [[COND_TRUE]] ], [ [[TMP12]], [[COND_FALSE]] ]
// CHECK15-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
// CHECK15-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK15-NEXT:    store i32 [[TMP13]], i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK15-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK15-NEXT:    br i1 [[CMP5]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK15:       omp.dispatch.body:
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK15:       omp.inner.for.cond:
// CHECK15-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK15-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP16]], [[TMP17]]
// CHECK15-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK15:       omp.inner.for.body:
// CHECK15-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP18]], 1
// CHECK15-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK15-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK15-NEXT:    [[TMP19:%.*]] = load i32, i32* [[I3]], align 4
// CHECK15-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i32 0, i32 [[TMP19]]
// CHECK15-NEXT:    store i32 1, i32* [[ARRAYIDX]], align 4
// CHECK15-NEXT:    [[TMP20:%.*]] = load i32, i32* [[I3]], align 4
// CHECK15-NEXT:    store i32 [[TMP20]], i32* [[L_ADDR]], align 4
// CHECK15-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK15:       omp.body.continue:
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK15:       omp.inner.for.inc:
// CHECK15-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP21]], 1
// CHECK15-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK15:       omp.inner.for.end:
// CHECK15-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK15:       omp.dispatch.inc:
// CHECK15-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK15-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK15-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_LB]], align 4
// CHECK15-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK15-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK15-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_UB]], align 4
// CHECK15-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK15:       omp.dispatch.end:
// CHECK15-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK15-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP27]])
// CHECK15-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK15-NEXT:    [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
// CHECK15-NEXT:    br i1 [[TMP29]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK15:       .omp.lastprivate.then:
// CHECK15-NEXT:    [[TMP30:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP30]], i32* [[L_ADDR]], align 4
// CHECK15-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK15:       .omp.lastprivate.done:
// CHECK15-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK15:       omp.precond.end:
// CHECK15-NEXT:    ret void
// CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l49
// CHECK15-SAME: (i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK15-NEXT:  entry:
// CHECK15-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK15-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK15-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK15-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK15-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK15-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK15:       .execute:
// CHECK15-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK15-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK15-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK15-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK15-NEXT:    call void @__omp_outlined__2(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [1000 x i16]* [[TMP0]]) #[[ATTR3]]
// CHECK15-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK15:       .omp.deinit:
// CHECK15-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK15-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK15:       .exit:
// CHECK15-NEXT:    ret void
// CHECK15-LABEL: define {{[^@]+}}@__omp_outlined__2
// CHECK15-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK15-NEXT:  entry:
// CHECK15-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK15-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK15-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK15-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK15-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK15-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK15-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK15-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK15-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK15-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK15-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK15-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK15-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK15-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK15-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK15:       omp.precond.then:
// CHECK15-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK15-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK15-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK15-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK15-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK15-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK15-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK15-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK15:       cond.true:
// CHECK15-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK15-NEXT:    br label [[COND_END:%.*]]
// CHECK15:       cond.false:
// CHECK15-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    br label [[COND_END]]
// CHECK15:       cond.end:
// CHECK15-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK15-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK15:       omp.inner.for.cond:
// CHECK15-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK15-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK15-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK15-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK15:       omp.inner.for.body:
// CHECK15-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP16:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP16]], i32* [[N_CASTED]], align 4
// CHECK15-NEXT:    [[TMP17:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK15-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK15-NEXT:    [[TMP19:%.*]] = inttoptr i32 [[TMP14]] to i8*
// CHECK15-NEXT:    store i8* [[TMP19]], i8** [[TMP18]], align 4
// CHECK15-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK15-NEXT:    [[TMP21:%.*]] = inttoptr i32 [[TMP15]] to i8*
// CHECK15-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 4
// CHECK15-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK15-NEXT:    [[TMP23:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK15-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 4
// CHECK15-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK15-NEXT:    [[TMP25:%.*]] = bitcast [1000 x i16]* [[TMP0]] to i8*
// CHECK15-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 4
// CHECK15-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK15-NEXT:    [[TMP28:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK15-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP27]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i16]*)* @__omp_outlined__3 to i8*), i8* null, i8** [[TMP28]], i32 4)
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK15:       omp.inner.for.inc:
// CHECK15-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP29]], [[TMP30]]
// CHECK15-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP31]], [[TMP32]]
// CHECK15-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP33]], [[TMP34]]
// CHECK15-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK15-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP35]], [[TMP36]]
// CHECK15-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
// CHECK15:       cond.true10:
// CHECK15-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK15-NEXT:    br label [[COND_END12:%.*]]
// CHECK15:       cond.false11:
// CHECK15-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    br label [[COND_END12]]
// CHECK15:       cond.end12:
// CHECK15-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP37]], [[COND_TRUE10]] ], [ [[TMP38]], [[COND_FALSE11]] ]
// CHECK15-NEXT:    store i32 [[COND13]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    store i32 [[TMP39]], i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK15:       omp.inner.for.end:
// CHECK15-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK15:       omp.loop.exit:
// CHECK15-NEXT:    [[TMP40:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    [[TMP41:%.*]] = load i32, i32* [[TMP40]], align 4
// CHECK15-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP41]])
// CHECK15-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK15:       omp.precond.end:
// CHECK15-NEXT:    ret void
// CHECK15-LABEL: define {{[^@]+}}@__omp_outlined__3
// CHECK15-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK15-NEXT:  entry:
// CHECK15-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK15-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK15-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK15-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK15-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK15-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK15-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK15-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK15-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK15-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK15-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK15-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK15-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK15-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK15-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK15-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK15:       omp.precond.then:
// CHECK15-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK15-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK15-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK15-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK15-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK15-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK15-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK15-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK15-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK15-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK15-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK15:       omp.inner.for.cond:
// CHECK15-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK15-NEXT:    [[CMP4:%.*]] = icmp ule i32 [[TMP10]], [[TMP11]]
// CHECK15-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK15:       omp.inner.for.body:
// CHECK15-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK15-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK15-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK15-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I3]], align 4
// CHECK15-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i16], [1000 x i16]* [[TMP0]], i32 0, i32 [[TMP13]]
// CHECK15-NEXT:    [[TMP14:%.*]] = load i16, i16* [[ARRAYIDX]], align 2
// CHECK15-NEXT:    [[CONV:%.*]] = sext i16 [[TMP14]] to i32
// CHECK15-NEXT:    [[ADD5:%.*]] = add nsw i32 [[CONV]], 1
// CHECK15-NEXT:    [[CONV6:%.*]] = trunc i32 [[ADD5]] to i16
// CHECK15-NEXT:    store i16 [[CONV6]], i16* [[ARRAYIDX]], align 2
// CHECK15-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK15:       omp.body.continue:
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK15:       omp.inner.for.inc:
// CHECK15-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP15]], [[TMP16]]
// CHECK15-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK15:       omp.inner.for.end:
// CHECK15-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK15:       omp.loop.exit:
// CHECK15-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
// CHECK15-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP18]])
// CHECK15-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK15:       omp.precond.end:
// CHECK15-NEXT:    ret void
// CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l54
// CHECK15-SAME: ([10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK15-NEXT:  entry:
// CHECK15-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK15-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK15-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK15-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK15-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK15-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK15-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK15:       .execute:
// CHECK15-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK15-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK15-NEXT:    call void @__omp_outlined__4(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x i32]* [[TMP0]]) #[[ATTR3]]
// CHECK15-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK15:       .omp.deinit:
// CHECK15-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK15-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK15:       .exit:
// CHECK15-NEXT:    ret void
// CHECK15-LABEL: define {{[^@]+}}@__omp_outlined__4
// CHECK15-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK15-NEXT:  entry:
// CHECK15-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK15-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x i8*], align 4
// CHECK15-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK15-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK15-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK15-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    store i32 9, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK15-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK15-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK15-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK15-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9
// CHECK15-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK15:       cond.true:
// CHECK15-NEXT:    br label [[COND_END:%.*]]
// CHECK15:       cond.false:
// CHECK15-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    br label [[COND_END]]
// CHECK15:       cond.end:
// CHECK15-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK15-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK15:       omp.inner.for.cond:
// CHECK15-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[CMP1:%.*]] = icmp slt i32 [[TMP6]], 10
// CHECK15-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK15:       omp.inner.for.body:
// CHECK15-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK15-NEXT:    [[TMP10:%.*]] = inttoptr i32 [[TMP7]] to i8*
// CHECK15-NEXT:    store i8* [[TMP10]], i8** [[TMP9]], align 4
// CHECK15-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK15-NEXT:    [[TMP12:%.*]] = inttoptr i32 [[TMP8]] to i8*
// CHECK15-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 4
// CHECK15-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK15-NEXT:    [[TMP14:%.*]] = bitcast [10 x i32]* [[TMP0]] to i8*
// CHECK15-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 4
// CHECK15-NEXT:    [[TMP15:%.*]] = bitcast [3 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK15-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, [10 x i32]*)* @__omp_outlined__5 to i8*), i8* null, i8** [[TMP15]], i32 3)
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK15:       omp.inner.for.inc:
// CHECK15-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP16]], [[TMP17]]
// CHECK15-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP18]], [[TMP19]]
// CHECK15-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK15-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP22]], 9
// CHECK15-NEXT:    br i1 [[CMP4]], label [[COND_TRUE5:%.*]], label [[COND_FALSE6:%.*]]
// CHECK15:       cond.true5:
// CHECK15-NEXT:    br label [[COND_END7:%.*]]
// CHECK15:       cond.false6:
// CHECK15-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    br label [[COND_END7]]
// CHECK15:       cond.end7:
// CHECK15-NEXT:    [[COND8:%.*]] = phi i32 [ 9, [[COND_TRUE5]] ], [ [[TMP23]], [[COND_FALSE6]] ]
// CHECK15-NEXT:    store i32 [[COND8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    store i32 [[TMP24]], i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK15:       omp.inner.for.end:
// CHECK15-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK15:       omp.loop.exit:
// CHECK15-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]])
// CHECK15-NEXT:    ret void
// CHECK15-LABEL: define {{[^@]+}}@__omp_outlined__5
// CHECK15-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK15-NEXT:  entry:
// CHECK15-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK15-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK15-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK15-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK15-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK15-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK15-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK15-NEXT:    store i32 9, i32* [[DOTOMP_UB]], align 4
// CHECK15-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK15-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP1]], i32* [[DOTOMP_LB]], align 4
// CHECK15-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_UB]], align 4
// CHECK15-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK15-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK15-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK15-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK15-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK15:       omp.inner.for.cond:
// CHECK15-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK15-NEXT:    [[CMP:%.*]] = icmp ule i32 [[TMP6]], [[TMP7]]
// CHECK15-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK15:       omp.inner.for.body:
// CHECK15-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP8]], 1
// CHECK15-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK15-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK15-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4
// CHECK15-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[TMP0]], i32 0, i32 [[TMP9]]
// CHECK15-NEXT:    [[TMP10:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK15-NEXT:    [[ADD1:%.*]] = add nsw i32 [[TMP10]], 1
// CHECK15-NEXT:    store i32 [[ADD1]], i32* [[ARRAYIDX]], align 4
// CHECK15-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK15:       omp.body.continue:
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK15:       omp.inner.for.inc:
// CHECK15-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
// CHECK15-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK15:       omp.inner.for.end:
// CHECK15-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK15:       omp.loop.exit:
// CHECK15-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK15-NEXT:    ret void
// CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l59
// CHECK15-SAME: ([10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK15-NEXT:  entry:
// CHECK15-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK15-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[F_CASTED:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK15-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK15-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK15-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK15-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK15-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK15:       .execute:
// CHECK15-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK15-NEXT:    [[TMP2:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP2]], i32* [[F_CASTED]], align 4
// CHECK15-NEXT:    [[TMP3:%.*]] = load i32, i32* [[F_CASTED]], align 4
// CHECK15-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK15-NEXT:    call void @__omp_outlined__6(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x [10 x i32]]* [[TMP0]], i32 [[TMP3]]) #[[ATTR3]]
// CHECK15-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK15:       .omp.deinit:
// CHECK15-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK15-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK15:       .exit:
// CHECK15-NEXT:    ret void
// CHECK15-LABEL: define {{[^@]+}}@__omp_outlined__6
// CHECK15-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK15-NEXT:  entry:
// CHECK15-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK15-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[F_CASTED:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK15-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK15-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK15-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK15-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    store i32 99, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK15-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK15-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK15-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK15-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 99
// CHECK15-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK15:       cond.true:
// CHECK15-NEXT:    br label [[COND_END:%.*]]
// CHECK15:       cond.false:
// CHECK15-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    br label [[COND_END]]
// CHECK15:       cond.end:
// CHECK15-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK15-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK15:       omp.inner.for.cond:
// CHECK15-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[TMP6]], 100
// CHECK15-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK15:       omp.inner.for.body:
// CHECK15-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP9:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP9]], i32* [[F_CASTED]], align 4
// CHECK15-NEXT:    [[TMP10:%.*]] = load i32, i32* [[F_CASTED]], align 4
// CHECK15-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK15-NEXT:    [[TMP12:%.*]] = inttoptr i32 [[TMP7]] to i8*
// CHECK15-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 4
// CHECK15-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK15-NEXT:    [[TMP14:%.*]] = inttoptr i32 [[TMP8]] to i8*
// CHECK15-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 4
// CHECK15-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK15-NEXT:    [[TMP16:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK15-NEXT:    store i8* [[TMP16]], i8** [[TMP15]], align 4
// CHECK15-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK15-NEXT:    [[TMP18:%.*]] = inttoptr i32 [[TMP10]] to i8*
// CHECK15-NEXT:    store i8* [[TMP18]], i8** [[TMP17]], align 4
// CHECK15-NEXT:    [[TMP19:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK15-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, [10 x [10 x i32]]*, i32)* @__omp_outlined__7 to i8*), i8* null, i8** [[TMP19]], i32 4)
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK15:       omp.inner.for.inc:
// CHECK15-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK15-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK15-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK15-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP26]], 99
// CHECK15-NEXT:    br i1 [[CMP5]], label [[COND_TRUE6:%.*]], label [[COND_FALSE7:%.*]]
// CHECK15:       cond.true6:
// CHECK15-NEXT:    br label [[COND_END8:%.*]]
// CHECK15:       cond.false7:
// CHECK15-NEXT:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    br label [[COND_END8]]
// CHECK15:       cond.end8:
// CHECK15-NEXT:    [[COND9:%.*]] = phi i32 [ 99, [[COND_TRUE6]] ], [ [[TMP27]], [[COND_FALSE7]] ]
// CHECK15-NEXT:    store i32 [[COND9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    store i32 [[TMP28]], i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK15:       omp.inner.for.end:
// CHECK15-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK15:       omp.loop.exit:
// CHECK15-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]])
// CHECK15-NEXT:    ret void
// CHECK15-LABEL: define {{[^@]+}}@__omp_outlined__7
// CHECK15-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK15-NEXT:  entry:
// CHECK15-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK15-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK15-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK15-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK15-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK15-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK15-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK15-NEXT:    store i32 99, i32* [[DOTOMP_UB]], align 4
// CHECK15-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK15-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP1]], i32* [[DOTOMP_LB]], align 4
// CHECK15-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_UB]], align 4
// CHECK15-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK15-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK15-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK15-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK15-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK15:       omp.inner.for.cond:
// CHECK15-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK15-NEXT:    [[CMP:%.*]] = icmp ule i32 [[TMP6]], [[TMP7]]
// CHECK15-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK15:       omp.inner.for.body:
// CHECK15-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP8]], 10
// CHECK15-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK15-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK15-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK15-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[DIV2:%.*]] = sdiv i32 [[TMP10]], 10
// CHECK15-NEXT:    [[MUL3:%.*]] = mul nsw i32 [[DIV2]], 10
// CHECK15-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP9]], [[MUL3]]
// CHECK15-NEXT:    [[MUL4:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK15-NEXT:    [[ADD5:%.*]] = add nsw i32 0, [[MUL4]]
// CHECK15-NEXT:    store i32 [[ADD5]], i32* [[J]], align 4
// CHECK15-NEXT:    store i32 10, i32* [[K]], align 4
// CHECK15-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
// CHECK15-NEXT:    [[TMP12:%.*]] = load i32, i32* [[J]], align 4
// CHECK15-NEXT:    [[TMP13:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK15-NEXT:    [[MUL6:%.*]] = mul nsw i32 [[TMP12]], [[TMP13]]
// CHECK15-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP11]], [[MUL6]]
// CHECK15-NEXT:    [[TMP14:%.*]] = load i32, i32* [[K]], align 4
// CHECK15-NEXT:    [[ADD8:%.*]] = add nsw i32 [[ADD7]], [[TMP14]]
// CHECK15-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
// CHECK15-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i32 0, i32 [[TMP15]]
// CHECK15-NEXT:    [[TMP16:%.*]] = load i32, i32* [[J]], align 4
// CHECK15-NEXT:    [[ARRAYIDX9:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i32 0, i32 [[TMP16]]
// CHECK15-NEXT:    store i32 [[ADD8]], i32* [[ARRAYIDX9]], align 4
// CHECK15-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK15:       omp.body.continue:
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK15:       omp.inner.for.inc:
// CHECK15-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK15-NEXT:    store i32 [[ADD10]], i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK15:       omp.inner.for.end:
// CHECK15-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK15:       omp.loop.exit:
// CHECK15-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK15-NEXT:    ret void
// CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l67
// CHECK15-SAME: (i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK15-NEXT:  entry:
// CHECK15-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK15-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK15-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK15-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK15-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK15-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK15:       .execute:
// CHECK15-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK15-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK15-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK15-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK15-NEXT:    call void @__omp_outlined__8(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [10 x [10 x i32]]* [[TMP0]]) #[[ATTR3]]
// CHECK15-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK15:       .omp.deinit:
// CHECK15-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK15-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK15:       .exit:
// CHECK15-NEXT:    ret void
// CHECK15-LABEL: define {{[^@]+}}@__omp_outlined__8
// CHECK15-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK15-NEXT:  entry:
// CHECK15-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK15-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK15-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK15-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i64, align 8
// CHECK15-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i64, align 8
// CHECK15-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK15-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[I9:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[J10:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK15-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK15-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK15-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK15-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK15-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK15-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK15-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK15-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK15-NEXT:    [[CONV:%.*]] = sext i32 [[DIV]] to i64
// CHECK15-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK15-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK15-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK15-NEXT:    [[CONV6:%.*]] = sext i32 [[DIV5]] to i64
// CHECK15-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV]], [[CONV6]]
// CHECK15-NEXT:    [[SUB7:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK15-NEXT:    store i64 [[SUB7]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK15-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK15-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK15-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK15-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK15-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK15:       land.lhs.true:
// CHECK15-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK15-NEXT:    [[CMP8:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK15-NEXT:    br i1 [[CMP8]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK15:       omp.precond.then:
// CHECK15-NEXT:    store i64 0, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK15-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK15-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK15-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK15-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK15-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK15-NEXT:    [[CONV11:%.*]] = zext i32 [[NVPTX_NUM_THREADS]] to i64
// CHECK15-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    [[TMP9:%.*]] = load i32, i32* [[TMP8]], align 4
// CHECK15-NEXT:    call void @__kmpc_for_static_init_8(%struct.ident_t* @[[GLOB1]], i32 [[TMP9]], i32 91, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_COMB_LB]], i64* [[DOTOMP_COMB_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 [[CONV11]])
// CHECK15-NEXT:    [[TMP10:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK15-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK15-NEXT:    [[CMP12:%.*]] = icmp sgt i64 [[TMP10]], [[TMP11]]
// CHECK15-NEXT:    br i1 [[CMP12]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK15:       cond.true:
// CHECK15-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK15-NEXT:    br label [[COND_END:%.*]]
// CHECK15:       cond.false:
// CHECK15-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK15-NEXT:    br label [[COND_END]]
// CHECK15:       cond.end:
// CHECK15-NEXT:    [[COND:%.*]] = phi i64 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
// CHECK15-NEXT:    store i64 [[COND]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK15-NEXT:    [[TMP14:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK15-NEXT:    store i64 [[TMP14]], i64* [[DOTOMP_IV]], align 8
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK15:       omp.inner.for.cond:
// CHECK15-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK15-NEXT:    [[TMP16:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK15-NEXT:    [[ADD:%.*]] = add nsw i64 [[TMP16]], 1
// CHECK15-NEXT:    [[CMP13:%.*]] = icmp slt i64 [[TMP15]], [[ADD]]
// CHECK15-NEXT:    br i1 [[CMP13]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK15:       omp.inner.for.body:
// CHECK15-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK15-NEXT:    [[TMP18:%.*]] = trunc i64 [[TMP17]] to i32
// CHECK15-NEXT:    [[TMP19:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK15-NEXT:    [[TMP20:%.*]] = trunc i64 [[TMP19]] to i32
// CHECK15-NEXT:    [[TMP21:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP21]], i32* [[N_CASTED]], align 4
// CHECK15-NEXT:    [[TMP22:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK15-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK15-NEXT:    [[TMP24:%.*]] = inttoptr i32 [[TMP18]] to i8*
// CHECK15-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 4
// CHECK15-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK15-NEXT:    [[TMP26:%.*]] = inttoptr i32 [[TMP20]] to i8*
// CHECK15-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 4
// CHECK15-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK15-NEXT:    [[TMP28:%.*]] = inttoptr i32 [[TMP22]] to i8*
// CHECK15-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 4
// CHECK15-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK15-NEXT:    [[TMP30:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK15-NEXT:    store i8* [[TMP30]], i8** [[TMP29]], align 4
// CHECK15-NEXT:    [[TMP31:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    [[TMP32:%.*]] = load i32, i32* [[TMP31]], align 4
// CHECK15-NEXT:    [[TMP33:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK15-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP32]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [10 x [10 x i32]]*)* @__omp_outlined__9 to i8*), i8* null, i8** [[TMP33]], i32 4)
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK15:       omp.inner.for.inc:
// CHECK15-NEXT:    [[TMP34:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK15-NEXT:    [[TMP35:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK15-NEXT:    [[ADD14:%.*]] = add nsw i64 [[TMP34]], [[TMP35]]
// CHECK15-NEXT:    store i64 [[ADD14]], i64* [[DOTOMP_IV]], align 8
// CHECK15-NEXT:    [[TMP36:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK15-NEXT:    [[TMP37:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK15-NEXT:    [[ADD15:%.*]] = add nsw i64 [[TMP36]], [[TMP37]]
// CHECK15-NEXT:    store i64 [[ADD15]], i64* [[DOTOMP_COMB_LB]], align 8
// CHECK15-NEXT:    [[TMP38:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK15-NEXT:    [[TMP39:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK15-NEXT:    [[ADD16:%.*]] = add nsw i64 [[TMP38]], [[TMP39]]
// CHECK15-NEXT:    store i64 [[ADD16]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK15-NEXT:    [[TMP40:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK15-NEXT:    [[TMP41:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK15-NEXT:    [[CMP17:%.*]] = icmp sgt i64 [[TMP40]], [[TMP41]]
// CHECK15-NEXT:    br i1 [[CMP17]], label [[COND_TRUE18:%.*]], label [[COND_FALSE19:%.*]]
// CHECK15:       cond.true18:
// CHECK15-NEXT:    [[TMP42:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK15-NEXT:    br label [[COND_END20:%.*]]
// CHECK15:       cond.false19:
// CHECK15-NEXT:    [[TMP43:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK15-NEXT:    br label [[COND_END20]]
// CHECK15:       cond.end20:
// CHECK15-NEXT:    [[COND21:%.*]] = phi i64 [ [[TMP42]], [[COND_TRUE18]] ], [ [[TMP43]], [[COND_FALSE19]] ]
// CHECK15-NEXT:    store i64 [[COND21]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK15-NEXT:    [[TMP44:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK15-NEXT:    store i64 [[TMP44]], i64* [[DOTOMP_IV]], align 8
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK15:       omp.inner.for.end:
// CHECK15-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK15:       omp.loop.exit:
// CHECK15-NEXT:    [[TMP45:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    [[TMP46:%.*]] = load i32, i32* [[TMP45]], align 4
// CHECK15-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP46]])
// CHECK15-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK15:       omp.precond.end:
// CHECK15-NEXT:    ret void
// CHECK15-LABEL: define {{[^@]+}}@__omp_outlined__9
// CHECK15-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK15-NEXT:  entry:
// CHECK15-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK15-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK15-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK15-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_LB:%.*]] = alloca i64, align 8
// CHECK15-NEXT:    [[DOTOMP_UB:%.*]] = alloca i64, align 8
// CHECK15-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK15-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[I11:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[J12:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK15-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK15-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK15-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK15-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK15-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK15-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK15-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK15-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK15-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK15-NEXT:    [[CONV:%.*]] = sext i32 [[DIV]] to i64
// CHECK15-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK15-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK15-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK15-NEXT:    [[CONV6:%.*]] = sext i32 [[DIV5]] to i64
// CHECK15-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV]], [[CONV6]]
// CHECK15-NEXT:    [[SUB7:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK15-NEXT:    store i64 [[SUB7]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK15-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK15-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK15-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK15-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK15-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK15:       land.lhs.true:
// CHECK15-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK15-NEXT:    [[CMP8:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK15-NEXT:    br i1 [[CMP8]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK15:       omp.precond.then:
// CHECK15-NEXT:    store i64 0, i64* [[DOTOMP_LB]], align 8
// CHECK15-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK15-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_UB]], align 8
// CHECK15-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK15-NEXT:    [[CONV9:%.*]] = zext i32 [[TMP8]] to i64
// CHECK15-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK15-NEXT:    [[CONV10:%.*]] = zext i32 [[TMP9]] to i64
// CHECK15-NEXT:    store i64 [[CONV9]], i64* [[DOTOMP_LB]], align 8
// CHECK15-NEXT:    store i64 [[CONV10]], i64* [[DOTOMP_UB]], align 8
// CHECK15-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK15-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK15-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK15-NEXT:    call void @__kmpc_for_static_init_8(%struct.ident_t* @[[GLOB2]], i32 [[TMP11]], i32 33, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_LB]], i64* [[DOTOMP_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 1)
// CHECK15-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTOMP_LB]], align 8
// CHECK15-NEXT:    store i64 [[TMP12]], i64* [[DOTOMP_IV]], align 8
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK15:       omp.inner.for.cond:
// CHECK15-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK15-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK15-NEXT:    [[CONV13:%.*]] = zext i32 [[TMP14]] to i64
// CHECK15-NEXT:    [[CMP14:%.*]] = icmp sle i64 [[TMP13]], [[CONV13]]
// CHECK15-NEXT:    br i1 [[CMP14]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK15:       omp.inner.for.body:
// CHECK15-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK15-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK15-NEXT:    [[SUB15:%.*]] = sub nsw i32 [[TMP16]], 0
// CHECK15-NEXT:    [[DIV16:%.*]] = sdiv i32 [[SUB15]], 1
// CHECK15-NEXT:    [[MUL17:%.*]] = mul nsw i32 1, [[DIV16]]
// CHECK15-NEXT:    [[CONV18:%.*]] = sext i32 [[MUL17]] to i64
// CHECK15-NEXT:    [[DIV19:%.*]] = sdiv i64 [[TMP15]], [[CONV18]]
// CHECK15-NEXT:    [[MUL20:%.*]] = mul nsw i64 [[DIV19]], 1
// CHECK15-NEXT:    [[ADD:%.*]] = add nsw i64 0, [[MUL20]]
// CHECK15-NEXT:    [[CONV21:%.*]] = trunc i64 [[ADD]] to i32
// CHECK15-NEXT:    store i32 [[CONV21]], i32* [[I11]], align 4
// CHECK15-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK15-NEXT:    [[TMP18:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK15-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK15-NEXT:    [[SUB22:%.*]] = sub nsw i32 [[TMP19]], 0
// CHECK15-NEXT:    [[DIV23:%.*]] = sdiv i32 [[SUB22]], 1
// CHECK15-NEXT:    [[MUL24:%.*]] = mul nsw i32 1, [[DIV23]]
// CHECK15-NEXT:    [[CONV25:%.*]] = sext i32 [[MUL24]] to i64
// CHECK15-NEXT:    [[DIV26:%.*]] = sdiv i64 [[TMP18]], [[CONV25]]
// CHECK15-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK15-NEXT:    [[SUB27:%.*]] = sub nsw i32 [[TMP20]], 0
// CHECK15-NEXT:    [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK15-NEXT:    [[MUL29:%.*]] = mul nsw i32 1, [[DIV28]]
// CHECK15-NEXT:    [[CONV30:%.*]] = sext i32 [[MUL29]] to i64
// CHECK15-NEXT:    [[MUL31:%.*]] = mul nsw i64 [[DIV26]], [[CONV30]]
// CHECK15-NEXT:    [[SUB32:%.*]] = sub nsw i64 [[TMP17]], [[MUL31]]
// CHECK15-NEXT:    [[MUL33:%.*]] = mul nsw i64 [[SUB32]], 1
// CHECK15-NEXT:    [[ADD34:%.*]] = add nsw i64 0, [[MUL33]]
// CHECK15-NEXT:    [[CONV35:%.*]] = trunc i64 [[ADD34]] to i32
// CHECK15-NEXT:    store i32 [[CONV35]], i32* [[J12]], align 4
// CHECK15-NEXT:    [[TMP21:%.*]] = load i32, i32* [[I11]], align 4
// CHECK15-NEXT:    [[TMP22:%.*]] = load i32, i32* [[J12]], align 4
// CHECK15-NEXT:    [[ADD36:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
// CHECK15-NEXT:    [[TMP23:%.*]] = load i32, i32* [[I11]], align 4
// CHECK15-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i32 0, i32 [[TMP23]]
// CHECK15-NEXT:    [[TMP24:%.*]] = load i32, i32* [[J12]], align 4
// CHECK15-NEXT:    [[ARRAYIDX37:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i32 0, i32 [[TMP24]]
// CHECK15-NEXT:    store i32 [[ADD36]], i32* [[ARRAYIDX37]], align 4
// CHECK15-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK15:       omp.body.continue:
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK15:       omp.inner.for.inc:
// CHECK15-NEXT:    [[TMP25:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK15-NEXT:    [[TMP26:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK15-NEXT:    [[ADD38:%.*]] = add nsw i64 [[TMP25]], [[TMP26]]
// CHECK15-NEXT:    store i64 [[ADD38]], i64* [[DOTOMP_IV]], align 8
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK15:       omp.inner.for.end:
// CHECK15-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK15:       omp.loop.exit:
// CHECK15-NEXT:    [[TMP27:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    [[TMP28:%.*]] = load i32, i32* [[TMP27]], align 4
// CHECK15-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP28]])
// CHECK15-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK15:       omp.precond.end:
// CHECK15-NEXT:    ret void
// CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l74
// CHECK15-SAME: (i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK15-NEXT:  entry:
// CHECK15-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK15-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK15-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK15-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK15-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK15-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK15-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK15:       .execute:
// CHECK15-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK15-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK15-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK15-NEXT:    [[TMP4:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK15-NEXT:    call void @__omp_outlined__10(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [1000 x i32]* [[TMP0]], i32* [[TMP4]]) #[[ATTR3]]
// CHECK15-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK15:       .omp.deinit:
// CHECK15-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK15-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK15:       .exit:
// CHECK15-NEXT:    ret void
// CHECK15-LABEL: define {{[^@]+}}@__omp_outlined__10
// CHECK15-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK15-NEXT:  entry:
// CHECK15-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK15-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 4
// CHECK15-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK15-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK15-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK15-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK15-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK15-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK15-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK15-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK15-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK15-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK15-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK15-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK15-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK15-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK15:       omp.precond.then:
// CHECK15-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK15-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK15-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK15-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK15-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK15-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK15-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK15-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK15:       cond.true:
// CHECK15-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK15-NEXT:    br label [[COND_END:%.*]]
// CHECK15:       cond.false:
// CHECK15-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    br label [[COND_END]]
// CHECK15:       cond.end:
// CHECK15-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK15-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK15:       omp.inner.for.cond:
// CHECK15-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK15-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK15-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK15-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK15:       omp.inner.for.body:
// CHECK15-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP16:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP16]], i32* [[N_CASTED]], align 4
// CHECK15-NEXT:    [[TMP17:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK15-NEXT:    [[TMP18:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK15-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK15-NEXT:    [[TMP20:%.*]] = inttoptr i32 [[TMP14]] to i8*
// CHECK15-NEXT:    store i8* [[TMP20]], i8** [[TMP19]], align 4
// CHECK15-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK15-NEXT:    [[TMP22:%.*]] = inttoptr i32 [[TMP15]] to i8*
// CHECK15-NEXT:    store i8* [[TMP22]], i8** [[TMP21]], align 4
// CHECK15-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK15-NEXT:    [[TMP24:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK15-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 4
// CHECK15-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK15-NEXT:    [[TMP26:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK15-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 4
// CHECK15-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 4
// CHECK15-NEXT:    [[TMP28:%.*]] = bitcast i32* [[TMP18]] to i8*
// CHECK15-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 4
// CHECK15-NEXT:    [[TMP29:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    [[TMP30:%.*]] = load i32, i32* [[TMP29]], align 4
// CHECK15-NEXT:    [[TMP31:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK15-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP30]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i32]*, i32*)* @__omp_outlined__11 to i8*), i8* null, i8** [[TMP31]], i32 5)
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK15:       omp.inner.for.inc:
// CHECK15-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP32]], [[TMP33]]
// CHECK15-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP34]], [[TMP35]]
// CHECK15-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP36]], [[TMP37]]
// CHECK15-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK15-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP38]], [[TMP39]]
// CHECK15-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
// CHECK15:       cond.true10:
// CHECK15-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK15-NEXT:    br label [[COND_END12:%.*]]
// CHECK15:       cond.false11:
// CHECK15-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    br label [[COND_END12]]
// CHECK15:       cond.end12:
// CHECK15-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP40]], [[COND_TRUE10]] ], [ [[TMP41]], [[COND_FALSE11]] ]
// CHECK15-NEXT:    store i32 [[COND13]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK15-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK15-NEXT:    store i32 [[TMP42]], i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK15:       omp.inner.for.end:
// CHECK15-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK15:       omp.loop.exit:
// CHECK15-NEXT:    [[TMP43:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    [[TMP44:%.*]] = load i32, i32* [[TMP43]], align 4
// CHECK15-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP44]])
// CHECK15-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK15:       omp.precond.end:
// CHECK15-NEXT:    ret void
// CHECK15-LABEL: define {{[^@]+}}@__omp_outlined__11
// CHECK15-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK15-NEXT:  entry:
// CHECK15-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK15-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK15-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK15-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK15-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK15-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK15-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK15-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK15-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK15-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK15-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK15-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK15-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK15-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK15-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK15-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK15-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK15-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK15-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK15:       omp.precond.then:
// CHECK15-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK15-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK15-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK15-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK15-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK15-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK15-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK15-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK15-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK15-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK15-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK15-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK15:       omp.inner.for.cond:
// CHECK15-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK15-NEXT:    [[CMP4:%.*]] = icmp ule i32 [[TMP10]], [[TMP11]]
// CHECK15-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK15:       omp.inner.for.body:
// CHECK15-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK15-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK15-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK15-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK15-NEXT:    [[TMP14:%.*]] = load i32, i32* [[I3]], align 4
// CHECK15-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[TMP13]], i32 [[TMP14]]
// CHECK15-NEXT:    [[TMP15:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK15-NEXT:    [[TMP16:%.*]] = load i32, i32* [[I3]], align 4
// CHECK15-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i32 0, i32 [[TMP16]]
// CHECK15-NEXT:    store i32 [[TMP15]], i32* [[ARRAYIDX5]], align 4
// CHECK15-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK15:       omp.body.continue:
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK15:       omp.inner.for.inc:
// CHECK15-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK15-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK15-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK15-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK15:       omp.inner.for.end:
// CHECK15-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK15:       omp.loop.exit:
// CHECK15-NEXT:    [[TMP19:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK15-NEXT:    [[TMP20:%.*]] = load i32, i32* [[TMP19]], align 4
// CHECK15-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP20]])
// CHECK15-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK15:       omp.precond.end:
// CHECK15-NEXT:    ret void
// CHECK16-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l43
// CHECK16-SAME: (i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK16-NEXT:  entry:
// CHECK16-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK16-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[L_CASTED:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK16-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK16-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK16-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK16-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK16:       .execute:
// CHECK16-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3:[0-9]+]])
// CHECK16-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK16-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK16-NEXT:    [[TMP4:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP4]], i32* [[L_CASTED]], align 4
// CHECK16-NEXT:    [[TMP5:%.*]] = load i32, i32* [[L_CASTED]], align 4
// CHECK16-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK16-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [1000 x i32]* [[TMP0]], i32 [[TMP5]]) #[[ATTR3:[0-9]+]]
// CHECK16-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK16:       .omp.deinit:
// CHECK16-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK16-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK16:       .exit:
// CHECK16-NEXT:    ret void
// CHECK16-LABEL: define {{[^@]+}}@__omp_outlined__
// CHECK16-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0]] {
// CHECK16-NEXT:  entry:
// CHECK16-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK16-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[I4:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[L_CASTED:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 4
// CHECK16-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK16-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK16-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK16-NEXT:    [[TMP1:%.*]] = call i8* @__kmpc_data_sharing_push_stack(i32 4, i16 1)
// CHECK16-NEXT:    [[TMP2:%.*]] = bitcast i8* [[TMP1]] to %struct._globalized_locals_ty*
// CHECK16-NEXT:    [[L1:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY:%.*]], %struct._globalized_locals_ty* [[TMP2]], i32 0, i32 0
// CHECK16-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP3]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK16-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK16-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK16-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK16-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK16-NEXT:    store i32 [[SUB3]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK16-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK16-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK16-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK16-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK16:       omp.precond.then:
// CHECK16-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK16-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK16-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK16-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1:[0-9]+]], i32 [[TMP8]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 128)
// CHECK16-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK16-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
// CHECK16-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK16:       cond.true:
// CHECK16-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK16-NEXT:    br label [[COND_END:%.*]]
// CHECK16:       cond.false:
// CHECK16-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    br label [[COND_END]]
// CHECK16:       cond.end:
// CHECK16-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP11]], [[COND_TRUE]] ], [ [[TMP12]], [[COND_FALSE]] ]
// CHECK16-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    store i32 [[TMP13]], i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK16:       omp.inner.for.cond:
// CHECK16-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK16-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP15]], 1
// CHECK16-NEXT:    [[CMP6:%.*]] = icmp slt i32 [[TMP14]], [[ADD]]
// CHECK16-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK16:       omp.inner.for.body:
// CHECK16-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP18:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP18]], i32* [[N_CASTED]], align 4
// CHECK16-NEXT:    [[TMP19:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK16-NEXT:    [[TMP20:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP20]], i32* [[L_CASTED]], align 4
// CHECK16-NEXT:    [[TMP21:%.*]] = load i32, i32* [[L_CASTED]], align 4
// CHECK16-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK16-NEXT:    [[TMP23:%.*]] = inttoptr i32 [[TMP16]] to i8*
// CHECK16-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 4
// CHECK16-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK16-NEXT:    [[TMP25:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK16-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 4
// CHECK16-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK16-NEXT:    [[TMP27:%.*]] = inttoptr i32 [[TMP19]] to i8*
// CHECK16-NEXT:    store i8* [[TMP27]], i8** [[TMP26]], align 4
// CHECK16-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK16-NEXT:    [[TMP29:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK16-NEXT:    store i8* [[TMP29]], i8** [[TMP28]], align 4
// CHECK16-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 4
// CHECK16-NEXT:    [[TMP31:%.*]] = inttoptr i32 [[TMP21]] to i8*
// CHECK16-NEXT:    store i8* [[TMP31]], i8** [[TMP30]], align 4
// CHECK16-NEXT:    [[TMP32:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    [[TMP33:%.*]] = load i32, i32* [[TMP32]], align 4
// CHECK16-NEXT:    [[TMP34:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK16-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP33]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i32]*, i32)* @__omp_outlined__1 to i8*), i8* null, i8** [[TMP34]], i32 5)
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK16:       omp.inner.for.inc:
// CHECK16-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP35]], [[TMP36]]
// CHECK16-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP37]], [[TMP38]]
// CHECK16-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP39]], [[TMP40]]
// CHECK16-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK16-NEXT:    [[CMP10:%.*]] = icmp sgt i32 [[TMP41]], [[TMP42]]
// CHECK16-NEXT:    br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]]
// CHECK16:       cond.true11:
// CHECK16-NEXT:    [[TMP43:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK16-NEXT:    br label [[COND_END13:%.*]]
// CHECK16:       cond.false12:
// CHECK16-NEXT:    [[TMP44:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    br label [[COND_END13]]
// CHECK16:       cond.end13:
// CHECK16-NEXT:    [[COND14:%.*]] = phi i32 [ [[TMP43]], [[COND_TRUE11]] ], [ [[TMP44]], [[COND_FALSE12]] ]
// CHECK16-NEXT:    store i32 [[COND14]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP45:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    store i32 [[TMP45]], i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK16:       omp.inner.for.end:
// CHECK16-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK16:       omp.loop.exit:
// CHECK16-NEXT:    [[TMP46:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    [[TMP47:%.*]] = load i32, i32* [[TMP46]], align 4
// CHECK16-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP47]])
// CHECK16-NEXT:    [[TMP48:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK16-NEXT:    [[TMP49:%.*]] = icmp ne i32 [[TMP48]], 0
// CHECK16-NEXT:    br i1 [[TMP49]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK16:       .omp.lastprivate.then:
// CHECK16-NEXT:    [[TMP50:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP50]], i32* [[L_ADDR]], align 4
// CHECK16-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK16:       .omp.lastprivate.done:
// CHECK16-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK16:       omp.precond.end:
// CHECK16-NEXT:    call void @__kmpc_data_sharing_pop_stack(i8* [[TMP1]])
// CHECK16-NEXT:    ret void
// CHECK16-LABEL: define {{[^@]+}}@__omp_outlined__1
// CHECK16-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0]] {
// CHECK16-NEXT:  entry:
// CHECK16-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK16-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK16-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK16-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK16-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK16-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK16-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK16-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK16-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK16-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK16-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK16-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK16-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK16-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK16-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK16-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK16:       omp.precond.then:
// CHECK16-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK16-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK16-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK16-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK16-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK16-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK16-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK16-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK16-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 32)
// CHECK16-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK16:       omp.dispatch.cond:
// CHECK16-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK16-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK16-NEXT:    [[CMP4:%.*]] = icmp ugt i32 [[TMP9]], [[TMP10]]
// CHECK16-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK16:       cond.true:
// CHECK16-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK16-NEXT:    br label [[COND_END:%.*]]
// CHECK16:       cond.false:
// CHECK16-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK16-NEXT:    br label [[COND_END]]
// CHECK16:       cond.end:
// CHECK16-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP11]], [[COND_TRUE]] ], [ [[TMP12]], [[COND_FALSE]] ]
// CHECK16-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
// CHECK16-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK16-NEXT:    store i32 [[TMP13]], i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK16-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK16-NEXT:    br i1 [[CMP5]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK16:       omp.dispatch.body:
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK16:       omp.inner.for.cond:
// CHECK16-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK16-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP16]], [[TMP17]]
// CHECK16-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK16:       omp.inner.for.body:
// CHECK16-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP18]], 1
// CHECK16-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK16-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK16-NEXT:    [[TMP19:%.*]] = load i32, i32* [[I3]], align 4
// CHECK16-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i32 0, i32 [[TMP19]]
// CHECK16-NEXT:    store i32 1, i32* [[ARRAYIDX]], align 4
// CHECK16-NEXT:    [[TMP20:%.*]] = load i32, i32* [[I3]], align 4
// CHECK16-NEXT:    store i32 [[TMP20]], i32* [[L_ADDR]], align 4
// CHECK16-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK16:       omp.body.continue:
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK16:       omp.inner.for.inc:
// CHECK16-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP21]], 1
// CHECK16-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK16:       omp.inner.for.end:
// CHECK16-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK16:       omp.dispatch.inc:
// CHECK16-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK16-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK16-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_LB]], align 4
// CHECK16-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK16-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK16-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_UB]], align 4
// CHECK16-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK16:       omp.dispatch.end:
// CHECK16-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK16-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP27]])
// CHECK16-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK16-NEXT:    [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
// CHECK16-NEXT:    br i1 [[TMP29]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK16:       .omp.lastprivate.then:
// CHECK16-NEXT:    [[TMP30:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP30]], i32* [[L_ADDR]], align 4
// CHECK16-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK16:       .omp.lastprivate.done:
// CHECK16-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK16:       omp.precond.end:
// CHECK16-NEXT:    ret void
// CHECK16-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l49
// CHECK16-SAME: (i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK16-NEXT:  entry:
// CHECK16-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK16-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK16-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK16-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK16-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK16-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK16:       .execute:
// CHECK16-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK16-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK16-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK16-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK16-NEXT:    call void @__omp_outlined__2(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [1000 x i16]* [[TMP0]]) #[[ATTR3]]
// CHECK16-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK16:       .omp.deinit:
// CHECK16-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK16-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK16:       .exit:
// CHECK16-NEXT:    ret void
// CHECK16-LABEL: define {{[^@]+}}@__omp_outlined__2
// CHECK16-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK16-NEXT:  entry:
// CHECK16-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK16-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK16-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK16-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK16-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK16-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK16-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK16-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK16-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK16-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK16-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK16-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK16-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK16-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK16-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK16:       omp.precond.then:
// CHECK16-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK16-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK16-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK16-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK16-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK16-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK16-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK16-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK16:       cond.true:
// CHECK16-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK16-NEXT:    br label [[COND_END:%.*]]
// CHECK16:       cond.false:
// CHECK16-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    br label [[COND_END]]
// CHECK16:       cond.end:
// CHECK16-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK16-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK16:       omp.inner.for.cond:
// CHECK16-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK16-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK16-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK16-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK16:       omp.inner.for.body:
// CHECK16-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP16:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP16]], i32* [[N_CASTED]], align 4
// CHECK16-NEXT:    [[TMP17:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK16-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK16-NEXT:    [[TMP19:%.*]] = inttoptr i32 [[TMP14]] to i8*
// CHECK16-NEXT:    store i8* [[TMP19]], i8** [[TMP18]], align 4
// CHECK16-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK16-NEXT:    [[TMP21:%.*]] = inttoptr i32 [[TMP15]] to i8*
// CHECK16-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 4
// CHECK16-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK16-NEXT:    [[TMP23:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK16-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 4
// CHECK16-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK16-NEXT:    [[TMP25:%.*]] = bitcast [1000 x i16]* [[TMP0]] to i8*
// CHECK16-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 4
// CHECK16-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK16-NEXT:    [[TMP28:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK16-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP27]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i16]*)* @__omp_outlined__3 to i8*), i8* null, i8** [[TMP28]], i32 4)
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK16:       omp.inner.for.inc:
// CHECK16-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP29]], [[TMP30]]
// CHECK16-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP31]], [[TMP32]]
// CHECK16-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP33]], [[TMP34]]
// CHECK16-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK16-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP35]], [[TMP36]]
// CHECK16-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
// CHECK16:       cond.true10:
// CHECK16-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK16-NEXT:    br label [[COND_END12:%.*]]
// CHECK16:       cond.false11:
// CHECK16-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    br label [[COND_END12]]
// CHECK16:       cond.end12:
// CHECK16-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP37]], [[COND_TRUE10]] ], [ [[TMP38]], [[COND_FALSE11]] ]
// CHECK16-NEXT:    store i32 [[COND13]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    store i32 [[TMP39]], i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK16:       omp.inner.for.end:
// CHECK16-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK16:       omp.loop.exit:
// CHECK16-NEXT:    [[TMP40:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    [[TMP41:%.*]] = load i32, i32* [[TMP40]], align 4
// CHECK16-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP41]])
// CHECK16-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK16:       omp.precond.end:
// CHECK16-NEXT:    ret void
// CHECK16-LABEL: define {{[^@]+}}@__omp_outlined__3
// CHECK16-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK16-NEXT:  entry:
// CHECK16-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK16-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK16-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK16-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK16-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK16-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK16-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK16-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK16-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK16-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK16-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK16-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK16-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK16-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK16-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK16-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK16:       omp.precond.then:
// CHECK16-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK16-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK16-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK16-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK16-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK16-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK16-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK16-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK16-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK16-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK16-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK16:       omp.inner.for.cond:
// CHECK16-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK16-NEXT:    [[CMP4:%.*]] = icmp ule i32 [[TMP10]], [[TMP11]]
// CHECK16-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK16:       omp.inner.for.body:
// CHECK16-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK16-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK16-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK16-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I3]], align 4
// CHECK16-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i16], [1000 x i16]* [[TMP0]], i32 0, i32 [[TMP13]]
// CHECK16-NEXT:    [[TMP14:%.*]] = load i16, i16* [[ARRAYIDX]], align 2
// CHECK16-NEXT:    [[CONV:%.*]] = sext i16 [[TMP14]] to i32
// CHECK16-NEXT:    [[ADD5:%.*]] = add nsw i32 [[CONV]], 1
// CHECK16-NEXT:    [[CONV6:%.*]] = trunc i32 [[ADD5]] to i16
// CHECK16-NEXT:    store i16 [[CONV6]], i16* [[ARRAYIDX]], align 2
// CHECK16-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK16:       omp.body.continue:
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK16:       omp.inner.for.inc:
// CHECK16-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP15]], [[TMP16]]
// CHECK16-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK16:       omp.inner.for.end:
// CHECK16-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK16:       omp.loop.exit:
// CHECK16-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
// CHECK16-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP18]])
// CHECK16-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK16:       omp.precond.end:
// CHECK16-NEXT:    ret void
// CHECK16-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l54
// CHECK16-SAME: ([10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK16-NEXT:  entry:
// CHECK16-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK16-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK16-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK16-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK16-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK16-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK16-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK16:       .execute:
// CHECK16-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK16-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK16-NEXT:    call void @__omp_outlined__4(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x i32]* [[TMP0]]) #[[ATTR3]]
// CHECK16-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK16:       .omp.deinit:
// CHECK16-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK16-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK16:       .exit:
// CHECK16-NEXT:    ret void
// CHECK16-LABEL: define {{[^@]+}}@__omp_outlined__4
// CHECK16-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK16-NEXT:  entry:
// CHECK16-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK16-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x i8*], align 4
// CHECK16-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK16-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK16-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK16-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    store i32 9, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK16-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK16-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK16-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK16-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9
// CHECK16-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK16:       cond.true:
// CHECK16-NEXT:    br label [[COND_END:%.*]]
// CHECK16:       cond.false:
// CHECK16-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    br label [[COND_END]]
// CHECK16:       cond.end:
// CHECK16-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK16-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK16:       omp.inner.for.cond:
// CHECK16-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[CMP1:%.*]] = icmp slt i32 [[TMP6]], 10
// CHECK16-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK16:       omp.inner.for.body:
// CHECK16-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK16-NEXT:    [[TMP10:%.*]] = inttoptr i32 [[TMP7]] to i8*
// CHECK16-NEXT:    store i8* [[TMP10]], i8** [[TMP9]], align 4
// CHECK16-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK16-NEXT:    [[TMP12:%.*]] = inttoptr i32 [[TMP8]] to i8*
// CHECK16-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 4
// CHECK16-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK16-NEXT:    [[TMP14:%.*]] = bitcast [10 x i32]* [[TMP0]] to i8*
// CHECK16-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 4
// CHECK16-NEXT:    [[TMP15:%.*]] = bitcast [3 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK16-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, [10 x i32]*)* @__omp_outlined__5 to i8*), i8* null, i8** [[TMP15]], i32 3)
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK16:       omp.inner.for.inc:
// CHECK16-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP16]], [[TMP17]]
// CHECK16-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP18]], [[TMP19]]
// CHECK16-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK16-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP22]], 9
// CHECK16-NEXT:    br i1 [[CMP4]], label [[COND_TRUE5:%.*]], label [[COND_FALSE6:%.*]]
// CHECK16:       cond.true5:
// CHECK16-NEXT:    br label [[COND_END7:%.*]]
// CHECK16:       cond.false6:
// CHECK16-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    br label [[COND_END7]]
// CHECK16:       cond.end7:
// CHECK16-NEXT:    [[COND8:%.*]] = phi i32 [ 9, [[COND_TRUE5]] ], [ [[TMP23]], [[COND_FALSE6]] ]
// CHECK16-NEXT:    store i32 [[COND8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    store i32 [[TMP24]], i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK16:       omp.inner.for.end:
// CHECK16-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK16:       omp.loop.exit:
// CHECK16-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]])
// CHECK16-NEXT:    ret void
// CHECK16-LABEL: define {{[^@]+}}@__omp_outlined__5
// CHECK16-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK16-NEXT:  entry:
// CHECK16-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK16-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK16-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK16-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK16-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK16-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK16-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK16-NEXT:    store i32 9, i32* [[DOTOMP_UB]], align 4
// CHECK16-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK16-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP1]], i32* [[DOTOMP_LB]], align 4
// CHECK16-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_UB]], align 4
// CHECK16-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK16-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK16-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK16-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK16-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK16:       omp.inner.for.cond:
// CHECK16-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK16-NEXT:    [[CMP:%.*]] = icmp ule i32 [[TMP6]], [[TMP7]]
// CHECK16-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK16:       omp.inner.for.body:
// CHECK16-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP8]], 1
// CHECK16-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK16-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK16-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4
// CHECK16-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[TMP0]], i32 0, i32 [[TMP9]]
// CHECK16-NEXT:    [[TMP10:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK16-NEXT:    [[ADD1:%.*]] = add nsw i32 [[TMP10]], 1
// CHECK16-NEXT:    store i32 [[ADD1]], i32* [[ARRAYIDX]], align 4
// CHECK16-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK16:       omp.body.continue:
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK16:       omp.inner.for.inc:
// CHECK16-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
// CHECK16-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK16:       omp.inner.for.end:
// CHECK16-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK16:       omp.loop.exit:
// CHECK16-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK16-NEXT:    ret void
// CHECK16-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l59
// CHECK16-SAME: ([10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK16-NEXT:  entry:
// CHECK16-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK16-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[F_CASTED:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK16-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK16-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK16-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK16-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK16-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK16:       .execute:
// CHECK16-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK16-NEXT:    [[TMP2:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP2]], i32* [[F_CASTED]], align 4
// CHECK16-NEXT:    [[TMP3:%.*]] = load i32, i32* [[F_CASTED]], align 4
// CHECK16-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK16-NEXT:    call void @__omp_outlined__6(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x [10 x i32]]* [[TMP0]], i32 [[TMP3]]) #[[ATTR3]]
// CHECK16-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK16:       .omp.deinit:
// CHECK16-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK16-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK16:       .exit:
// CHECK16-NEXT:    ret void
// CHECK16-LABEL: define {{[^@]+}}@__omp_outlined__6
// CHECK16-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK16-NEXT:  entry:
// CHECK16-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK16-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[F_CASTED:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK16-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK16-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK16-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK16-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    store i32 99, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK16-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK16-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK16-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK16-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 99
// CHECK16-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK16:       cond.true:
// CHECK16-NEXT:    br label [[COND_END:%.*]]
// CHECK16:       cond.false:
// CHECK16-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    br label [[COND_END]]
// CHECK16:       cond.end:
// CHECK16-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK16-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK16:       omp.inner.for.cond:
// CHECK16-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[TMP6]], 100
// CHECK16-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK16:       omp.inner.for.body:
// CHECK16-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP9:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP9]], i32* [[F_CASTED]], align 4
// CHECK16-NEXT:    [[TMP10:%.*]] = load i32, i32* [[F_CASTED]], align 4
// CHECK16-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK16-NEXT:    [[TMP12:%.*]] = inttoptr i32 [[TMP7]] to i8*
// CHECK16-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 4
// CHECK16-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK16-NEXT:    [[TMP14:%.*]] = inttoptr i32 [[TMP8]] to i8*
// CHECK16-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 4
// CHECK16-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK16-NEXT:    [[TMP16:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK16-NEXT:    store i8* [[TMP16]], i8** [[TMP15]], align 4
// CHECK16-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK16-NEXT:    [[TMP18:%.*]] = inttoptr i32 [[TMP10]] to i8*
// CHECK16-NEXT:    store i8* [[TMP18]], i8** [[TMP17]], align 4
// CHECK16-NEXT:    [[TMP19:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK16-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, [10 x [10 x i32]]*, i32)* @__omp_outlined__7 to i8*), i8* null, i8** [[TMP19]], i32 4)
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK16:       omp.inner.for.inc:
// CHECK16-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK16-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK16-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK16-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP26]], 99
// CHECK16-NEXT:    br i1 [[CMP5]], label [[COND_TRUE6:%.*]], label [[COND_FALSE7:%.*]]
// CHECK16:       cond.true6:
// CHECK16-NEXT:    br label [[COND_END8:%.*]]
// CHECK16:       cond.false7:
// CHECK16-NEXT:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    br label [[COND_END8]]
// CHECK16:       cond.end8:
// CHECK16-NEXT:    [[COND9:%.*]] = phi i32 [ 99, [[COND_TRUE6]] ], [ [[TMP27]], [[COND_FALSE7]] ]
// CHECK16-NEXT:    store i32 [[COND9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    store i32 [[TMP28]], i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK16:       omp.inner.for.end:
// CHECK16-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK16:       omp.loop.exit:
// CHECK16-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]])
// CHECK16-NEXT:    ret void
// CHECK16-LABEL: define {{[^@]+}}@__omp_outlined__7
// CHECK16-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK16-NEXT:  entry:
// CHECK16-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK16-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK16-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK16-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK16-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK16-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK16-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK16-NEXT:    store i32 99, i32* [[DOTOMP_UB]], align 4
// CHECK16-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK16-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP1]], i32* [[DOTOMP_LB]], align 4
// CHECK16-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_UB]], align 4
// CHECK16-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK16-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK16-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK16-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK16-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK16:       omp.inner.for.cond:
// CHECK16-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK16-NEXT:    [[CMP:%.*]] = icmp ule i32 [[TMP6]], [[TMP7]]
// CHECK16-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK16:       omp.inner.for.body:
// CHECK16-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP8]], 10
// CHECK16-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK16-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK16-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK16-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[DIV2:%.*]] = sdiv i32 [[TMP10]], 10
// CHECK16-NEXT:    [[MUL3:%.*]] = mul nsw i32 [[DIV2]], 10
// CHECK16-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP9]], [[MUL3]]
// CHECK16-NEXT:    [[MUL4:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK16-NEXT:    [[ADD5:%.*]] = add nsw i32 0, [[MUL4]]
// CHECK16-NEXT:    store i32 [[ADD5]], i32* [[J]], align 4
// CHECK16-NEXT:    store i32 10, i32* [[K]], align 4
// CHECK16-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
// CHECK16-NEXT:    [[TMP12:%.*]] = load i32, i32* [[J]], align 4
// CHECK16-NEXT:    [[TMP13:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK16-NEXT:    [[MUL6:%.*]] = mul nsw i32 [[TMP12]], [[TMP13]]
// CHECK16-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP11]], [[MUL6]]
// CHECK16-NEXT:    [[TMP14:%.*]] = load i32, i32* [[K]], align 4
// CHECK16-NEXT:    [[ADD8:%.*]] = add nsw i32 [[ADD7]], [[TMP14]]
// CHECK16-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
// CHECK16-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i32 0, i32 [[TMP15]]
// CHECK16-NEXT:    [[TMP16:%.*]] = load i32, i32* [[J]], align 4
// CHECK16-NEXT:    [[ARRAYIDX9:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i32 0, i32 [[TMP16]]
// CHECK16-NEXT:    store i32 [[ADD8]], i32* [[ARRAYIDX9]], align 4
// CHECK16-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK16:       omp.body.continue:
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK16:       omp.inner.for.inc:
// CHECK16-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK16-NEXT:    store i32 [[ADD10]], i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK16:       omp.inner.for.end:
// CHECK16-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK16:       omp.loop.exit:
// CHECK16-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK16-NEXT:    ret void
// CHECK16-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l67
// CHECK16-SAME: (i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK16-NEXT:  entry:
// CHECK16-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK16-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK16-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK16-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK16-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK16-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK16:       .execute:
// CHECK16-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK16-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK16-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK16-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK16-NEXT:    call void @__omp_outlined__8(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [10 x [10 x i32]]* [[TMP0]]) #[[ATTR3]]
// CHECK16-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK16:       .omp.deinit:
// CHECK16-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK16-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK16:       .exit:
// CHECK16-NEXT:    ret void
// CHECK16-LABEL: define {{[^@]+}}@__omp_outlined__8
// CHECK16-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK16-NEXT:  entry:
// CHECK16-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK16-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK16-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK16-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i64, align 8
// CHECK16-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i64, align 8
// CHECK16-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK16-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[I9:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[J10:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK16-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK16-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK16-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK16-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK16-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK16-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK16-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK16-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK16-NEXT:    [[CONV:%.*]] = sext i32 [[DIV]] to i64
// CHECK16-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK16-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK16-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK16-NEXT:    [[CONV6:%.*]] = sext i32 [[DIV5]] to i64
// CHECK16-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV]], [[CONV6]]
// CHECK16-NEXT:    [[SUB7:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK16-NEXT:    store i64 [[SUB7]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK16-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK16-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK16-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK16-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK16-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK16:       land.lhs.true:
// CHECK16-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK16-NEXT:    [[CMP8:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK16-NEXT:    br i1 [[CMP8]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK16:       omp.precond.then:
// CHECK16-NEXT:    store i64 0, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK16-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK16-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK16-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK16-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK16-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK16-NEXT:    [[CONV11:%.*]] = zext i32 [[NVPTX_NUM_THREADS]] to i64
// CHECK16-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    [[TMP9:%.*]] = load i32, i32* [[TMP8]], align 4
// CHECK16-NEXT:    call void @__kmpc_for_static_init_8(%struct.ident_t* @[[GLOB1]], i32 [[TMP9]], i32 91, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_COMB_LB]], i64* [[DOTOMP_COMB_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 [[CONV11]])
// CHECK16-NEXT:    [[TMP10:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK16-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK16-NEXT:    [[CMP12:%.*]] = icmp sgt i64 [[TMP10]], [[TMP11]]
// CHECK16-NEXT:    br i1 [[CMP12]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK16:       cond.true:
// CHECK16-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK16-NEXT:    br label [[COND_END:%.*]]
// CHECK16:       cond.false:
// CHECK16-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK16-NEXT:    br label [[COND_END]]
// CHECK16:       cond.end:
// CHECK16-NEXT:    [[COND:%.*]] = phi i64 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
// CHECK16-NEXT:    store i64 [[COND]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK16-NEXT:    [[TMP14:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK16-NEXT:    store i64 [[TMP14]], i64* [[DOTOMP_IV]], align 8
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK16:       omp.inner.for.cond:
// CHECK16-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK16-NEXT:    [[TMP16:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK16-NEXT:    [[ADD:%.*]] = add nsw i64 [[TMP16]], 1
// CHECK16-NEXT:    [[CMP13:%.*]] = icmp slt i64 [[TMP15]], [[ADD]]
// CHECK16-NEXT:    br i1 [[CMP13]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK16:       omp.inner.for.body:
// CHECK16-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK16-NEXT:    [[TMP18:%.*]] = trunc i64 [[TMP17]] to i32
// CHECK16-NEXT:    [[TMP19:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK16-NEXT:    [[TMP20:%.*]] = trunc i64 [[TMP19]] to i32
// CHECK16-NEXT:    [[TMP21:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP21]], i32* [[N_CASTED]], align 4
// CHECK16-NEXT:    [[TMP22:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK16-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK16-NEXT:    [[TMP24:%.*]] = inttoptr i32 [[TMP18]] to i8*
// CHECK16-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 4
// CHECK16-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK16-NEXT:    [[TMP26:%.*]] = inttoptr i32 [[TMP20]] to i8*
// CHECK16-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 4
// CHECK16-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK16-NEXT:    [[TMP28:%.*]] = inttoptr i32 [[TMP22]] to i8*
// CHECK16-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 4
// CHECK16-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK16-NEXT:    [[TMP30:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK16-NEXT:    store i8* [[TMP30]], i8** [[TMP29]], align 4
// CHECK16-NEXT:    [[TMP31:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    [[TMP32:%.*]] = load i32, i32* [[TMP31]], align 4
// CHECK16-NEXT:    [[TMP33:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK16-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP32]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [10 x [10 x i32]]*)* @__omp_outlined__9 to i8*), i8* null, i8** [[TMP33]], i32 4)
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK16:       omp.inner.for.inc:
// CHECK16-NEXT:    [[TMP34:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK16-NEXT:    [[TMP35:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK16-NEXT:    [[ADD14:%.*]] = add nsw i64 [[TMP34]], [[TMP35]]
// CHECK16-NEXT:    store i64 [[ADD14]], i64* [[DOTOMP_IV]], align 8
// CHECK16-NEXT:    [[TMP36:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK16-NEXT:    [[TMP37:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK16-NEXT:    [[ADD15:%.*]] = add nsw i64 [[TMP36]], [[TMP37]]
// CHECK16-NEXT:    store i64 [[ADD15]], i64* [[DOTOMP_COMB_LB]], align 8
// CHECK16-NEXT:    [[TMP38:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK16-NEXT:    [[TMP39:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK16-NEXT:    [[ADD16:%.*]] = add nsw i64 [[TMP38]], [[TMP39]]
// CHECK16-NEXT:    store i64 [[ADD16]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK16-NEXT:    [[TMP40:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK16-NEXT:    [[TMP41:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK16-NEXT:    [[CMP17:%.*]] = icmp sgt i64 [[TMP40]], [[TMP41]]
// CHECK16-NEXT:    br i1 [[CMP17]], label [[COND_TRUE18:%.*]], label [[COND_FALSE19:%.*]]
// CHECK16:       cond.true18:
// CHECK16-NEXT:    [[TMP42:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK16-NEXT:    br label [[COND_END20:%.*]]
// CHECK16:       cond.false19:
// CHECK16-NEXT:    [[TMP43:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK16-NEXT:    br label [[COND_END20]]
// CHECK16:       cond.end20:
// CHECK16-NEXT:    [[COND21:%.*]] = phi i64 [ [[TMP42]], [[COND_TRUE18]] ], [ [[TMP43]], [[COND_FALSE19]] ]
// CHECK16-NEXT:    store i64 [[COND21]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK16-NEXT:    [[TMP44:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK16-NEXT:    store i64 [[TMP44]], i64* [[DOTOMP_IV]], align 8
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK16:       omp.inner.for.end:
// CHECK16-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK16:       omp.loop.exit:
// CHECK16-NEXT:    [[TMP45:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    [[TMP46:%.*]] = load i32, i32* [[TMP45]], align 4
// CHECK16-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP46]])
// CHECK16-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK16:       omp.precond.end:
// CHECK16-NEXT:    ret void
// CHECK16-LABEL: define {{[^@]+}}@__omp_outlined__9
// CHECK16-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK16-NEXT:  entry:
// CHECK16-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK16-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK16-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK16-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_LB:%.*]] = alloca i64, align 8
// CHECK16-NEXT:    [[DOTOMP_UB:%.*]] = alloca i64, align 8
// CHECK16-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK16-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[I11:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[J12:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK16-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK16-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK16-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK16-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK16-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK16-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK16-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK16-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK16-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK16-NEXT:    [[CONV:%.*]] = sext i32 [[DIV]] to i64
// CHECK16-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK16-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK16-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK16-NEXT:    [[CONV6:%.*]] = sext i32 [[DIV5]] to i64
// CHECK16-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV]], [[CONV6]]
// CHECK16-NEXT:    [[SUB7:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK16-NEXT:    store i64 [[SUB7]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK16-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK16-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK16-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK16-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK16-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK16:       land.lhs.true:
// CHECK16-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK16-NEXT:    [[CMP8:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK16-NEXT:    br i1 [[CMP8]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK16:       omp.precond.then:
// CHECK16-NEXT:    store i64 0, i64* [[DOTOMP_LB]], align 8
// CHECK16-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK16-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_UB]], align 8
// CHECK16-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK16-NEXT:    [[CONV9:%.*]] = zext i32 [[TMP8]] to i64
// CHECK16-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK16-NEXT:    [[CONV10:%.*]] = zext i32 [[TMP9]] to i64
// CHECK16-NEXT:    store i64 [[CONV9]], i64* [[DOTOMP_LB]], align 8
// CHECK16-NEXT:    store i64 [[CONV10]], i64* [[DOTOMP_UB]], align 8
// CHECK16-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK16-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK16-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK16-NEXT:    call void @__kmpc_for_static_init_8(%struct.ident_t* @[[GLOB2]], i32 [[TMP11]], i32 33, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_LB]], i64* [[DOTOMP_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 1)
// CHECK16-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTOMP_LB]], align 8
// CHECK16-NEXT:    store i64 [[TMP12]], i64* [[DOTOMP_IV]], align 8
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK16:       omp.inner.for.cond:
// CHECK16-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK16-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK16-NEXT:    [[CONV13:%.*]] = zext i32 [[TMP14]] to i64
// CHECK16-NEXT:    [[CMP14:%.*]] = icmp sle i64 [[TMP13]], [[CONV13]]
// CHECK16-NEXT:    br i1 [[CMP14]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK16:       omp.inner.for.body:
// CHECK16-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK16-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK16-NEXT:    [[SUB15:%.*]] = sub nsw i32 [[TMP16]], 0
// CHECK16-NEXT:    [[DIV16:%.*]] = sdiv i32 [[SUB15]], 1
// CHECK16-NEXT:    [[MUL17:%.*]] = mul nsw i32 1, [[DIV16]]
// CHECK16-NEXT:    [[CONV18:%.*]] = sext i32 [[MUL17]] to i64
// CHECK16-NEXT:    [[DIV19:%.*]] = sdiv i64 [[TMP15]], [[CONV18]]
// CHECK16-NEXT:    [[MUL20:%.*]] = mul nsw i64 [[DIV19]], 1
// CHECK16-NEXT:    [[ADD:%.*]] = add nsw i64 0, [[MUL20]]
// CHECK16-NEXT:    [[CONV21:%.*]] = trunc i64 [[ADD]] to i32
// CHECK16-NEXT:    store i32 [[CONV21]], i32* [[I11]], align 4
// CHECK16-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK16-NEXT:    [[TMP18:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK16-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK16-NEXT:    [[SUB22:%.*]] = sub nsw i32 [[TMP19]], 0
// CHECK16-NEXT:    [[DIV23:%.*]] = sdiv i32 [[SUB22]], 1
// CHECK16-NEXT:    [[MUL24:%.*]] = mul nsw i32 1, [[DIV23]]
// CHECK16-NEXT:    [[CONV25:%.*]] = sext i32 [[MUL24]] to i64
// CHECK16-NEXT:    [[DIV26:%.*]] = sdiv i64 [[TMP18]], [[CONV25]]
// CHECK16-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK16-NEXT:    [[SUB27:%.*]] = sub nsw i32 [[TMP20]], 0
// CHECK16-NEXT:    [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK16-NEXT:    [[MUL29:%.*]] = mul nsw i32 1, [[DIV28]]
// CHECK16-NEXT:    [[CONV30:%.*]] = sext i32 [[MUL29]] to i64
// CHECK16-NEXT:    [[MUL31:%.*]] = mul nsw i64 [[DIV26]], [[CONV30]]
// CHECK16-NEXT:    [[SUB32:%.*]] = sub nsw i64 [[TMP17]], [[MUL31]]
// CHECK16-NEXT:    [[MUL33:%.*]] = mul nsw i64 [[SUB32]], 1
// CHECK16-NEXT:    [[ADD34:%.*]] = add nsw i64 0, [[MUL33]]
// CHECK16-NEXT:    [[CONV35:%.*]] = trunc i64 [[ADD34]] to i32
// CHECK16-NEXT:    store i32 [[CONV35]], i32* [[J12]], align 4
// CHECK16-NEXT:    [[TMP21:%.*]] = load i32, i32* [[I11]], align 4
// CHECK16-NEXT:    [[TMP22:%.*]] = load i32, i32* [[J12]], align 4
// CHECK16-NEXT:    [[ADD36:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
// CHECK16-NEXT:    [[TMP23:%.*]] = load i32, i32* [[I11]], align 4
// CHECK16-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i32 0, i32 [[TMP23]]
// CHECK16-NEXT:    [[TMP24:%.*]] = load i32, i32* [[J12]], align 4
// CHECK16-NEXT:    [[ARRAYIDX37:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i32 0, i32 [[TMP24]]
// CHECK16-NEXT:    store i32 [[ADD36]], i32* [[ARRAYIDX37]], align 4
// CHECK16-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK16:       omp.body.continue:
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK16:       omp.inner.for.inc:
// CHECK16-NEXT:    [[TMP25:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK16-NEXT:    [[TMP26:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK16-NEXT:    [[ADD38:%.*]] = add nsw i64 [[TMP25]], [[TMP26]]
// CHECK16-NEXT:    store i64 [[ADD38]], i64* [[DOTOMP_IV]], align 8
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK16:       omp.inner.for.end:
// CHECK16-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK16:       omp.loop.exit:
// CHECK16-NEXT:    [[TMP27:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    [[TMP28:%.*]] = load i32, i32* [[TMP27]], align 4
// CHECK16-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP28]])
// CHECK16-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK16:       omp.precond.end:
// CHECK16-NEXT:    ret void
// CHECK16-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l74
// CHECK16-SAME: (i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK16-NEXT:  entry:
// CHECK16-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK16-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK16-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK16-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK16-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK16-NEXT:    call void @__kmpc_spmd_kernel_init(i32 [[NVPTX_NUM_THREADS]], i16 0)
// CHECK16-NEXT:    br label [[DOTEXECUTE:%.*]]
// CHECK16:       .execute:
// CHECK16-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK16-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP2]], i32* [[N_CASTED]], align 4
// CHECK16-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK16-NEXT:    [[TMP4:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK16-NEXT:    call void @__omp_outlined__10(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]], [1000 x i32]* [[TMP0]], i32* [[TMP4]]) #[[ATTR3]]
// CHECK16-NEXT:    br label [[DOTOMP_DEINIT:%.*]]
// CHECK16:       .omp.deinit:
// CHECK16-NEXT:    call void @__kmpc_spmd_kernel_deinit_v2(i16 0)
// CHECK16-NEXT:    br label [[DOTEXIT:%.*]]
// CHECK16:       .exit:
// CHECK16-NEXT:    ret void
// CHECK16-LABEL: define {{[^@]+}}@__omp_outlined__10
// CHECK16-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK16-NEXT:  entry:
// CHECK16-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK16-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 4
// CHECK16-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK16-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK16-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK16-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK16-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK16-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK16-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK16-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK16-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK16-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK16-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK16-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK16-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK16-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK16:       omp.precond.then:
// CHECK16-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK16-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK16-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK16-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK16-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK16-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK16-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK16-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK16:       cond.true:
// CHECK16-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK16-NEXT:    br label [[COND_END:%.*]]
// CHECK16:       cond.false:
// CHECK16-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    br label [[COND_END]]
// CHECK16:       cond.end:
// CHECK16-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK16-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK16:       omp.inner.for.cond:
// CHECK16-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK16-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK16-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK16-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK16:       omp.inner.for.body:
// CHECK16-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP16:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP16]], i32* [[N_CASTED]], align 4
// CHECK16-NEXT:    [[TMP17:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK16-NEXT:    [[TMP18:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK16-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK16-NEXT:    [[TMP20:%.*]] = inttoptr i32 [[TMP14]] to i8*
// CHECK16-NEXT:    store i8* [[TMP20]], i8** [[TMP19]], align 4
// CHECK16-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK16-NEXT:    [[TMP22:%.*]] = inttoptr i32 [[TMP15]] to i8*
// CHECK16-NEXT:    store i8* [[TMP22]], i8** [[TMP21]], align 4
// CHECK16-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK16-NEXT:    [[TMP24:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK16-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 4
// CHECK16-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK16-NEXT:    [[TMP26:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK16-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 4
// CHECK16-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 4
// CHECK16-NEXT:    [[TMP28:%.*]] = bitcast i32* [[TMP18]] to i8*
// CHECK16-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 4
// CHECK16-NEXT:    [[TMP29:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    [[TMP30:%.*]] = load i32, i32* [[TMP29]], align 4
// CHECK16-NEXT:    [[TMP31:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK16-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB3]], i32 [[TMP30]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i32]*, i32*)* @__omp_outlined__11 to i8*), i8* null, i8** [[TMP31]], i32 5)
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK16:       omp.inner.for.inc:
// CHECK16-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP32]], [[TMP33]]
// CHECK16-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP34]], [[TMP35]]
// CHECK16-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP36]], [[TMP37]]
// CHECK16-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK16-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP38]], [[TMP39]]
// CHECK16-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
// CHECK16:       cond.true10:
// CHECK16-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK16-NEXT:    br label [[COND_END12:%.*]]
// CHECK16:       cond.false11:
// CHECK16-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    br label [[COND_END12]]
// CHECK16:       cond.end12:
// CHECK16-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP40]], [[COND_TRUE10]] ], [ [[TMP41]], [[COND_FALSE11]] ]
// CHECK16-NEXT:    store i32 [[COND13]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK16-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK16-NEXT:    store i32 [[TMP42]], i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK16:       omp.inner.for.end:
// CHECK16-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK16:       omp.loop.exit:
// CHECK16-NEXT:    [[TMP43:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    [[TMP44:%.*]] = load i32, i32* [[TMP43]], align 4
// CHECK16-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP44]])
// CHECK16-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK16:       omp.precond.end:
// CHECK16-NEXT:    ret void
// CHECK16-LABEL: define {{[^@]+}}@__omp_outlined__11
// CHECK16-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK16-NEXT:  entry:
// CHECK16-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK16-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK16-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK16-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK16-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK16-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK16-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK16-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK16-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK16-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK16-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK16-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK16-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK16-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK16-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK16-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK16-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK16-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK16-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK16:       omp.precond.then:
// CHECK16-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK16-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK16-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK16-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK16-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK16-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK16-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK16-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK16-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK16-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK16-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK16-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK16:       omp.inner.for.cond:
// CHECK16-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK16-NEXT:    [[CMP4:%.*]] = icmp ule i32 [[TMP10]], [[TMP11]]
// CHECK16-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK16:       omp.inner.for.body:
// CHECK16-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK16-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK16-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK16-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK16-NEXT:    [[TMP14:%.*]] = load i32, i32* [[I3]], align 4
// CHECK16-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[TMP13]], i32 [[TMP14]]
// CHECK16-NEXT:    [[TMP15:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK16-NEXT:    [[TMP16:%.*]] = load i32, i32* [[I3]], align 4
// CHECK16-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i32 0, i32 [[TMP16]]
// CHECK16-NEXT:    store i32 [[TMP15]], i32* [[ARRAYIDX5]], align 4
// CHECK16-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK16:       omp.body.continue:
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK16:       omp.inner.for.inc:
// CHECK16-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK16-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK16-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK16-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK16:       omp.inner.for.end:
// CHECK16-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK16:       omp.loop.exit:
// CHECK16-NEXT:    [[TMP19:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK16-NEXT:    [[TMP20:%.*]] = load i32, i32* [[TMP19]], align 4
// CHECK16-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP20]])
// CHECK16-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK16:       omp.precond.end:
// CHECK16-NEXT:    ret void
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l28
// CHECK1-SAME: (i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i64 [[L:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK1-NEXT:    [[L_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[L_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK1-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK1-NEXT:    store i64 [[L]], i64* [[L_ADDR]], align 8
// CHECK1-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK1-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK1-NEXT:    [[CONV1:%.*]] = bitcast i64* [[L_ADDR]] to i32*
// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i8 2, i1 false, i1 false)
// CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK1:       user_code.entry:
// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB4:[0-9]+]])
// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK1-NEXT:    [[CONV2:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK1-NEXT:    store i32 [[TMP3]], i32* [[CONV2]], align 4
// CHECK1-NEXT:    [[TMP4:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK1-NEXT:    [[CONV3:%.*]] = bitcast i64* [[L_CASTED]] to i32*
// CHECK1-NEXT:    store i32 [[TMP5]], i32* [[CONV3]], align 4
// CHECK1-NEXT:    [[TMP6:%.*]] = load i64, i64* [[L_CASTED]], align 8
// CHECK1-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK1-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK1-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP4]], [1000 x i32]* [[TMP0]], i64 [[TMP6]]) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
// CHECK1-NEXT:    ret void
// CHECK1:       worker.exit:
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@__omp_outlined__
// CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i64 [[L:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK1-NEXT:    [[L_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I5:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[L_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 8
// CHECK1-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK1-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK1-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK1-NEXT:    store i64 [[L]], i64* [[L_ADDR]], align 8
// CHECK1-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK1-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK1-NEXT:    [[CONV1:%.*]] = bitcast i64* [[L_ADDR]] to i32*
// CHECK1-NEXT:    [[L2:%.*]] = call i8* @__kmpc_alloc_shared(i64 4)
// CHECK1-NEXT:    [[L_ON_STACK:%.*]] = bitcast i8* [[L2]] to i32*
// CHECK1-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK1-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK1-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK1-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK1-NEXT:    store i32 [[SUB4]], i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK1-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK1:       omp.precond.then:
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK1-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK1-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK1-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 128)
// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK1-NEXT:    [[CMP6:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK1-NEXT:    br i1 [[CMP6]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK1:       cond.true:
// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK1-NEXT:    br label [[COND_END:%.*]]
// CHECK1:       cond.false:
// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    br label [[COND_END]]
// CHECK1:       cond.end:
// CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK1-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK1:       omp.inner.for.cond:
// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK1-NEXT:    [[CMP7:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK1-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK1:       omp.inner.for.body:
// CHECK1-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    [[TMP15:%.*]] = zext i32 [[TMP14]] to i64
// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP17:%.*]] = zext i32 [[TMP16]] to i64
// CHECK1-NEXT:    [[TMP18:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK1-NEXT:    [[CONV8:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK1-NEXT:    store i32 [[TMP18]], i32* [[CONV8]], align 4
// CHECK1-NEXT:    [[TMP19:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK1-NEXT:    [[TMP20:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK1-NEXT:    [[CONV9:%.*]] = bitcast i64* [[L_CASTED]] to i32*
// CHECK1-NEXT:    store i32 [[TMP20]], i32* [[CONV9]], align 4
// CHECK1-NEXT:    [[TMP21:%.*]] = load i64, i64* [[L_CASTED]], align 8
// CHECK1-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK1-NEXT:    [[TMP23:%.*]] = inttoptr i64 [[TMP15]] to i8*
// CHECK1-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 8
// CHECK1-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK1-NEXT:    [[TMP25:%.*]] = inttoptr i64 [[TMP17]] to i8*
// CHECK1-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 8
// CHECK1-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK1-NEXT:    [[TMP27:%.*]] = inttoptr i64 [[TMP19]] to i8*
// CHECK1-NEXT:    store i8* [[TMP27]], i8** [[TMP26]], align 8
// CHECK1-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK1-NEXT:    [[TMP29:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK1-NEXT:    store i8* [[TMP29]], i8** [[TMP28]], align 8
// CHECK1-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 4
// CHECK1-NEXT:    [[TMP31:%.*]] = inttoptr i64 [[TMP21]] to i8*
// CHECK1-NEXT:    store i8* [[TMP31]], i8** [[TMP30]], align 8
// CHECK1-NEXT:    [[TMP32:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP33:%.*]] = load i32, i32* [[TMP32]], align 4
// CHECK1-NEXT:    [[TMP34:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK1-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB4]], i32 [[TMP33]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, i64, [1000 x i32]*, i64)* @__omp_outlined__1 to i8*), i8* null, i8** [[TMP34]], i64 5)
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK1:       omp.inner.for.inc:
// CHECK1-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP35]], [[TMP36]]
// CHECK1-NEXT:    store i32 [[ADD10]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP37]], [[TMP38]]
// CHECK1-NEXT:    store i32 [[ADD11]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP39]], [[TMP40]]
// CHECK1-NEXT:    store i32 [[ADD12]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK1-NEXT:    [[CMP13:%.*]] = icmp sgt i32 [[TMP41]], [[TMP42]]
// CHECK1-NEXT:    br i1 [[CMP13]], label [[COND_TRUE14:%.*]], label [[COND_FALSE15:%.*]]
// CHECK1:       cond.true14:
// CHECK1-NEXT:    [[TMP43:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK1-NEXT:    br label [[COND_END16:%.*]]
// CHECK1:       cond.false15:
// CHECK1-NEXT:    [[TMP44:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    br label [[COND_END16]]
// CHECK1:       cond.end16:
// CHECK1-NEXT:    [[COND17:%.*]] = phi i32 [ [[TMP43]], [[COND_TRUE14]] ], [ [[TMP44]], [[COND_FALSE15]] ]
// CHECK1-NEXT:    store i32 [[COND17]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP45:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    store i32 [[TMP45]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK1:       omp.inner.for.end:
// CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK1:       omp.loop.exit:
// CHECK1-NEXT:    [[TMP46:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP47:%.*]] = load i32, i32* [[TMP46]], align 4
// CHECK1-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP47]])
// CHECK1-NEXT:    [[TMP48:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK1-NEXT:    [[TMP49:%.*]] = icmp ne i32 [[TMP48]], 0
// CHECK1-NEXT:    br i1 [[TMP49]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK1:       .omp.lastprivate.then:
// CHECK1-NEXT:    [[TMP50:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK1-NEXT:    store i32 [[TMP50]], i32* [[CONV1]], align 8
// CHECK1-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK1:       .omp.lastprivate.done:
// CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK1:       omp.precond.end:
// CHECK1-NEXT:    call void @__kmpc_free_shared(i8* [[L2]], i64 4)
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@__omp_outlined__1
// CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i64 [[L:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK1-NEXT:    [[L_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I6:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK1-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK1-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK1-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK1-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK1-NEXT:    store i64 [[L]], i64* [[L_ADDR]], align 8
// CHECK1-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK1-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK1-NEXT:    [[CONV1:%.*]] = bitcast i64* [[L_ADDR]] to i32*
// CHECK1-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK1-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK1-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK1-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK1-NEXT:    store i32 [[SUB3]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK1-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK1:       omp.precond.then:
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK1-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK1-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK1-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK1-NEXT:    [[CONV4:%.*]] = trunc i64 [[TMP5]] to i32
// CHECK1-NEXT:    [[TMP6:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK1-NEXT:    [[CONV5:%.*]] = trunc i64 [[TMP6]] to i32
// CHECK1-NEXT:    store i32 [[CONV4]], i32* [[DOTOMP_LB]], align 4
// CHECK1-NEXT:    store i32 [[CONV5]], i32* [[DOTOMP_UB]], align 4
// CHECK1-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK1-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK1-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB3:[0-9]+]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 32)
// CHECK1-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK1:       omp.dispatch.cond:
// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK1-NEXT:    [[TMP10:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK1-NEXT:    [[CONV7:%.*]] = trunc i64 [[TMP10]] to i32
// CHECK1-NEXT:    [[CMP8:%.*]] = icmp sgt i32 [[TMP9]], [[CONV7]]
// CHECK1-NEXT:    br i1 [[CMP8]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK1:       cond.true:
// CHECK1-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK1-NEXT:    [[CONV9:%.*]] = trunc i64 [[TMP11]] to i32
// CHECK1-NEXT:    br label [[COND_END:%.*]]
// CHECK1:       cond.false:
// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK1-NEXT:    br label [[COND_END]]
// CHECK1:       cond.end:
// CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ [[CONV9]], [[COND_TRUE]] ], [ [[TMP12]], [[COND_FALSE]] ]
// CHECK1-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
// CHECK1-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK1-NEXT:    store i32 [[TMP13]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK1-NEXT:    [[CMP10:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK1-NEXT:    br i1 [[CMP10]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK1:       omp.dispatch.body:
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK1:       omp.inner.for.cond:
// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK1-NEXT:    [[CMP11:%.*]] = icmp sle i32 [[TMP16]], [[TMP17]]
// CHECK1-NEXT:    br i1 [[CMP11]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK1:       omp.inner.for.body:
// CHECK1-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP18]], 1
// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK1-NEXT:    store i32 [[ADD]], i32* [[I6]], align 4
// CHECK1-NEXT:    [[TMP19:%.*]] = load i32, i32* [[I6]], align 4
// CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP19]] to i64
// CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK1-NEXT:    store i32 1, i32* [[ARRAYIDX]], align 4
// CHECK1-NEXT:    [[TMP20:%.*]] = load i32, i32* [[I6]], align 4
// CHECK1-NEXT:    store i32 [[TMP20]], i32* [[CONV1]], align 8
// CHECK1-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK1:       omp.body.continue:
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK1:       omp.inner.for.inc:
// CHECK1-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP21]], 1
// CHECK1-NEXT:    store i32 [[ADD12]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK1:       omp.inner.for.end:
// CHECK1-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK1:       omp.dispatch.inc:
// CHECK1-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK1-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    [[ADD13:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK1-NEXT:    store i32 [[ADD13]], i32* [[DOTOMP_LB]], align 4
// CHECK1-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK1-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    [[ADD14:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK1-NEXT:    store i32 [[ADD14]], i32* [[DOTOMP_UB]], align 4
// CHECK1-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK1:       omp.dispatch.end:
// CHECK1-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK1-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP27]])
// CHECK1-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK1-NEXT:    [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
// CHECK1-NEXT:    br i1 [[TMP29]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK1:       .omp.lastprivate.then:
// CHECK1-NEXT:    [[TMP30:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK1-NEXT:    store i32 [[TMP30]], i32* [[CONV1]], align 8
// CHECK1-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK1:       .omp.lastprivate.done:
// CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK1:       omp.precond.end:
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34
// CHECK1-SAME: (i64 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR4:[0-9]+]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 8
// CHECK1-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK1-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 8
// CHECK1-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK1-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 8
// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false, i1 false)
// CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK1:       user_code.entry:
// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB4]])
// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK1-NEXT:    [[CONV1:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK1-NEXT:    store i32 [[TMP3]], i32* [[CONV1]], align 4
// CHECK1-NEXT:    [[TMP4:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK1-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK1-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK1-NEXT:    call void @__omp_outlined__2(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP4]], [1000 x i16]* [[TMP0]]) #[[ATTR2]]
// CHECK1-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
// CHECK1-NEXT:    ret void
// CHECK1:       worker.exit:
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@__omp_outlined__2
// CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 8
// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 8
// CHECK1-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK1-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK1-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 8
// CHECK1-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK1-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 8
// CHECK1-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK1-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK1-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK1-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK1-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK1:       omp.precond.then:
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK1-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK1-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK1-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK1-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK1:       cond.true:
// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT:    br label [[COND_END:%.*]]
// CHECK1:       cond.false:
// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    br label [[COND_END]]
// CHECK1:       cond.end:
// CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK1-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK1:       omp.inner.for.cond:
// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK1-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK1-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK1:       omp.inner.for.body:
// CHECK1-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    [[TMP15:%.*]] = zext i32 [[TMP14]] to i64
// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP17:%.*]] = zext i32 [[TMP16]] to i64
// CHECK1-NEXT:    [[TMP18:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK1-NEXT:    [[CONV6:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK1-NEXT:    store i32 [[TMP18]], i32* [[CONV6]], align 4
// CHECK1-NEXT:    [[TMP19:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK1-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK1-NEXT:    [[TMP21:%.*]] = inttoptr i64 [[TMP15]] to i8*
// CHECK1-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 8
// CHECK1-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK1-NEXT:    [[TMP23:%.*]] = inttoptr i64 [[TMP17]] to i8*
// CHECK1-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 8
// CHECK1-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK1-NEXT:    [[TMP25:%.*]] = inttoptr i64 [[TMP19]] to i8*
// CHECK1-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 8
// CHECK1-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK1-NEXT:    [[TMP27:%.*]] = bitcast [1000 x i16]* [[TMP0]] to i8*
// CHECK1-NEXT:    store i8* [[TMP27]], i8** [[TMP26]], align 8
// CHECK1-NEXT:    [[TMP28:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP29:%.*]] = load i32, i32* [[TMP28]], align 4
// CHECK1-NEXT:    [[TMP30:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK1-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB4]], i32 [[TMP29]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, i64, [1000 x i16]*)* @__omp_outlined__3 to i8*), i8* null, i8** [[TMP30]], i64 4)
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK1:       omp.inner.for.inc:
// CHECK1-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP31]], [[TMP32]]
// CHECK1-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP33]], [[TMP34]]
// CHECK1-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP35]], [[TMP36]]
// CHECK1-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT:    [[CMP10:%.*]] = icmp sgt i32 [[TMP37]], [[TMP38]]
// CHECK1-NEXT:    br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]]
// CHECK1:       cond.true11:
// CHECK1-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT:    br label [[COND_END13:%.*]]
// CHECK1:       cond.false12:
// CHECK1-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    br label [[COND_END13]]
// CHECK1:       cond.end13:
// CHECK1-NEXT:    [[COND14:%.*]] = phi i32 [ [[TMP39]], [[COND_TRUE11]] ], [ [[TMP40]], [[COND_FALSE12]] ]
// CHECK1-NEXT:    store i32 [[COND14]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    store i32 [[TMP41]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK1:       omp.inner.for.end:
// CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK1:       omp.loop.exit:
// CHECK1-NEXT:    [[TMP42:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP43:%.*]] = load i32, i32* [[TMP42]], align 4
// CHECK1-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP43]])
// CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK1:       omp.precond.end:
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@__omp_outlined__3
// CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], i64 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 8
// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I5:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK1-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK1-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK1-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK1-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 8
// CHECK1-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK1-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 8
// CHECK1-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK1-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK1-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK1-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK1-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK1:       omp.precond.then:
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK1-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK1-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK1-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP5]] to i32
// CHECK1-NEXT:    [[TMP6:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK1-NEXT:    [[CONV4:%.*]] = trunc i64 [[TMP6]] to i32
// CHECK1-NEXT:    store i32 [[CONV3]], i32* [[DOTOMP_LB]], align 4
// CHECK1-NEXT:    store i32 [[CONV4]], i32* [[DOTOMP_UB]], align 4
// CHECK1-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK1-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK1-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB3]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK1-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK1:       omp.inner.for.cond:
// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[CONV6:%.*]] = sext i32 [[TMP10]] to i64
// CHECK1-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK1-NEXT:    [[CMP7:%.*]] = icmp ule i64 [[CONV6]], [[TMP11]]
// CHECK1-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK1:       omp.inner.for.body:
// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK1-NEXT:    store i32 [[ADD]], i32* [[I5]], align 4
// CHECK1-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I5]], align 4
// CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP13]] to i64
// CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i16], [1000 x i16]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK1-NEXT:    [[TMP14:%.*]] = load i16, i16* [[ARRAYIDX]], align 2
// CHECK1-NEXT:    [[CONV8:%.*]] = sext i16 [[TMP14]] to i32
// CHECK1-NEXT:    [[ADD9:%.*]] = add nsw i32 [[CONV8]], 1
// CHECK1-NEXT:    [[CONV10:%.*]] = trunc i32 [[ADD9]] to i16
// CHECK1-NEXT:    store i16 [[CONV10]], i16* [[ARRAYIDX]], align 2
// CHECK1-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK1:       omp.body.continue:
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK1:       omp.inner.for.inc:
// CHECK1-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP15]], [[TMP16]]
// CHECK1-NEXT:    store i32 [[ADD11]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK1:       omp.inner.for.end:
// CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK1:       omp.loop.exit:
// CHECK1-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
// CHECK1-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP18]])
// CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK1:       omp.precond.end:
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l39
// CHECK1-SAME: ([10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 8
// CHECK1-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 8
// CHECK1-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 8
// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false, i1 false)
// CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK1:       user_code.entry:
// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB4]])
// CHECK1-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK1-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK1-NEXT:    call void @__omp_outlined__4(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x i32]* [[TMP0]]) #[[ATTR2]]
// CHECK1-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
// CHECK1-NEXT:    ret void
// CHECK1:       worker.exit:
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@__omp_outlined__4
// CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 8
// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x i8*], align 8
// CHECK1-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK1-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 8
// CHECK1-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 8
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    store i32 9, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK1-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK1-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK1-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9
// CHECK1-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK1:       cond.true:
// CHECK1-NEXT:    br label [[COND_END:%.*]]
// CHECK1:       cond.false:
// CHECK1-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    br label [[COND_END]]
// CHECK1:       cond.end:
// CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK1-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK1:       omp.inner.for.cond:
// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[CMP1:%.*]] = icmp slt i32 [[TMP6]], 10
// CHECK1-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK1:       omp.inner.for.body:
// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
// CHECK1-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK1-NEXT:    [[TMP12:%.*]] = inttoptr i64 [[TMP8]] to i8*
// CHECK1-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 8
// CHECK1-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK1-NEXT:    [[TMP14:%.*]] = inttoptr i64 [[TMP10]] to i8*
// CHECK1-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 8
// CHECK1-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK1-NEXT:    [[TMP16:%.*]] = bitcast [10 x i32]* [[TMP0]] to i8*
// CHECK1-NEXT:    store i8* [[TMP16]], i8** [[TMP15]], align 8
// CHECK1-NEXT:    [[TMP17:%.*]] = bitcast [3 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK1-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, [10 x i32]*)* @__omp_outlined__5 to i8*), i8* null, i8** [[TMP17]], i64 3)
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK1:       omp.inner.for.inc:
// CHECK1-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP18]], [[TMP19]]
// CHECK1-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK1-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK1-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP24]], 9
// CHECK1-NEXT:    br i1 [[CMP4]], label [[COND_TRUE5:%.*]], label [[COND_FALSE6:%.*]]
// CHECK1:       cond.true5:
// CHECK1-NEXT:    br label [[COND_END7:%.*]]
// CHECK1:       cond.false6:
// CHECK1-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    br label [[COND_END7]]
// CHECK1:       cond.end7:
// CHECK1-NEXT:    [[COND8:%.*]] = phi i32 [ 9, [[COND_TRUE5]] ], [ [[TMP25]], [[COND_FALSE6]] ]
// CHECK1-NEXT:    store i32 [[COND8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    store i32 [[TMP26]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK1:       omp.inner.for.end:
// CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK1:       omp.loop.exit:
// CHECK1-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]])
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@__omp_outlined__5
// CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 8
// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK1-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK1-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK1-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 8
// CHECK1-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 8
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK1-NEXT:    store i32 9, i32* [[DOTOMP_UB]], align 4
// CHECK1-NEXT:    [[TMP1:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK1-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP1]] to i32
// CHECK1-NEXT:    [[TMP2:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK1-NEXT:    [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32
// CHECK1-NEXT:    store i32 [[CONV]], i32* [[DOTOMP_LB]], align 4
// CHECK1-NEXT:    store i32 [[CONV1]], i32* [[DOTOMP_UB]], align 4
// CHECK1-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK1-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK1-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB3]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK1-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK1:       omp.inner.for.cond:
// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[CONV2:%.*]] = sext i32 [[TMP6]] to i64
// CHECK1-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK1-NEXT:    [[CMP:%.*]] = icmp ule i64 [[CONV2]], [[TMP7]]
// CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK1:       omp.inner.for.body:
// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP8]], 1
// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK1-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4
// CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP9]] to i64
// CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK1-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP10]], 1
// CHECK1-NEXT:    store i32 [[ADD3]], i32* [[ARRAYIDX]], align 4
// CHECK1-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK1:       omp.body.continue:
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK1:       omp.inner.for.inc:
// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
// CHECK1-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK1:       omp.inner.for.end:
// CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK1:       omp.loop.exit:
// CHECK1-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]])
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l44
// CHECK1-SAME: ([10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i64 [[F:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK1-NEXT:    [[F_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[F_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK1-NEXT:    store i64 [[F]], i64* [[F_ADDR]], align 8
// CHECK1-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK1-NEXT:    [[CONV:%.*]] = bitcast i64* [[F_ADDR]] to i32*
// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false, i1 false)
// CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK1:       user_code.entry:
// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB4]])
// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK1-NEXT:    [[CONV1:%.*]] = bitcast i64* [[F_CASTED]] to i32*
// CHECK1-NEXT:    store i32 [[TMP3]], i32* [[CONV1]], align 4
// CHECK1-NEXT:    [[TMP4:%.*]] = load i64, i64* [[F_CASTED]], align 8
// CHECK1-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK1-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK1-NEXT:    call void @__omp_outlined__6(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x [10 x i32]]* [[TMP0]], i64 [[TMP4]]) #[[ATTR2]]
// CHECK1-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
// CHECK1-NEXT:    ret void
// CHECK1:       worker.exit:
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@__omp_outlined__6
// CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i64 [[F:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK1-NEXT:    [[F_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[F_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 8
// CHECK1-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK1-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK1-NEXT:    store i64 [[F]], i64* [[F_ADDR]], align 8
// CHECK1-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK1-NEXT:    [[CONV:%.*]] = bitcast i64* [[F_ADDR]] to i32*
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    store i32 99, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK1-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK1-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK1-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 99
// CHECK1-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK1:       cond.true:
// CHECK1-NEXT:    br label [[COND_END:%.*]]
// CHECK1:       cond.false:
// CHECK1-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    br label [[COND_END]]
// CHECK1:       cond.end:
// CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK1-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK1:       omp.inner.for.cond:
// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[TMP6]], 100
// CHECK1-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK1:       omp.inner.for.body:
// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK1-NEXT:    [[CONV3:%.*]] = bitcast i64* [[F_CASTED]] to i32*
// CHECK1-NEXT:    store i32 [[TMP11]], i32* [[CONV3]], align 4
// CHECK1-NEXT:    [[TMP12:%.*]] = load i64, i64* [[F_CASTED]], align 8
// CHECK1-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK1-NEXT:    [[TMP14:%.*]] = inttoptr i64 [[TMP8]] to i8*
// CHECK1-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 8
// CHECK1-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK1-NEXT:    [[TMP16:%.*]] = inttoptr i64 [[TMP10]] to i8*
// CHECK1-NEXT:    store i8* [[TMP16]], i8** [[TMP15]], align 8
// CHECK1-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK1-NEXT:    [[TMP18:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK1-NEXT:    store i8* [[TMP18]], i8** [[TMP17]], align 8
// CHECK1-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK1-NEXT:    [[TMP20:%.*]] = inttoptr i64 [[TMP12]] to i8*
// CHECK1-NEXT:    store i8* [[TMP20]], i8** [[TMP19]], align 8
// CHECK1-NEXT:    [[TMP21:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK1-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, [10 x [10 x i32]]*, i64)* @__omp_outlined__7 to i8*), i8* null, i8** [[TMP21]], i64 4)
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK1:       omp.inner.for.inc:
// CHECK1-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK1-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK1-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP26]], [[TMP27]]
// CHECK1-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[CMP6:%.*]] = icmp sgt i32 [[TMP28]], 99
// CHECK1-NEXT:    br i1 [[CMP6]], label [[COND_TRUE7:%.*]], label [[COND_FALSE8:%.*]]
// CHECK1:       cond.true7:
// CHECK1-NEXT:    br label [[COND_END9:%.*]]
// CHECK1:       cond.false8:
// CHECK1-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    br label [[COND_END9]]
// CHECK1:       cond.end9:
// CHECK1-NEXT:    [[COND10:%.*]] = phi i32 [ 99, [[COND_TRUE7]] ], [ [[TMP29]], [[COND_FALSE8]] ]
// CHECK1-NEXT:    store i32 [[COND10]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    store i32 [[TMP30]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK1:       omp.inner.for.end:
// CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK1:       omp.loop.exit:
// CHECK1-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]])
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@__omp_outlined__7
// CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i64 [[F:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK1-NEXT:    [[F_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK1-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK1-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK1-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK1-NEXT:    store i64 [[F]], i64* [[F_ADDR]], align 8
// CHECK1-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK1-NEXT:    [[CONV:%.*]] = bitcast i64* [[F_ADDR]] to i32*
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK1-NEXT:    store i32 99, i32* [[DOTOMP_UB]], align 4
// CHECK1-NEXT:    [[TMP1:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK1-NEXT:    [[CONV2:%.*]] = trunc i64 [[TMP1]] to i32
// CHECK1-NEXT:    [[TMP2:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK1-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP2]] to i32
// CHECK1-NEXT:    store i32 [[CONV2]], i32* [[DOTOMP_LB]], align 4
// CHECK1-NEXT:    store i32 [[CONV3]], i32* [[DOTOMP_UB]], align 4
// CHECK1-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK1-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK1-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB3]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK1-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK1:       omp.inner.for.cond:
// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[CONV4:%.*]] = sext i32 [[TMP6]] to i64
// CHECK1-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK1-NEXT:    [[CMP:%.*]] = icmp ule i64 [[CONV4]], [[TMP7]]
// CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK1:       omp.inner.for.body:
// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP8]], 10
// CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK1-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[DIV5:%.*]] = sdiv i32 [[TMP10]], 10
// CHECK1-NEXT:    [[MUL6:%.*]] = mul nsw i32 [[DIV5]], 10
// CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP9]], [[MUL6]]
// CHECK1-NEXT:    [[MUL7:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK1-NEXT:    [[ADD8:%.*]] = add nsw i32 0, [[MUL7]]
// CHECK1-NEXT:    store i32 [[ADD8]], i32* [[J]], align 4
// CHECK1-NEXT:    store i32 10, i32* [[K]], align 4
// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, i32* [[J]], align 4
// CHECK1-NEXT:    [[TMP13:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK1-NEXT:    [[MUL9:%.*]] = mul nsw i32 [[TMP12]], [[TMP13]]
// CHECK1-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP11]], [[MUL9]]
// CHECK1-NEXT:    [[TMP14:%.*]] = load i32, i32* [[K]], align 4
// CHECK1-NEXT:    [[ADD11:%.*]] = add nsw i32 [[ADD10]], [[TMP14]]
// CHECK1-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
// CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP15]] to i64
// CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, i32* [[J]], align 4
// CHECK1-NEXT:    [[IDXPROM12:%.*]] = sext i32 [[TMP16]] to i64
// CHECK1-NEXT:    [[ARRAYIDX13:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i64 0, i64 [[IDXPROM12]]
// CHECK1-NEXT:    store i32 [[ADD11]], i32* [[ARRAYIDX13]], align 4
// CHECK1-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK1:       omp.body.continue:
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK1:       omp.inner.for.inc:
// CHECK1-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    [[ADD14:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK1-NEXT:    store i32 [[ADD14]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK1:       omp.inner.for.end:
// CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK1:       omp.loop.exit:
// CHECK1-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]])
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l52
// CHECK1-SAME: (i64 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK1-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK1-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK1-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK1-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false, i1 false)
// CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK1:       user_code.entry:
// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB4]])
// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK1-NEXT:    [[CONV1:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK1-NEXT:    store i32 [[TMP3]], i32* [[CONV1]], align 4
// CHECK1-NEXT:    [[TMP4:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK1-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK1-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK1-NEXT:    call void @__omp_outlined__8(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP4]], [10 x [10 x i32]]* [[TMP0]]) #[[ATTR2]]
// CHECK1-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
// CHECK1-NEXT:    ret void
// CHECK1:       worker.exit:
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@__omp_outlined__8
// CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I10:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[J11:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 8
// CHECK1-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK1-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK1-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK1-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK1-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK1-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK1-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK1-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK1-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK1-NEXT:    [[CONV4:%.*]] = sext i32 [[DIV]] to i64
// CHECK1-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK1-NEXT:    [[SUB5:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK1-NEXT:    [[DIV6:%.*]] = sdiv i32 [[SUB5]], 1
// CHECK1-NEXT:    [[CONV7:%.*]] = sext i32 [[DIV6]] to i64
// CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV4]], [[CONV7]]
// CHECK1-NEXT:    [[SUB8:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK1-NEXT:    store i64 [[SUB8]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK1-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK1-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK1-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK1:       land.lhs.true:
// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK1-NEXT:    [[CMP9:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK1-NEXT:    br i1 [[CMP9]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK1:       omp.precond.then:
// CHECK1-NEXT:    store i64 0, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK1-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK1-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK1-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK1-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK1-NEXT:    [[CONV12:%.*]] = zext i32 [[NVPTX_NUM_THREADS]] to i64
// CHECK1-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[TMP8]], align 4
// CHECK1-NEXT:    call void @__kmpc_distribute_static_init_8(%struct.ident_t* @[[GLOB2]], i32 [[TMP9]], i32 91, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_COMB_LB]], i64* [[DOTOMP_COMB_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 [[CONV12]])
// CHECK1-NEXT:    [[TMP10:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK1-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK1-NEXT:    [[CMP13:%.*]] = icmp sgt i64 [[TMP10]], [[TMP11]]
// CHECK1-NEXT:    br i1 [[CMP13]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK1:       cond.true:
// CHECK1-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK1-NEXT:    br label [[COND_END:%.*]]
// CHECK1:       cond.false:
// CHECK1-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK1-NEXT:    br label [[COND_END]]
// CHECK1:       cond.end:
// CHECK1-NEXT:    [[COND:%.*]] = phi i64 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
// CHECK1-NEXT:    store i64 [[COND]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK1-NEXT:    [[TMP14:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK1-NEXT:    store i64 [[TMP14]], i64* [[DOTOMP_IV]], align 8
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK1:       omp.inner.for.cond:
// CHECK1-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK1-NEXT:    [[TMP16:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i64 [[TMP16]], 1
// CHECK1-NEXT:    [[CMP14:%.*]] = icmp slt i64 [[TMP15]], [[ADD]]
// CHECK1-NEXT:    br i1 [[CMP14]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK1:       omp.inner.for.body:
// CHECK1-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK1-NEXT:    [[TMP18:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK1-NEXT:    [[TMP19:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK1-NEXT:    [[CONV15:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK1-NEXT:    store i32 [[TMP19]], i32* [[CONV15]], align 4
// CHECK1-NEXT:    [[TMP20:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK1-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK1-NEXT:    [[TMP22:%.*]] = inttoptr i64 [[TMP17]] to i8*
// CHECK1-NEXT:    store i8* [[TMP22]], i8** [[TMP21]], align 8
// CHECK1-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK1-NEXT:    [[TMP24:%.*]] = inttoptr i64 [[TMP18]] to i8*
// CHECK1-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 8
// CHECK1-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK1-NEXT:    [[TMP26:%.*]] = inttoptr i64 [[TMP20]] to i8*
// CHECK1-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 8
// CHECK1-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK1-NEXT:    [[TMP28:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK1-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 8
// CHECK1-NEXT:    [[TMP29:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP30:%.*]] = load i32, i32* [[TMP29]], align 4
// CHECK1-NEXT:    [[TMP31:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK1-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB4]], i32 [[TMP30]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, i64, [10 x [10 x i32]]*)* @__omp_outlined__9 to i8*), i8* null, i8** [[TMP31]], i64 4)
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK1:       omp.inner.for.inc:
// CHECK1-NEXT:    [[TMP32:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK1-NEXT:    [[TMP33:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK1-NEXT:    [[ADD16:%.*]] = add nsw i64 [[TMP32]], [[TMP33]]
// CHECK1-NEXT:    store i64 [[ADD16]], i64* [[DOTOMP_IV]], align 8
// CHECK1-NEXT:    [[TMP34:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK1-NEXT:    [[TMP35:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK1-NEXT:    [[ADD17:%.*]] = add nsw i64 [[TMP34]], [[TMP35]]
// CHECK1-NEXT:    store i64 [[ADD17]], i64* [[DOTOMP_COMB_LB]], align 8
// CHECK1-NEXT:    [[TMP36:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK1-NEXT:    [[TMP37:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK1-NEXT:    [[ADD18:%.*]] = add nsw i64 [[TMP36]], [[TMP37]]
// CHECK1-NEXT:    store i64 [[ADD18]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK1-NEXT:    [[TMP38:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK1-NEXT:    [[TMP39:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK1-NEXT:    [[CMP19:%.*]] = icmp sgt i64 [[TMP38]], [[TMP39]]
// CHECK1-NEXT:    br i1 [[CMP19]], label [[COND_TRUE20:%.*]], label [[COND_FALSE21:%.*]]
// CHECK1:       cond.true20:
// CHECK1-NEXT:    [[TMP40:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK1-NEXT:    br label [[COND_END22:%.*]]
// CHECK1:       cond.false21:
// CHECK1-NEXT:    [[TMP41:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK1-NEXT:    br label [[COND_END22]]
// CHECK1:       cond.end22:
// CHECK1-NEXT:    [[COND23:%.*]] = phi i64 [ [[TMP40]], [[COND_TRUE20]] ], [ [[TMP41]], [[COND_FALSE21]] ]
// CHECK1-NEXT:    store i64 [[COND23]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK1-NEXT:    [[TMP42:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK1-NEXT:    store i64 [[TMP42]], i64* [[DOTOMP_IV]], align 8
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK1:       omp.inner.for.end:
// CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK1:       omp.loop.exit:
// CHECK1-NEXT:    [[TMP43:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP44:%.*]] = load i32, i32* [[TMP43]], align 4
// CHECK1-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP44]])
// CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK1:       omp.precond.end:
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@__omp_outlined__9
// CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], i64 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_LB:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTOMP_UB:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I10:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[J11:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK1-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK1-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK1-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK1-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK1-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK1-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK1-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK1-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK1-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK1-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK1-NEXT:    [[CONV4:%.*]] = sext i32 [[DIV]] to i64
// CHECK1-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK1-NEXT:    [[SUB5:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK1-NEXT:    [[DIV6:%.*]] = sdiv i32 [[SUB5]], 1
// CHECK1-NEXT:    [[CONV7:%.*]] = sext i32 [[DIV6]] to i64
// CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV4]], [[CONV7]]
// CHECK1-NEXT:    [[SUB8:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK1-NEXT:    store i64 [[SUB8]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK1-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK1-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK1-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK1:       land.lhs.true:
// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK1-NEXT:    [[CMP9:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK1-NEXT:    br i1 [[CMP9]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK1:       omp.precond.then:
// CHECK1-NEXT:    store i64 0, i64* [[DOTOMP_LB]], align 8
// CHECK1-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK1-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_UB]], align 8
// CHECK1-NEXT:    [[TMP8:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK1-NEXT:    [[TMP9:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK1-NEXT:    store i64 [[TMP8]], i64* [[DOTOMP_LB]], align 8
// CHECK1-NEXT:    store i64 [[TMP9]], i64* [[DOTOMP_UB]], align 8
// CHECK1-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK1-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK1-NEXT:    call void @__kmpc_for_static_init_8(%struct.ident_t* @[[GLOB3]], i32 [[TMP11]], i32 33, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_LB]], i64* [[DOTOMP_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 1)
// CHECK1-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTOMP_LB]], align 8
// CHECK1-NEXT:    store i64 [[TMP12]], i64* [[DOTOMP_IV]], align 8
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK1:       omp.inner.for.cond:
// CHECK1-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK1-NEXT:    [[TMP14:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK1-NEXT:    [[CMP12:%.*]] = icmp ule i64 [[TMP13]], [[TMP14]]
// CHECK1-NEXT:    br i1 [[CMP12]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK1:       omp.inner.for.body:
// CHECK1-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK1-NEXT:    [[SUB13:%.*]] = sub nsw i32 [[TMP16]], 0
// CHECK1-NEXT:    [[DIV14:%.*]] = sdiv i32 [[SUB13]], 1
// CHECK1-NEXT:    [[MUL15:%.*]] = mul nsw i32 1, [[DIV14]]
// CHECK1-NEXT:    [[CONV16:%.*]] = sext i32 [[MUL15]] to i64
// CHECK1-NEXT:    [[DIV17:%.*]] = sdiv i64 [[TMP15]], [[CONV16]]
// CHECK1-NEXT:    [[MUL18:%.*]] = mul nsw i64 [[DIV17]], 1
// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i64 0, [[MUL18]]
// CHECK1-NEXT:    [[CONV19:%.*]] = trunc i64 [[ADD]] to i32
// CHECK1-NEXT:    store i32 [[CONV19]], i32* [[I10]], align 4
// CHECK1-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK1-NEXT:    [[TMP18:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK1-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK1-NEXT:    [[SUB20:%.*]] = sub nsw i32 [[TMP19]], 0
// CHECK1-NEXT:    [[DIV21:%.*]] = sdiv i32 [[SUB20]], 1
// CHECK1-NEXT:    [[MUL22:%.*]] = mul nsw i32 1, [[DIV21]]
// CHECK1-NEXT:    [[CONV23:%.*]] = sext i32 [[MUL22]] to i64
// CHECK1-NEXT:    [[DIV24:%.*]] = sdiv i64 [[TMP18]], [[CONV23]]
// CHECK1-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK1-NEXT:    [[SUB25:%.*]] = sub nsw i32 [[TMP20]], 0
// CHECK1-NEXT:    [[DIV26:%.*]] = sdiv i32 [[SUB25]], 1
// CHECK1-NEXT:    [[MUL27:%.*]] = mul nsw i32 1, [[DIV26]]
// CHECK1-NEXT:    [[CONV28:%.*]] = sext i32 [[MUL27]] to i64
// CHECK1-NEXT:    [[MUL29:%.*]] = mul nsw i64 [[DIV24]], [[CONV28]]
// CHECK1-NEXT:    [[SUB30:%.*]] = sub nsw i64 [[TMP17]], [[MUL29]]
// CHECK1-NEXT:    [[MUL31:%.*]] = mul nsw i64 [[SUB30]], 1
// CHECK1-NEXT:    [[ADD32:%.*]] = add nsw i64 0, [[MUL31]]
// CHECK1-NEXT:    [[CONV33:%.*]] = trunc i64 [[ADD32]] to i32
// CHECK1-NEXT:    store i32 [[CONV33]], i32* [[J11]], align 4
// CHECK1-NEXT:    [[TMP21:%.*]] = load i32, i32* [[I10]], align 4
// CHECK1-NEXT:    [[TMP22:%.*]] = load i32, i32* [[J11]], align 4
// CHECK1-NEXT:    [[ADD34:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
// CHECK1-NEXT:    [[TMP23:%.*]] = load i32, i32* [[I10]], align 4
// CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP23]] to i64
// CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK1-NEXT:    [[TMP24:%.*]] = load i32, i32* [[J11]], align 4
// CHECK1-NEXT:    [[IDXPROM35:%.*]] = sext i32 [[TMP24]] to i64
// CHECK1-NEXT:    [[ARRAYIDX36:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i64 0, i64 [[IDXPROM35]]
// CHECK1-NEXT:    store i32 [[ADD34]], i32* [[ARRAYIDX36]], align 4
// CHECK1-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK1:       omp.body.continue:
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK1:       omp.inner.for.inc:
// CHECK1-NEXT:    [[TMP25:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK1-NEXT:    [[TMP26:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK1-NEXT:    [[ADD37:%.*]] = add nsw i64 [[TMP25]], [[TMP26]]
// CHECK1-NEXT:    store i64 [[ADD37]], i64* [[DOTOMP_IV]], align 8
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK1:       omp.inner.for.end:
// CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK1:       omp.loop.exit:
// CHECK1-NEXT:    [[TMP27:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP28:%.*]] = load i32, i32* [[TMP27]], align 4
// CHECK1-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP28]])
// CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK1:       omp.precond.end:
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l59
// CHECK1-SAME: (i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK1-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK1-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK1-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 8
// CHECK1-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK1-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false, i1 false)
// CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK1:       user_code.entry:
// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB4]])
// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK1-NEXT:    [[CONV1:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK1-NEXT:    store i32 [[TMP3]], i32* [[CONV1]], align 4
// CHECK1-NEXT:    [[TMP4:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK1-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[V_ADDR]], align 8
// CHECK1-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK1-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK1-NEXT:    call void @__omp_outlined__10(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP4]], [1000 x i32]* [[TMP0]], i32* [[TMP5]]) #[[ATTR2]]
// CHECK1-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
// CHECK1-NEXT:    ret void
// CHECK1:       worker.exit:
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@__omp_outlined__10
// CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK1-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 8
// CHECK1-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK1-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK1-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK1-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 8
// CHECK1-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK1-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK1-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK1-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK1-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK1-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK1-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK1:       omp.precond.then:
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK1-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK1-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK1-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK1-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK1:       cond.true:
// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT:    br label [[COND_END:%.*]]
// CHECK1:       cond.false:
// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    br label [[COND_END]]
// CHECK1:       cond.end:
// CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK1-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK1:       omp.inner.for.cond:
// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK1-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK1-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK1:       omp.inner.for.body:
// CHECK1-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    [[TMP15:%.*]] = zext i32 [[TMP14]] to i64
// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP17:%.*]] = zext i32 [[TMP16]] to i64
// CHECK1-NEXT:    [[TMP18:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK1-NEXT:    [[CONV6:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK1-NEXT:    store i32 [[TMP18]], i32* [[CONV6]], align 4
// CHECK1-NEXT:    [[TMP19:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK1-NEXT:    [[TMP20:%.*]] = load i32*, i32** [[V_ADDR]], align 8
// CHECK1-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK1-NEXT:    [[TMP22:%.*]] = inttoptr i64 [[TMP15]] to i8*
// CHECK1-NEXT:    store i8* [[TMP22]], i8** [[TMP21]], align 8
// CHECK1-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK1-NEXT:    [[TMP24:%.*]] = inttoptr i64 [[TMP17]] to i8*
// CHECK1-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 8
// CHECK1-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK1-NEXT:    [[TMP26:%.*]] = inttoptr i64 [[TMP19]] to i8*
// CHECK1-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 8
// CHECK1-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK1-NEXT:    [[TMP28:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK1-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 8
// CHECK1-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 4
// CHECK1-NEXT:    [[TMP30:%.*]] = bitcast i32* [[TMP20]] to i8*
// CHECK1-NEXT:    store i8* [[TMP30]], i8** [[TMP29]], align 8
// CHECK1-NEXT:    [[TMP31:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP32:%.*]] = load i32, i32* [[TMP31]], align 4
// CHECK1-NEXT:    [[TMP33:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK1-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB4]], i32 [[TMP32]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, i64, [1000 x i32]*, i32*)* @__omp_outlined__11 to i8*), i8* null, i8** [[TMP33]], i64 5)
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK1:       omp.inner.for.inc:
// CHECK1-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP34]], [[TMP35]]
// CHECK1-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP36]], [[TMP37]]
// CHECK1-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP38]], [[TMP39]]
// CHECK1-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT:    [[CMP10:%.*]] = icmp sgt i32 [[TMP40]], [[TMP41]]
// CHECK1-NEXT:    br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]]
// CHECK1:       cond.true11:
// CHECK1-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT:    br label [[COND_END13:%.*]]
// CHECK1:       cond.false12:
// CHECK1-NEXT:    [[TMP43:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    br label [[COND_END13]]
// CHECK1:       cond.end13:
// CHECK1-NEXT:    [[COND14:%.*]] = phi i32 [ [[TMP42]], [[COND_TRUE11]] ], [ [[TMP43]], [[COND_FALSE12]] ]
// CHECK1-NEXT:    store i32 [[COND14]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK1-NEXT:    [[TMP44:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK1-NEXT:    store i32 [[TMP44]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK1:       omp.inner.for.end:
// CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK1:       omp.loop.exit:
// CHECK1-NEXT:    [[TMP45:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP46:%.*]] = load i32, i32* [[TMP45]], align 4
// CHECK1-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP46]])
// CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK1:       omp.precond.end:
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@__omp_outlined__11
// CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK1-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I5:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK1-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK1-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK1-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK1-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK1-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 8
// CHECK1-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK1-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK1-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK1-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK1-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK1-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK1-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK1:       omp.precond.then:
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK1-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK1-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK1-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP5]] to i32
// CHECK1-NEXT:    [[TMP6:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK1-NEXT:    [[CONV4:%.*]] = trunc i64 [[TMP6]] to i32
// CHECK1-NEXT:    store i32 [[CONV3]], i32* [[DOTOMP_LB]], align 4
// CHECK1-NEXT:    store i32 [[CONV4]], i32* [[DOTOMP_UB]], align 4
// CHECK1-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK1-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK1-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB3]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK1-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK1:       omp.inner.for.cond:
// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[CONV6:%.*]] = sext i32 [[TMP10]] to i64
// CHECK1-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK1-NEXT:    [[CMP7:%.*]] = icmp ule i64 [[CONV6]], [[TMP11]]
// CHECK1-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK1:       omp.inner.for.body:
// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK1-NEXT:    store i32 [[ADD]], i32* [[I5]], align 4
// CHECK1-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[V_ADDR]], align 8
// CHECK1-NEXT:    [[TMP14:%.*]] = load i32, i32* [[I5]], align 4
// CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP14]] to i64
// CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[TMP13]], i64 [[IDXPROM]]
// CHECK1-NEXT:    [[TMP15:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, i32* [[I5]], align 4
// CHECK1-NEXT:    [[IDXPROM8:%.*]] = sext i32 [[TMP16]] to i64
// CHECK1-NEXT:    [[ARRAYIDX9:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i64 0, i64 [[IDXPROM8]]
// CHECK1-NEXT:    store i32 [[TMP15]], i32* [[ARRAYIDX9]], align 4
// CHECK1-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK1:       omp.body.continue:
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK1:       omp.inner.for.inc:
// CHECK1-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK1-NEXT:    store i32 [[ADD10]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK1:       omp.inner.for.end:
// CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK1:       omp.loop.exit:
// CHECK1-NEXT:    [[TMP19:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:    [[TMP20:%.*]] = load i32, i32* [[TMP19]], align 4
// CHECK1-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP20]])
// CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK1:       omp.precond.end:
// CHECK1-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l28
// CHECK2-SAME: (i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i64 [[L:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK2-NEXT:    [[L_ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[L_CASTED:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK2-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK2-NEXT:    store i64 [[L]], i64* [[L_ADDR]], align 8
// CHECK2-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK2-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK2-NEXT:    [[CONV1:%.*]] = bitcast i64* [[L_ADDR]] to i32*
// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i8 2, i1 false, i1 false)
// CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK2:       user_code.entry:
// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB4:[0-9]+]])
// CHECK2-NEXT:    [[TMP3:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK2-NEXT:    [[CONV2:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK2-NEXT:    store i32 [[TMP3]], i32* [[CONV2]], align 4
// CHECK2-NEXT:    [[TMP4:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK2-NEXT:    [[CONV3:%.*]] = bitcast i64* [[L_CASTED]] to i32*
// CHECK2-NEXT:    store i32 [[TMP5]], i32* [[CONV3]], align 4
// CHECK2-NEXT:    [[TMP6:%.*]] = load i64, i64* [[L_CASTED]], align 8
// CHECK2-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK2-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK2-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP4]], [1000 x i32]* [[TMP0]], i64 [[TMP6]]) #[[ATTR2:[0-9]+]]
// CHECK2-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
// CHECK2-NEXT:    ret void
// CHECK2:       worker.exit:
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@__omp_outlined__
// CHECK2-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i64 [[L:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK2-NEXT:    [[L_ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I5:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[L_CASTED:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 8
// CHECK2-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK2-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK2-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK2-NEXT:    store i64 [[L]], i64* [[L_ADDR]], align 8
// CHECK2-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK2-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK2-NEXT:    [[CONV1:%.*]] = bitcast i64* [[L_ADDR]] to i32*
// CHECK2-NEXT:    [[L2:%.*]] = call i8* @__kmpc_alloc_shared(i64 4)
// CHECK2-NEXT:    [[L_ON_STACK:%.*]] = bitcast i8* [[L2]] to i32*
// CHECK2-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK2-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK2-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK2-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK2-NEXT:    store i32 [[SUB4]], i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK2-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK2-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK2:       omp.precond.then:
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK2-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK2-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK2-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 128)
// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK2-NEXT:    [[CMP6:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK2-NEXT:    br i1 [[CMP6]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK2:       cond.true:
// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK2-NEXT:    br label [[COND_END:%.*]]
// CHECK2:       cond.false:
// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    br label [[COND_END]]
// CHECK2:       cond.end:
// CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK2-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK2:       omp.inner.for.cond:
// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK2-NEXT:    [[CMP7:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK2-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK2:       omp.inner.for.body:
// CHECK2-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    [[TMP15:%.*]] = zext i32 [[TMP14]] to i64
// CHECK2-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP17:%.*]] = zext i32 [[TMP16]] to i64
// CHECK2-NEXT:    [[TMP18:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK2-NEXT:    [[CONV8:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK2-NEXT:    store i32 [[TMP18]], i32* [[CONV8]], align 4
// CHECK2-NEXT:    [[TMP19:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK2-NEXT:    [[TMP20:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK2-NEXT:    [[CONV9:%.*]] = bitcast i64* [[L_CASTED]] to i32*
// CHECK2-NEXT:    store i32 [[TMP20]], i32* [[CONV9]], align 4
// CHECK2-NEXT:    [[TMP21:%.*]] = load i64, i64* [[L_CASTED]], align 8
// CHECK2-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK2-NEXT:    [[TMP23:%.*]] = inttoptr i64 [[TMP15]] to i8*
// CHECK2-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 8
// CHECK2-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK2-NEXT:    [[TMP25:%.*]] = inttoptr i64 [[TMP17]] to i8*
// CHECK2-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 8
// CHECK2-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK2-NEXT:    [[TMP27:%.*]] = inttoptr i64 [[TMP19]] to i8*
// CHECK2-NEXT:    store i8* [[TMP27]], i8** [[TMP26]], align 8
// CHECK2-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK2-NEXT:    [[TMP29:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK2-NEXT:    store i8* [[TMP29]], i8** [[TMP28]], align 8
// CHECK2-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 4
// CHECK2-NEXT:    [[TMP31:%.*]] = inttoptr i64 [[TMP21]] to i8*
// CHECK2-NEXT:    store i8* [[TMP31]], i8** [[TMP30]], align 8
// CHECK2-NEXT:    [[TMP32:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP33:%.*]] = load i32, i32* [[TMP32]], align 4
// CHECK2-NEXT:    [[TMP34:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK2-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB4]], i32 [[TMP33]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, i64, [1000 x i32]*, i64)* @__omp_outlined__1 to i8*), i8* null, i8** [[TMP34]], i64 5)
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK2:       omp.inner.for.inc:
// CHECK2-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP35]], [[TMP36]]
// CHECK2-NEXT:    store i32 [[ADD10]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP37]], [[TMP38]]
// CHECK2-NEXT:    store i32 [[ADD11]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP39]], [[TMP40]]
// CHECK2-NEXT:    store i32 [[ADD12]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK2-NEXT:    [[CMP13:%.*]] = icmp sgt i32 [[TMP41]], [[TMP42]]
// CHECK2-NEXT:    br i1 [[CMP13]], label [[COND_TRUE14:%.*]], label [[COND_FALSE15:%.*]]
// CHECK2:       cond.true14:
// CHECK2-NEXT:    [[TMP43:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK2-NEXT:    br label [[COND_END16:%.*]]
// CHECK2:       cond.false15:
// CHECK2-NEXT:    [[TMP44:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    br label [[COND_END16]]
// CHECK2:       cond.end16:
// CHECK2-NEXT:    [[COND17:%.*]] = phi i32 [ [[TMP43]], [[COND_TRUE14]] ], [ [[TMP44]], [[COND_FALSE15]] ]
// CHECK2-NEXT:    store i32 [[COND17]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP45:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    store i32 [[TMP45]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK2:       omp.inner.for.end:
// CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK2:       omp.loop.exit:
// CHECK2-NEXT:    [[TMP46:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP47:%.*]] = load i32, i32* [[TMP46]], align 4
// CHECK2-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP47]])
// CHECK2-NEXT:    [[TMP48:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK2-NEXT:    [[TMP49:%.*]] = icmp ne i32 [[TMP48]], 0
// CHECK2-NEXT:    br i1 [[TMP49]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK2:       .omp.lastprivate.then:
// CHECK2-NEXT:    [[TMP50:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK2-NEXT:    store i32 [[TMP50]], i32* [[CONV1]], align 8
// CHECK2-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK2:       .omp.lastprivate.done:
// CHECK2-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK2:       omp.precond.end:
// CHECK2-NEXT:    call void @__kmpc_free_shared(i8* [[L2]], i64 4)
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@__omp_outlined__1
// CHECK2-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i64 [[L:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK2-NEXT:    [[L_ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I6:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK2-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK2-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK2-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK2-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK2-NEXT:    store i64 [[L]], i64* [[L_ADDR]], align 8
// CHECK2-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK2-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK2-NEXT:    [[CONV1:%.*]] = bitcast i64* [[L_ADDR]] to i32*
// CHECK2-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK2-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK2-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK2-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK2-NEXT:    store i32 [[SUB3]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK2-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK2-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK2:       omp.precond.then:
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK2-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK2-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK2-NEXT:    [[CONV4:%.*]] = trunc i64 [[TMP5]] to i32
// CHECK2-NEXT:    [[TMP6:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK2-NEXT:    [[CONV5:%.*]] = trunc i64 [[TMP6]] to i32
// CHECK2-NEXT:    store i32 [[CONV4]], i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    store i32 [[CONV5]], i32* [[DOTOMP_UB]], align 4
// CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK2-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK2-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB3:[0-9]+]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 32)
// CHECK2-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK2:       omp.dispatch.cond:
// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK2-NEXT:    [[TMP10:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK2-NEXT:    [[CONV7:%.*]] = trunc i64 [[TMP10]] to i32
// CHECK2-NEXT:    [[CMP8:%.*]] = icmp sgt i32 [[TMP9]], [[CONV7]]
// CHECK2-NEXT:    br i1 [[CMP8]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK2:       cond.true:
// CHECK2-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK2-NEXT:    [[CONV9:%.*]] = trunc i64 [[TMP11]] to i32
// CHECK2-NEXT:    br label [[COND_END:%.*]]
// CHECK2:       cond.false:
// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK2-NEXT:    br label [[COND_END]]
// CHECK2:       cond.end:
// CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ [[CONV9]], [[COND_TRUE]] ], [ [[TMP12]], [[COND_FALSE]] ]
// CHECK2-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
// CHECK2-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    store i32 [[TMP13]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK2-NEXT:    [[CMP10:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK2-NEXT:    br i1 [[CMP10]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK2:       omp.dispatch.body:
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK2:       omp.inner.for.cond:
// CHECK2-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK2-NEXT:    [[CMP11:%.*]] = icmp sle i32 [[TMP16]], [[TMP17]]
// CHECK2-NEXT:    br i1 [[CMP11]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK2:       omp.inner.for.body:
// CHECK2-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP18]], 1
// CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK2-NEXT:    store i32 [[ADD]], i32* [[I6]], align 4
// CHECK2-NEXT:    [[TMP19:%.*]] = load i32, i32* [[I6]], align 4
// CHECK2-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP19]] to i64
// CHECK2-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK2-NEXT:    store i32 1, i32* [[ARRAYIDX]], align 4
// CHECK2-NEXT:    [[TMP20:%.*]] = load i32, i32* [[I6]], align 4
// CHECK2-NEXT:    store i32 [[TMP20]], i32* [[CONV1]], align 8
// CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK2:       omp.body.continue:
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK2:       omp.inner.for.inc:
// CHECK2-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP21]], 1
// CHECK2-NEXT:    store i32 [[ADD12]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK2:       omp.inner.for.end:
// CHECK2-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK2:       omp.dispatch.inc:
// CHECK2-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    [[ADD13:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK2-NEXT:    store i32 [[ADD13]], i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK2-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    [[ADD14:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK2-NEXT:    store i32 [[ADD14]], i32* [[DOTOMP_UB]], align 4
// CHECK2-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK2:       omp.dispatch.end:
// CHECK2-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK2-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP27]])
// CHECK2-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK2-NEXT:    [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
// CHECK2-NEXT:    br i1 [[TMP29]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK2:       .omp.lastprivate.then:
// CHECK2-NEXT:    [[TMP30:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK2-NEXT:    store i32 [[TMP30]], i32* [[CONV1]], align 8
// CHECK2-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK2:       .omp.lastprivate.done:
// CHECK2-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK2:       omp.precond.end:
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34
// CHECK2-SAME: (i64 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR4:[0-9]+]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 8
// CHECK2-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK2-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 8
// CHECK2-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK2-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 8
// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false, i1 false)
// CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK2:       user_code.entry:
// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB4]])
// CHECK2-NEXT:    [[TMP3:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK2-NEXT:    [[CONV1:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK2-NEXT:    store i32 [[TMP3]], i32* [[CONV1]], align 4
// CHECK2-NEXT:    [[TMP4:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK2-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK2-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK2-NEXT:    call void @__omp_outlined__2(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP4]], [1000 x i16]* [[TMP0]]) #[[ATTR2]]
// CHECK2-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
// CHECK2-NEXT:    ret void
// CHECK2:       worker.exit:
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@__omp_outlined__2
// CHECK2-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 8
// CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 8
// CHECK2-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK2-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK2-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 8
// CHECK2-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK2-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 8
// CHECK2-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK2-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK2-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK2-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK2-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK2-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK2-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK2:       omp.precond.then:
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK2-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK2-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK2-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK2-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK2-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK2-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK2:       cond.true:
// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK2-NEXT:    br label [[COND_END:%.*]]
// CHECK2:       cond.false:
// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    br label [[COND_END]]
// CHECK2:       cond.end:
// CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK2-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK2:       omp.inner.for.cond:
// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK2-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK2-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK2:       omp.inner.for.body:
// CHECK2-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    [[TMP15:%.*]] = zext i32 [[TMP14]] to i64
// CHECK2-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP17:%.*]] = zext i32 [[TMP16]] to i64
// CHECK2-NEXT:    [[TMP18:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK2-NEXT:    [[CONV6:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK2-NEXT:    store i32 [[TMP18]], i32* [[CONV6]], align 4
// CHECK2-NEXT:    [[TMP19:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK2-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK2-NEXT:    [[TMP21:%.*]] = inttoptr i64 [[TMP15]] to i8*
// CHECK2-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 8
// CHECK2-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK2-NEXT:    [[TMP23:%.*]] = inttoptr i64 [[TMP17]] to i8*
// CHECK2-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 8
// CHECK2-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK2-NEXT:    [[TMP25:%.*]] = inttoptr i64 [[TMP19]] to i8*
// CHECK2-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 8
// CHECK2-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK2-NEXT:    [[TMP27:%.*]] = bitcast [1000 x i16]* [[TMP0]] to i8*
// CHECK2-NEXT:    store i8* [[TMP27]], i8** [[TMP26]], align 8
// CHECK2-NEXT:    [[TMP28:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP29:%.*]] = load i32, i32* [[TMP28]], align 4
// CHECK2-NEXT:    [[TMP30:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK2-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB4]], i32 [[TMP29]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, i64, [1000 x i16]*)* @__omp_outlined__3 to i8*), i8* null, i8** [[TMP30]], i64 4)
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK2:       omp.inner.for.inc:
// CHECK2-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP31]], [[TMP32]]
// CHECK2-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP33]], [[TMP34]]
// CHECK2-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP35]], [[TMP36]]
// CHECK2-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK2-NEXT:    [[CMP10:%.*]] = icmp sgt i32 [[TMP37]], [[TMP38]]
// CHECK2-NEXT:    br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]]
// CHECK2:       cond.true11:
// CHECK2-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK2-NEXT:    br label [[COND_END13:%.*]]
// CHECK2:       cond.false12:
// CHECK2-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    br label [[COND_END13]]
// CHECK2:       cond.end13:
// CHECK2-NEXT:    [[COND14:%.*]] = phi i32 [ [[TMP39]], [[COND_TRUE11]] ], [ [[TMP40]], [[COND_FALSE12]] ]
// CHECK2-NEXT:    store i32 [[COND14]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    store i32 [[TMP41]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK2:       omp.inner.for.end:
// CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK2:       omp.loop.exit:
// CHECK2-NEXT:    [[TMP42:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP43:%.*]] = load i32, i32* [[TMP42]], align 4
// CHECK2-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP43]])
// CHECK2-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK2:       omp.precond.end:
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@__omp_outlined__3
// CHECK2-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], i64 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 8
// CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I5:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK2-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK2-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK2-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK2-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 8
// CHECK2-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK2-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 8
// CHECK2-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK2-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK2-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK2-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK2-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK2-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK2-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK2:       omp.precond.then:
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK2-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK2-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK2-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP5]] to i32
// CHECK2-NEXT:    [[TMP6:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK2-NEXT:    [[CONV4:%.*]] = trunc i64 [[TMP6]] to i32
// CHECK2-NEXT:    store i32 [[CONV3]], i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    store i32 [[CONV4]], i32* [[DOTOMP_UB]], align 4
// CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK2-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK2-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB3]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK2:       omp.inner.for.cond:
// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[CONV6:%.*]] = sext i32 [[TMP10]] to i64
// CHECK2-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK2-NEXT:    [[CMP7:%.*]] = icmp ule i64 [[CONV6]], [[TMP11]]
// CHECK2-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK2:       omp.inner.for.body:
// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK2-NEXT:    store i32 [[ADD]], i32* [[I5]], align 4
// CHECK2-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I5]], align 4
// CHECK2-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP13]] to i64
// CHECK2-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i16], [1000 x i16]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK2-NEXT:    [[TMP14:%.*]] = load i16, i16* [[ARRAYIDX]], align 2
// CHECK2-NEXT:    [[CONV8:%.*]] = sext i16 [[TMP14]] to i32
// CHECK2-NEXT:    [[ADD9:%.*]] = add nsw i32 [[CONV8]], 1
// CHECK2-NEXT:    [[CONV10:%.*]] = trunc i32 [[ADD9]] to i16
// CHECK2-NEXT:    store i16 [[CONV10]], i16* [[ARRAYIDX]], align 2
// CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK2:       omp.body.continue:
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK2:       omp.inner.for.inc:
// CHECK2-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP15]], [[TMP16]]
// CHECK2-NEXT:    store i32 [[ADD11]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK2:       omp.inner.for.end:
// CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK2:       omp.loop.exit:
// CHECK2-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
// CHECK2-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP18]])
// CHECK2-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK2:       omp.precond.end:
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l39
// CHECK2-SAME: ([10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 8
// CHECK2-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 8
// CHECK2-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 8
// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false, i1 false)
// CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK2:       user_code.entry:
// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB4]])
// CHECK2-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK2-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK2-NEXT:    call void @__omp_outlined__4(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x i32]* [[TMP0]]) #[[ATTR2]]
// CHECK2-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
// CHECK2-NEXT:    ret void
// CHECK2:       worker.exit:
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@__omp_outlined__4
// CHECK2-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 8
// CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x i8*], align 8
// CHECK2-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK2-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 8
// CHECK2-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 8
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    store i32 9, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK2-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK2-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK2-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK2-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9
// CHECK2-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK2:       cond.true:
// CHECK2-NEXT:    br label [[COND_END:%.*]]
// CHECK2:       cond.false:
// CHECK2-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    br label [[COND_END]]
// CHECK2:       cond.end:
// CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK2-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK2:       omp.inner.for.cond:
// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[CMP1:%.*]] = icmp slt i32 [[TMP6]], 10
// CHECK2-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK2:       omp.inner.for.body:
// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
// CHECK2-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK2-NEXT:    [[TMP12:%.*]] = inttoptr i64 [[TMP8]] to i8*
// CHECK2-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 8
// CHECK2-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK2-NEXT:    [[TMP14:%.*]] = inttoptr i64 [[TMP10]] to i8*
// CHECK2-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 8
// CHECK2-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK2-NEXT:    [[TMP16:%.*]] = bitcast [10 x i32]* [[TMP0]] to i8*
// CHECK2-NEXT:    store i8* [[TMP16]], i8** [[TMP15]], align 8
// CHECK2-NEXT:    [[TMP17:%.*]] = bitcast [3 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK2-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, [10 x i32]*)* @__omp_outlined__5 to i8*), i8* null, i8** [[TMP17]], i64 3)
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK2:       omp.inner.for.inc:
// CHECK2-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP18]], [[TMP19]]
// CHECK2-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK2-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK2-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP24]], 9
// CHECK2-NEXT:    br i1 [[CMP4]], label [[COND_TRUE5:%.*]], label [[COND_FALSE6:%.*]]
// CHECK2:       cond.true5:
// CHECK2-NEXT:    br label [[COND_END7:%.*]]
// CHECK2:       cond.false6:
// CHECK2-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    br label [[COND_END7]]
// CHECK2:       cond.end7:
// CHECK2-NEXT:    [[COND8:%.*]] = phi i32 [ 9, [[COND_TRUE5]] ], [ [[TMP25]], [[COND_FALSE6]] ]
// CHECK2-NEXT:    store i32 [[COND8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    store i32 [[TMP26]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK2:       omp.inner.for.end:
// CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK2:       omp.loop.exit:
// CHECK2-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]])
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@__omp_outlined__5
// CHECK2-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 8
// CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK2-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK2-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK2-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 8
// CHECK2-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 8
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    store i32 9, i32* [[DOTOMP_UB]], align 4
// CHECK2-NEXT:    [[TMP1:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK2-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP1]] to i32
// CHECK2-NEXT:    [[TMP2:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK2-NEXT:    [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32
// CHECK2-NEXT:    store i32 [[CONV]], i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    store i32 [[CONV1]], i32* [[DOTOMP_UB]], align 4
// CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK2-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK2-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB3]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK2:       omp.inner.for.cond:
// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[CONV2:%.*]] = sext i32 [[TMP6]] to i64
// CHECK2-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK2-NEXT:    [[CMP:%.*]] = icmp ule i64 [[CONV2]], [[TMP7]]
// CHECK2-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK2:       omp.inner.for.body:
// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP8]], 1
// CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK2-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4
// CHECK2-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP9]] to i64
// CHECK2-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK2-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP10]], 1
// CHECK2-NEXT:    store i32 [[ADD3]], i32* [[ARRAYIDX]], align 4
// CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK2:       omp.body.continue:
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK2:       omp.inner.for.inc:
// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
// CHECK2-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK2:       omp.inner.for.end:
// CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK2:       omp.loop.exit:
// CHECK2-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]])
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l44
// CHECK2-SAME: ([10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i64 [[F:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK2-NEXT:    [[F_ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[F_CASTED:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK2-NEXT:    store i64 [[F]], i64* [[F_ADDR]], align 8
// CHECK2-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK2-NEXT:    [[CONV:%.*]] = bitcast i64* [[F_ADDR]] to i32*
// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false, i1 false)
// CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK2:       user_code.entry:
// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB4]])
// CHECK2-NEXT:    [[TMP3:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK2-NEXT:    [[CONV1:%.*]] = bitcast i64* [[F_CASTED]] to i32*
// CHECK2-NEXT:    store i32 [[TMP3]], i32* [[CONV1]], align 4
// CHECK2-NEXT:    [[TMP4:%.*]] = load i64, i64* [[F_CASTED]], align 8
// CHECK2-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK2-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK2-NEXT:    call void @__omp_outlined__6(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x [10 x i32]]* [[TMP0]], i64 [[TMP4]]) #[[ATTR2]]
// CHECK2-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
// CHECK2-NEXT:    ret void
// CHECK2:       worker.exit:
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@__omp_outlined__6
// CHECK2-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i64 [[F:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK2-NEXT:    [[F_ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[F_CASTED:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 8
// CHECK2-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK2-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK2-NEXT:    store i64 [[F]], i64* [[F_ADDR]], align 8
// CHECK2-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK2-NEXT:    [[CONV:%.*]] = bitcast i64* [[F_ADDR]] to i32*
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    store i32 99, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK2-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK2-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK2-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK2-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 99
// CHECK2-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK2:       cond.true:
// CHECK2-NEXT:    br label [[COND_END:%.*]]
// CHECK2:       cond.false:
// CHECK2-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    br label [[COND_END]]
// CHECK2:       cond.end:
// CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK2-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK2:       omp.inner.for.cond:
// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[TMP6]], 100
// CHECK2-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK2:       omp.inner.for.body:
// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP9]] to i64
// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK2-NEXT:    [[CONV3:%.*]] = bitcast i64* [[F_CASTED]] to i32*
// CHECK2-NEXT:    store i32 [[TMP11]], i32* [[CONV3]], align 4
// CHECK2-NEXT:    [[TMP12:%.*]] = load i64, i64* [[F_CASTED]], align 8
// CHECK2-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK2-NEXT:    [[TMP14:%.*]] = inttoptr i64 [[TMP8]] to i8*
// CHECK2-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 8
// CHECK2-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK2-NEXT:    [[TMP16:%.*]] = inttoptr i64 [[TMP10]] to i8*
// CHECK2-NEXT:    store i8* [[TMP16]], i8** [[TMP15]], align 8
// CHECK2-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK2-NEXT:    [[TMP18:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK2-NEXT:    store i8* [[TMP18]], i8** [[TMP17]], align 8
// CHECK2-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK2-NEXT:    [[TMP20:%.*]] = inttoptr i64 [[TMP12]] to i8*
// CHECK2-NEXT:    store i8* [[TMP20]], i8** [[TMP19]], align 8
// CHECK2-NEXT:    [[TMP21:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK2-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, [10 x [10 x i32]]*, i64)* @__omp_outlined__7 to i8*), i8* null, i8** [[TMP21]], i64 4)
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK2:       omp.inner.for.inc:
// CHECK2-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK2-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK2-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP26]], [[TMP27]]
// CHECK2-NEXT:    store i32 [[ADD5]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[CMP6:%.*]] = icmp sgt i32 [[TMP28]], 99
// CHECK2-NEXT:    br i1 [[CMP6]], label [[COND_TRUE7:%.*]], label [[COND_FALSE8:%.*]]
// CHECK2:       cond.true7:
// CHECK2-NEXT:    br label [[COND_END9:%.*]]
// CHECK2:       cond.false8:
// CHECK2-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    br label [[COND_END9]]
// CHECK2:       cond.end9:
// CHECK2-NEXT:    [[COND10:%.*]] = phi i32 [ 99, [[COND_TRUE7]] ], [ [[TMP29]], [[COND_FALSE8]] ]
// CHECK2-NEXT:    store i32 [[COND10]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    store i32 [[TMP30]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK2:       omp.inner.for.end:
// CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK2:       omp.loop.exit:
// CHECK2-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]])
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@__omp_outlined__7
// CHECK2-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i64 [[F:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK2-NEXT:    [[F_ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK2-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK2-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK2-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK2-NEXT:    store i64 [[F]], i64* [[F_ADDR]], align 8
// CHECK2-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK2-NEXT:    [[CONV:%.*]] = bitcast i64* [[F_ADDR]] to i32*
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    store i32 99, i32* [[DOTOMP_UB]], align 4
// CHECK2-NEXT:    [[TMP1:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK2-NEXT:    [[CONV2:%.*]] = trunc i64 [[TMP1]] to i32
// CHECK2-NEXT:    [[TMP2:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK2-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP2]] to i32
// CHECK2-NEXT:    store i32 [[CONV2]], i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    store i32 [[CONV3]], i32* [[DOTOMP_UB]], align 4
// CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK2-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK2-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB3]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK2:       omp.inner.for.cond:
// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[CONV4:%.*]] = sext i32 [[TMP6]] to i64
// CHECK2-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK2-NEXT:    [[CMP:%.*]] = icmp ule i64 [[CONV4]], [[TMP7]]
// CHECK2-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK2:       omp.inner.for.body:
// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP8]], 10
// CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK2-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[DIV5:%.*]] = sdiv i32 [[TMP10]], 10
// CHECK2-NEXT:    [[MUL6:%.*]] = mul nsw i32 [[DIV5]], 10
// CHECK2-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP9]], [[MUL6]]
// CHECK2-NEXT:    [[MUL7:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK2-NEXT:    [[ADD8:%.*]] = add nsw i32 0, [[MUL7]]
// CHECK2-NEXT:    store i32 [[ADD8]], i32* [[J]], align 4
// CHECK2-NEXT:    store i32 10, i32* [[K]], align 4
// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[J]], align 4
// CHECK2-NEXT:    [[TMP13:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK2-NEXT:    [[MUL9:%.*]] = mul nsw i32 [[TMP12]], [[TMP13]]
// CHECK2-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP11]], [[MUL9]]
// CHECK2-NEXT:    [[TMP14:%.*]] = load i32, i32* [[K]], align 4
// CHECK2-NEXT:    [[ADD11:%.*]] = add nsw i32 [[ADD10]], [[TMP14]]
// CHECK2-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
// CHECK2-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP15]] to i64
// CHECK2-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK2-NEXT:    [[TMP16:%.*]] = load i32, i32* [[J]], align 4
// CHECK2-NEXT:    [[IDXPROM12:%.*]] = sext i32 [[TMP16]] to i64
// CHECK2-NEXT:    [[ARRAYIDX13:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i64 0, i64 [[IDXPROM12]]
// CHECK2-NEXT:    store i32 [[ADD11]], i32* [[ARRAYIDX13]], align 4
// CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK2:       omp.body.continue:
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK2:       omp.inner.for.inc:
// CHECK2-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    [[ADD14:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK2-NEXT:    store i32 [[ADD14]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK2:       omp.inner.for.end:
// CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK2:       omp.loop.exit:
// CHECK2-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]])
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l52
// CHECK2-SAME: (i64 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK2-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK2-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK2-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK2-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false, i1 false)
// CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK2:       user_code.entry:
// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB4]])
// CHECK2-NEXT:    [[TMP3:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK2-NEXT:    [[CONV1:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK2-NEXT:    store i32 [[TMP3]], i32* [[CONV1]], align 4
// CHECK2-NEXT:    [[TMP4:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK2-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK2-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK2-NEXT:    call void @__omp_outlined__8(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP4]], [10 x [10 x i32]]* [[TMP0]]) #[[ATTR2]]
// CHECK2-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
// CHECK2-NEXT:    ret void
// CHECK2:       worker.exit:
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@__omp_outlined__8
// CHECK2-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I8:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[J9:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 8
// CHECK2-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK2-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK2-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK2-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK2-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK2-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK2-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK2-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK2-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK2-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK2-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK2-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK2-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], [[DIV5]]
// CHECK2-NEXT:    [[SUB6:%.*]] = sub nsw i32 [[MUL]], 1
// CHECK2-NEXT:    store i32 [[SUB6]], i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK2-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK2:       land.lhs.true:
// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK2-NEXT:    [[CMP7:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK2-NEXT:    br i1 [[CMP7]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK2:       omp.precond.then:
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK2-NEXT:    store i32 [[TMP7]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK2-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK2-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[TMP8]], align 4
// CHECK2-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP9]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK2-NEXT:    [[CMP10:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
// CHECK2-NEXT:    br i1 [[CMP10]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK2:       cond.true:
// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK2-NEXT:    br label [[COND_END:%.*]]
// CHECK2:       cond.false:
// CHECK2-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    br label [[COND_END]]
// CHECK2:       cond.end:
// CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
// CHECK2-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    store i32 [[TMP14]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK2:       omp.inner.for.cond:
// CHECK2-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP16]], 1
// CHECK2-NEXT:    [[CMP11:%.*]] = icmp slt i32 [[TMP15]], [[ADD]]
// CHECK2-NEXT:    br i1 [[CMP11]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK2:       omp.inner.for.body:
// CHECK2-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    [[TMP18:%.*]] = zext i32 [[TMP17]] to i64
// CHECK2-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
// CHECK2-NEXT:    [[TMP21:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK2-NEXT:    [[CONV12:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK2-NEXT:    store i32 [[TMP21]], i32* [[CONV12]], align 4
// CHECK2-NEXT:    [[TMP22:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK2-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK2-NEXT:    [[TMP24:%.*]] = inttoptr i64 [[TMP18]] to i8*
// CHECK2-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 8
// CHECK2-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK2-NEXT:    [[TMP26:%.*]] = inttoptr i64 [[TMP20]] to i8*
// CHECK2-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 8
// CHECK2-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK2-NEXT:    [[TMP28:%.*]] = inttoptr i64 [[TMP22]] to i8*
// CHECK2-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 8
// CHECK2-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK2-NEXT:    [[TMP30:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK2-NEXT:    store i8* [[TMP30]], i8** [[TMP29]], align 8
// CHECK2-NEXT:    [[TMP31:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP32:%.*]] = load i32, i32* [[TMP31]], align 4
// CHECK2-NEXT:    [[TMP33:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK2-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB4]], i32 [[TMP32]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, i64, [10 x [10 x i32]]*)* @__omp_outlined__9 to i8*), i8* null, i8** [[TMP33]], i64 4)
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK2:       omp.inner.for.inc:
// CHECK2-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    [[ADD13:%.*]] = add nsw i32 [[TMP34]], [[TMP35]]
// CHECK2-NEXT:    store i32 [[ADD13]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    [[ADD14:%.*]] = add nsw i32 [[TMP36]], [[TMP37]]
// CHECK2-NEXT:    store i32 [[ADD14]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    [[ADD15:%.*]] = add nsw i32 [[TMP38]], [[TMP39]]
// CHECK2-NEXT:    store i32 [[ADD15]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK2-NEXT:    [[CMP16:%.*]] = icmp sgt i32 [[TMP40]], [[TMP41]]
// CHECK2-NEXT:    br i1 [[CMP16]], label [[COND_TRUE17:%.*]], label [[COND_FALSE18:%.*]]
// CHECK2:       cond.true17:
// CHECK2-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK2-NEXT:    br label [[COND_END19:%.*]]
// CHECK2:       cond.false18:
// CHECK2-NEXT:    [[TMP43:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    br label [[COND_END19]]
// CHECK2:       cond.end19:
// CHECK2-NEXT:    [[COND20:%.*]] = phi i32 [ [[TMP42]], [[COND_TRUE17]] ], [ [[TMP43]], [[COND_FALSE18]] ]
// CHECK2-NEXT:    store i32 [[COND20]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP44:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    store i32 [[TMP44]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK2:       omp.inner.for.end:
// CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK2:       omp.loop.exit:
// CHECK2-NEXT:    [[TMP45:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP46:%.*]] = load i32, i32* [[TMP45]], align 4
// CHECK2-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP46]])
// CHECK2-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK2:       omp.precond.end:
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@__omp_outlined__9
// CHECK2-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], i64 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 8
// CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I10:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[J11:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK2-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK2-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK2-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK2-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK2-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK2-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 8
// CHECK2-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK2-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[TMP2:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK2-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK2-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK2-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK2-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK2-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK2-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], [[DIV5]]
// CHECK2-NEXT:    [[SUB6:%.*]] = sub nsw i32 [[MUL]], 1
// CHECK2-NEXT:    store i32 [[SUB6]], i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK2-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK2:       land.lhs.true:
// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK2-NEXT:    [[CMP7:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK2-NEXT:    br i1 [[CMP7]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK2:       omp.precond.then:
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_3]], align 4
// CHECK2-NEXT:    store i32 [[TMP7]], i32* [[DOTOMP_UB]], align 4
// CHECK2-NEXT:    [[TMP8:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK2-NEXT:    [[CONV8:%.*]] = trunc i64 [[TMP8]] to i32
// CHECK2-NEXT:    [[TMP9:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK2-NEXT:    [[CONV9:%.*]] = trunc i64 [[TMP9]] to i32
// CHECK2-NEXT:    store i32 [[CONV8]], i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    store i32 [[CONV9]], i32* [[DOTOMP_UB]], align 4
// CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK2-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK2-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB3]], i32 [[TMP11]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    store i32 [[TMP12]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK2:       omp.inner.for.cond:
// CHECK2-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[CONV12:%.*]] = sext i32 [[TMP13]] to i64
// CHECK2-NEXT:    [[TMP14:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK2-NEXT:    [[CMP13:%.*]] = icmp ule i64 [[CONV12]], [[TMP14]]
// CHECK2-NEXT:    br i1 [[CMP13]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK2:       omp.inner.for.body:
// CHECK2-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK2-NEXT:    [[SUB14:%.*]] = sub nsw i32 [[TMP16]], 0
// CHECK2-NEXT:    [[DIV15:%.*]] = sdiv i32 [[SUB14]], 1
// CHECK2-NEXT:    [[MUL16:%.*]] = mul nsw i32 1, [[DIV15]]
// CHECK2-NEXT:    [[DIV17:%.*]] = sdiv i32 [[TMP15]], [[MUL16]]
// CHECK2-NEXT:    [[MUL18:%.*]] = mul nsw i32 [[DIV17]], 1
// CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL18]]
// CHECK2-NEXT:    store i32 [[ADD]], i32* [[I10]], align 4
// CHECK2-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK2-NEXT:    [[SUB19:%.*]] = sub nsw i32 [[TMP19]], 0
// CHECK2-NEXT:    [[DIV20:%.*]] = sdiv i32 [[SUB19]], 1
// CHECK2-NEXT:    [[MUL21:%.*]] = mul nsw i32 1, [[DIV20]]
// CHECK2-NEXT:    [[DIV22:%.*]] = sdiv i32 [[TMP18]], [[MUL21]]
// CHECK2-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK2-NEXT:    [[SUB23:%.*]] = sub nsw i32 [[TMP20]], 0
// CHECK2-NEXT:    [[DIV24:%.*]] = sdiv i32 [[SUB23]], 1
// CHECK2-NEXT:    [[MUL25:%.*]] = mul nsw i32 1, [[DIV24]]
// CHECK2-NEXT:    [[MUL26:%.*]] = mul nsw i32 [[DIV22]], [[MUL25]]
// CHECK2-NEXT:    [[SUB27:%.*]] = sub nsw i32 [[TMP17]], [[MUL26]]
// CHECK2-NEXT:    [[MUL28:%.*]] = mul nsw i32 [[SUB27]], 1
// CHECK2-NEXT:    [[ADD29:%.*]] = add nsw i32 0, [[MUL28]]
// CHECK2-NEXT:    store i32 [[ADD29]], i32* [[J11]], align 4
// CHECK2-NEXT:    [[TMP21:%.*]] = load i32, i32* [[I10]], align 4
// CHECK2-NEXT:    [[TMP22:%.*]] = load i32, i32* [[J11]], align 4
// CHECK2-NEXT:    [[ADD30:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
// CHECK2-NEXT:    [[TMP23:%.*]] = load i32, i32* [[I10]], align 4
// CHECK2-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP23]] to i64
// CHECK2-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i64 0, i64 [[IDXPROM]]
// CHECK2-NEXT:    [[TMP24:%.*]] = load i32, i32* [[J11]], align 4
// CHECK2-NEXT:    [[IDXPROM31:%.*]] = sext i32 [[TMP24]] to i64
// CHECK2-NEXT:    [[ARRAYIDX32:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i64 0, i64 [[IDXPROM31]]
// CHECK2-NEXT:    store i32 [[ADD30]], i32* [[ARRAYIDX32]], align 4
// CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK2:       omp.body.continue:
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK2:       omp.inner.for.inc:
// CHECK2-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    [[ADD33:%.*]] = add nsw i32 [[TMP25]], [[TMP26]]
// CHECK2-NEXT:    store i32 [[ADD33]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK2:       omp.inner.for.end:
// CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK2:       omp.loop.exit:
// CHECK2-NEXT:    [[TMP27:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP28:%.*]] = load i32, i32* [[TMP27]], align 4
// CHECK2-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP28]])
// CHECK2-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK2:       omp.precond.end:
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l59
// CHECK2-SAME: (i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK2-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK2-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK2-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 8
// CHECK2-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK2-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false, i1 false)
// CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK2:       user_code.entry:
// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB4]])
// CHECK2-NEXT:    [[TMP3:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK2-NEXT:    [[CONV1:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK2-NEXT:    store i32 [[TMP3]], i32* [[CONV1]], align 4
// CHECK2-NEXT:    [[TMP4:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK2-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[V_ADDR]], align 8
// CHECK2-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK2-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK2-NEXT:    call void @__omp_outlined__10(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP4]], [1000 x i32]* [[TMP0]], i32* [[TMP5]]) #[[ATTR2]]
// CHECK2-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
// CHECK2-NEXT:    ret void
// CHECK2:       worker.exit:
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@__omp_outlined__10
// CHECK2-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK2-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 8
// CHECK2-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK2-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK2-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK2-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 8
// CHECK2-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK2-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK2-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK2-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK2-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK2-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK2-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK2-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK2-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK2:       omp.precond.then:
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK2-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK2-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK2-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK2-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK2-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK2-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK2:       cond.true:
// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK2-NEXT:    br label [[COND_END:%.*]]
// CHECK2:       cond.false:
// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    br label [[COND_END]]
// CHECK2:       cond.end:
// CHECK2-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK2-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK2:       omp.inner.for.cond:
// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK2-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK2-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK2:       omp.inner.for.body:
// CHECK2-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    [[TMP15:%.*]] = zext i32 [[TMP14]] to i64
// CHECK2-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP17:%.*]] = zext i32 [[TMP16]] to i64
// CHECK2-NEXT:    [[TMP18:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK2-NEXT:    [[CONV6:%.*]] = bitcast i64* [[N_CASTED]] to i32*
// CHECK2-NEXT:    store i32 [[TMP18]], i32* [[CONV6]], align 4
// CHECK2-NEXT:    [[TMP19:%.*]] = load i64, i64* [[N_CASTED]], align 8
// CHECK2-NEXT:    [[TMP20:%.*]] = load i32*, i32** [[V_ADDR]], align 8
// CHECK2-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK2-NEXT:    [[TMP22:%.*]] = inttoptr i64 [[TMP15]] to i8*
// CHECK2-NEXT:    store i8* [[TMP22]], i8** [[TMP21]], align 8
// CHECK2-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
// CHECK2-NEXT:    [[TMP24:%.*]] = inttoptr i64 [[TMP17]] to i8*
// CHECK2-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 8
// CHECK2-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
// CHECK2-NEXT:    [[TMP26:%.*]] = inttoptr i64 [[TMP19]] to i8*
// CHECK2-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 8
// CHECK2-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 3
// CHECK2-NEXT:    [[TMP28:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK2-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 8
// CHECK2-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 4
// CHECK2-NEXT:    [[TMP30:%.*]] = bitcast i32* [[TMP20]] to i8*
// CHECK2-NEXT:    store i8* [[TMP30]], i8** [[TMP29]], align 8
// CHECK2-NEXT:    [[TMP31:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP32:%.*]] = load i32, i32* [[TMP31]], align 4
// CHECK2-NEXT:    [[TMP33:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK2-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB4]], i32 [[TMP32]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i64, i64, i64, [1000 x i32]*, i32*)* @__omp_outlined__11 to i8*), i8* null, i8** [[TMP33]], i64 5)
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK2:       omp.inner.for.inc:
// CHECK2-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP34]], [[TMP35]]
// CHECK2-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP36]], [[TMP37]]
// CHECK2-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP38]], [[TMP39]]
// CHECK2-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK2-NEXT:    [[CMP10:%.*]] = icmp sgt i32 [[TMP40]], [[TMP41]]
// CHECK2-NEXT:    br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]]
// CHECK2:       cond.true11:
// CHECK2-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK2-NEXT:    br label [[COND_END13:%.*]]
// CHECK2:       cond.false12:
// CHECK2-NEXT:    [[TMP43:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    br label [[COND_END13]]
// CHECK2:       cond.end13:
// CHECK2-NEXT:    [[COND14:%.*]] = phi i32 [ [[TMP42]], [[COND_TRUE11]] ], [ [[TMP43]], [[COND_FALSE12]] ]
// CHECK2-NEXT:    store i32 [[COND14]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK2-NEXT:    [[TMP44:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK2-NEXT:    store i32 [[TMP44]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK2:       omp.inner.for.end:
// CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK2:       omp.loop.exit:
// CHECK2-NEXT:    [[TMP45:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP46:%.*]] = load i32, i32* [[TMP45]], align 4
// CHECK2-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP46]])
// CHECK2-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK2:       omp.precond.end:
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@__omp_outlined__11
// CHECK2-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i64 [[DOTPREVIOUS_LB_:%.*]], i64 [[DOTPREVIOUS_UB_:%.*]], i64 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 8
// CHECK2-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I5:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
// CHECK2-NEXT:    store i64 [[DOTPREVIOUS_LB_]], i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK2-NEXT:    store i64 [[DOTPREVIOUS_UB_]], i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK2-NEXT:    store i64 [[N]], i64* [[N_ADDR]], align 8
// CHECK2-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 8
// CHECK2-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 8
// CHECK2-NEXT:    [[CONV:%.*]] = bitcast i64* [[N_ADDR]] to i32*
// CHECK2-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 8
// CHECK2-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CONV]], align 8
// CHECK2-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK2-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK2-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK2-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK2-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK2-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK2:       omp.precond.then:
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK2-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK2-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTPREVIOUS_LB__ADDR]], align 8
// CHECK2-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP5]] to i32
// CHECK2-NEXT:    [[TMP6:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK2-NEXT:    [[CONV4:%.*]] = trunc i64 [[TMP6]] to i32
// CHECK2-NEXT:    store i32 [[CONV3]], i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    store i32 [[CONV4]], i32* [[DOTOMP_UB]], align 4
// CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK2-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK2-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB3]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK2:       omp.inner.for.cond:
// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[CONV6:%.*]] = sext i32 [[TMP10]] to i64
// CHECK2-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTPREVIOUS_UB__ADDR]], align 8
// CHECK2-NEXT:    [[CMP7:%.*]] = icmp ule i64 [[CONV6]], [[TMP11]]
// CHECK2-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK2:       omp.inner.for.body:
// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK2-NEXT:    store i32 [[ADD]], i32* [[I5]], align 4
// CHECK2-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[V_ADDR]], align 8
// CHECK2-NEXT:    [[TMP14:%.*]] = load i32, i32* [[I5]], align 4
// CHECK2-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP14]] to i64
// CHECK2-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[TMP13]], i64 [[IDXPROM]]
// CHECK2-NEXT:    [[TMP15:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK2-NEXT:    [[TMP16:%.*]] = load i32, i32* [[I5]], align 4
// CHECK2-NEXT:    [[IDXPROM8:%.*]] = sext i32 [[TMP16]] to i64
// CHECK2-NEXT:    [[ARRAYIDX9:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i64 0, i64 [[IDXPROM8]]
// CHECK2-NEXT:    store i32 [[TMP15]], i32* [[ARRAYIDX9]], align 4
// CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK2:       omp.body.continue:
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK2:       omp.inner.for.inc:
// CHECK2-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK2-NEXT:    store i32 [[ADD10]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK2:       omp.inner.for.end:
// CHECK2-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK2:       omp.loop.exit:
// CHECK2-NEXT:    [[TMP19:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT:    [[TMP20:%.*]] = load i32, i32* [[TMP19]], align 4
// CHECK2-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP20]])
// CHECK2-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK2:       omp.precond.end:
// CHECK2-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l28
// CHECK3-SAME: (i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK3-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[L_CASTED:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK3-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK3-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i8 2, i1 false, i1 false)
// CHECK3-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK3-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK3:       user_code.entry:
// CHECK3-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB4:[0-9]+]])
// CHECK3-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP3]], i32* [[N_CASTED]], align 4
// CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP5]], i32* [[L_CASTED]], align 4
// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[L_CASTED]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK3-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP4]], [1000 x i32]* [[TMP0]], i32 [[TMP6]]) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
// CHECK3-NEXT:    ret void
// CHECK3:       worker.exit:
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@__omp_outlined__
// CHECK3-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK3-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I4:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[L_CASTED:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 4
// CHECK3-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK3-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK3-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK3-NEXT:    [[L1:%.*]] = call i8* @__kmpc_alloc_shared(i32 4)
// CHECK3-NEXT:    [[L_ON_STACK:%.*]] = bitcast i8* [[L1]] to i32*
// CHECK3-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK3-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK3-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK3-NEXT:    store i32 [[SUB3]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK3-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK3:       omp.precond.then:
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK3-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK3-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK3-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 128)
// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK3-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK3-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK3:       cond.true:
// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK3-NEXT:    br label [[COND_END:%.*]]
// CHECK3:       cond.false:
// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    br label [[COND_END]]
// CHECK3:       cond.end:
// CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK3-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK3:       omp.inner.for.cond:
// CHECK3-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK3-NEXT:    [[CMP6:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK3-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK3:       omp.inner.for.body:
// CHECK3-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP16:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP16]], i32* [[N_CASTED]], align 4
// CHECK3-NEXT:    [[TMP17:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK3-NEXT:    [[TMP18:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP18]], i32* [[L_CASTED]], align 4
// CHECK3-NEXT:    [[TMP19:%.*]] = load i32, i32* [[L_CASTED]], align 4
// CHECK3-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK3-NEXT:    [[TMP21:%.*]] = inttoptr i32 [[TMP14]] to i8*
// CHECK3-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 4
// CHECK3-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK3-NEXT:    [[TMP23:%.*]] = inttoptr i32 [[TMP15]] to i8*
// CHECK3-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 4
// CHECK3-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK3-NEXT:    [[TMP25:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK3-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 4
// CHECK3-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK3-NEXT:    [[TMP27:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK3-NEXT:    store i8* [[TMP27]], i8** [[TMP26]], align 4
// CHECK3-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 4
// CHECK3-NEXT:    [[TMP29:%.*]] = inttoptr i32 [[TMP19]] to i8*
// CHECK3-NEXT:    store i8* [[TMP29]], i8** [[TMP28]], align 4
// CHECK3-NEXT:    [[TMP30:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    [[TMP31:%.*]] = load i32, i32* [[TMP30]], align 4
// CHECK3-NEXT:    [[TMP32:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK3-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB4]], i32 [[TMP31]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i32]*, i32)* @__omp_outlined__1 to i8*), i8* null, i8** [[TMP32]], i32 5)
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK3:       omp.inner.for.inc:
// CHECK3-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP33]], [[TMP34]]
// CHECK3-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP35]], [[TMP36]]
// CHECK3-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP37]], [[TMP38]]
// CHECK3-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK3-NEXT:    [[CMP10:%.*]] = icmp sgt i32 [[TMP39]], [[TMP40]]
// CHECK3-NEXT:    br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]]
// CHECK3:       cond.true11:
// CHECK3-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK3-NEXT:    br label [[COND_END13:%.*]]
// CHECK3:       cond.false12:
// CHECK3-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    br label [[COND_END13]]
// CHECK3:       cond.end13:
// CHECK3-NEXT:    [[COND14:%.*]] = phi i32 [ [[TMP41]], [[COND_TRUE11]] ], [ [[TMP42]], [[COND_FALSE12]] ]
// CHECK3-NEXT:    store i32 [[COND14]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP43:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    store i32 [[TMP43]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK3:       omp.inner.for.end:
// CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK3:       omp.loop.exit:
// CHECK3-NEXT:    [[TMP44:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    [[TMP45:%.*]] = load i32, i32* [[TMP44]], align 4
// CHECK3-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP45]])
// CHECK3-NEXT:    [[TMP46:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK3-NEXT:    [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
// CHECK3-NEXT:    br i1 [[TMP47]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK3:       .omp.lastprivate.then:
// CHECK3-NEXT:    [[TMP48:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP48]], i32* [[L_ADDR]], align 4
// CHECK3-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK3:       .omp.lastprivate.done:
// CHECK3-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK3:       omp.precond.end:
// CHECK3-NEXT:    call void @__kmpc_free_shared(i8* [[L1]], i32 4)
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@__omp_outlined__1
// CHECK3-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK3-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK3-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK3-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK3-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK3-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK3-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK3-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK3-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK3-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK3-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK3:       omp.precond.then:
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK3-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK3-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK3-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK3-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB3:[0-9]+]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 32)
// CHECK3-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK3:       omp.dispatch.cond:
// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK3-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
// CHECK3-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK3:       cond.true:
// CHECK3-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK3-NEXT:    br label [[COND_END:%.*]]
// CHECK3:       cond.false:
// CHECK3-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK3-NEXT:    br label [[COND_END]]
// CHECK3:       cond.end:
// CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP11]], [[COND_TRUE]] ], [ [[TMP12]], [[COND_FALSE]] ]
// CHECK3-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
// CHECK3-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK3-NEXT:    store i32 [[TMP13]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK3-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK3-NEXT:    br i1 [[CMP5]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK3:       omp.dispatch.body:
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK3:       omp.inner.for.cond:
// CHECK3-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK3-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP16]], [[TMP17]]
// CHECK3-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK3:       omp.inner.for.body:
// CHECK3-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP18]], 1
// CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK3-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK3-NEXT:    [[TMP19:%.*]] = load i32, i32* [[I3]], align 4
// CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i32 0, i32 [[TMP19]]
// CHECK3-NEXT:    store i32 1, i32* [[ARRAYIDX]], align 4
// CHECK3-NEXT:    [[TMP20:%.*]] = load i32, i32* [[I3]], align 4
// CHECK3-NEXT:    store i32 [[TMP20]], i32* [[L_ADDR]], align 4
// CHECK3-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK3:       omp.body.continue:
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK3:       omp.inner.for.inc:
// CHECK3-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP21]], 1
// CHECK3-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK3:       omp.inner.for.end:
// CHECK3-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK3:       omp.dispatch.inc:
// CHECK3-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK3-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK3-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_LB]], align 4
// CHECK3-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK3-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK3-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_UB]], align 4
// CHECK3-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK3:       omp.dispatch.end:
// CHECK3-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK3-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP27]])
// CHECK3-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK3-NEXT:    [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
// CHECK3-NEXT:    br i1 [[TMP29]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK3:       .omp.lastprivate.then:
// CHECK3-NEXT:    [[TMP30:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP30]], i32* [[L_ADDR]], align 4
// CHECK3-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK3:       .omp.lastprivate.done:
// CHECK3-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK3:       omp.precond.end:
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34
// CHECK3-SAME: (i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR4:[0-9]+]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK3-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK3-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK3-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false, i1 false)
// CHECK3-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK3-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK3:       user_code.entry:
// CHECK3-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB4]])
// CHECK3-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP3]], i32* [[N_CASTED]], align 4
// CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK3-NEXT:    call void @__omp_outlined__2(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP4]], [1000 x i16]* [[TMP0]]) #[[ATTR2]]
// CHECK3-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
// CHECK3-NEXT:    ret void
// CHECK3:       worker.exit:
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@__omp_outlined__2
// CHECK3-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK3-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK3-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK3-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK3-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK3-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK3-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK3-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK3-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK3:       omp.precond.then:
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK3-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK3-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK3-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK3-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK3:       cond.true:
// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT:    br label [[COND_END:%.*]]
// CHECK3:       cond.false:
// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    br label [[COND_END]]
// CHECK3:       cond.end:
// CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK3-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK3:       omp.inner.for.cond:
// CHECK3-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK3-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK3-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK3:       omp.inner.for.body:
// CHECK3-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP16:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP16]], i32* [[N_CASTED]], align 4
// CHECK3-NEXT:    [[TMP17:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK3-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK3-NEXT:    [[TMP19:%.*]] = inttoptr i32 [[TMP14]] to i8*
// CHECK3-NEXT:    store i8* [[TMP19]], i8** [[TMP18]], align 4
// CHECK3-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK3-NEXT:    [[TMP21:%.*]] = inttoptr i32 [[TMP15]] to i8*
// CHECK3-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 4
// CHECK3-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK3-NEXT:    [[TMP23:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK3-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 4
// CHECK3-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK3-NEXT:    [[TMP25:%.*]] = bitcast [1000 x i16]* [[TMP0]] to i8*
// CHECK3-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 4
// CHECK3-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK3-NEXT:    [[TMP28:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK3-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB4]], i32 [[TMP27]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i16]*)* @__omp_outlined__3 to i8*), i8* null, i8** [[TMP28]], i32 4)
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK3:       omp.inner.for.inc:
// CHECK3-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP29]], [[TMP30]]
// CHECK3-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP31]], [[TMP32]]
// CHECK3-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP33]], [[TMP34]]
// CHECK3-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP35]], [[TMP36]]
// CHECK3-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
// CHECK3:       cond.true10:
// CHECK3-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT:    br label [[COND_END12:%.*]]
// CHECK3:       cond.false11:
// CHECK3-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    br label [[COND_END12]]
// CHECK3:       cond.end12:
// CHECK3-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP37]], [[COND_TRUE10]] ], [ [[TMP38]], [[COND_FALSE11]] ]
// CHECK3-NEXT:    store i32 [[COND13]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    store i32 [[TMP39]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK3:       omp.inner.for.end:
// CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK3:       omp.loop.exit:
// CHECK3-NEXT:    [[TMP40:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    [[TMP41:%.*]] = load i32, i32* [[TMP40]], align 4
// CHECK3-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP41]])
// CHECK3-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK3:       omp.precond.end:
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@__omp_outlined__3
// CHECK3-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK3-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK3-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK3-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK3-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK3-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK3-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK3-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK3-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK3-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK3:       omp.precond.then:
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK3-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK3-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK3-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK3-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB3]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK3-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK3:       omp.inner.for.cond:
// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK3-NEXT:    [[CMP4:%.*]] = icmp ule i32 [[TMP10]], [[TMP11]]
// CHECK3-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK3:       omp.inner.for.body:
// CHECK3-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK3-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK3-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I3]], align 4
// CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i16], [1000 x i16]* [[TMP0]], i32 0, i32 [[TMP13]]
// CHECK3-NEXT:    [[TMP14:%.*]] = load i16, i16* [[ARRAYIDX]], align 2
// CHECK3-NEXT:    [[CONV:%.*]] = sext i16 [[TMP14]] to i32
// CHECK3-NEXT:    [[ADD5:%.*]] = add nsw i32 [[CONV]], 1
// CHECK3-NEXT:    [[CONV6:%.*]] = trunc i32 [[ADD5]] to i16
// CHECK3-NEXT:    store i16 [[CONV6]], i16* [[ARRAYIDX]], align 2
// CHECK3-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK3:       omp.body.continue:
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK3:       omp.inner.for.inc:
// CHECK3-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP15]], [[TMP16]]
// CHECK3-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK3:       omp.inner.for.end:
// CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK3:       omp.loop.exit:
// CHECK3-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
// CHECK3-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP18]])
// CHECK3-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK3:       omp.precond.end:
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l39
// CHECK3-SAME: ([10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK3-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK3-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK3-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false, i1 false)
// CHECK3-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK3-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK3:       user_code.entry:
// CHECK3-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB4]])
// CHECK3-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK3-NEXT:    call void @__omp_outlined__4(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x i32]* [[TMP0]]) #[[ATTR2]]
// CHECK3-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
// CHECK3-NEXT:    ret void
// CHECK3:       worker.exit:
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@__omp_outlined__4
// CHECK3-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x i8*], align 4
// CHECK3-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK3-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK3-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    store i32 9, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK3-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK3-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK3-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK3-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9
// CHECK3-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK3:       cond.true:
// CHECK3-NEXT:    br label [[COND_END:%.*]]
// CHECK3:       cond.false:
// CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    br label [[COND_END]]
// CHECK3:       cond.end:
// CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK3-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK3:       omp.inner.for.cond:
// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[CMP1:%.*]] = icmp slt i32 [[TMP6]], 10
// CHECK3-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK3:       omp.inner.for.body:
// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK3-NEXT:    [[TMP10:%.*]] = inttoptr i32 [[TMP7]] to i8*
// CHECK3-NEXT:    store i8* [[TMP10]], i8** [[TMP9]], align 4
// CHECK3-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK3-NEXT:    [[TMP12:%.*]] = inttoptr i32 [[TMP8]] to i8*
// CHECK3-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 4
// CHECK3-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK3-NEXT:    [[TMP14:%.*]] = bitcast [10 x i32]* [[TMP0]] to i8*
// CHECK3-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 4
// CHECK3-NEXT:    [[TMP15:%.*]] = bitcast [3 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK3-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, [10 x i32]*)* @__omp_outlined__5 to i8*), i8* null, i8** [[TMP15]], i32 3)
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK3:       omp.inner.for.inc:
// CHECK3-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP16]], [[TMP17]]
// CHECK3-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP18]], [[TMP19]]
// CHECK3-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK3-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP22]], 9
// CHECK3-NEXT:    br i1 [[CMP4]], label [[COND_TRUE5:%.*]], label [[COND_FALSE6:%.*]]
// CHECK3:       cond.true5:
// CHECK3-NEXT:    br label [[COND_END7:%.*]]
// CHECK3:       cond.false6:
// CHECK3-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    br label [[COND_END7]]
// CHECK3:       cond.end7:
// CHECK3-NEXT:    [[COND8:%.*]] = phi i32 [ 9, [[COND_TRUE5]] ], [ [[TMP23]], [[COND_FALSE6]] ]
// CHECK3-NEXT:    store i32 [[COND8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    store i32 [[TMP24]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK3:       omp.inner.for.end:
// CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK3:       omp.loop.exit:
// CHECK3-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]])
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@__omp_outlined__5
// CHECK3-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK3-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK3-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK3-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK3-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK3-NEXT:    store i32 9, i32* [[DOTOMP_UB]], align 4
// CHECK3-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK3-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP1]], i32* [[DOTOMP_LB]], align 4
// CHECK3-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_UB]], align 4
// CHECK3-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK3-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK3-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB3]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK3-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK3:       omp.inner.for.cond:
// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK3-NEXT:    [[CMP:%.*]] = icmp ule i32 [[TMP6]], [[TMP7]]
// CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK3:       omp.inner.for.body:
// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP8]], 1
// CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK3-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4
// CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[TMP0]], i32 0, i32 [[TMP9]]
// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK3-NEXT:    [[ADD1:%.*]] = add nsw i32 [[TMP10]], 1
// CHECK3-NEXT:    store i32 [[ADD1]], i32* [[ARRAYIDX]], align 4
// CHECK3-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK3:       omp.body.continue:
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK3:       omp.inner.for.inc:
// CHECK3-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
// CHECK3-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK3:       omp.inner.for.end:
// CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK3:       omp.loop.exit:
// CHECK3-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]])
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l44
// CHECK3-SAME: ([10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK3-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[F_CASTED:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK3-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK3-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false, i1 false)
// CHECK3-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK3-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK3:       user_code.entry:
// CHECK3-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB4]])
// CHECK3-NEXT:    [[TMP3:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP3]], i32* [[F_CASTED]], align 4
// CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[F_CASTED]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK3-NEXT:    call void @__omp_outlined__6(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x [10 x i32]]* [[TMP0]], i32 [[TMP4]]) #[[ATTR2]]
// CHECK3-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
// CHECK3-NEXT:    ret void
// CHECK3:       worker.exit:
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@__omp_outlined__6
// CHECK3-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK3-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[F_CASTED:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK3-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK3-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK3-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    store i32 99, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK3-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK3-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK3-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK3-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 99
// CHECK3-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK3:       cond.true:
// CHECK3-NEXT:    br label [[COND_END:%.*]]
// CHECK3:       cond.false:
// CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    br label [[COND_END]]
// CHECK3:       cond.end:
// CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK3-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK3:       omp.inner.for.cond:
// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[TMP6]], 100
// CHECK3-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK3:       omp.inner.for.body:
// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP9]], i32* [[F_CASTED]], align 4
// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[F_CASTED]], align 4
// CHECK3-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK3-NEXT:    [[TMP12:%.*]] = inttoptr i32 [[TMP7]] to i8*
// CHECK3-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 4
// CHECK3-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK3-NEXT:    [[TMP14:%.*]] = inttoptr i32 [[TMP8]] to i8*
// CHECK3-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 4
// CHECK3-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK3-NEXT:    [[TMP16:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK3-NEXT:    store i8* [[TMP16]], i8** [[TMP15]], align 4
// CHECK3-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK3-NEXT:    [[TMP18:%.*]] = inttoptr i32 [[TMP10]] to i8*
// CHECK3-NEXT:    store i8* [[TMP18]], i8** [[TMP17]], align 4
// CHECK3-NEXT:    [[TMP19:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK3-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, [10 x [10 x i32]]*, i32)* @__omp_outlined__7 to i8*), i8* null, i8** [[TMP19]], i32 4)
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK3:       omp.inner.for.inc:
// CHECK3-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK3-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK3-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK3-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP26]], 99
// CHECK3-NEXT:    br i1 [[CMP5]], label [[COND_TRUE6:%.*]], label [[COND_FALSE7:%.*]]
// CHECK3:       cond.true6:
// CHECK3-NEXT:    br label [[COND_END8:%.*]]
// CHECK3:       cond.false7:
// CHECK3-NEXT:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    br label [[COND_END8]]
// CHECK3:       cond.end8:
// CHECK3-NEXT:    [[COND9:%.*]] = phi i32 [ 99, [[COND_TRUE6]] ], [ [[TMP27]], [[COND_FALSE7]] ]
// CHECK3-NEXT:    store i32 [[COND9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    store i32 [[TMP28]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK3:       omp.inner.for.end:
// CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK3:       omp.loop.exit:
// CHECK3-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]])
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@__omp_outlined__7
// CHECK3-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK3-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK3-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK3-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK3-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK3-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK3-NEXT:    store i32 99, i32* [[DOTOMP_UB]], align 4
// CHECK3-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK3-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP1]], i32* [[DOTOMP_LB]], align 4
// CHECK3-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_UB]], align 4
// CHECK3-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK3-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK3-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB3]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK3-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK3:       omp.inner.for.cond:
// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK3-NEXT:    [[CMP:%.*]] = icmp ule i32 [[TMP6]], [[TMP7]]
// CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK3:       omp.inner.for.body:
// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP8]], 10
// CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK3-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[DIV2:%.*]] = sdiv i32 [[TMP10]], 10
// CHECK3-NEXT:    [[MUL3:%.*]] = mul nsw i32 [[DIV2]], 10
// CHECK3-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP9]], [[MUL3]]
// CHECK3-NEXT:    [[MUL4:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK3-NEXT:    [[ADD5:%.*]] = add nsw i32 0, [[MUL4]]
// CHECK3-NEXT:    store i32 [[ADD5]], i32* [[J]], align 4
// CHECK3-NEXT:    store i32 10, i32* [[K]], align 4
// CHECK3-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
// CHECK3-NEXT:    [[TMP12:%.*]] = load i32, i32* [[J]], align 4
// CHECK3-NEXT:    [[TMP13:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK3-NEXT:    [[MUL6:%.*]] = mul nsw i32 [[TMP12]], [[TMP13]]
// CHECK3-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP11]], [[MUL6]]
// CHECK3-NEXT:    [[TMP14:%.*]] = load i32, i32* [[K]], align 4
// CHECK3-NEXT:    [[ADD8:%.*]] = add nsw i32 [[ADD7]], [[TMP14]]
// CHECK3-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
// CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i32 0, i32 [[TMP15]]
// CHECK3-NEXT:    [[TMP16:%.*]] = load i32, i32* [[J]], align 4
// CHECK3-NEXT:    [[ARRAYIDX9:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i32 0, i32 [[TMP16]]
// CHECK3-NEXT:    store i32 [[ADD8]], i32* [[ARRAYIDX9]], align 4
// CHECK3-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK3:       omp.body.continue:
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK3:       omp.inner.for.inc:
// CHECK3-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK3-NEXT:    store i32 [[ADD10]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK3:       omp.inner.for.end:
// CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK3:       omp.loop.exit:
// CHECK3-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]])
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l52
// CHECK3-SAME: (i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK3-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK3-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK3-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false, i1 false)
// CHECK3-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK3-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK3:       user_code.entry:
// CHECK3-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB4]])
// CHECK3-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP3]], i32* [[N_CASTED]], align 4
// CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK3-NEXT:    call void @__omp_outlined__8(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP4]], [10 x [10 x i32]]* [[TMP0]]) #[[ATTR2]]
// CHECK3-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
// CHECK3-NEXT:    ret void
// CHECK3:       worker.exit:
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@__omp_outlined__8
// CHECK3-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK3-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i64, align 8
// CHECK3-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i64, align 8
// CHECK3-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK3-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I9:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[J10:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK3-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK3-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK3-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK3-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK3-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK3-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK3-NEXT:    [[CONV:%.*]] = sext i32 [[DIV]] to i64
// CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK3-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK3-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK3-NEXT:    [[CONV6:%.*]] = sext i32 [[DIV5]] to i64
// CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV]], [[CONV6]]
// CHECK3-NEXT:    [[SUB7:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK3-NEXT:    store i64 [[SUB7]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK3-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK3-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK3:       land.lhs.true:
// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK3-NEXT:    [[CMP8:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK3-NEXT:    br i1 [[CMP8]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK3:       omp.precond.then:
// CHECK3-NEXT:    store i64 0, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK3-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK3-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK3-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK3-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK3-NEXT:    [[CONV11:%.*]] = zext i32 [[NVPTX_NUM_THREADS]] to i64
// CHECK3-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[TMP8]], align 4
// CHECK3-NEXT:    call void @__kmpc_distribute_static_init_8(%struct.ident_t* @[[GLOB2]], i32 [[TMP9]], i32 91, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_COMB_LB]], i64* [[DOTOMP_COMB_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 [[CONV11]])
// CHECK3-NEXT:    [[TMP10:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK3-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK3-NEXT:    [[CMP12:%.*]] = icmp sgt i64 [[TMP10]], [[TMP11]]
// CHECK3-NEXT:    br i1 [[CMP12]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK3:       cond.true:
// CHECK3-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK3-NEXT:    br label [[COND_END:%.*]]
// CHECK3:       cond.false:
// CHECK3-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK3-NEXT:    br label [[COND_END]]
// CHECK3:       cond.end:
// CHECK3-NEXT:    [[COND:%.*]] = phi i64 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
// CHECK3-NEXT:    store i64 [[COND]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK3-NEXT:    [[TMP14:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK3-NEXT:    store i64 [[TMP14]], i64* [[DOTOMP_IV]], align 8
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK3:       omp.inner.for.cond:
// CHECK3-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK3-NEXT:    [[TMP16:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK3-NEXT:    [[ADD:%.*]] = add nsw i64 [[TMP16]], 1
// CHECK3-NEXT:    [[CMP13:%.*]] = icmp slt i64 [[TMP15]], [[ADD]]
// CHECK3-NEXT:    br i1 [[CMP13]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK3:       omp.inner.for.body:
// CHECK3-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK3-NEXT:    [[TMP18:%.*]] = trunc i64 [[TMP17]] to i32
// CHECK3-NEXT:    [[TMP19:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK3-NEXT:    [[TMP20:%.*]] = trunc i64 [[TMP19]] to i32
// CHECK3-NEXT:    [[TMP21:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP21]], i32* [[N_CASTED]], align 4
// CHECK3-NEXT:    [[TMP22:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK3-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK3-NEXT:    [[TMP24:%.*]] = inttoptr i32 [[TMP18]] to i8*
// CHECK3-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 4
// CHECK3-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK3-NEXT:    [[TMP26:%.*]] = inttoptr i32 [[TMP20]] to i8*
// CHECK3-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 4
// CHECK3-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK3-NEXT:    [[TMP28:%.*]] = inttoptr i32 [[TMP22]] to i8*
// CHECK3-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 4
// CHECK3-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK3-NEXT:    [[TMP30:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK3-NEXT:    store i8* [[TMP30]], i8** [[TMP29]], align 4
// CHECK3-NEXT:    [[TMP31:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    [[TMP32:%.*]] = load i32, i32* [[TMP31]], align 4
// CHECK3-NEXT:    [[TMP33:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK3-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB4]], i32 [[TMP32]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [10 x [10 x i32]]*)* @__omp_outlined__9 to i8*), i8* null, i8** [[TMP33]], i32 4)
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK3:       omp.inner.for.inc:
// CHECK3-NEXT:    [[TMP34:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK3-NEXT:    [[TMP35:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK3-NEXT:    [[ADD14:%.*]] = add nsw i64 [[TMP34]], [[TMP35]]
// CHECK3-NEXT:    store i64 [[ADD14]], i64* [[DOTOMP_IV]], align 8
// CHECK3-NEXT:    [[TMP36:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK3-NEXT:    [[TMP37:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK3-NEXT:    [[ADD15:%.*]] = add nsw i64 [[TMP36]], [[TMP37]]
// CHECK3-NEXT:    store i64 [[ADD15]], i64* [[DOTOMP_COMB_LB]], align 8
// CHECK3-NEXT:    [[TMP38:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK3-NEXT:    [[TMP39:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK3-NEXT:    [[ADD16:%.*]] = add nsw i64 [[TMP38]], [[TMP39]]
// CHECK3-NEXT:    store i64 [[ADD16]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK3-NEXT:    [[TMP40:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK3-NEXT:    [[TMP41:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK3-NEXT:    [[CMP17:%.*]] = icmp sgt i64 [[TMP40]], [[TMP41]]
// CHECK3-NEXT:    br i1 [[CMP17]], label [[COND_TRUE18:%.*]], label [[COND_FALSE19:%.*]]
// CHECK3:       cond.true18:
// CHECK3-NEXT:    [[TMP42:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK3-NEXT:    br label [[COND_END20:%.*]]
// CHECK3:       cond.false19:
// CHECK3-NEXT:    [[TMP43:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK3-NEXT:    br label [[COND_END20]]
// CHECK3:       cond.end20:
// CHECK3-NEXT:    [[COND21:%.*]] = phi i64 [ [[TMP42]], [[COND_TRUE18]] ], [ [[TMP43]], [[COND_FALSE19]] ]
// CHECK3-NEXT:    store i64 [[COND21]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK3-NEXT:    [[TMP44:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK3-NEXT:    store i64 [[TMP44]], i64* [[DOTOMP_IV]], align 8
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK3:       omp.inner.for.end:
// CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK3:       omp.loop.exit:
// CHECK3-NEXT:    [[TMP45:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    [[TMP46:%.*]] = load i32, i32* [[TMP45]], align 4
// CHECK3-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP46]])
// CHECK3-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK3:       omp.precond.end:
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@__omp_outlined__9
// CHECK3-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK3-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_LB:%.*]] = alloca i64, align 8
// CHECK3-NEXT:    [[DOTOMP_UB:%.*]] = alloca i64, align 8
// CHECK3-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK3-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I11:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[J12:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK3-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK3-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK3-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK3-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK3-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK3-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK3-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK3-NEXT:    [[CONV:%.*]] = sext i32 [[DIV]] to i64
// CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK3-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK3-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK3-NEXT:    [[CONV6:%.*]] = sext i32 [[DIV5]] to i64
// CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV]], [[CONV6]]
// CHECK3-NEXT:    [[SUB7:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK3-NEXT:    store i64 [[SUB7]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK3-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK3-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK3:       land.lhs.true:
// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK3-NEXT:    [[CMP8:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK3-NEXT:    br i1 [[CMP8]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK3:       omp.precond.then:
// CHECK3-NEXT:    store i64 0, i64* [[DOTOMP_LB]], align 8
// CHECK3-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK3-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_UB]], align 8
// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK3-NEXT:    [[CONV9:%.*]] = zext i32 [[TMP8]] to i64
// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK3-NEXT:    [[CONV10:%.*]] = zext i32 [[TMP9]] to i64
// CHECK3-NEXT:    store i64 [[CONV9]], i64* [[DOTOMP_LB]], align 8
// CHECK3-NEXT:    store i64 [[CONV10]], i64* [[DOTOMP_UB]], align 8
// CHECK3-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK3-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK3-NEXT:    call void @__kmpc_for_static_init_8(%struct.ident_t* @[[GLOB3]], i32 [[TMP11]], i32 33, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_LB]], i64* [[DOTOMP_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 1)
// CHECK3-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTOMP_LB]], align 8
// CHECK3-NEXT:    store i64 [[TMP12]], i64* [[DOTOMP_IV]], align 8
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK3:       omp.inner.for.cond:
// CHECK3-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK3-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK3-NEXT:    [[CONV13:%.*]] = zext i32 [[TMP14]] to i64
// CHECK3-NEXT:    [[CMP14:%.*]] = icmp sle i64 [[TMP13]], [[CONV13]]
// CHECK3-NEXT:    br i1 [[CMP14]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK3:       omp.inner.for.body:
// CHECK3-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK3-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK3-NEXT:    [[SUB15:%.*]] = sub nsw i32 [[TMP16]], 0
// CHECK3-NEXT:    [[DIV16:%.*]] = sdiv i32 [[SUB15]], 1
// CHECK3-NEXT:    [[MUL17:%.*]] = mul nsw i32 1, [[DIV16]]
// CHECK3-NEXT:    [[CONV18:%.*]] = sext i32 [[MUL17]] to i64
// CHECK3-NEXT:    [[DIV19:%.*]] = sdiv i64 [[TMP15]], [[CONV18]]
// CHECK3-NEXT:    [[MUL20:%.*]] = mul nsw i64 [[DIV19]], 1
// CHECK3-NEXT:    [[ADD:%.*]] = add nsw i64 0, [[MUL20]]
// CHECK3-NEXT:    [[CONV21:%.*]] = trunc i64 [[ADD]] to i32
// CHECK3-NEXT:    store i32 [[CONV21]], i32* [[I11]], align 4
// CHECK3-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK3-NEXT:    [[TMP18:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK3-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK3-NEXT:    [[SUB22:%.*]] = sub nsw i32 [[TMP19]], 0
// CHECK3-NEXT:    [[DIV23:%.*]] = sdiv i32 [[SUB22]], 1
// CHECK3-NEXT:    [[MUL24:%.*]] = mul nsw i32 1, [[DIV23]]
// CHECK3-NEXT:    [[CONV25:%.*]] = sext i32 [[MUL24]] to i64
// CHECK3-NEXT:    [[DIV26:%.*]] = sdiv i64 [[TMP18]], [[CONV25]]
// CHECK3-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK3-NEXT:    [[SUB27:%.*]] = sub nsw i32 [[TMP20]], 0
// CHECK3-NEXT:    [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK3-NEXT:    [[MUL29:%.*]] = mul nsw i32 1, [[DIV28]]
// CHECK3-NEXT:    [[CONV30:%.*]] = sext i32 [[MUL29]] to i64
// CHECK3-NEXT:    [[MUL31:%.*]] = mul nsw i64 [[DIV26]], [[CONV30]]
// CHECK3-NEXT:    [[SUB32:%.*]] = sub nsw i64 [[TMP17]], [[MUL31]]
// CHECK3-NEXT:    [[MUL33:%.*]] = mul nsw i64 [[SUB32]], 1
// CHECK3-NEXT:    [[ADD34:%.*]] = add nsw i64 0, [[MUL33]]
// CHECK3-NEXT:    [[CONV35:%.*]] = trunc i64 [[ADD34]] to i32
// CHECK3-NEXT:    store i32 [[CONV35]], i32* [[J12]], align 4
// CHECK3-NEXT:    [[TMP21:%.*]] = load i32, i32* [[I11]], align 4
// CHECK3-NEXT:    [[TMP22:%.*]] = load i32, i32* [[J12]], align 4
// CHECK3-NEXT:    [[ADD36:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
// CHECK3-NEXT:    [[TMP23:%.*]] = load i32, i32* [[I11]], align 4
// CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i32 0, i32 [[TMP23]]
// CHECK3-NEXT:    [[TMP24:%.*]] = load i32, i32* [[J12]], align 4
// CHECK3-NEXT:    [[ARRAYIDX37:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i32 0, i32 [[TMP24]]
// CHECK3-NEXT:    store i32 [[ADD36]], i32* [[ARRAYIDX37]], align 4
// CHECK3-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK3:       omp.body.continue:
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK3:       omp.inner.for.inc:
// CHECK3-NEXT:    [[TMP25:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK3-NEXT:    [[TMP26:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK3-NEXT:    [[ADD38:%.*]] = add nsw i64 [[TMP25]], [[TMP26]]
// CHECK3-NEXT:    store i64 [[ADD38]], i64* [[DOTOMP_IV]], align 8
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK3:       omp.inner.for.end:
// CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK3:       omp.loop.exit:
// CHECK3-NEXT:    [[TMP27:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    [[TMP28:%.*]] = load i32, i32* [[TMP27]], align 4
// CHECK3-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP28]])
// CHECK3-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK3:       omp.precond.end:
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l59
// CHECK3-SAME: (i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK3-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK3-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK3-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK3-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false, i1 false)
// CHECK3-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK3-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK3:       user_code.entry:
// CHECK3-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB4]])
// CHECK3-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP3]], i32* [[N_CASTED]], align 4
// CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK3-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK3-NEXT:    call void @__omp_outlined__10(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP4]], [1000 x i32]* [[TMP0]], i32* [[TMP5]]) #[[ATTR2]]
// CHECK3-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
// CHECK3-NEXT:    ret void
// CHECK3:       worker.exit:
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@__omp_outlined__10
// CHECK3-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK3-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 4
// CHECK3-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK3-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK3-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK3-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK3-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK3-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK3-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK3-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK3-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK3:       omp.precond.then:
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK3-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK3-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK3-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK3-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK3:       cond.true:
// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT:    br label [[COND_END:%.*]]
// CHECK3:       cond.false:
// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    br label [[COND_END]]
// CHECK3:       cond.end:
// CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK3-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK3:       omp.inner.for.cond:
// CHECK3-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK3-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK3-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK3:       omp.inner.for.body:
// CHECK3-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP16:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP16]], i32* [[N_CASTED]], align 4
// CHECK3-NEXT:    [[TMP17:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK3-NEXT:    [[TMP18:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK3-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK3-NEXT:    [[TMP20:%.*]] = inttoptr i32 [[TMP14]] to i8*
// CHECK3-NEXT:    store i8* [[TMP20]], i8** [[TMP19]], align 4
// CHECK3-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK3-NEXT:    [[TMP22:%.*]] = inttoptr i32 [[TMP15]] to i8*
// CHECK3-NEXT:    store i8* [[TMP22]], i8** [[TMP21]], align 4
// CHECK3-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK3-NEXT:    [[TMP24:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK3-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 4
// CHECK3-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK3-NEXT:    [[TMP26:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK3-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 4
// CHECK3-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 4
// CHECK3-NEXT:    [[TMP28:%.*]] = bitcast i32* [[TMP18]] to i8*
// CHECK3-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 4
// CHECK3-NEXT:    [[TMP29:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    [[TMP30:%.*]] = load i32, i32* [[TMP29]], align 4
// CHECK3-NEXT:    [[TMP31:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK3-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB4]], i32 [[TMP30]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i32]*, i32*)* @__omp_outlined__11 to i8*), i8* null, i8** [[TMP31]], i32 5)
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK3:       omp.inner.for.inc:
// CHECK3-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP32]], [[TMP33]]
// CHECK3-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP34]], [[TMP35]]
// CHECK3-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP36]], [[TMP37]]
// CHECK3-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP38]], [[TMP39]]
// CHECK3-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
// CHECK3:       cond.true10:
// CHECK3-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT:    br label [[COND_END12:%.*]]
// CHECK3:       cond.false11:
// CHECK3-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    br label [[COND_END12]]
// CHECK3:       cond.end12:
// CHECK3-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP40]], [[COND_TRUE10]] ], [ [[TMP41]], [[COND_FALSE11]] ]
// CHECK3-NEXT:    store i32 [[COND13]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK3-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK3-NEXT:    store i32 [[TMP42]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK3:       omp.inner.for.end:
// CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK3:       omp.loop.exit:
// CHECK3-NEXT:    [[TMP43:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    [[TMP44:%.*]] = load i32, i32* [[TMP43]], align 4
// CHECK3-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP44]])
// CHECK3-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK3:       omp.precond.end:
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@__omp_outlined__11
// CHECK3-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK3-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK3-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK3-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK3-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK3-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK3-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK3-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK3-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK3-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK3-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK3-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK3:       omp.precond.then:
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK3-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK3-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK3-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK3-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB3]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK3-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK3:       omp.inner.for.cond:
// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK3-NEXT:    [[CMP4:%.*]] = icmp ule i32 [[TMP10]], [[TMP11]]
// CHECK3-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK3:       omp.inner.for.body:
// CHECK3-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK3-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK3-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK3-NEXT:    [[TMP14:%.*]] = load i32, i32* [[I3]], align 4
// CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[TMP13]], i32 [[TMP14]]
// CHECK3-NEXT:    [[TMP15:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK3-NEXT:    [[TMP16:%.*]] = load i32, i32* [[I3]], align 4
// CHECK3-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i32 0, i32 [[TMP16]]
// CHECK3-NEXT:    store i32 [[TMP15]], i32* [[ARRAYIDX5]], align 4
// CHECK3-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK3:       omp.body.continue:
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK3:       omp.inner.for.inc:
// CHECK3-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK3-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK3:       omp.inner.for.end:
// CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK3:       omp.loop.exit:
// CHECK3-NEXT:    [[TMP19:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT:    [[TMP20:%.*]] = load i32, i32* [[TMP19]], align 4
// CHECK3-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP20]])
// CHECK3-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK3:       omp.precond.end:
// CHECK3-NEXT:    ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l28
// CHECK4-SAME: (i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK4-NEXT:  entry:
// CHECK4-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK4-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[L_CASTED:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK4-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK4-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i8 2, i1 false, i1 false)
// CHECK4-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK4-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK4:       user_code.entry:
// CHECK4-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB4:[0-9]+]])
// CHECK4-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP3]], i32* [[N_CASTED]], align 4
// CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP5]], i32* [[L_CASTED]], align 4
// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[L_CASTED]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK4-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP4]], [1000 x i32]* [[TMP0]], i32 [[TMP6]]) #[[ATTR2:[0-9]+]]
// CHECK4-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
// CHECK4-NEXT:    ret void
// CHECK4:       worker.exit:
// CHECK4-NEXT:    ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@__omp_outlined__
// CHECK4-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0]] {
// CHECK4-NEXT:  entry:
// CHECK4-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK4-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I4:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[L_CASTED:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 4
// CHECK4-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK4-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK4-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK4-NEXT:    [[L1:%.*]] = call i8* @__kmpc_alloc_shared(i32 4)
// CHECK4-NEXT:    [[L_ON_STACK:%.*]] = bitcast i8* [[L1]] to i32*
// CHECK4-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK4-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK4-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK4-NEXT:    store i32 [[SUB3]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK4-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK4-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK4:       omp.precond.then:
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK4-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK4-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK4-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 128)
// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK4-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK4-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK4:       cond.true:
// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK4-NEXT:    br label [[COND_END:%.*]]
// CHECK4:       cond.false:
// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    br label [[COND_END]]
// CHECK4:       cond.end:
// CHECK4-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK4-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK4:       omp.inner.for.cond:
// CHECK4-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK4-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK4-NEXT:    [[CMP6:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK4-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK4:       omp.inner.for.body:
// CHECK4-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP16:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP16]], i32* [[N_CASTED]], align 4
// CHECK4-NEXT:    [[TMP17:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK4-NEXT:    [[TMP18:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP18]], i32* [[L_CASTED]], align 4
// CHECK4-NEXT:    [[TMP19:%.*]] = load i32, i32* [[L_CASTED]], align 4
// CHECK4-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK4-NEXT:    [[TMP21:%.*]] = inttoptr i32 [[TMP14]] to i8*
// CHECK4-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 4
// CHECK4-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK4-NEXT:    [[TMP23:%.*]] = inttoptr i32 [[TMP15]] to i8*
// CHECK4-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 4
// CHECK4-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK4-NEXT:    [[TMP25:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK4-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 4
// CHECK4-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK4-NEXT:    [[TMP27:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK4-NEXT:    store i8* [[TMP27]], i8** [[TMP26]], align 4
// CHECK4-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 4
// CHECK4-NEXT:    [[TMP29:%.*]] = inttoptr i32 [[TMP19]] to i8*
// CHECK4-NEXT:    store i8* [[TMP29]], i8** [[TMP28]], align 4
// CHECK4-NEXT:    [[TMP30:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    [[TMP31:%.*]] = load i32, i32* [[TMP30]], align 4
// CHECK4-NEXT:    [[TMP32:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK4-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB4]], i32 [[TMP31]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i32]*, i32)* @__omp_outlined__1 to i8*), i8* null, i8** [[TMP32]], i32 5)
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK4:       omp.inner.for.inc:
// CHECK4-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP33]], [[TMP34]]
// CHECK4-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP35]], [[TMP36]]
// CHECK4-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP37]], [[TMP38]]
// CHECK4-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK4-NEXT:    [[CMP10:%.*]] = icmp sgt i32 [[TMP39]], [[TMP40]]
// CHECK4-NEXT:    br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]]
// CHECK4:       cond.true11:
// CHECK4-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK4-NEXT:    br label [[COND_END13:%.*]]
// CHECK4:       cond.false12:
// CHECK4-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    br label [[COND_END13]]
// CHECK4:       cond.end13:
// CHECK4-NEXT:    [[COND14:%.*]] = phi i32 [ [[TMP41]], [[COND_TRUE11]] ], [ [[TMP42]], [[COND_FALSE12]] ]
// CHECK4-NEXT:    store i32 [[COND14]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP43:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    store i32 [[TMP43]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK4:       omp.inner.for.end:
// CHECK4-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK4:       omp.loop.exit:
// CHECK4-NEXT:    [[TMP44:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    [[TMP45:%.*]] = load i32, i32* [[TMP44]], align 4
// CHECK4-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP45]])
// CHECK4-NEXT:    [[TMP46:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK4-NEXT:    [[TMP47:%.*]] = icmp ne i32 [[TMP46]], 0
// CHECK4-NEXT:    br i1 [[TMP47]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK4:       .omp.lastprivate.then:
// CHECK4-NEXT:    [[TMP48:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP48]], i32* [[L_ADDR]], align 4
// CHECK4-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK4:       .omp.lastprivate.done:
// CHECK4-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK4:       omp.precond.end:
// CHECK4-NEXT:    call void @__kmpc_free_shared(i8* [[L1]], i32 4)
// CHECK4-NEXT:    ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@__omp_outlined__1
// CHECK4-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32 [[L:%.*]]) #[[ATTR0]] {
// CHECK4-NEXT:  entry:
// CHECK4-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK4-NEXT:    [[L_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK4-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK4-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK4-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[L]], i32* [[L_ADDR]], align 4
// CHECK4-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK4-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK4-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK4-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK4-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK4-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK4-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK4:       omp.precond.then:
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK4-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK4-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK4-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK4-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK4-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB3:[0-9]+]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 32)
// CHECK4-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK4:       omp.dispatch.cond:
// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK4-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
// CHECK4-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK4:       cond.true:
// CHECK4-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK4-NEXT:    br label [[COND_END:%.*]]
// CHECK4:       cond.false:
// CHECK4-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK4-NEXT:    br label [[COND_END]]
// CHECK4:       cond.end:
// CHECK4-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP11]], [[COND_TRUE]] ], [ [[TMP12]], [[COND_FALSE]] ]
// CHECK4-NEXT:    store i32 [[COND]], i32* [[DOTOMP_UB]], align 4
// CHECK4-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK4-NEXT:    store i32 [[TMP13]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK4-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]]
// CHECK4-NEXT:    br i1 [[CMP5]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK4:       omp.dispatch.body:
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK4:       omp.inner.for.cond:
// CHECK4-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK4-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP16]], [[TMP17]]
// CHECK4-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK4:       omp.inner.for.body:
// CHECK4-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP18]], 1
// CHECK4-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK4-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK4-NEXT:    [[TMP19:%.*]] = load i32, i32* [[I3]], align 4
// CHECK4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i32 0, i32 [[TMP19]]
// CHECK4-NEXT:    store i32 1, i32* [[ARRAYIDX]], align 4
// CHECK4-NEXT:    [[TMP20:%.*]] = load i32, i32* [[I3]], align 4
// CHECK4-NEXT:    store i32 [[TMP20]], i32* [[L_ADDR]], align 4
// CHECK4-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK4:       omp.body.continue:
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK4:       omp.inner.for.inc:
// CHECK4-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP21]], 1
// CHECK4-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK4:       omp.inner.for.end:
// CHECK4-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK4:       omp.dispatch.inc:
// CHECK4-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK4-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK4-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_LB]], align 4
// CHECK4-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK4-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK4-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_UB]], align 4
// CHECK4-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK4:       omp.dispatch.end:
// CHECK4-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK4-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP27]])
// CHECK4-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK4-NEXT:    [[TMP29:%.*]] = icmp ne i32 [[TMP28]], 0
// CHECK4-NEXT:    br i1 [[TMP29]], label [[DOTOMP_LASTPRIVATE_THEN:%.*]], label [[DOTOMP_LASTPRIVATE_DONE:%.*]]
// CHECK4:       .omp.lastprivate.then:
// CHECK4-NEXT:    [[TMP30:%.*]] = load i32, i32* [[L_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP30]], i32* [[L_ADDR]], align 4
// CHECK4-NEXT:    br label [[DOTOMP_LASTPRIVATE_DONE]]
// CHECK4:       .omp.lastprivate.done:
// CHECK4-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK4:       omp.precond.end:
// CHECK4-NEXT:    ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34
// CHECK4-SAME: (i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR4:[0-9]+]] {
// CHECK4-NEXT:  entry:
// CHECK4-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK4-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK4-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK4-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false, i1 false)
// CHECK4-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK4-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK4:       user_code.entry:
// CHECK4-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB4]])
// CHECK4-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP3]], i32* [[N_CASTED]], align 4
// CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK4-NEXT:    call void @__omp_outlined__2(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP4]], [1000 x i16]* [[TMP0]]) #[[ATTR2]]
// CHECK4-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
// CHECK4-NEXT:    ret void
// CHECK4:       worker.exit:
// CHECK4-NEXT:    ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@__omp_outlined__2
// CHECK4-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK4-NEXT:  entry:
// CHECK4-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK4-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK4-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK4-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK4-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK4-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK4-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK4-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK4-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK4-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK4-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK4:       omp.precond.then:
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK4-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK4-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK4-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK4-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK4-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK4-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK4:       cond.true:
// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK4-NEXT:    br label [[COND_END:%.*]]
// CHECK4:       cond.false:
// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    br label [[COND_END]]
// CHECK4:       cond.end:
// CHECK4-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK4-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK4:       omp.inner.for.cond:
// CHECK4-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK4-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK4-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK4-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK4:       omp.inner.for.body:
// CHECK4-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP16:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP16]], i32* [[N_CASTED]], align 4
// CHECK4-NEXT:    [[TMP17:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK4-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK4-NEXT:    [[TMP19:%.*]] = inttoptr i32 [[TMP14]] to i8*
// CHECK4-NEXT:    store i8* [[TMP19]], i8** [[TMP18]], align 4
// CHECK4-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK4-NEXT:    [[TMP21:%.*]] = inttoptr i32 [[TMP15]] to i8*
// CHECK4-NEXT:    store i8* [[TMP21]], i8** [[TMP20]], align 4
// CHECK4-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK4-NEXT:    [[TMP23:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK4-NEXT:    store i8* [[TMP23]], i8** [[TMP22]], align 4
// CHECK4-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK4-NEXT:    [[TMP25:%.*]] = bitcast [1000 x i16]* [[TMP0]] to i8*
// CHECK4-NEXT:    store i8* [[TMP25]], i8** [[TMP24]], align 4
// CHECK4-NEXT:    [[TMP26:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    [[TMP27:%.*]] = load i32, i32* [[TMP26]], align 4
// CHECK4-NEXT:    [[TMP28:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK4-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB4]], i32 [[TMP27]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i16]*)* @__omp_outlined__3 to i8*), i8* null, i8** [[TMP28]], i32 4)
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK4:       omp.inner.for.inc:
// CHECK4-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP29]], [[TMP30]]
// CHECK4-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP31]], [[TMP32]]
// CHECK4-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP33]], [[TMP34]]
// CHECK4-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK4-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP35]], [[TMP36]]
// CHECK4-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
// CHECK4:       cond.true10:
// CHECK4-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK4-NEXT:    br label [[COND_END12:%.*]]
// CHECK4:       cond.false11:
// CHECK4-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    br label [[COND_END12]]
// CHECK4:       cond.end12:
// CHECK4-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP37]], [[COND_TRUE10]] ], [ [[TMP38]], [[COND_FALSE11]] ]
// CHECK4-NEXT:    store i32 [[COND13]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    store i32 [[TMP39]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK4:       omp.inner.for.end:
// CHECK4-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK4:       omp.loop.exit:
// CHECK4-NEXT:    [[TMP40:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    [[TMP41:%.*]] = load i32, i32* [[TMP40]], align 4
// CHECK4-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP41]])
// CHECK4-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK4:       omp.precond.end:
// CHECK4-NEXT:    ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@__omp_outlined__3
// CHECK4-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i16]* nonnull align 2 dereferenceable(2000) [[AA:%.*]]) #[[ATTR0]] {
// CHECK4-NEXT:  entry:
// CHECK4-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[AA_ADDR:%.*]] = alloca [1000 x i16]*, align 4
// CHECK4-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK4-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK4-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK4-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store [1000 x i16]* [[AA]], [1000 x i16]** [[AA_ADDR]], align 4
// CHECK4-NEXT:    [[TMP0:%.*]] = load [1000 x i16]*, [1000 x i16]** [[AA_ADDR]], align 4
// CHECK4-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK4-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK4-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK4-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK4-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK4-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK4:       omp.precond.then:
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK4-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK4-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK4-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK4-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK4-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB3]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK4-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK4:       omp.inner.for.cond:
// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK4-NEXT:    [[CMP4:%.*]] = icmp ule i32 [[TMP10]], [[TMP11]]
// CHECK4-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK4:       omp.inner.for.body:
// CHECK4-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK4-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK4-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK4-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I3]], align 4
// CHECK4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x i16], [1000 x i16]* [[TMP0]], i32 0, i32 [[TMP13]]
// CHECK4-NEXT:    [[TMP14:%.*]] = load i16, i16* [[ARRAYIDX]], align 2
// CHECK4-NEXT:    [[CONV:%.*]] = sext i16 [[TMP14]] to i32
// CHECK4-NEXT:    [[ADD5:%.*]] = add nsw i32 [[CONV]], 1
// CHECK4-NEXT:    [[CONV6:%.*]] = trunc i32 [[ADD5]] to i16
// CHECK4-NEXT:    store i16 [[CONV6]], i16* [[ARRAYIDX]], align 2
// CHECK4-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK4:       omp.body.continue:
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK4:       omp.inner.for.inc:
// CHECK4-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP15]], [[TMP16]]
// CHECK4-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK4:       omp.inner.for.end:
// CHECK4-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK4:       omp.loop.exit:
// CHECK4-NEXT:    [[TMP17:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    [[TMP18:%.*]] = load i32, i32* [[TMP17]], align 4
// CHECK4-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP18]])
// CHECK4-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK4:       omp.precond.end:
// CHECK4-NEXT:    ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l39
// CHECK4-SAME: ([10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK4-NEXT:  entry:
// CHECK4-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK4-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK4-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK4-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false, i1 false)
// CHECK4-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK4-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK4:       user_code.entry:
// CHECK4-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB4]])
// CHECK4-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK4-NEXT:    call void @__omp_outlined__4(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x i32]* [[TMP0]]) #[[ATTR2]]
// CHECK4-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
// CHECK4-NEXT:    ret void
// CHECK4:       worker.exit:
// CHECK4-NEXT:    ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@__omp_outlined__4
// CHECK4-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK4-NEXT:  entry:
// CHECK4-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK4-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x i8*], align 4
// CHECK4-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK4-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK4-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    store i32 9, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK4-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK4-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK4-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK4-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9
// CHECK4-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK4:       cond.true:
// CHECK4-NEXT:    br label [[COND_END:%.*]]
// CHECK4:       cond.false:
// CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    br label [[COND_END]]
// CHECK4:       cond.end:
// CHECK4-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK4-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK4:       omp.inner.for.cond:
// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[CMP1:%.*]] = icmp slt i32 [[TMP6]], 10
// CHECK4-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK4:       omp.inner.for.body:
// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK4-NEXT:    [[TMP10:%.*]] = inttoptr i32 [[TMP7]] to i8*
// CHECK4-NEXT:    store i8* [[TMP10]], i8** [[TMP9]], align 4
// CHECK4-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK4-NEXT:    [[TMP12:%.*]] = inttoptr i32 [[TMP8]] to i8*
// CHECK4-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 4
// CHECK4-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK4-NEXT:    [[TMP14:%.*]] = bitcast [10 x i32]* [[TMP0]] to i8*
// CHECK4-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 4
// CHECK4-NEXT:    [[TMP15:%.*]] = bitcast [3 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK4-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, [10 x i32]*)* @__omp_outlined__5 to i8*), i8* null, i8** [[TMP15]], i32 3)
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK4:       omp.inner.for.inc:
// CHECK4-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP16]], [[TMP17]]
// CHECK4-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP18]], [[TMP19]]
// CHECK4-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK4-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP22]], 9
// CHECK4-NEXT:    br i1 [[CMP4]], label [[COND_TRUE5:%.*]], label [[COND_FALSE6:%.*]]
// CHECK4:       cond.true5:
// CHECK4-NEXT:    br label [[COND_END7:%.*]]
// CHECK4:       cond.false6:
// CHECK4-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    br label [[COND_END7]]
// CHECK4:       cond.end7:
// CHECK4-NEXT:    [[COND8:%.*]] = phi i32 [ 9, [[COND_TRUE5]] ], [ [[TMP23]], [[COND_FALSE6]] ]
// CHECK4-NEXT:    store i32 [[COND8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    store i32 [[TMP24]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK4:       omp.inner.for.end:
// CHECK4-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK4:       omp.loop.exit:
// CHECK4-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]])
// CHECK4-NEXT:    ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@__omp_outlined__5
// CHECK4-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], [10 x i32]* nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
// CHECK4-NEXT:  entry:
// CHECK4-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
// CHECK4-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK4-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK4-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK4-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
// CHECK4-NEXT:    [[TMP0:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK4-NEXT:    store i32 9, i32* [[DOTOMP_UB]], align 4
// CHECK4-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK4-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP1]], i32* [[DOTOMP_LB]], align 4
// CHECK4-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_UB]], align 4
// CHECK4-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK4-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK4-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB3]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK4-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK4:       omp.inner.for.cond:
// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK4-NEXT:    [[CMP:%.*]] = icmp ule i32 [[TMP6]], [[TMP7]]
// CHECK4-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK4:       omp.inner.for.body:
// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP8]], 1
// CHECK4-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK4-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[I]], align 4
// CHECK4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[TMP0]], i32 0, i32 [[TMP9]]
// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK4-NEXT:    [[ADD1:%.*]] = add nsw i32 [[TMP10]], 1
// CHECK4-NEXT:    store i32 [[ADD1]], i32* [[ARRAYIDX]], align 4
// CHECK4-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK4:       omp.body.continue:
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK4:       omp.inner.for.inc:
// CHECK4-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP11]], [[TMP12]]
// CHECK4-NEXT:    store i32 [[ADD2]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK4:       omp.inner.for.end:
// CHECK4-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK4:       omp.loop.exit:
// CHECK4-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]])
// CHECK4-NEXT:    ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l44
// CHECK4-SAME: ([10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK4-NEXT:  entry:
// CHECK4-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK4-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[F_CASTED:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK4-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK4-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false, i1 false)
// CHECK4-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK4-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK4:       user_code.entry:
// CHECK4-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB4]])
// CHECK4-NEXT:    [[TMP3:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP3]], i32* [[F_CASTED]], align 4
// CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[F_CASTED]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK4-NEXT:    call void @__omp_outlined__6(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], [10 x [10 x i32]]* [[TMP0]], i32 [[TMP4]]) #[[ATTR2]]
// CHECK4-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
// CHECK4-NEXT:    ret void
// CHECK4:       worker.exit:
// CHECK4-NEXT:    ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@__omp_outlined__6
// CHECK4-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK4-NEXT:  entry:
// CHECK4-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK4-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[F_CASTED:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK4-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK4-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK4-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    store i32 99, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK4-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK4-NEXT:    [[TMP1:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    [[TMP2:%.*]] = load i32, i32* [[TMP1]], align 4
// CHECK4-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK4-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 99
// CHECK4-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK4:       cond.true:
// CHECK4-NEXT:    br label [[COND_END:%.*]]
// CHECK4:       cond.false:
// CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    br label [[COND_END]]
// CHECK4:       cond.end:
// CHECK4-NEXT:    [[COND:%.*]] = phi i32 [ 99, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
// CHECK4-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK4:       omp.inner.for.cond:
// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[TMP6]], 100
// CHECK4-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK4:       omp.inner.for.body:
// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP9]], i32* [[F_CASTED]], align 4
// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[F_CASTED]], align 4
// CHECK4-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK4-NEXT:    [[TMP12:%.*]] = inttoptr i32 [[TMP7]] to i8*
// CHECK4-NEXT:    store i8* [[TMP12]], i8** [[TMP11]], align 4
// CHECK4-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK4-NEXT:    [[TMP14:%.*]] = inttoptr i32 [[TMP8]] to i8*
// CHECK4-NEXT:    store i8* [[TMP14]], i8** [[TMP13]], align 4
// CHECK4-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK4-NEXT:    [[TMP16:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK4-NEXT:    store i8* [[TMP16]], i8** [[TMP15]], align 4
// CHECK4-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK4-NEXT:    [[TMP18:%.*]] = inttoptr i32 [[TMP10]] to i8*
// CHECK4-NEXT:    store i8* [[TMP18]], i8** [[TMP17]], align 4
// CHECK4-NEXT:    [[TMP19:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK4-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB4]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, [10 x [10 x i32]]*, i32)* @__omp_outlined__7 to i8*), i8* null, i8** [[TMP19]], i32 4)
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK4:       omp.inner.for.inc:
// CHECK4-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
// CHECK4-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP22]], [[TMP23]]
// CHECK4-NEXT:    store i32 [[ADD3]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP24]], [[TMP25]]
// CHECK4-NEXT:    store i32 [[ADD4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP26]], 99
// CHECK4-NEXT:    br i1 [[CMP5]], label [[COND_TRUE6:%.*]], label [[COND_FALSE7:%.*]]
// CHECK4:       cond.true6:
// CHECK4-NEXT:    br label [[COND_END8:%.*]]
// CHECK4:       cond.false7:
// CHECK4-NEXT:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    br label [[COND_END8]]
// CHECK4:       cond.end8:
// CHECK4-NEXT:    [[COND9:%.*]] = phi i32 [ 99, [[COND_TRUE6]] ], [ [[TMP27]], [[COND_FALSE7]] ]
// CHECK4-NEXT:    store i32 [[COND9]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    store i32 [[TMP28]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK4:       omp.inner.for.end:
// CHECK4-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK4:       omp.loop.exit:
// CHECK4-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]])
// CHECK4-NEXT:    ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@__omp_outlined__7
// CHECK4-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]], i32 [[F:%.*]]) #[[ATTR0]] {
// CHECK4-NEXT:  entry:
// CHECK4-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK4-NEXT:    [[F_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[K:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK4-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK4-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK4-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[F]], i32* [[F_ADDR]], align 4
// CHECK4-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK4-NEXT:    store i32 99, i32* [[DOTOMP_UB]], align 4
// CHECK4-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK4-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP1]], i32* [[DOTOMP_LB]], align 4
// CHECK4-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_UB]], align 4
// CHECK4-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK4-NEXT:    [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK4-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB3]], i32 [[TMP4]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK4-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK4:       omp.inner.for.cond:
// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK4-NEXT:    [[CMP:%.*]] = icmp ule i32 [[TMP6]], [[TMP7]]
// CHECK4-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK4:       omp.inner.for.body:
// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP8]], 10
// CHECK4-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK4-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK4-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[DIV2:%.*]] = sdiv i32 [[TMP10]], 10
// CHECK4-NEXT:    [[MUL3:%.*]] = mul nsw i32 [[DIV2]], 10
// CHECK4-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP9]], [[MUL3]]
// CHECK4-NEXT:    [[MUL4:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK4-NEXT:    [[ADD5:%.*]] = add nsw i32 0, [[MUL4]]
// CHECK4-NEXT:    store i32 [[ADD5]], i32* [[J]], align 4
// CHECK4-NEXT:    store i32 10, i32* [[K]], align 4
// CHECK4-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
// CHECK4-NEXT:    [[TMP12:%.*]] = load i32, i32* [[J]], align 4
// CHECK4-NEXT:    [[TMP13:%.*]] = load i32, i32* [[F_ADDR]], align 4
// CHECK4-NEXT:    [[MUL6:%.*]] = mul nsw i32 [[TMP12]], [[TMP13]]
// CHECK4-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP11]], [[MUL6]]
// CHECK4-NEXT:    [[TMP14:%.*]] = load i32, i32* [[K]], align 4
// CHECK4-NEXT:    [[ADD8:%.*]] = add nsw i32 [[ADD7]], [[TMP14]]
// CHECK4-NEXT:    [[TMP15:%.*]] = load i32, i32* [[I]], align 4
// CHECK4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i32 0, i32 [[TMP15]]
// CHECK4-NEXT:    [[TMP16:%.*]] = load i32, i32* [[J]], align 4
// CHECK4-NEXT:    [[ARRAYIDX9:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i32 0, i32 [[TMP16]]
// CHECK4-NEXT:    store i32 [[ADD8]], i32* [[ARRAYIDX9]], align 4
// CHECK4-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK4:       omp.body.continue:
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK4:       omp.inner.for.inc:
// CHECK4-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    [[ADD10:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK4-NEXT:    store i32 [[ADD10]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK4:       omp.inner.for.end:
// CHECK4-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK4:       omp.loop.exit:
// CHECK4-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]])
// CHECK4-NEXT:    ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l52
// CHECK4-SAME: (i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK4-NEXT:  entry:
// CHECK4-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK4-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK4-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK4-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false, i1 false)
// CHECK4-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK4-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK4:       user_code.entry:
// CHECK4-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB4]])
// CHECK4-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP3]], i32* [[N_CASTED]], align 4
// CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK4-NEXT:    call void @__omp_outlined__8(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP4]], [10 x [10 x i32]]* [[TMP0]]) #[[ATTR2]]
// CHECK4-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
// CHECK4-NEXT:    ret void
// CHECK4:       worker.exit:
// CHECK4-NEXT:    ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@__omp_outlined__8
// CHECK4-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK4-NEXT:  entry:
// CHECK4-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK4-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK4-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK4-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i64, align 8
// CHECK4-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i64, align 8
// CHECK4-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK4-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I9:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[J10:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x i8*], align 4
// CHECK4-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK4-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK4-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK4-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK4-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK4-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK4-NEXT:    [[CONV:%.*]] = sext i32 [[DIV]] to i64
// CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK4-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK4-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK4-NEXT:    [[CONV6:%.*]] = sext i32 [[DIV5]] to i64
// CHECK4-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV]], [[CONV6]]
// CHECK4-NEXT:    [[SUB7:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK4-NEXT:    store i64 [[SUB7]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK4-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK4-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK4:       land.lhs.true:
// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK4-NEXT:    [[CMP8:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK4-NEXT:    br i1 [[CMP8]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK4:       omp.precond.then:
// CHECK4-NEXT:    store i64 0, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK4-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK4-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK4-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK4-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK4-NEXT:    [[CONV11:%.*]] = zext i32 [[NVPTX_NUM_THREADS]] to i64
// CHECK4-NEXT:    [[TMP8:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[TMP8]], align 4
// CHECK4-NEXT:    call void @__kmpc_distribute_static_init_8(%struct.ident_t* @[[GLOB2]], i32 [[TMP9]], i32 91, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_COMB_LB]], i64* [[DOTOMP_COMB_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 [[CONV11]])
// CHECK4-NEXT:    [[TMP10:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK4-NEXT:    [[TMP11:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK4-NEXT:    [[CMP12:%.*]] = icmp sgt i64 [[TMP10]], [[TMP11]]
// CHECK4-NEXT:    br i1 [[CMP12]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK4:       cond.true:
// CHECK4-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK4-NEXT:    br label [[COND_END:%.*]]
// CHECK4:       cond.false:
// CHECK4-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK4-NEXT:    br label [[COND_END]]
// CHECK4:       cond.end:
// CHECK4-NEXT:    [[COND:%.*]] = phi i64 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
// CHECK4-NEXT:    store i64 [[COND]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK4-NEXT:    [[TMP14:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK4-NEXT:    store i64 [[TMP14]], i64* [[DOTOMP_IV]], align 8
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK4:       omp.inner.for.cond:
// CHECK4-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK4-NEXT:    [[TMP16:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK4-NEXT:    [[ADD:%.*]] = add nsw i64 [[TMP16]], 1
// CHECK4-NEXT:    [[CMP13:%.*]] = icmp slt i64 [[TMP15]], [[ADD]]
// CHECK4-NEXT:    br i1 [[CMP13]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK4:       omp.inner.for.body:
// CHECK4-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK4-NEXT:    [[TMP18:%.*]] = trunc i64 [[TMP17]] to i32
// CHECK4-NEXT:    [[TMP19:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK4-NEXT:    [[TMP20:%.*]] = trunc i64 [[TMP19]] to i32
// CHECK4-NEXT:    [[TMP21:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP21]], i32* [[N_CASTED]], align 4
// CHECK4-NEXT:    [[TMP22:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK4-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK4-NEXT:    [[TMP24:%.*]] = inttoptr i32 [[TMP18]] to i8*
// CHECK4-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 4
// CHECK4-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK4-NEXT:    [[TMP26:%.*]] = inttoptr i32 [[TMP20]] to i8*
// CHECK4-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 4
// CHECK4-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK4-NEXT:    [[TMP28:%.*]] = inttoptr i32 [[TMP22]] to i8*
// CHECK4-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 4
// CHECK4-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK4-NEXT:    [[TMP30:%.*]] = bitcast [10 x [10 x i32]]* [[TMP0]] to i8*
// CHECK4-NEXT:    store i8* [[TMP30]], i8** [[TMP29]], align 4
// CHECK4-NEXT:    [[TMP31:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    [[TMP32:%.*]] = load i32, i32* [[TMP31]], align 4
// CHECK4-NEXT:    [[TMP33:%.*]] = bitcast [4 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK4-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB4]], i32 [[TMP32]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [10 x [10 x i32]]*)* @__omp_outlined__9 to i8*), i8* null, i8** [[TMP33]], i32 4)
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK4:       omp.inner.for.inc:
// CHECK4-NEXT:    [[TMP34:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK4-NEXT:    [[TMP35:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK4-NEXT:    [[ADD14:%.*]] = add nsw i64 [[TMP34]], [[TMP35]]
// CHECK4-NEXT:    store i64 [[ADD14]], i64* [[DOTOMP_IV]], align 8
// CHECK4-NEXT:    [[TMP36:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK4-NEXT:    [[TMP37:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK4-NEXT:    [[ADD15:%.*]] = add nsw i64 [[TMP36]], [[TMP37]]
// CHECK4-NEXT:    store i64 [[ADD15]], i64* [[DOTOMP_COMB_LB]], align 8
// CHECK4-NEXT:    [[TMP38:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK4-NEXT:    [[TMP39:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK4-NEXT:    [[ADD16:%.*]] = add nsw i64 [[TMP38]], [[TMP39]]
// CHECK4-NEXT:    store i64 [[ADD16]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK4-NEXT:    [[TMP40:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK4-NEXT:    [[TMP41:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK4-NEXT:    [[CMP17:%.*]] = icmp sgt i64 [[TMP40]], [[TMP41]]
// CHECK4-NEXT:    br i1 [[CMP17]], label [[COND_TRUE18:%.*]], label [[COND_FALSE19:%.*]]
// CHECK4:       cond.true18:
// CHECK4-NEXT:    [[TMP42:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK4-NEXT:    br label [[COND_END20:%.*]]
// CHECK4:       cond.false19:
// CHECK4-NEXT:    [[TMP43:%.*]] = load i64, i64* [[DOTOMP_COMB_UB]], align 8
// CHECK4-NEXT:    br label [[COND_END20]]
// CHECK4:       cond.end20:
// CHECK4-NEXT:    [[COND21:%.*]] = phi i64 [ [[TMP42]], [[COND_TRUE18]] ], [ [[TMP43]], [[COND_FALSE19]] ]
// CHECK4-NEXT:    store i64 [[COND21]], i64* [[DOTOMP_COMB_UB]], align 8
// CHECK4-NEXT:    [[TMP44:%.*]] = load i64, i64* [[DOTOMP_COMB_LB]], align 8
// CHECK4-NEXT:    store i64 [[TMP44]], i64* [[DOTOMP_IV]], align 8
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK4:       omp.inner.for.end:
// CHECK4-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK4:       omp.loop.exit:
// CHECK4-NEXT:    [[TMP45:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    [[TMP46:%.*]] = load i32, i32* [[TMP45]], align 4
// CHECK4-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP46]])
// CHECK4-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK4:       omp.precond.end:
// CHECK4-NEXT:    ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@__omp_outlined__9
// CHECK4-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [10 x [10 x i32]]* nonnull align 4 dereferenceable(400) [[C:%.*]]) #[[ATTR0]] {
// CHECK4-NEXT:  entry:
// CHECK4-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[C_ADDR:%.*]] = alloca [10 x [10 x i32]]*, align 4
// CHECK4-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK4-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_3:%.*]] = alloca i64, align 8
// CHECK4-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[J:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_LB:%.*]] = alloca i64, align 8
// CHECK4-NEXT:    [[DOTOMP_UB:%.*]] = alloca i64, align 8
// CHECK4-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK4-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I11:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[J12:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK4-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK4-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK4-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store [10 x [10 x i32]]* [[C]], [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK4-NEXT:    [[TMP0:%.*]] = load [10 x [10 x i32]]*, [10 x [10 x i32]]** [[C_ADDR]], align 4
// CHECK4-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[TMP2:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK4-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0
// CHECK4-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK4-NEXT:    [[CONV:%.*]] = sext i32 [[DIV]] to i64
// CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK4-NEXT:    [[SUB4:%.*]] = sub nsw i32 [[TMP4]], 0
// CHECK4-NEXT:    [[DIV5:%.*]] = sdiv i32 [[SUB4]], 1
// CHECK4-NEXT:    [[CONV6:%.*]] = sext i32 [[DIV5]] to i64
// CHECK4-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV]], [[CONV6]]
// CHECK4-NEXT:    [[SUB7:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK4-NEXT:    store i64 [[SUB7]], i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK4-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[J]], align 4
// CHECK4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP5]]
// CHECK4-NEXT:    br i1 [[CMP]], label [[LAND_LHS_TRUE:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK4:       land.lhs.true:
// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK4-NEXT:    [[CMP8:%.*]] = icmp slt i32 0, [[TMP6]]
// CHECK4-NEXT:    br i1 [[CMP8]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END]]
// CHECK4:       omp.precond.then:
// CHECK4-NEXT:    store i64 0, i64* [[DOTOMP_LB]], align 8
// CHECK4-NEXT:    [[TMP7:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_3]], align 8
// CHECK4-NEXT:    store i64 [[TMP7]], i64* [[DOTOMP_UB]], align 8
// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK4-NEXT:    [[CONV9:%.*]] = zext i32 [[TMP8]] to i64
// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK4-NEXT:    [[CONV10:%.*]] = zext i32 [[TMP9]] to i64
// CHECK4-NEXT:    store i64 [[CONV9]], i64* [[DOTOMP_LB]], align 8
// CHECK4-NEXT:    store i64 [[CONV10]], i64* [[DOTOMP_UB]], align 8
// CHECK4-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK4-NEXT:    [[TMP10:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK4-NEXT:    call void @__kmpc_for_static_init_8(%struct.ident_t* @[[GLOB3]], i32 [[TMP11]], i32 33, i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_LB]], i64* [[DOTOMP_UB]], i64* [[DOTOMP_STRIDE]], i64 1, i64 1)
// CHECK4-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTOMP_LB]], align 8
// CHECK4-NEXT:    store i64 [[TMP12]], i64* [[DOTOMP_IV]], align 8
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK4:       omp.inner.for.cond:
// CHECK4-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK4-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK4-NEXT:    [[CONV13:%.*]] = zext i32 [[TMP14]] to i64
// CHECK4-NEXT:    [[CMP14:%.*]] = icmp sle i64 [[TMP13]], [[CONV13]]
// CHECK4-NEXT:    br i1 [[CMP14]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK4:       omp.inner.for.body:
// CHECK4-NEXT:    [[TMP15:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK4-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK4-NEXT:    [[SUB15:%.*]] = sub nsw i32 [[TMP16]], 0
// CHECK4-NEXT:    [[DIV16:%.*]] = sdiv i32 [[SUB15]], 1
// CHECK4-NEXT:    [[MUL17:%.*]] = mul nsw i32 1, [[DIV16]]
// CHECK4-NEXT:    [[CONV18:%.*]] = sext i32 [[MUL17]] to i64
// CHECK4-NEXT:    [[DIV19:%.*]] = sdiv i64 [[TMP15]], [[CONV18]]
// CHECK4-NEXT:    [[MUL20:%.*]] = mul nsw i64 [[DIV19]], 1
// CHECK4-NEXT:    [[ADD:%.*]] = add nsw i64 0, [[MUL20]]
// CHECK4-NEXT:    [[CONV21:%.*]] = trunc i64 [[ADD]] to i32
// CHECK4-NEXT:    store i32 [[CONV21]], i32* [[I11]], align 4
// CHECK4-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK4-NEXT:    [[TMP18:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK4-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK4-NEXT:    [[SUB22:%.*]] = sub nsw i32 [[TMP19]], 0
// CHECK4-NEXT:    [[DIV23:%.*]] = sdiv i32 [[SUB22]], 1
// CHECK4-NEXT:    [[MUL24:%.*]] = mul nsw i32 1, [[DIV23]]
// CHECK4-NEXT:    [[CONV25:%.*]] = sext i32 [[MUL24]] to i64
// CHECK4-NEXT:    [[DIV26:%.*]] = sdiv i64 [[TMP18]], [[CONV25]]
// CHECK4-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK4-NEXT:    [[SUB27:%.*]] = sub nsw i32 [[TMP20]], 0
// CHECK4-NEXT:    [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1
// CHECK4-NEXT:    [[MUL29:%.*]] = mul nsw i32 1, [[DIV28]]
// CHECK4-NEXT:    [[CONV30:%.*]] = sext i32 [[MUL29]] to i64
// CHECK4-NEXT:    [[MUL31:%.*]] = mul nsw i64 [[DIV26]], [[CONV30]]
// CHECK4-NEXT:    [[SUB32:%.*]] = sub nsw i64 [[TMP17]], [[MUL31]]
// CHECK4-NEXT:    [[MUL33:%.*]] = mul nsw i64 [[SUB32]], 1
// CHECK4-NEXT:    [[ADD34:%.*]] = add nsw i64 0, [[MUL33]]
// CHECK4-NEXT:    [[CONV35:%.*]] = trunc i64 [[ADD34]] to i32
// CHECK4-NEXT:    store i32 [[CONV35]], i32* [[J12]], align 4
// CHECK4-NEXT:    [[TMP21:%.*]] = load i32, i32* [[I11]], align 4
// CHECK4-NEXT:    [[TMP22:%.*]] = load i32, i32* [[J12]], align 4
// CHECK4-NEXT:    [[ADD36:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
// CHECK4-NEXT:    [[TMP23:%.*]] = load i32, i32* [[I11]], align 4
// CHECK4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x i32]], [10 x [10 x i32]]* [[TMP0]], i32 0, i32 [[TMP23]]
// CHECK4-NEXT:    [[TMP24:%.*]] = load i32, i32* [[J12]], align 4
// CHECK4-NEXT:    [[ARRAYIDX37:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[ARRAYIDX]], i32 0, i32 [[TMP24]]
// CHECK4-NEXT:    store i32 [[ADD36]], i32* [[ARRAYIDX37]], align 4
// CHECK4-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK4:       omp.body.continue:
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK4:       omp.inner.for.inc:
// CHECK4-NEXT:    [[TMP25:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK4-NEXT:    [[TMP26:%.*]] = load i64, i64* [[DOTOMP_STRIDE]], align 8
// CHECK4-NEXT:    [[ADD38:%.*]] = add nsw i64 [[TMP25]], [[TMP26]]
// CHECK4-NEXT:    store i64 [[ADD38]], i64* [[DOTOMP_IV]], align 8
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK4:       omp.inner.for.end:
// CHECK4-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK4:       omp.loop.exit:
// CHECK4-NEXT:    [[TMP27:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    [[TMP28:%.*]] = load i32, i32* [[TMP27]], align 4
// CHECK4-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP28]])
// CHECK4-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK4:       omp.precond.end:
// CHECK4-NEXT:    ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l59
// CHECK4-SAME: (i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK4-NEXT:  entry:
// CHECK4-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK4-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK4-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK4-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK4-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false, i1 false)
// CHECK4-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK4-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK4:       user_code.entry:
// CHECK4-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB4]])
// CHECK4-NEXT:    [[TMP3:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP3]], i32* [[N_CASTED]], align 4
// CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK4-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP2]], i32* [[DOTTHREADID_TEMP_]], align 4
// CHECK4-NEXT:    call void @__omp_outlined__10(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP4]], [1000 x i32]* [[TMP0]], i32* [[TMP5]]) #[[ATTR2]]
// CHECK4-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
// CHECK4-NEXT:    ret void
// CHECK4:       worker.exit:
// CHECK4-NEXT:    ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@__omp_outlined__10
// CHECK4-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK4-NEXT:  entry:
// CHECK4-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK4-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [5 x i8*], align 4
// CHECK4-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK4-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK4-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK4-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK4-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK4-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK4-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK4-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK4-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK4-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK4:       omp.precond.then:
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK4-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK4-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK4-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP5]], align 4
// CHECK4-NEXT:    call void @__kmpc_distribute_static_init_4(%struct.ident_t* @[[GLOB2]], i32 [[TMP6]], i32 91, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_COMB_LB]], i32* [[DOTOMP_COMB_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 [[NVPTX_NUM_THREADS]])
// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK4-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP7]], [[TMP8]]
// CHECK4-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK4:       cond.true:
// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK4-NEXT:    br label [[COND_END:%.*]]
// CHECK4:       cond.false:
// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    br label [[COND_END]]
// CHECK4:       cond.end:
// CHECK4-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP9]], [[COND_TRUE]] ], [ [[TMP10]], [[COND_FALSE]] ]
// CHECK4-NEXT:    store i32 [[COND]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    store i32 [[TMP11]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK4:       omp.inner.for.cond:
// CHECK4-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK4-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK4-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP12]], [[ADD]]
// CHECK4-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK4:       omp.inner.for.body:
// CHECK4-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP16:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP16]], i32* [[N_CASTED]], align 4
// CHECK4-NEXT:    [[TMP17:%.*]] = load i32, i32* [[N_CASTED]], align 4
// CHECK4-NEXT:    [[TMP18:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK4-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
// CHECK4-NEXT:    [[TMP20:%.*]] = inttoptr i32 [[TMP14]] to i8*
// CHECK4-NEXT:    store i8* [[TMP20]], i8** [[TMP19]], align 4
// CHECK4-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
// CHECK4-NEXT:    [[TMP22:%.*]] = inttoptr i32 [[TMP15]] to i8*
// CHECK4-NEXT:    store i8* [[TMP22]], i8** [[TMP21]], align 4
// CHECK4-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
// CHECK4-NEXT:    [[TMP24:%.*]] = inttoptr i32 [[TMP17]] to i8*
// CHECK4-NEXT:    store i8* [[TMP24]], i8** [[TMP23]], align 4
// CHECK4-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 3
// CHECK4-NEXT:    [[TMP26:%.*]] = bitcast [1000 x i32]* [[TMP0]] to i8*
// CHECK4-NEXT:    store i8* [[TMP26]], i8** [[TMP25]], align 4
// CHECK4-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 4
// CHECK4-NEXT:    [[TMP28:%.*]] = bitcast i32* [[TMP18]] to i8*
// CHECK4-NEXT:    store i8* [[TMP28]], i8** [[TMP27]], align 4
// CHECK4-NEXT:    [[TMP29:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    [[TMP30:%.*]] = load i32, i32* [[TMP29]], align 4
// CHECK4-NEXT:    [[TMP31:%.*]] = bitcast [5 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
// CHECK4-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB4]], i32 [[TMP30]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32, i32, i32, [1000 x i32]*, i32*)* @__omp_outlined__11 to i8*), i8* null, i8** [[TMP31]], i32 5)
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK4:       omp.inner.for.inc:
// CHECK4-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP32]], [[TMP33]]
// CHECK4-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP34]], [[TMP35]]
// CHECK4-NEXT:    store i32 [[ADD7]], i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP36]], [[TMP37]]
// CHECK4-NEXT:    store i32 [[ADD8]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP39:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK4-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP38]], [[TMP39]]
// CHECK4-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
// CHECK4:       cond.true10:
// CHECK4-NEXT:    [[TMP40:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK4-NEXT:    br label [[COND_END12:%.*]]
// CHECK4:       cond.false11:
// CHECK4-NEXT:    [[TMP41:%.*]] = load i32, i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    br label [[COND_END12]]
// CHECK4:       cond.end12:
// CHECK4-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP40]], [[COND_TRUE10]] ], [ [[TMP41]], [[COND_FALSE11]] ]
// CHECK4-NEXT:    store i32 [[COND13]], i32* [[DOTOMP_COMB_UB]], align 4
// CHECK4-NEXT:    [[TMP42:%.*]] = load i32, i32* [[DOTOMP_COMB_LB]], align 4
// CHECK4-NEXT:    store i32 [[TMP42]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK4:       omp.inner.for.end:
// CHECK4-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK4:       omp.loop.exit:
// CHECK4-NEXT:    [[TMP43:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    [[TMP44:%.*]] = load i32, i32* [[TMP43]], align 4
// CHECK4-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP44]])
// CHECK4-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK4:       omp.precond.end:
// CHECK4-NEXT:    ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@__omp_outlined__11
// CHECK4-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], i32 [[DOTPREVIOUS_LB_:%.*]], i32 [[DOTPREVIOUS_UB_:%.*]], i32 [[N:%.*]], [1000 x i32]* nonnull align 4 dereferenceable(4000) [[A:%.*]], i32* [[V:%.*]]) #[[ATTR0]] {
// CHECK4-NEXT:  entry:
// CHECK4-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[A_ADDR:%.*]] = alloca [1000 x i32]*, align 4
// CHECK4-NEXT:    [[V_ADDR:%.*]] = alloca i32*, align 4
// CHECK4-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I3:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
// CHECK4-NEXT:    store i32 [[DOTPREVIOUS_LB_]], i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK4-NEXT:    store i32 [[DOTPREVIOUS_UB_]], i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK4-NEXT:    store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store [1000 x i32]* [[A]], [1000 x i32]** [[A_ADDR]], align 4
// CHECK4-NEXT:    store i32* [[V]], i32** [[V_ADDR]], align 4
// CHECK4-NEXT:    [[TMP0:%.*]] = load [1000 x i32]*, [1000 x i32]** [[A_ADDR]], align 4
// CHECK4-NEXT:    [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP2]], 0
// CHECK4-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK4-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK4-NEXT:    store i32 [[SUB2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[I]], align 4
// CHECK4-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP3]]
// CHECK4-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK4:       omp.precond.then:
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK4-NEXT:    store i32 [[TMP4]], i32* [[DOTOMP_UB]], align 4
// CHECK4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTPREVIOUS_LB__ADDR]], align 4
// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP5]], i32* [[DOTOMP_LB]], align 4
// CHECK4-NEXT:    store i32 [[TMP6]], i32* [[DOTOMP_UB]], align 4
// CHECK4-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK4-NEXT:    [[TMP7:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK4-NEXT:    call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB3]], i32 [[TMP8]], i32 33, i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK4-NEXT:    store i32 [[TMP9]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK4:       omp.inner.for.cond:
// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTPREVIOUS_UB__ADDR]], align 4
// CHECK4-NEXT:    [[CMP4:%.*]] = icmp ule i32 [[TMP10]], [[TMP11]]
// CHECK4-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK4:       omp.inner.for.body:
// CHECK4-NEXT:    [[TMP12:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1
// CHECK4-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK4-NEXT:    store i32 [[ADD]], i32* [[I3]], align 4
// CHECK4-NEXT:    [[TMP13:%.*]] = load i32*, i32** [[V_ADDR]], align 4
// CHECK4-NEXT:    [[TMP14:%.*]] = load i32, i32* [[I3]], align 4
// CHECK4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[TMP13]], i32 [[TMP14]]
// CHECK4-NEXT:    [[TMP15:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
// CHECK4-NEXT:    [[TMP16:%.*]] = load i32, i32* [[I3]], align 4
// CHECK4-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds [1000 x i32], [1000 x i32]* [[TMP0]], i32 0, i32 [[TMP16]]
// CHECK4-NEXT:    store i32 [[TMP15]], i32* [[ARRAYIDX5]], align 4
// CHECK4-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK4:       omp.body.continue:
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK4:       omp.inner.for.inc:
// CHECK4-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP18:%.*]] = load i32, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
// CHECK4-NEXT:    store i32 [[ADD6]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK4:       omp.inner.for.end:
// CHECK4-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK4:       omp.loop.exit:
// CHECK4-NEXT:    [[TMP19:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK4-NEXT:    [[TMP20:%.*]] = load i32, i32* [[TMP19]], align 4
// CHECK4-NEXT:    call void @__kmpc_distribute_static_fini(%struct.ident_t* @[[GLOB2]], i32 [[TMP20]])
// CHECK4-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK4:       omp.precond.end:
// CHECK4-NEXT:    ret void
//