// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --include-generated-funcs --replace-value-regex "__omp_offloading_[0-9a-z]+_[0-9a-z]+" "reduction_size[.].+[.]" "pl_cond[.].+[.|,]" --prefix-filecheck-ir-name _
// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -fopenmp-cuda-mode -emit-llvm-bc %s -o %t-ppc-host.bc
// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -fopenmp-cuda-mode -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - -debug-info-kind=limited | FileCheck %s --check-prefix=CHECK1
// expected-no-diagnostics

int main() {
  /* int(*b)[a]; */
  /* int *(**c)[a]; */
  bool bb;
  int a;
  int b[10][10];
  int c[10][10][10];
#pragma omp target parallel loop firstprivate(a, b) map(tofrom          \
                                                       : c) map(tofrom \
                                                                : bb) if (a)
  for (int i = 0; i < 10; ++i) {
    int &f = c[1][1][1];
    int &g = a;
    int &h = b[1][1];
    int d = 15;
    a = 5;
    b[0][a] = 10;
    c[0][0][a] = 11;
    b[0][a] = c[0][0][a];
    bb |= b[0][a];
  }
#pragma omp target parallel loop firstprivate(a) map(tofrom         \
                                                    : c, b) map(to \
                                                                : bb)
  for (int i = 0; i < 10; ++i) {
    int &f = c[1][1][1];
    int &g = a;
    int &h = b[1][1];
    int d = 15;
    a = 5;
    b[0][a] = 10;
    c[0][0][a] = 11;
    b[0][a] = c[0][0][a];
    d = bb;
  }
#pragma omp target parallel loop map(tofrom              \
                                    : a, c, b) map(from \
                                                   : bb)
  for (int i = 0; i < 10; ++i) {
    int &f = c[1][1][1];
    int &g = a;
    int &h = b[1][1];
    int d = 15;
    a = 5;
    b[0][a] = 10;
    c[0][0][a] = 11;
    b[0][a] = c[0][0][a];
    bb = b[0][a];
  }
  return 0;
}
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l13_debug__
// CHECK1-SAME: (ptr addrspace(1) noalias noundef [[C:%.*]], i32 noundef [[A:%.*]], ptr noalias noundef [[B:%.*]], ptr addrspace(1) noalias noundef [[BB:%.*]], i1 noundef zeroext [[DOTCAPTURE_EXPR_:%.*]], ptr noalias noundef [[DYN_PTR:%.*]]) #[[ATTR0:[0-9]+]] !dbg [[DBG9:![0-9]+]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr addrspace(1), align 8
// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[BB_ADDR:%.*]] = alloca ptr addrspace(1), align 8
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i8, align 1
// CHECK1-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[TMP:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[_TMP1:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[_TMP2:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x ptr], align 8
// CHECK1-NEXT:    store ptr addrspace(1) [[C]], ptr [[C_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[C_ADDR]], [[META30:![0-9]+]], !DIExpression(), [[META31:![0-9]+]])
// CHECK1-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK1-NEXT:      #dbg_declare(ptr [[A_ADDR]], [[META32:![0-9]+]], !DIExpression(), [[META33:![0-9]+]])
// CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[B_ADDR]], [[META34:![0-9]+]], !DIExpression(), [[META35:![0-9]+]])
// CHECK1-NEXT:    store ptr addrspace(1) [[BB]], ptr [[BB_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[BB_ADDR]], [[META36:![0-9]+]], !DIExpression(), [[META37:![0-9]+]])
// CHECK1-NEXT:    [[STOREDV:%.*]] = zext i1 [[DOTCAPTURE_EXPR_]] to i8
// CHECK1-NEXT:    store i8 [[STOREDV]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 1
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTCAPTURE_EXPR__ADDR]], [[META38:![0-9]+]], !DIExpression(), [[META39:![0-9]+]])
// CHECK1-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[DYN_PTR_ADDR]], [[META40:![0-9]+]], !DIExpression(), [[META41:![0-9]+]])
// CHECK1-NEXT:    [[TMP0:%.*]] = load ptr addrspace(1), ptr [[C_ADDR]], align 8, !dbg [[DBG42:![0-9]+]]
// CHECK1-NEXT:    [[TMP1:%.*]] = addrspacecast ptr addrspace(1) [[TMP0]] to ptr, !dbg [[DBG42]]
// CHECK1-NEXT:    store ptr [[TMP1]], ptr [[TMP]], align 8, !dbg [[DBG42]]
// CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP]], align 8, !dbg [[DBG42]], !nonnull [[META29:![0-9]+]], !align [[META43:![0-9]+]]
// CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B_ADDR]], align 8, !dbg [[DBG42]]
// CHECK1-NEXT:    store ptr [[TMP3]], ptr [[_TMP1]], align 8, !dbg [[DBG42]]
// CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[_TMP1]], align 8, !dbg [[DBG42]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP5:%.*]] = load ptr addrspace(1), ptr [[BB_ADDR]], align 8, !dbg [[DBG42]]
// CHECK1-NEXT:    [[TMP6:%.*]] = addrspacecast ptr addrspace(1) [[TMP5]] to ptr, !dbg [[DBG42]]
// CHECK1-NEXT:    store ptr [[TMP6]], ptr [[_TMP2]], align 8, !dbg [[DBG42]]
// CHECK1-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[_TMP2]], align 8, !dbg [[DBG42]], !nonnull [[META29]]
// CHECK1-NEXT:    [[TMP8:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l13_kernel_environment, ptr [[DYN_PTR]]), !dbg [[DBG42]]
// CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP8]], -1, !dbg [[DBG42]]
// CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]], !dbg [[DBG42]]
// CHECK1:       user_code.entry:
// CHECK1-NEXT:    [[TMP9:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB6:[0-9]+]]), !dbg [[DBG44:![0-9]+]]
// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG46:![0-9]+]]
// CHECK1-NEXT:    store i32 [[TMP10]], ptr [[A_CASTED]], align 4, !dbg [[DBG46]]
// CHECK1-NEXT:    [[TMP11:%.*]] = load i64, ptr [[A_CASTED]], align 8, !dbg [[DBG46]]
// CHECK1-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 0, !dbg [[DBG46]]
// CHECK1-NEXT:    store ptr [[TMP2]], ptr [[TMP12]], align 8, !dbg [[DBG46]]
// CHECK1-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 1, !dbg [[DBG46]]
// CHECK1-NEXT:    [[TMP14:%.*]] = inttoptr i64 [[TMP11]] to ptr, !dbg [[DBG46]]
// CHECK1-NEXT:    store ptr [[TMP14]], ptr [[TMP13]], align 8, !dbg [[DBG46]]
// CHECK1-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 2, !dbg [[DBG46]]
// CHECK1-NEXT:    store ptr [[TMP4]], ptr [[TMP15]], align 8, !dbg [[DBG46]]
// CHECK1-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 3, !dbg [[DBG46]]
// CHECK1-NEXT:    store ptr [[TMP7]], ptr [[TMP16]], align 8, !dbg [[DBG46]]
// CHECK1-NEXT:    [[TMP17:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR__ADDR]], align 1, !dbg [[DBG47:![0-9]+]]
// CHECK1-NEXT:    [[LOADEDV:%.*]] = trunc i8 [[TMP17]] to i1, !dbg [[DBG47]]
// CHECK1-NEXT:    [[TMP18:%.*]] = zext i1 [[LOADEDV]] to i32, !dbg [[DBG46]]
// CHECK1-NEXT:    call void @__kmpc_parallel_60(ptr @[[GLOB6]], i32 [[TMP9]], i32 [[TMP18]], i32 -1, i32 -1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l13_debug___omp_outlined, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 4, i32 0), !dbg [[DBG46]]
// CHECK1-NEXT:    call void @__kmpc_target_deinit(), !dbg [[DBG48:![0-9]+]]
// CHECK1-NEXT:    ret void, !dbg [[DBG49:![0-9]+]]
// CHECK1:       worker.exit:
// CHECK1-NEXT:    ret void, !dbg [[DBG42]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l13
// CHECK1-SAME: (ptr noundef nonnull align 4 dereferenceable(4000) [[C:%.*]], i64 noundef [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(400) [[B:%.*]], ptr noundef nonnull align 1 dereferenceable(1) [[BB:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]], ptr noalias noundef [[DYN_PTR:%.*]]) #[[ATTR1:[0-9]+]] !dbg [[DBG50:![0-9]+]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[BB_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[C_ADDR]], [[META57:![0-9]+]], !DIExpression(), [[META58:![0-9]+]])
// CHECK1-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[A_ADDR]], [[META59:![0-9]+]], !DIExpression(), [[META58]])
// CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[B_ADDR]], [[META60:![0-9]+]], !DIExpression(), [[META58]])
// CHECK1-NEXT:    store ptr [[BB]], ptr [[BB_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[BB_ADDR]], [[META61:![0-9]+]], !DIExpression(), [[META58]])
// CHECK1-NEXT:    store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTCAPTURE_EXPR__ADDR]], [[META62:![0-9]+]], !DIExpression(), [[META58]])
// CHECK1-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[DYN_PTR_ADDR]], [[META63:![0-9]+]], !DIExpression(), [[META58]])
// CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 8, !dbg [[DBG64:![0-9]+]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[B_ADDR]], align 8, !dbg [[DBG64]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[BB_ADDR]], align 8, !dbg [[DBG64]], !nonnull [[META29]]
// CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8, !dbg [[DBG64]]
// CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG64]]
// CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[B_ADDR]], align 8, !dbg [[DBG64]]
// CHECK1-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[BB_ADDR]], align 8, !dbg [[DBG64]]
// CHECK1-NEXT:    [[TMP7:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR__ADDR]], align 1, !dbg [[DBG64]]
// CHECK1-NEXT:    [[LOADEDV:%.*]] = trunc i8 [[TMP7]] to i1, !dbg [[DBG64]]
// CHECK1-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[DYN_PTR_ADDR]], align 8, !dbg [[DBG64]]
// CHECK1-NEXT:    [[TMP9:%.*]] = addrspacecast ptr [[TMP3]] to ptr addrspace(1), !dbg [[DBG64]]
// CHECK1-NEXT:    [[TMP10:%.*]] = addrspacecast ptr [[TMP6]] to ptr addrspace(1), !dbg [[DBG64]]
// CHECK1-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l13_debug__(ptr addrspace(1) [[TMP9]], i32 [[TMP4]], ptr [[TMP5]], ptr addrspace(1) [[TMP10]], i1 [[LOADEDV]], ptr [[TMP8]]) #[[ATTR3:[0-9]+]], !dbg [[DBG64]]
// CHECK1-NEXT:    ret void, !dbg [[DBG64]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l13_debug___omp_outlined_debug__
// CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr addrspace(1) noalias noundef [[C:%.*]], i32 noundef [[A:%.*]], ptr noalias noundef [[B:%.*]], ptr addrspace(1) noalias noundef [[BB:%.*]]) #[[ATTR0]] !dbg [[DBG65:![0-9]+]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr addrspace(1), align 8
// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[BB_ADDR:%.*]] = alloca ptr addrspace(1), align 8
// CHECK1-NEXT:    [[TMP:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[_TMP1:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[_TMP2:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[_TMP3:%.*]] = 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:    [[B4:%.*]] = alloca [10 x [10 x i32]], align 4
// CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[F:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[G:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[H:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[D:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTGLOBAL_TID__ADDR]], [[META72:![0-9]+]], !DIExpression(), [[META73:![0-9]+]])
// CHECK1-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTBOUND_TID__ADDR]], [[META74:![0-9]+]], !DIExpression(), [[META73]])
// CHECK1-NEXT:    store ptr addrspace(1) [[C]], ptr [[C_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[C_ADDR]], [[META75:![0-9]+]], !DIExpression(), [[META76:![0-9]+]])
// CHECK1-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK1-NEXT:      #dbg_declare(ptr [[A_ADDR]], [[META77:![0-9]+]], !DIExpression(), [[META78:![0-9]+]])
// CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[B_ADDR]], [[META79:![0-9]+]], !DIExpression(), [[META80:![0-9]+]])
// CHECK1-NEXT:    store ptr addrspace(1) [[BB]], ptr [[BB_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[BB_ADDR]], [[META81:![0-9]+]], !DIExpression(), [[META82:![0-9]+]])
// CHECK1-NEXT:    [[TMP0:%.*]] = load ptr addrspace(1), ptr [[C_ADDR]], align 8, !dbg [[DBG83:![0-9]+]]
// CHECK1-NEXT:    [[TMP1:%.*]] = addrspacecast ptr addrspace(1) [[TMP0]] to ptr, !dbg [[DBG83]]
// CHECK1-NEXT:    store ptr [[TMP1]], ptr [[TMP]], align 8, !dbg [[DBG83]]
// CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP]], align 8, !dbg [[DBG83]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B_ADDR]], align 8, !dbg [[DBG83]]
// CHECK1-NEXT:    store ptr [[TMP3]], ptr [[_TMP1]], align 8, !dbg [[DBG83]]
// CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[_TMP1]], align 8, !dbg [[DBG83]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP5:%.*]] = load ptr addrspace(1), ptr [[BB_ADDR]], align 8, !dbg [[DBG83]]
// CHECK1-NEXT:    [[TMP6:%.*]] = addrspacecast ptr addrspace(1) [[TMP5]] to ptr, !dbg [[DBG83]]
// CHECK1-NEXT:    store ptr [[TMP6]], ptr [[_TMP2]], align 8, !dbg [[DBG83]]
// CHECK1-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[_TMP2]], align 8, !dbg [[DBG83]], !nonnull [[META29]]
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTOMP_IV]], [[META84:![0-9]+]], !DIExpression(), [[META73]])
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTOMP_LB]], [[META85:![0-9]+]], !DIExpression(), [[META73]])
// CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4, !dbg [[DBG86:![0-9]+]]
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTOMP_UB]], [[META87:![0-9]+]], !DIExpression(), [[META73]])
// CHECK1-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4, !dbg [[DBG86]]
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTOMP_STRIDE]], [[META88:![0-9]+]], !DIExpression(), [[META73]])
// CHECK1-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4, !dbg [[DBG86]]
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTOMP_IS_LAST]], [[META89:![0-9]+]], !DIExpression(), [[META73]])
// CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4, !dbg [[DBG86]]
// CHECK1-NEXT:      #dbg_declare(ptr [[B4]], [[META90:![0-9]+]], !DIExpression(), [[META73]])
// CHECK1-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 4 [[B4]], ptr align 4 [[TMP4]], i64 400, i1 false), !dbg [[DBG83]]
// CHECK1-NEXT:      #dbg_declare(ptr [[I]], [[META91:![0-9]+]], !DIExpression(), [[META73]])
// CHECK1-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8, !dbg [[DBG83]]
// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4, !dbg [[DBG83]]
// CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB3:[0-9]+]], i32 [[TMP9]], i32 33, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1), !dbg [[DBG92:![0-9]+]]
// CHECK1-NEXT:    br label [[OMP_DISPATCH_COND:%.*]], !dbg [[DBG83]]
// CHECK1:       omp.dispatch.cond:
// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !dbg [[DBG86]]
// CHECK1-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP10]], 9, !dbg [[DBG86]]
// CHECK1-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]], !dbg [[DBG86]]
// CHECK1:       cond.true:
// CHECK1-NEXT:    br label [[COND_END:%.*]], !dbg [[DBG86]]
// CHECK1:       cond.false:
// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !dbg [[DBG86]]
// CHECK1-NEXT:    br label [[COND_END]], !dbg [[DBG86]]
// CHECK1:       cond.end:
// CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP11]], [[COND_FALSE]] ], !dbg [[DBG86]]
// CHECK1-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4, !dbg [[DBG86]]
// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4, !dbg [[DBG86]]
// CHECK1-NEXT:    store i32 [[TMP12]], ptr [[DOTOMP_IV]], align 4, !dbg [[DBG86]]
// CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !dbg [[DBG86]]
// CHECK1-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !dbg [[DBG86]]
// CHECK1-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP13]], [[TMP14]], !dbg [[DBG83]]
// CHECK1-NEXT:    br i1 [[CMP5]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]], !dbg [[DBG83]]
// CHECK1:       omp.dispatch.body:
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]], !dbg [[DBG83]]
// CHECK1:       omp.inner.for.cond:
// CHECK1-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !dbg [[DBG86]]
// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !dbg [[DBG86]]
// CHECK1-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]], !dbg [[DBG83]]
// CHECK1-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]], !dbg [[DBG83]]
// CHECK1:       omp.inner.for.body:
// CHECK1-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !dbg [[DBG86]]
// CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP17]], 1, !dbg [[DBG93:![0-9]+]]
// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]], !dbg [[DBG93]]
// CHECK1-NEXT:    store i32 [[ADD]], ptr [[I]], align 4, !dbg [[DBG93]]
// CHECK1-NEXT:      #dbg_declare(ptr [[F]], [[META94:![0-9]+]], !DIExpression(), [[META97:![0-9]+]])
// CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x [10 x i32]]], ptr [[TMP2]], i64 0, i64 1, !dbg [[DBG98:![0-9]+]]
// CHECK1-NEXT:    [[ARRAYIDX7:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[ARRAYIDX]], i64 0, i64 1, !dbg [[DBG98]]
// CHECK1-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX7]], i64 0, i64 1, !dbg [[DBG98]]
// CHECK1-NEXT:    store ptr [[ARRAYIDX8]], ptr [[F]], align 8, !dbg [[META97]]
// CHECK1-NEXT:      #dbg_declare(ptr [[G]], [[META99:![0-9]+]], !DIExpression(), [[META100:![0-9]+]])
// CHECK1-NEXT:    store ptr [[A_ADDR]], ptr [[G]], align 8, !dbg [[META100]]
// CHECK1-NEXT:      #dbg_declare(ptr [[H]], [[META101:![0-9]+]], !DIExpression(), [[META102:![0-9]+]])
// CHECK1-NEXT:    [[ARRAYIDX9:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[B4]], i64 0, i64 1, !dbg [[DBG103:![0-9]+]]
// CHECK1-NEXT:    [[ARRAYIDX10:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX9]], i64 0, i64 1, !dbg [[DBG103]]
// CHECK1-NEXT:    store ptr [[ARRAYIDX10]], ptr [[H]], align 8, !dbg [[META102]]
// CHECK1-NEXT:      #dbg_declare(ptr [[D]], [[META104:![0-9]+]], !DIExpression(), [[META105:![0-9]+]])
// CHECK1-NEXT:    store i32 15, ptr [[D]], align 4, !dbg [[META105]]
// CHECK1-NEXT:    store i32 5, ptr [[A_ADDR]], align 4, !dbg [[DBG106:![0-9]+]]
// CHECK1-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[B4]], i64 0, i64 0, !dbg [[DBG107:![0-9]+]]
// CHECK1-NEXT:    [[TMP18:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG108:![0-9]+]]
// CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP18]] to i64, !dbg [[DBG107]]
// CHECK1-NEXT:    [[ARRAYIDX12:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX11]], i64 0, i64 [[IDXPROM]], !dbg [[DBG107]]
// CHECK1-NEXT:    store i32 10, ptr [[ARRAYIDX12]], align 4, !dbg [[DBG109:![0-9]+]]
// CHECK1-NEXT:    [[ARRAYIDX13:%.*]] = getelementptr inbounds [10 x [10 x [10 x i32]]], ptr [[TMP2]], i64 0, i64 0, !dbg [[DBG110:![0-9]+]]
// CHECK1-NEXT:    [[ARRAYIDX14:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[ARRAYIDX13]], i64 0, i64 0, !dbg [[DBG110]]
// CHECK1-NEXT:    [[TMP19:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG111:![0-9]+]]
// CHECK1-NEXT:    [[IDXPROM15:%.*]] = sext i32 [[TMP19]] to i64, !dbg [[DBG110]]
// CHECK1-NEXT:    [[ARRAYIDX16:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX14]], i64 0, i64 [[IDXPROM15]], !dbg [[DBG110]]
// CHECK1-NEXT:    store i32 11, ptr [[ARRAYIDX16]], align 4, !dbg [[DBG112:![0-9]+]]
// CHECK1-NEXT:    [[ARRAYIDX17:%.*]] = getelementptr inbounds [10 x [10 x [10 x i32]]], ptr [[TMP2]], i64 0, i64 0, !dbg [[DBG113:![0-9]+]]
// CHECK1-NEXT:    [[ARRAYIDX18:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[ARRAYIDX17]], i64 0, i64 0, !dbg [[DBG113]]
// CHECK1-NEXT:    [[TMP20:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG114:![0-9]+]]
// CHECK1-NEXT:    [[IDXPROM19:%.*]] = sext i32 [[TMP20]] to i64, !dbg [[DBG113]]
// CHECK1-NEXT:    [[ARRAYIDX20:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX18]], i64 0, i64 [[IDXPROM19]], !dbg [[DBG113]]
// CHECK1-NEXT:    [[TMP21:%.*]] = load i32, ptr [[ARRAYIDX20]], align 4, !dbg [[DBG113]]
// CHECK1-NEXT:    [[ARRAYIDX21:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[B4]], i64 0, i64 0, !dbg [[DBG115:![0-9]+]]
// CHECK1-NEXT:    [[TMP22:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG116:![0-9]+]]
// CHECK1-NEXT:    [[IDXPROM22:%.*]] = sext i32 [[TMP22]] to i64, !dbg [[DBG115]]
// CHECK1-NEXT:    [[ARRAYIDX23:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX21]], i64 0, i64 [[IDXPROM22]], !dbg [[DBG115]]
// CHECK1-NEXT:    store i32 [[TMP21]], ptr [[ARRAYIDX23]], align 4, !dbg [[DBG117:![0-9]+]]
// CHECK1-NEXT:    [[ARRAYIDX24:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[B4]], i64 0, i64 0, !dbg [[DBG118:![0-9]+]]
// CHECK1-NEXT:    [[TMP23:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG119:![0-9]+]]
// CHECK1-NEXT:    [[IDXPROM25:%.*]] = sext i32 [[TMP23]] to i64, !dbg [[DBG118]]
// CHECK1-NEXT:    [[ARRAYIDX26:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX24]], i64 0, i64 [[IDXPROM25]], !dbg [[DBG118]]
// CHECK1-NEXT:    [[TMP24:%.*]] = load i32, ptr [[ARRAYIDX26]], align 4, !dbg [[DBG118]]
// CHECK1-NEXT:    [[TMP25:%.*]] = load i8, ptr [[TMP7]], align 1, !dbg [[DBG120:![0-9]+]]
// CHECK1-NEXT:    [[LOADEDV:%.*]] = trunc i8 [[TMP25]] to i1, !dbg [[DBG120]]
// CHECK1-NEXT:    [[CONV:%.*]] = zext i1 [[LOADEDV]] to i32, !dbg [[DBG120]]
// CHECK1-NEXT:    [[OR:%.*]] = or i32 [[CONV]], [[TMP24]], !dbg [[DBG120]]
// CHECK1-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[OR]], 0, !dbg [[DBG120]]
// CHECK1-NEXT:    [[STOREDV:%.*]] = zext i1 [[TOBOOL]] to i8, !dbg [[DBG120]]
// CHECK1-NEXT:    store i8 [[STOREDV]], ptr [[TMP7]], align 1, !dbg [[DBG120]]
// CHECK1-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]], !dbg [[DBG121:![0-9]+]]
// CHECK1:       omp.body.continue:
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]], !dbg [[DBG92]]
// CHECK1:       omp.inner.for.inc:
// CHECK1-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !dbg [[DBG86]]
// CHECK1-NEXT:    [[ADD27:%.*]] = add nsw i32 [[TMP26]], 1, !dbg [[DBG83]]
// CHECK1-NEXT:    store i32 [[ADD27]], ptr [[DOTOMP_IV]], align 4, !dbg [[DBG83]]
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]], !dbg [[DBG92]], !llvm.loop [[LOOP122:![0-9]+]]
// CHECK1:       omp.inner.for.end:
// CHECK1-NEXT:    br label [[OMP_DISPATCH_INC:%.*]], !dbg [[DBG92]]
// CHECK1:       omp.dispatch.inc:
// CHECK1-NEXT:    [[TMP27:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4, !dbg [[DBG86]]
// CHECK1-NEXT:    [[TMP28:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !dbg [[DBG86]]
// CHECK1-NEXT:    [[ADD28:%.*]] = add nsw i32 [[TMP27]], [[TMP28]], !dbg [[DBG83]]
// CHECK1-NEXT:    store i32 [[ADD28]], ptr [[DOTOMP_LB]], align 4, !dbg [[DBG83]]
// CHECK1-NEXT:    [[TMP29:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !dbg [[DBG86]]
// CHECK1-NEXT:    [[TMP30:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !dbg [[DBG86]]
// CHECK1-NEXT:    [[ADD29:%.*]] = add nsw i32 [[TMP29]], [[TMP30]], !dbg [[DBG83]]
// CHECK1-NEXT:    store i32 [[ADD29]], ptr [[DOTOMP_UB]], align 4, !dbg [[DBG83]]
// CHECK1-NEXT:    br label [[OMP_DISPATCH_COND]], !dbg [[DBG92]], !llvm.loop [[LOOP124:![0-9]+]]
// CHECK1:       omp.dispatch.end:
// CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB5:[0-9]+]], i32 [[TMP9]]), !dbg [[DBG123:![0-9]+]]
// CHECK1-NEXT:    ret void, !dbg [[DBG125:![0-9]+]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l13_debug___omp_outlined
// CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4000) [[C:%.*]], i64 noundef [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(400) [[B:%.*]], ptr noundef nonnull align 1 dereferenceable(1) [[BB:%.*]]) #[[ATTR0]] !dbg [[DBG126:![0-9]+]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[BB_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTGLOBAL_TID__ADDR]], [[META129:![0-9]+]], !DIExpression(), [[META130:![0-9]+]])
// CHECK1-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTBOUND_TID__ADDR]], [[META131:![0-9]+]], !DIExpression(), [[META130]])
// CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[C_ADDR]], [[META132:![0-9]+]], !DIExpression(), [[META130]])
// CHECK1-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[A_ADDR]], [[META133:![0-9]+]], !DIExpression(), [[META130]])
// CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[B_ADDR]], [[META134:![0-9]+]], !DIExpression(), [[META130]])
// CHECK1-NEXT:    store ptr [[BB]], ptr [[BB_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[BB_ADDR]], [[META135:![0-9]+]], !DIExpression(), [[META130]])
// CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 8, !dbg [[DBG136:![0-9]+]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[B_ADDR]], align 8, !dbg [[DBG136]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[BB_ADDR]], align 8, !dbg [[DBG136]], !nonnull [[META29]]
// CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8, !dbg [[DBG136]]
// CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTBOUND_TID__ADDR]], align 8, !dbg [[DBG136]]
// CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[C_ADDR]], align 8, !dbg [[DBG136]]
// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG136]]
// CHECK1-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[B_ADDR]], align 8, !dbg [[DBG136]]
// CHECK1-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[BB_ADDR]], align 8, !dbg [[DBG136]]
// CHECK1-NEXT:    [[TMP9:%.*]] = addrspacecast ptr [[TMP5]] to ptr addrspace(1), !dbg [[DBG136]]
// CHECK1-NEXT:    [[TMP10:%.*]] = addrspacecast ptr [[TMP8]] to ptr addrspace(1), !dbg [[DBG136]]
// CHECK1-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l13_debug___omp_outlined_debug__(ptr [[TMP3]], ptr [[TMP4]], ptr addrspace(1) [[TMP9]], i32 [[TMP6]], ptr [[TMP7]], ptr addrspace(1) [[TMP10]]) #[[ATTR3]], !dbg [[DBG136]]
// CHECK1-NEXT:    ret void, !dbg [[DBG136]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l27_debug__
// CHECK1-SAME: (ptr addrspace(1) noalias noundef [[C:%.*]], i32 noundef [[A:%.*]], ptr addrspace(1) noalias noundef [[B:%.*]], ptr addrspace(1) noalias noundef [[BB:%.*]], ptr noalias noundef [[DYN_PTR:%.*]]) #[[ATTR0]] !dbg [[DBG137:![0-9]+]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr addrspace(1), align 8
// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr addrspace(1), align 8
// CHECK1-NEXT:    [[BB_ADDR:%.*]] = alloca ptr addrspace(1), align 8
// CHECK1-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[TMP:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[_TMP1:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[_TMP2:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x ptr], align 8
// CHECK1-NEXT:    store ptr addrspace(1) [[C]], ptr [[C_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[C_ADDR]], [[META142:![0-9]+]], !DIExpression(), [[META143:![0-9]+]])
// CHECK1-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK1-NEXT:      #dbg_declare(ptr [[A_ADDR]], [[META144:![0-9]+]], !DIExpression(), [[META145:![0-9]+]])
// CHECK1-NEXT:    store ptr addrspace(1) [[B]], ptr [[B_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[B_ADDR]], [[META146:![0-9]+]], !DIExpression(), [[META147:![0-9]+]])
// CHECK1-NEXT:    store ptr addrspace(1) [[BB]], ptr [[BB_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[BB_ADDR]], [[META148:![0-9]+]], !DIExpression(), [[META149:![0-9]+]])
// CHECK1-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[DYN_PTR_ADDR]], [[META150:![0-9]+]], !DIExpression(), [[META151:![0-9]+]])
// CHECK1-NEXT:    [[TMP0:%.*]] = load ptr addrspace(1), ptr [[C_ADDR]], align 8, !dbg [[DBG152:![0-9]+]]
// CHECK1-NEXT:    [[TMP1:%.*]] = addrspacecast ptr addrspace(1) [[TMP0]] to ptr, !dbg [[DBG152]]
// CHECK1-NEXT:    store ptr [[TMP1]], ptr [[TMP]], align 8, !dbg [[DBG152]]
// CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP]], align 8, !dbg [[DBG152]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP3:%.*]] = load ptr addrspace(1), ptr [[B_ADDR]], align 8, !dbg [[DBG152]]
// CHECK1-NEXT:    [[TMP4:%.*]] = addrspacecast ptr addrspace(1) [[TMP3]] to ptr, !dbg [[DBG152]]
// CHECK1-NEXT:    store ptr [[TMP4]], ptr [[_TMP1]], align 8, !dbg [[DBG152]]
// CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[_TMP1]], align 8, !dbg [[DBG152]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP6:%.*]] = load ptr addrspace(1), ptr [[BB_ADDR]], align 8, !dbg [[DBG152]]
// CHECK1-NEXT:    [[TMP7:%.*]] = addrspacecast ptr addrspace(1) [[TMP6]] to ptr, !dbg [[DBG152]]
// CHECK1-NEXT:    store ptr [[TMP7]], ptr [[_TMP2]], align 8, !dbg [[DBG152]]
// CHECK1-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[_TMP2]], align 8, !dbg [[DBG152]], !nonnull [[META29]]
// CHECK1-NEXT:    [[TMP9:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l27_kernel_environment, ptr [[DYN_PTR]]), !dbg [[DBG152]]
// CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP9]], -1, !dbg [[DBG152]]
// CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]], !dbg [[DBG152]]
// CHECK1:       user_code.entry:
// CHECK1-NEXT:    [[TMP10:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB13:[0-9]+]]), !dbg [[DBG153:![0-9]+]]
// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG155:![0-9]+]]
// CHECK1-NEXT:    store i32 [[TMP11]], ptr [[A_CASTED]], align 4, !dbg [[DBG155]]
// CHECK1-NEXT:    [[TMP12:%.*]] = load i64, ptr [[A_CASTED]], align 8, !dbg [[DBG155]]
// CHECK1-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 0, !dbg [[DBG155]]
// CHECK1-NEXT:    store ptr [[TMP2]], ptr [[TMP13]], align 8, !dbg [[DBG155]]
// CHECK1-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 1, !dbg [[DBG155]]
// CHECK1-NEXT:    [[TMP15:%.*]] = inttoptr i64 [[TMP12]] to ptr, !dbg [[DBG155]]
// CHECK1-NEXT:    store ptr [[TMP15]], ptr [[TMP14]], align 8, !dbg [[DBG155]]
// CHECK1-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 2, !dbg [[DBG155]]
// CHECK1-NEXT:    store ptr [[TMP5]], ptr [[TMP16]], align 8, !dbg [[DBG155]]
// CHECK1-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 3, !dbg [[DBG155]]
// CHECK1-NEXT:    store ptr [[TMP8]], ptr [[TMP17]], align 8, !dbg [[DBG155]]
// CHECK1-NEXT:    call void @__kmpc_parallel_60(ptr @[[GLOB13]], i32 [[TMP10]], i32 1, i32 -1, i32 -1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l27_debug___omp_outlined, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 4, i32 0), !dbg [[DBG155]]
// CHECK1-NEXT:    call void @__kmpc_target_deinit(), !dbg [[DBG156:![0-9]+]]
// CHECK1-NEXT:    ret void, !dbg [[DBG157:![0-9]+]]
// CHECK1:       worker.exit:
// CHECK1-NEXT:    ret void, !dbg [[DBG152]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l27
// CHECK1-SAME: (ptr noundef nonnull align 4 dereferenceable(4000) [[C:%.*]], i64 noundef [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(400) [[B:%.*]], ptr noundef nonnull align 1 dereferenceable(1) [[BB:%.*]], ptr noalias noundef [[DYN_PTR:%.*]]) #[[ATTR1]] !dbg [[DBG158:![0-9]+]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[BB_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[C_ADDR]], [[META161:![0-9]+]], !DIExpression(), [[META162:![0-9]+]])
// CHECK1-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[A_ADDR]], [[META163:![0-9]+]], !DIExpression(), [[META162]])
// CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[B_ADDR]], [[META164:![0-9]+]], !DIExpression(), [[META162]])
// CHECK1-NEXT:    store ptr [[BB]], ptr [[BB_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[BB_ADDR]], [[META165:![0-9]+]], !DIExpression(), [[META162]])
// CHECK1-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[DYN_PTR_ADDR]], [[META166:![0-9]+]], !DIExpression(), [[META162]])
// CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 8, !dbg [[DBG167:![0-9]+]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[B_ADDR]], align 8, !dbg [[DBG167]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[BB_ADDR]], align 8, !dbg [[DBG167]], !nonnull [[META29]]
// CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8, !dbg [[DBG167]]
// CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG167]]
// CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[B_ADDR]], align 8, !dbg [[DBG167]]
// CHECK1-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[BB_ADDR]], align 8, !dbg [[DBG167]]
// CHECK1-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[DYN_PTR_ADDR]], align 8, !dbg [[DBG167]]
// CHECK1-NEXT:    [[TMP8:%.*]] = addrspacecast ptr [[TMP3]] to ptr addrspace(1), !dbg [[DBG167]]
// CHECK1-NEXT:    [[TMP9:%.*]] = addrspacecast ptr [[TMP5]] to ptr addrspace(1), !dbg [[DBG167]]
// CHECK1-NEXT:    [[TMP10:%.*]] = addrspacecast ptr [[TMP6]] to ptr addrspace(1), !dbg [[DBG167]]
// CHECK1-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l27_debug__(ptr addrspace(1) [[TMP8]], i32 [[TMP4]], ptr addrspace(1) [[TMP9]], ptr addrspace(1) [[TMP10]], ptr [[TMP7]]) #[[ATTR3]], !dbg [[DBG167]]
// CHECK1-NEXT:    ret void, !dbg [[DBG167]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l27_debug___omp_outlined_debug__
// CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr addrspace(1) noalias noundef [[C:%.*]], i32 noundef [[A:%.*]], ptr addrspace(1) noalias noundef [[B:%.*]], ptr addrspace(1) noalias noundef [[BB:%.*]]) #[[ATTR0]] !dbg [[DBG168:![0-9]+]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr addrspace(1), align 8
// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr addrspace(1), align 8
// CHECK1-NEXT:    [[BB_ADDR:%.*]] = alloca ptr addrspace(1), align 8
// CHECK1-NEXT:    [[TMP:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[_TMP1:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[_TMP2:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[_TMP3:%.*]] = 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:    [[F:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[G:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[H:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[D:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTGLOBAL_TID__ADDR]], [[META171:![0-9]+]], !DIExpression(), [[META172:![0-9]+]])
// CHECK1-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTBOUND_TID__ADDR]], [[META173:![0-9]+]], !DIExpression(), [[META172]])
// CHECK1-NEXT:    store ptr addrspace(1) [[C]], ptr [[C_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[C_ADDR]], [[META174:![0-9]+]], !DIExpression(), [[META175:![0-9]+]])
// CHECK1-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK1-NEXT:      #dbg_declare(ptr [[A_ADDR]], [[META176:![0-9]+]], !DIExpression(), [[META177:![0-9]+]])
// CHECK1-NEXT:    store ptr addrspace(1) [[B]], ptr [[B_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[B_ADDR]], [[META178:![0-9]+]], !DIExpression(), [[META179:![0-9]+]])
// CHECK1-NEXT:    store ptr addrspace(1) [[BB]], ptr [[BB_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[BB_ADDR]], [[META180:![0-9]+]], !DIExpression(), [[META181:![0-9]+]])
// CHECK1-NEXT:    [[TMP0:%.*]] = load ptr addrspace(1), ptr [[C_ADDR]], align 8, !dbg [[DBG182:![0-9]+]]
// CHECK1-NEXT:    [[TMP1:%.*]] = addrspacecast ptr addrspace(1) [[TMP0]] to ptr, !dbg [[DBG182]]
// CHECK1-NEXT:    store ptr [[TMP1]], ptr [[TMP]], align 8, !dbg [[DBG182]]
// CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP]], align 8, !dbg [[DBG182]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP3:%.*]] = load ptr addrspace(1), ptr [[B_ADDR]], align 8, !dbg [[DBG182]]
// CHECK1-NEXT:    [[TMP4:%.*]] = addrspacecast ptr addrspace(1) [[TMP3]] to ptr, !dbg [[DBG182]]
// CHECK1-NEXT:    store ptr [[TMP4]], ptr [[_TMP1]], align 8, !dbg [[DBG182]]
// CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[_TMP1]], align 8, !dbg [[DBG182]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP6:%.*]] = load ptr addrspace(1), ptr [[BB_ADDR]], align 8, !dbg [[DBG182]]
// CHECK1-NEXT:    [[TMP7:%.*]] = addrspacecast ptr addrspace(1) [[TMP6]] to ptr, !dbg [[DBG182]]
// CHECK1-NEXT:    store ptr [[TMP7]], ptr [[_TMP2]], align 8, !dbg [[DBG182]]
// CHECK1-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[_TMP2]], align 8, !dbg [[DBG182]], !nonnull [[META29]]
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTOMP_IV]], [[META183:![0-9]+]], !DIExpression(), [[META172]])
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTOMP_LB]], [[META184:![0-9]+]], !DIExpression(), [[META172]])
// CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4, !dbg [[DBG185:![0-9]+]]
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTOMP_UB]], [[META186:![0-9]+]], !DIExpression(), [[META172]])
// CHECK1-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4, !dbg [[DBG185]]
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTOMP_STRIDE]], [[META187:![0-9]+]], !DIExpression(), [[META172]])
// CHECK1-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4, !dbg [[DBG185]]
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTOMP_IS_LAST]], [[META188:![0-9]+]], !DIExpression(), [[META172]])
// CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4, !dbg [[DBG185]]
// CHECK1-NEXT:      #dbg_declare(ptr [[I]], [[META189:![0-9]+]], !DIExpression(), [[META172]])
// CHECK1-NEXT:    [[TMP9:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8, !dbg [[DBG182]]
// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, ptr [[TMP9]], align 4, !dbg [[DBG182]]
// CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB10:[0-9]+]], i32 [[TMP10]], i32 33, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1), !dbg [[DBG190:![0-9]+]]
// CHECK1-NEXT:    br label [[OMP_DISPATCH_COND:%.*]], !dbg [[DBG182]]
// CHECK1:       omp.dispatch.cond:
// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !dbg [[DBG185]]
// CHECK1-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP11]], 9, !dbg [[DBG185]]
// CHECK1-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]], !dbg [[DBG185]]
// CHECK1:       cond.true:
// CHECK1-NEXT:    br label [[COND_END:%.*]], !dbg [[DBG185]]
// CHECK1:       cond.false:
// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !dbg [[DBG185]]
// CHECK1-NEXT:    br label [[COND_END]], !dbg [[DBG185]]
// CHECK1:       cond.end:
// CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP12]], [[COND_FALSE]] ], !dbg [[DBG185]]
// CHECK1-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4, !dbg [[DBG185]]
// CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4, !dbg [[DBG185]]
// CHECK1-NEXT:    store i32 [[TMP13]], ptr [[DOTOMP_IV]], align 4, !dbg [[DBG185]]
// CHECK1-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !dbg [[DBG185]]
// CHECK1-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !dbg [[DBG185]]
// CHECK1-NEXT:    [[CMP4:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]], !dbg [[DBG182]]
// CHECK1-NEXT:    br i1 [[CMP4]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]], !dbg [[DBG182]]
// CHECK1:       omp.dispatch.body:
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]], !dbg [[DBG182]]
// CHECK1:       omp.inner.for.cond:
// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !dbg [[DBG185]]
// CHECK1-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !dbg [[DBG185]]
// CHECK1-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP16]], [[TMP17]], !dbg [[DBG182]]
// CHECK1-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]], !dbg [[DBG182]]
// CHECK1:       omp.inner.for.body:
// CHECK1-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !dbg [[DBG185]]
// CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP18]], 1, !dbg [[DBG191:![0-9]+]]
// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]], !dbg [[DBG191]]
// CHECK1-NEXT:    store i32 [[ADD]], ptr [[I]], align 4, !dbg [[DBG191]]
// CHECK1-NEXT:      #dbg_declare(ptr [[F]], [[META192:![0-9]+]], !DIExpression(), [[META194:![0-9]+]])
// CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x [10 x i32]]], ptr [[TMP2]], i64 0, i64 1, !dbg [[DBG195:![0-9]+]]
// CHECK1-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[ARRAYIDX]], i64 0, i64 1, !dbg [[DBG195]]
// CHECK1-NEXT:    [[ARRAYIDX7:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX6]], i64 0, i64 1, !dbg [[DBG195]]
// CHECK1-NEXT:    store ptr [[ARRAYIDX7]], ptr [[F]], align 8, !dbg [[META194]]
// CHECK1-NEXT:      #dbg_declare(ptr [[G]], [[META196:![0-9]+]], !DIExpression(), [[META197:![0-9]+]])
// CHECK1-NEXT:    store ptr [[A_ADDR]], ptr [[G]], align 8, !dbg [[META197]]
// CHECK1-NEXT:      #dbg_declare(ptr [[H]], [[META198:![0-9]+]], !DIExpression(), [[META199:![0-9]+]])
// CHECK1-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[TMP5]], i64 0, i64 1, !dbg [[DBG200:![0-9]+]]
// CHECK1-NEXT:    [[ARRAYIDX9:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX8]], i64 0, i64 1, !dbg [[DBG200]]
// CHECK1-NEXT:    store ptr [[ARRAYIDX9]], ptr [[H]], align 8, !dbg [[META199]]
// CHECK1-NEXT:      #dbg_declare(ptr [[D]], [[META201:![0-9]+]], !DIExpression(), [[META202:![0-9]+]])
// CHECK1-NEXT:    store i32 15, ptr [[D]], align 4, !dbg [[META202]]
// CHECK1-NEXT:    store i32 5, ptr [[A_ADDR]], align 4, !dbg [[DBG203:![0-9]+]]
// CHECK1-NEXT:    [[ARRAYIDX10:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[TMP5]], i64 0, i64 0, !dbg [[DBG204:![0-9]+]]
// CHECK1-NEXT:    [[TMP19:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG205:![0-9]+]]
// CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP19]] to i64, !dbg [[DBG204]]
// CHECK1-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX10]], i64 0, i64 [[IDXPROM]], !dbg [[DBG204]]
// CHECK1-NEXT:    store i32 10, ptr [[ARRAYIDX11]], align 4, !dbg [[DBG206:![0-9]+]]
// CHECK1-NEXT:    [[ARRAYIDX12:%.*]] = getelementptr inbounds [10 x [10 x [10 x i32]]], ptr [[TMP2]], i64 0, i64 0, !dbg [[DBG207:![0-9]+]]
// CHECK1-NEXT:    [[ARRAYIDX13:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[ARRAYIDX12]], i64 0, i64 0, !dbg [[DBG207]]
// CHECK1-NEXT:    [[TMP20:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG208:![0-9]+]]
// CHECK1-NEXT:    [[IDXPROM14:%.*]] = sext i32 [[TMP20]] to i64, !dbg [[DBG207]]
// CHECK1-NEXT:    [[ARRAYIDX15:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX13]], i64 0, i64 [[IDXPROM14]], !dbg [[DBG207]]
// CHECK1-NEXT:    store i32 11, ptr [[ARRAYIDX15]], align 4, !dbg [[DBG209:![0-9]+]]
// CHECK1-NEXT:    [[ARRAYIDX16:%.*]] = getelementptr inbounds [10 x [10 x [10 x i32]]], ptr [[TMP2]], i64 0, i64 0, !dbg [[DBG210:![0-9]+]]
// CHECK1-NEXT:    [[ARRAYIDX17:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[ARRAYIDX16]], i64 0, i64 0, !dbg [[DBG210]]
// CHECK1-NEXT:    [[TMP21:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG211:![0-9]+]]
// CHECK1-NEXT:    [[IDXPROM18:%.*]] = sext i32 [[TMP21]] to i64, !dbg [[DBG210]]
// CHECK1-NEXT:    [[ARRAYIDX19:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX17]], i64 0, i64 [[IDXPROM18]], !dbg [[DBG210]]
// CHECK1-NEXT:    [[TMP22:%.*]] = load i32, ptr [[ARRAYIDX19]], align 4, !dbg [[DBG210]]
// CHECK1-NEXT:    [[ARRAYIDX20:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[TMP5]], i64 0, i64 0, !dbg [[DBG212:![0-9]+]]
// CHECK1-NEXT:    [[TMP23:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG213:![0-9]+]]
// CHECK1-NEXT:    [[IDXPROM21:%.*]] = sext i32 [[TMP23]] to i64, !dbg [[DBG212]]
// CHECK1-NEXT:    [[ARRAYIDX22:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX20]], i64 0, i64 [[IDXPROM21]], !dbg [[DBG212]]
// CHECK1-NEXT:    store i32 [[TMP22]], ptr [[ARRAYIDX22]], align 4, !dbg [[DBG214:![0-9]+]]
// CHECK1-NEXT:    [[TMP24:%.*]] = load i8, ptr [[TMP8]], align 1, !dbg [[DBG215:![0-9]+]]
// CHECK1-NEXT:    [[LOADEDV:%.*]] = trunc i8 [[TMP24]] to i1, !dbg [[DBG215]]
// CHECK1-NEXT:    [[CONV:%.*]] = zext i1 [[LOADEDV]] to i32, !dbg [[DBG215]]
// CHECK1-NEXT:    store i32 [[CONV]], ptr [[D]], align 4, !dbg [[DBG216:![0-9]+]]
// CHECK1-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]], !dbg [[DBG217:![0-9]+]]
// CHECK1:       omp.body.continue:
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]], !dbg [[DBG190]]
// CHECK1:       omp.inner.for.inc:
// CHECK1-NEXT:    [[TMP25:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !dbg [[DBG185]]
// CHECK1-NEXT:    [[ADD23:%.*]] = add nsw i32 [[TMP25]], 1, !dbg [[DBG182]]
// CHECK1-NEXT:    store i32 [[ADD23]], ptr [[DOTOMP_IV]], align 4, !dbg [[DBG182]]
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]], !dbg [[DBG190]], !llvm.loop [[LOOP218:![0-9]+]]
// CHECK1:       omp.inner.for.end:
// CHECK1-NEXT:    br label [[OMP_DISPATCH_INC:%.*]], !dbg [[DBG190]]
// CHECK1:       omp.dispatch.inc:
// CHECK1-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4, !dbg [[DBG185]]
// CHECK1-NEXT:    [[TMP27:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !dbg [[DBG185]]
// CHECK1-NEXT:    [[ADD24:%.*]] = add nsw i32 [[TMP26]], [[TMP27]], !dbg [[DBG182]]
// CHECK1-NEXT:    store i32 [[ADD24]], ptr [[DOTOMP_LB]], align 4, !dbg [[DBG182]]
// CHECK1-NEXT:    [[TMP28:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !dbg [[DBG185]]
// CHECK1-NEXT:    [[TMP29:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !dbg [[DBG185]]
// CHECK1-NEXT:    [[ADD25:%.*]] = add nsw i32 [[TMP28]], [[TMP29]], !dbg [[DBG182]]
// CHECK1-NEXT:    store i32 [[ADD25]], ptr [[DOTOMP_UB]], align 4, !dbg [[DBG182]]
// CHECK1-NEXT:    br label [[OMP_DISPATCH_COND]], !dbg [[DBG190]], !llvm.loop [[LOOP220:![0-9]+]]
// CHECK1:       omp.dispatch.end:
// CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB12:[0-9]+]], i32 [[TMP10]]), !dbg [[DBG219:![0-9]+]]
// CHECK1-NEXT:    ret void, !dbg [[DBG221:![0-9]+]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l27_debug___omp_outlined
// CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4000) [[C:%.*]], i64 noundef [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(400) [[B:%.*]], ptr noundef nonnull align 1 dereferenceable(1) [[BB:%.*]]) #[[ATTR0]] !dbg [[DBG222:![0-9]+]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[BB_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTGLOBAL_TID__ADDR]], [[META223:![0-9]+]], !DIExpression(), [[META224:![0-9]+]])
// CHECK1-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTBOUND_TID__ADDR]], [[META225:![0-9]+]], !DIExpression(), [[META224]])
// CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[C_ADDR]], [[META226:![0-9]+]], !DIExpression(), [[META224]])
// CHECK1-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[A_ADDR]], [[META227:![0-9]+]], !DIExpression(), [[META224]])
// CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[B_ADDR]], [[META228:![0-9]+]], !DIExpression(), [[META224]])
// CHECK1-NEXT:    store ptr [[BB]], ptr [[BB_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[BB_ADDR]], [[META229:![0-9]+]], !DIExpression(), [[META224]])
// CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 8, !dbg [[DBG230:![0-9]+]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[B_ADDR]], align 8, !dbg [[DBG230]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[BB_ADDR]], align 8, !dbg [[DBG230]], !nonnull [[META29]]
// CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8, !dbg [[DBG230]]
// CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTBOUND_TID__ADDR]], align 8, !dbg [[DBG230]]
// CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[C_ADDR]], align 8, !dbg [[DBG230]]
// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG230]]
// CHECK1-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[B_ADDR]], align 8, !dbg [[DBG230]]
// CHECK1-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[BB_ADDR]], align 8, !dbg [[DBG230]]
// CHECK1-NEXT:    [[TMP9:%.*]] = addrspacecast ptr [[TMP5]] to ptr addrspace(1), !dbg [[DBG230]]
// CHECK1-NEXT:    [[TMP10:%.*]] = addrspacecast ptr [[TMP7]] to ptr addrspace(1), !dbg [[DBG230]]
// CHECK1-NEXT:    [[TMP11:%.*]] = addrspacecast ptr [[TMP8]] to ptr addrspace(1), !dbg [[DBG230]]
// CHECK1-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l27_debug___omp_outlined_debug__(ptr [[TMP3]], ptr [[TMP4]], ptr addrspace(1) [[TMP9]], i32 [[TMP6]], ptr addrspace(1) [[TMP10]], ptr addrspace(1) [[TMP11]]) #[[ATTR3]], !dbg [[DBG230]]
// CHECK1-NEXT:    ret void, !dbg [[DBG230]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l41_debug__
// CHECK1-SAME: (ptr addrspace(1) noalias noundef [[C:%.*]], ptr addrspace(1) noalias noundef [[A:%.*]], ptr addrspace(1) noalias noundef [[B:%.*]], ptr addrspace(1) noalias noundef [[BB:%.*]], ptr noalias noundef [[DYN_PTR:%.*]]) #[[ATTR0]] !dbg [[DBG231:![0-9]+]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr addrspace(1), align 8
// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr addrspace(1), align 8
// CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr addrspace(1), align 8
// CHECK1-NEXT:    [[BB_ADDR:%.*]] = alloca ptr addrspace(1), align 8
// CHECK1-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[TMP:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[_TMP1:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[_TMP2:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[_TMP3:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [4 x ptr], align 8
// CHECK1-NEXT:    store ptr addrspace(1) [[C]], ptr [[C_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[C_ADDR]], [[META236:![0-9]+]], !DIExpression(), [[META237:![0-9]+]])
// CHECK1-NEXT:    store ptr addrspace(1) [[A]], ptr [[A_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[A_ADDR]], [[META238:![0-9]+]], !DIExpression(), [[META239:![0-9]+]])
// CHECK1-NEXT:    store ptr addrspace(1) [[B]], ptr [[B_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[B_ADDR]], [[META240:![0-9]+]], !DIExpression(), [[META241:![0-9]+]])
// CHECK1-NEXT:    store ptr addrspace(1) [[BB]], ptr [[BB_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[BB_ADDR]], [[META242:![0-9]+]], !DIExpression(), [[META243:![0-9]+]])
// CHECK1-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[DYN_PTR_ADDR]], [[META244:![0-9]+]], !DIExpression(), [[META245:![0-9]+]])
// CHECK1-NEXT:    [[TMP0:%.*]] = load ptr addrspace(1), ptr [[C_ADDR]], align 8, !dbg [[DBG246:![0-9]+]]
// CHECK1-NEXT:    [[TMP1:%.*]] = addrspacecast ptr addrspace(1) [[TMP0]] to ptr, !dbg [[DBG246]]
// CHECK1-NEXT:    store ptr [[TMP1]], ptr [[TMP]], align 8, !dbg [[DBG246]]
// CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP]], align 8, !dbg [[DBG246]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP3:%.*]] = load ptr addrspace(1), ptr [[A_ADDR]], align 8, !dbg [[DBG246]]
// CHECK1-NEXT:    [[TMP4:%.*]] = addrspacecast ptr addrspace(1) [[TMP3]] to ptr, !dbg [[DBG246]]
// CHECK1-NEXT:    store ptr [[TMP4]], ptr [[_TMP1]], align 8, !dbg [[DBG246]]
// CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[_TMP1]], align 8, !dbg [[DBG246]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP6:%.*]] = load ptr addrspace(1), ptr [[B_ADDR]], align 8, !dbg [[DBG246]]
// CHECK1-NEXT:    [[TMP7:%.*]] = addrspacecast ptr addrspace(1) [[TMP6]] to ptr, !dbg [[DBG246]]
// CHECK1-NEXT:    store ptr [[TMP7]], ptr [[_TMP2]], align 8, !dbg [[DBG246]]
// CHECK1-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[_TMP2]], align 8, !dbg [[DBG246]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP9:%.*]] = load ptr addrspace(1), ptr [[BB_ADDR]], align 8, !dbg [[DBG246]]
// CHECK1-NEXT:    [[TMP10:%.*]] = addrspacecast ptr addrspace(1) [[TMP9]] to ptr, !dbg [[DBG246]]
// CHECK1-NEXT:    store ptr [[TMP10]], ptr [[_TMP3]], align 8, !dbg [[DBG246]]
// CHECK1-NEXT:    [[TMP11:%.*]] = load ptr, ptr [[_TMP3]], align 8, !dbg [[DBG246]], !nonnull [[META29]]
// CHECK1-NEXT:    [[TMP12:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l41_kernel_environment, ptr [[DYN_PTR]]), !dbg [[DBG246]]
// CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP12]], -1, !dbg [[DBG246]]
// CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]], !dbg [[DBG246]]
// CHECK1:       user_code.entry:
// CHECK1-NEXT:    [[TMP13:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB20:[0-9]+]]), !dbg [[DBG247:![0-9]+]]
// CHECK1-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 0, !dbg [[DBG249:![0-9]+]]
// CHECK1-NEXT:    store ptr [[TMP2]], ptr [[TMP14]], align 8, !dbg [[DBG249]]
// CHECK1-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 1, !dbg [[DBG249]]
// CHECK1-NEXT:    store ptr [[TMP5]], ptr [[TMP15]], align 8, !dbg [[DBG249]]
// CHECK1-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 2, !dbg [[DBG249]]
// CHECK1-NEXT:    store ptr [[TMP8]], ptr [[TMP16]], align 8, !dbg [[DBG249]]
// CHECK1-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 3, !dbg [[DBG249]]
// CHECK1-NEXT:    store ptr [[TMP11]], ptr [[TMP17]], align 8, !dbg [[DBG249]]
// CHECK1-NEXT:    call void @__kmpc_parallel_60(ptr @[[GLOB20]], i32 [[TMP13]], i32 1, i32 -1, i32 -1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l41_debug___omp_outlined, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 4, i32 0), !dbg [[DBG249]]
// CHECK1-NEXT:    call void @__kmpc_target_deinit(), !dbg [[DBG250:![0-9]+]]
// CHECK1-NEXT:    ret void, !dbg [[DBG251:![0-9]+]]
// CHECK1:       worker.exit:
// CHECK1-NEXT:    ret void, !dbg [[DBG246]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l41
// CHECK1-SAME: (ptr noundef nonnull align 4 dereferenceable(4000) [[C:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(400) [[B:%.*]], ptr noundef nonnull align 1 dereferenceable(1) [[BB:%.*]], ptr noalias noundef [[DYN_PTR:%.*]]) #[[ATTR1]] !dbg [[DBG252:![0-9]+]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[BB_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[C_ADDR]], [[META255:![0-9]+]], !DIExpression(), [[META256:![0-9]+]])
// CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[A_ADDR]], [[META257:![0-9]+]], !DIExpression(), [[META256]])
// CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[B_ADDR]], [[META258:![0-9]+]], !DIExpression(), [[META256]])
// CHECK1-NEXT:    store ptr [[BB]], ptr [[BB_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[BB_ADDR]], [[META259:![0-9]+]], !DIExpression(), [[META256]])
// CHECK1-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[DYN_PTR_ADDR]], [[META260:![0-9]+]], !DIExpression(), [[META256]])
// CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 8, !dbg [[DBG261:![0-9]+]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8, !dbg [[DBG261]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8, !dbg [[DBG261]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[BB_ADDR]], align 8, !dbg [[DBG261]], !nonnull [[META29]]
// CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[C_ADDR]], align 8, !dbg [[DBG261]]
// CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[A_ADDR]], align 8, !dbg [[DBG261]]
// CHECK1-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[B_ADDR]], align 8, !dbg [[DBG261]]
// CHECK1-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[BB_ADDR]], align 8, !dbg [[DBG261]]
// CHECK1-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[DYN_PTR_ADDR]], align 8, !dbg [[DBG261]]
// CHECK1-NEXT:    [[TMP9:%.*]] = addrspacecast ptr [[TMP4]] to ptr addrspace(1), !dbg [[DBG261]]
// CHECK1-NEXT:    [[TMP10:%.*]] = addrspacecast ptr [[TMP5]] to ptr addrspace(1), !dbg [[DBG261]]
// CHECK1-NEXT:    [[TMP11:%.*]] = addrspacecast ptr [[TMP6]] to ptr addrspace(1), !dbg [[DBG261]]
// CHECK1-NEXT:    [[TMP12:%.*]] = addrspacecast ptr [[TMP7]] to ptr addrspace(1), !dbg [[DBG261]]
// CHECK1-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l41_debug__(ptr addrspace(1) [[TMP9]], ptr addrspace(1) [[TMP10]], ptr addrspace(1) [[TMP11]], ptr addrspace(1) [[TMP12]], ptr [[TMP8]]) #[[ATTR3]], !dbg [[DBG261]]
// CHECK1-NEXT:    ret void, !dbg [[DBG261]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l41_debug___omp_outlined_debug__
// CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr addrspace(1) noalias noundef [[C:%.*]], ptr addrspace(1) noalias noundef [[A:%.*]], ptr addrspace(1) noalias noundef [[B:%.*]], ptr addrspace(1) noalias noundef [[BB:%.*]]) #[[ATTR0]] !dbg [[DBG262:![0-9]+]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr addrspace(1), align 8
// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr addrspace(1), align 8
// CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr addrspace(1), align 8
// CHECK1-NEXT:    [[BB_ADDR:%.*]] = alloca ptr addrspace(1), align 8
// CHECK1-NEXT:    [[TMP:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[_TMP1:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[_TMP2:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[_TMP3:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[_TMP4:%.*]] = 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:    [[F:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[G:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[H:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[D:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTGLOBAL_TID__ADDR]], [[META265:![0-9]+]], !DIExpression(), [[META266:![0-9]+]])
// CHECK1-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTBOUND_TID__ADDR]], [[META267:![0-9]+]], !DIExpression(), [[META266]])
// CHECK1-NEXT:    store ptr addrspace(1) [[C]], ptr [[C_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[C_ADDR]], [[META268:![0-9]+]], !DIExpression(), [[META269:![0-9]+]])
// CHECK1-NEXT:    store ptr addrspace(1) [[A]], ptr [[A_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[A_ADDR]], [[META270:![0-9]+]], !DIExpression(), [[META271:![0-9]+]])
// CHECK1-NEXT:    store ptr addrspace(1) [[B]], ptr [[B_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[B_ADDR]], [[META272:![0-9]+]], !DIExpression(), [[META273:![0-9]+]])
// CHECK1-NEXT:    store ptr addrspace(1) [[BB]], ptr [[BB_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[BB_ADDR]], [[META274:![0-9]+]], !DIExpression(), [[META275:![0-9]+]])
// CHECK1-NEXT:    [[TMP0:%.*]] = load ptr addrspace(1), ptr [[C_ADDR]], align 8, !dbg [[DBG276:![0-9]+]]
// CHECK1-NEXT:    [[TMP1:%.*]] = addrspacecast ptr addrspace(1) [[TMP0]] to ptr, !dbg [[DBG276]]
// CHECK1-NEXT:    store ptr [[TMP1]], ptr [[TMP]], align 8, !dbg [[DBG276]]
// CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP]], align 8, !dbg [[DBG276]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP3:%.*]] = load ptr addrspace(1), ptr [[A_ADDR]], align 8, !dbg [[DBG276]]
// CHECK1-NEXT:    [[TMP4:%.*]] = addrspacecast ptr addrspace(1) [[TMP3]] to ptr, !dbg [[DBG276]]
// CHECK1-NEXT:    store ptr [[TMP4]], ptr [[_TMP1]], align 8, !dbg [[DBG276]]
// CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[_TMP1]], align 8, !dbg [[DBG276]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP6:%.*]] = load ptr addrspace(1), ptr [[B_ADDR]], align 8, !dbg [[DBG276]]
// CHECK1-NEXT:    [[TMP7:%.*]] = addrspacecast ptr addrspace(1) [[TMP6]] to ptr, !dbg [[DBG276]]
// CHECK1-NEXT:    store ptr [[TMP7]], ptr [[_TMP2]], align 8, !dbg [[DBG276]]
// CHECK1-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[_TMP2]], align 8, !dbg [[DBG276]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP9:%.*]] = load ptr addrspace(1), ptr [[BB_ADDR]], align 8, !dbg [[DBG276]]
// CHECK1-NEXT:    [[TMP10:%.*]] = addrspacecast ptr addrspace(1) [[TMP9]] to ptr, !dbg [[DBG276]]
// CHECK1-NEXT:    store ptr [[TMP10]], ptr [[_TMP3]], align 8, !dbg [[DBG276]]
// CHECK1-NEXT:    [[TMP11:%.*]] = load ptr, ptr [[_TMP3]], align 8, !dbg [[DBG276]], !nonnull [[META29]]
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTOMP_IV]], [[META277:![0-9]+]], !DIExpression(), [[META266]])
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTOMP_LB]], [[META278:![0-9]+]], !DIExpression(), [[META266]])
// CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4, !dbg [[DBG279:![0-9]+]]
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTOMP_UB]], [[META280:![0-9]+]], !DIExpression(), [[META266]])
// CHECK1-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4, !dbg [[DBG279]]
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTOMP_STRIDE]], [[META281:![0-9]+]], !DIExpression(), [[META266]])
// CHECK1-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4, !dbg [[DBG279]]
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTOMP_IS_LAST]], [[META282:![0-9]+]], !DIExpression(), [[META266]])
// CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4, !dbg [[DBG279]]
// CHECK1-NEXT:      #dbg_declare(ptr [[I]], [[META283:![0-9]+]], !DIExpression(), [[META266]])
// CHECK1-NEXT:    [[TMP12:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8, !dbg [[DBG276]]
// CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[TMP12]], align 4, !dbg [[DBG276]]
// CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB17:[0-9]+]], i32 [[TMP13]], i32 33, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1), !dbg [[DBG284:![0-9]+]]
// CHECK1-NEXT:    br label [[OMP_DISPATCH_COND:%.*]], !dbg [[DBG276]]
// CHECK1:       omp.dispatch.cond:
// CHECK1-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !dbg [[DBG279]]
// CHECK1-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP14]], 9, !dbg [[DBG279]]
// CHECK1-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]], !dbg [[DBG279]]
// CHECK1:       cond.true:
// CHECK1-NEXT:    br label [[COND_END:%.*]], !dbg [[DBG279]]
// CHECK1:       cond.false:
// CHECK1-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !dbg [[DBG279]]
// CHECK1-NEXT:    br label [[COND_END]], !dbg [[DBG279]]
// CHECK1:       cond.end:
// CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ], !dbg [[DBG279]]
// CHECK1-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4, !dbg [[DBG279]]
// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4, !dbg [[DBG279]]
// CHECK1-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4, !dbg [[DBG279]]
// CHECK1-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !dbg [[DBG279]]
// CHECK1-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !dbg [[DBG279]]
// CHECK1-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]], !dbg [[DBG276]]
// CHECK1-NEXT:    br i1 [[CMP5]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]], !dbg [[DBG276]]
// CHECK1:       omp.dispatch.body:
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]], !dbg [[DBG276]]
// CHECK1:       omp.inner.for.cond:
// CHECK1-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !dbg [[DBG279]]
// CHECK1-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !dbg [[DBG279]]
// CHECK1-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP19]], [[TMP20]], !dbg [[DBG276]]
// CHECK1-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]], !dbg [[DBG276]]
// CHECK1:       omp.inner.for.body:
// CHECK1-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !dbg [[DBG279]]
// CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP21]], 1, !dbg [[DBG285:![0-9]+]]
// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]], !dbg [[DBG285]]
// CHECK1-NEXT:    store i32 [[ADD]], ptr [[I]], align 4, !dbg [[DBG285]]
// CHECK1-NEXT:      #dbg_declare(ptr [[F]], [[META286:![0-9]+]], !DIExpression(), [[META288:![0-9]+]])
// CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x [10 x [10 x i32]]], ptr [[TMP2]], i64 0, i64 1, !dbg [[DBG289:![0-9]+]]
// CHECK1-NEXT:    [[ARRAYIDX7:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[ARRAYIDX]], i64 0, i64 1, !dbg [[DBG289]]
// CHECK1-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX7]], i64 0, i64 1, !dbg [[DBG289]]
// CHECK1-NEXT:    store ptr [[ARRAYIDX8]], ptr [[F]], align 8, !dbg [[META288]]
// CHECK1-NEXT:      #dbg_declare(ptr [[G]], [[META290:![0-9]+]], !DIExpression(), [[META291:![0-9]+]])
// CHECK1-NEXT:    store ptr [[TMP5]], ptr [[G]], align 8, !dbg [[META291]]
// CHECK1-NEXT:      #dbg_declare(ptr [[H]], [[META292:![0-9]+]], !DIExpression(), [[META293:![0-9]+]])
// CHECK1-NEXT:    [[ARRAYIDX9:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[TMP8]], i64 0, i64 1, !dbg [[DBG294:![0-9]+]]
// CHECK1-NEXT:    [[ARRAYIDX10:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX9]], i64 0, i64 1, !dbg [[DBG294]]
// CHECK1-NEXT:    store ptr [[ARRAYIDX10]], ptr [[H]], align 8, !dbg [[META293]]
// CHECK1-NEXT:      #dbg_declare(ptr [[D]], [[META295:![0-9]+]], !DIExpression(), [[META296:![0-9]+]])
// CHECK1-NEXT:    store i32 15, ptr [[D]], align 4, !dbg [[META296]]
// CHECK1-NEXT:    store i32 5, ptr [[TMP5]], align 4, !dbg [[DBG297:![0-9]+]]
// CHECK1-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[TMP8]], i64 0, i64 0, !dbg [[DBG298:![0-9]+]]
// CHECK1-NEXT:    [[TMP22:%.*]] = load i32, ptr [[TMP5]], align 4, !dbg [[DBG299:![0-9]+]]
// CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP22]] to i64, !dbg [[DBG298]]
// CHECK1-NEXT:    [[ARRAYIDX12:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX11]], i64 0, i64 [[IDXPROM]], !dbg [[DBG298]]
// CHECK1-NEXT:    store i32 10, ptr [[ARRAYIDX12]], align 4, !dbg [[DBG300:![0-9]+]]
// CHECK1-NEXT:    [[ARRAYIDX13:%.*]] = getelementptr inbounds [10 x [10 x [10 x i32]]], ptr [[TMP2]], i64 0, i64 0, !dbg [[DBG301:![0-9]+]]
// CHECK1-NEXT:    [[ARRAYIDX14:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[ARRAYIDX13]], i64 0, i64 0, !dbg [[DBG301]]
// CHECK1-NEXT:    [[TMP23:%.*]] = load i32, ptr [[TMP5]], align 4, !dbg [[DBG302:![0-9]+]]
// CHECK1-NEXT:    [[IDXPROM15:%.*]] = sext i32 [[TMP23]] to i64, !dbg [[DBG301]]
// CHECK1-NEXT:    [[ARRAYIDX16:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX14]], i64 0, i64 [[IDXPROM15]], !dbg [[DBG301]]
// CHECK1-NEXT:    store i32 11, ptr [[ARRAYIDX16]], align 4, !dbg [[DBG303:![0-9]+]]
// CHECK1-NEXT:    [[ARRAYIDX17:%.*]] = getelementptr inbounds [10 x [10 x [10 x i32]]], ptr [[TMP2]], i64 0, i64 0, !dbg [[DBG304:![0-9]+]]
// CHECK1-NEXT:    [[ARRAYIDX18:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[ARRAYIDX17]], i64 0, i64 0, !dbg [[DBG304]]
// CHECK1-NEXT:    [[TMP24:%.*]] = load i32, ptr [[TMP5]], align 4, !dbg [[DBG305:![0-9]+]]
// CHECK1-NEXT:    [[IDXPROM19:%.*]] = sext i32 [[TMP24]] to i64, !dbg [[DBG304]]
// CHECK1-NEXT:    [[ARRAYIDX20:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX18]], i64 0, i64 [[IDXPROM19]], !dbg [[DBG304]]
// CHECK1-NEXT:    [[TMP25:%.*]] = load i32, ptr [[ARRAYIDX20]], align 4, !dbg [[DBG304]]
// CHECK1-NEXT:    [[ARRAYIDX21:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[TMP8]], i64 0, i64 0, !dbg [[DBG306:![0-9]+]]
// CHECK1-NEXT:    [[TMP26:%.*]] = load i32, ptr [[TMP5]], align 4, !dbg [[DBG307:![0-9]+]]
// CHECK1-NEXT:    [[IDXPROM22:%.*]] = sext i32 [[TMP26]] to i64, !dbg [[DBG306]]
// CHECK1-NEXT:    [[ARRAYIDX23:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX21]], i64 0, i64 [[IDXPROM22]], !dbg [[DBG306]]
// CHECK1-NEXT:    store i32 [[TMP25]], ptr [[ARRAYIDX23]], align 4, !dbg [[DBG308:![0-9]+]]
// CHECK1-NEXT:    [[ARRAYIDX24:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[TMP8]], i64 0, i64 0, !dbg [[DBG309:![0-9]+]]
// CHECK1-NEXT:    [[TMP27:%.*]] = load i32, ptr [[TMP5]], align 4, !dbg [[DBG310:![0-9]+]]
// CHECK1-NEXT:    [[IDXPROM25:%.*]] = sext i32 [[TMP27]] to i64, !dbg [[DBG309]]
// CHECK1-NEXT:    [[ARRAYIDX26:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX24]], i64 0, i64 [[IDXPROM25]], !dbg [[DBG309]]
// CHECK1-NEXT:    [[TMP28:%.*]] = load i32, ptr [[ARRAYIDX26]], align 4, !dbg [[DBG309]]
// CHECK1-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP28]], 0, !dbg [[DBG309]]
// CHECK1-NEXT:    [[STOREDV:%.*]] = zext i1 [[TOBOOL]] to i8, !dbg [[DBG311:![0-9]+]]
// CHECK1-NEXT:    store i8 [[STOREDV]], ptr [[TMP11]], align 1, !dbg [[DBG311]]
// CHECK1-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]], !dbg [[DBG312:![0-9]+]]
// CHECK1:       omp.body.continue:
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]], !dbg [[DBG284]]
// CHECK1:       omp.inner.for.inc:
// CHECK1-NEXT:    [[TMP29:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !dbg [[DBG279]]
// CHECK1-NEXT:    [[ADD27:%.*]] = add nsw i32 [[TMP29]], 1, !dbg [[DBG276]]
// CHECK1-NEXT:    store i32 [[ADD27]], ptr [[DOTOMP_IV]], align 4, !dbg [[DBG276]]
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]], !dbg [[DBG284]], !llvm.loop [[LOOP313:![0-9]+]]
// CHECK1:       omp.inner.for.end:
// CHECK1-NEXT:    br label [[OMP_DISPATCH_INC:%.*]], !dbg [[DBG284]]
// CHECK1:       omp.dispatch.inc:
// CHECK1-NEXT:    [[TMP30:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4, !dbg [[DBG279]]
// CHECK1-NEXT:    [[TMP31:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !dbg [[DBG279]]
// CHECK1-NEXT:    [[ADD28:%.*]] = add nsw i32 [[TMP30]], [[TMP31]], !dbg [[DBG276]]
// CHECK1-NEXT:    store i32 [[ADD28]], ptr [[DOTOMP_LB]], align 4, !dbg [[DBG276]]
// CHECK1-NEXT:    [[TMP32:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !dbg [[DBG279]]
// CHECK1-NEXT:    [[TMP33:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !dbg [[DBG279]]
// CHECK1-NEXT:    [[ADD29:%.*]] = add nsw i32 [[TMP32]], [[TMP33]], !dbg [[DBG276]]
// CHECK1-NEXT:    store i32 [[ADD29]], ptr [[DOTOMP_UB]], align 4, !dbg [[DBG276]]
// CHECK1-NEXT:    br label [[OMP_DISPATCH_COND]], !dbg [[DBG284]], !llvm.loop [[LOOP315:![0-9]+]]
// CHECK1:       omp.dispatch.end:
// CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB19:[0-9]+]], i32 [[TMP13]]), !dbg [[DBG314:![0-9]+]]
// CHECK1-NEXT:    ret void, !dbg [[DBG316:![0-9]+]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l41_debug___omp_outlined
// CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4000) [[C:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(400) [[B:%.*]], ptr noundef nonnull align 1 dereferenceable(1) [[BB:%.*]]) #[[ATTR0]] !dbg [[DBG317:![0-9]+]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    [[BB_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTGLOBAL_TID__ADDR]], [[META320:![0-9]+]], !DIExpression(), [[META321:![0-9]+]])
// CHECK1-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[DOTBOUND_TID__ADDR]], [[META322:![0-9]+]], !DIExpression(), [[META321]])
// CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[C_ADDR]], [[META323:![0-9]+]], !DIExpression(), [[META321]])
// CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[A_ADDR]], [[META324:![0-9]+]], !DIExpression(), [[META321]])
// CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[B_ADDR]], [[META325:![0-9]+]], !DIExpression(), [[META321]])
// CHECK1-NEXT:    store ptr [[BB]], ptr [[BB_ADDR]], align 8
// CHECK1-NEXT:      #dbg_declare(ptr [[BB_ADDR]], [[META326:![0-9]+]], !DIExpression(), [[META321]])
// CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 8, !dbg [[DBG327:![0-9]+]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8, !dbg [[DBG327]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8, !dbg [[DBG327]], !nonnull [[META29]], !align [[META43]]
// CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[BB_ADDR]], align 8, !dbg [[DBG327]], !nonnull [[META29]]
// CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8, !dbg [[DBG327]]
// CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTBOUND_TID__ADDR]], align 8, !dbg [[DBG327]]
// CHECK1-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[C_ADDR]], align 8, !dbg [[DBG327]]
// CHECK1-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[A_ADDR]], align 8, !dbg [[DBG327]]
// CHECK1-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[B_ADDR]], align 8, !dbg [[DBG327]]
// CHECK1-NEXT:    [[TMP9:%.*]] = load ptr, ptr [[BB_ADDR]], align 8, !dbg [[DBG327]]
// CHECK1-NEXT:    [[TMP10:%.*]] = addrspacecast ptr [[TMP6]] to ptr addrspace(1), !dbg [[DBG327]]
// CHECK1-NEXT:    [[TMP11:%.*]] = addrspacecast ptr [[TMP7]] to ptr addrspace(1), !dbg [[DBG327]]
// CHECK1-NEXT:    [[TMP12:%.*]] = addrspacecast ptr [[TMP8]] to ptr addrspace(1), !dbg [[DBG327]]
// CHECK1-NEXT:    [[TMP13:%.*]] = addrspacecast ptr [[TMP9]] to ptr addrspace(1), !dbg [[DBG327]]
// CHECK1-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l41_debug___omp_outlined_debug__(ptr [[TMP4]], ptr [[TMP5]], ptr addrspace(1) [[TMP10]], ptr addrspace(1) [[TMP11]], ptr addrspace(1) [[TMP12]], ptr addrspace(1) [[TMP13]]) #[[ATTR3]], !dbg [[DBG327]]
// CHECK1-NEXT:    ret void, !dbg [[DBG327]]
//