// 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 -verify -fopenmp -fopenmp-enable-irbuilder -x c++ -emit-llvm %s -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefixes=CHECK %s
// RUN: %clang_cc1 -fopenmp -fopenmp-enable-irbuilder -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -verify %s -emit-llvm -o - | FileCheck --check-prefixes=CHECK-DEBUG %s

// expected-no-diagnostics

// TODO: Teach the update script to check new functions too.

#ifndef HEADER
#define HEADER

void parallel_for_0(void) {
#pragma omp parallel
  {
#pragma omp for
    for (int i = 0; i < 100; ++i) {
    }
  }
}

void parallel_for_1(float *r, int a, double b) {
#pragma omp parallel
  {
#pragma omp parallel
    {
#pragma omp for
      for (int i = 0; i < 100; ++i) {
        *r = a + b;
      }
    }
  }
}

void parallel_for_2(float *r, int a, double b) {
#pragma omp parallel
  {
#pragma omp for
    for (int i = 0; i < 100; ++i)
      *r = a + b;
#pragma omp parallel
    {
#pragma omp for
      for (int i = 0; i < 100; ++i)
        *r = a + b;
#pragma omp parallel
      {
#pragma omp for
        for (int i = 0; i < 100; ++i)
          *r = a + b;
      }
#pragma omp for
      for (int i = 0; i < 100; ++i)
        *r = a + b;
#pragma omp parallel
      {
#pragma omp for
        for (int i = 0; i < 100; ++i)
          *r = a + b;
      }
#pragma omp for
      for (int i = 0; i < 100; ++i)
        *r = a + b;
    }
#pragma omp for
    for (int i = 0; i < 100; ++i)
      *r = a + b;
  }
#pragma omp for
  for (int i = 0; i < 100; ++i)
    *r = a + b;
}

#endif
// CHECK-LABEL: define {{[^@]+}}@_Z14parallel_for_0v
// CHECK-SAME: () #[[ATTR0:[0-9]+]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
// CHECK-NEXT:    br label [[OMP_PARALLEL:%.*]]
// CHECK:       omp_parallel:
// CHECK-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 0, ptr @_Z14parallel_for_0v..omp_par)
// CHECK-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT:%.*]]
// CHECK:       omp.par.outlined.exit:
// CHECK-NEXT:    br label [[OMP_PAR_EXIT_SPLIT:%.*]]
// CHECK:       omp.par.exit.split:
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@_Z14parallel_for_0v..omp_par
// CHECK-SAME: (ptr noalias [[TID_ADDR:%.*]], ptr noalias [[ZERO_ADDR:%.*]]) #[[ATTR1:[0-9]+]] {
// CHECK-NEXT:  omp.par.entry:
// CHECK-NEXT:    [[TID_ADDR_LOCAL:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[TMP0:%.*]] = load i32, ptr [[TID_ADDR]], align 4
// CHECK-NEXT:    store i32 [[TMP0]], ptr [[TID_ADDR_LOCAL]], align 4
// CHECK-NEXT:    [[TID:%.*]] = load i32, ptr [[TID_ADDR_LOCAL]], align 4
// CHECK-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[AGG_CAPTURED:%.*]] = alloca [[STRUCT_ANON:%.*]], align 8
// CHECK-NEXT:    [[AGG_CAPTURED1:%.*]] = alloca [[STRUCT_ANON_0:%.*]], align 4
// CHECK-NEXT:    [[DOTCOUNT_ADDR:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_LASTITER:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_LOWERBOUND:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_UPPERBOUND:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NEXT:    br label [[OMP_PAR_REGION:%.*]]
// CHECK:       omp.par.region:
// CHECK-NEXT:    store i32 0, ptr [[I]], align 4
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 0
// CHECK-NEXT:    store ptr [[I]], ptr [[TMP1]], align 8
// CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_ANON_0]], ptr [[AGG_CAPTURED1]], i32 0, i32 0
// CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[I]], align 4
// CHECK-NEXT:    store i32 [[TMP3]], ptr [[TMP2]], align 4
// CHECK-NEXT:    call void @__captured_stmt(ptr [[DOTCOUNT_ADDR]], ptr [[AGG_CAPTURED]])
// CHECK-NEXT:    [[DOTCOUNT:%.*]] = load i32, ptr [[DOTCOUNT_ADDR]], align 4
// CHECK-NEXT:    br label [[OMP_LOOP_PREHEADER:%.*]]
// CHECK:       omp_loop.preheader:
// CHECK-NEXT:    store i32 0, ptr [[P_LOWERBOUND]], align 4
// CHECK-NEXT:    [[TMP4:%.*]] = sub i32 [[DOTCOUNT]], 1
// CHECK-NEXT:    store i32 [[TMP4]], ptr [[P_UPPERBOUND]], align 4
// CHECK-NEXT:    store i32 1, ptr [[P_STRIDE]], align 4
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    call void @__kmpc_for_static_init_4u(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM2]], i32 34, ptr [[P_LASTITER]], ptr [[P_LOWERBOUND]], ptr [[P_UPPERBOUND]], ptr [[P_STRIDE]], i32 1, i32 0)
// CHECK-NEXT:    [[TMP5:%.*]] = load i32, ptr [[P_LOWERBOUND]], align 4
// CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[P_UPPERBOUND]], align 4
// CHECK-NEXT:    [[TMP7:%.*]] = sub i32 [[TMP6]], [[TMP5]]
// CHECK-NEXT:    [[TMP8:%.*]] = add i32 [[TMP7]], 1
// CHECK-NEXT:    br label [[OMP_LOOP_HEADER:%.*]]
// CHECK:       omp_loop.header:
// CHECK-NEXT:    [[OMP_LOOP_IV:%.*]] = phi i32 [ 0, [[OMP_LOOP_PREHEADER]] ], [ [[OMP_LOOP_NEXT:%.*]], [[OMP_LOOP_INC:%.*]] ]
// CHECK-NEXT:    br label [[OMP_LOOP_COND:%.*]]
// CHECK:       omp_loop.cond:
// CHECK-NEXT:    [[OMP_LOOP_CMP:%.*]] = icmp ult i32 [[OMP_LOOP_IV]], [[TMP8]]
// CHECK-NEXT:    br i1 [[OMP_LOOP_CMP]], label [[OMP_LOOP_BODY:%.*]], label [[OMP_LOOP_EXIT:%.*]]
// CHECK:       omp_loop.exit:
// CHECK-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM2]])
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM3:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2:[0-9]+]], i32 [[OMP_GLOBAL_THREAD_NUM3]])
// CHECK-NEXT:    br label [[OMP_LOOP_AFTER:%.*]]
// CHECK:       omp_loop.after:
// CHECK-NEXT:    br label [[OMP_PAR_REGION_PARALLEL_AFTER:%.*]]
// CHECK:       omp.par.region.parallel.after:
// CHECK-NEXT:    br label [[OMP_PAR_PRE_FINALIZE:%.*]]
// CHECK:       omp.par.pre_finalize:
// CHECK-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT_EXITSTUB:%.*]]
// CHECK:       omp_loop.body:
// CHECK-NEXT:    [[TMP9:%.*]] = add i32 [[OMP_LOOP_IV]], [[TMP5]]
// CHECK-NEXT:    call void @__captured_stmt.1(ptr [[I]], i32 [[TMP9]], ptr [[AGG_CAPTURED1]])
// CHECK-NEXT:    br label [[OMP_LOOP_INC]]
// CHECK:       omp_loop.inc:
// CHECK-NEXT:    [[OMP_LOOP_NEXT]] = add nuw i32 [[OMP_LOOP_IV]], 1
// CHECK-NEXT:    br label [[OMP_LOOP_HEADER]]
// CHECK:       omp.par.outlined.exit.exitStub:
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@__captured_stmt
// CHECK-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[DISTANCE:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR3:[0-9]+]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[DISTANCE_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[DOTSTART:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[DOTSTOP:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[DOTSTEP:%.*]] = alloca i32, align 4
// CHECK-NEXT:    store ptr [[DISTANCE]], ptr [[DISTANCE_ADDR]], align 8
// CHECK-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON:%.*]], ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP1]], align 8
// CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[TMP2]], align 4
// CHECK-NEXT:    store i32 [[TMP3]], ptr [[DOTSTART]], align 4
// CHECK-NEXT:    store i32 100, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    store i32 1, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTSTART]], align 4
// CHECK-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP4]], [[TMP5]]
// CHECK-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK:       cond.true:
// CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTSTART]], align 4
// CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], [[TMP7]]
// CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[SUB1:%.*]] = sub i32 [[TMP8]], 1
// CHECK-NEXT:    [[ADD:%.*]] = add i32 [[SUB]], [[SUB1]]
// CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], [[TMP9]]
// CHECK-NEXT:    br label [[COND_END:%.*]]
// CHECK:       cond.false:
// CHECK-NEXT:    br label [[COND_END]]
// CHECK:       cond.end:
// CHECK-NEXT:    [[COND:%.*]] = phi i32 [ [[DIV]], [[COND_TRUE]] ], [ 0, [[COND_FALSE]] ]
// CHECK-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DISTANCE_ADDR]], align 8
// CHECK-NEXT:    store i32 [[COND]], ptr [[TMP10]], align 4
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@__captured_stmt.1
// CHECK-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[LOOPVAR:%.*]], i32 noundef [[LOGICAL:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[LOOPVAR_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[LOGICAL_ADDR:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    store ptr [[LOOPVAR]], ptr [[LOOPVAR_ADDR]], align 8
// CHECK-NEXT:    store i32 [[LOGICAL]], ptr [[LOGICAL_ADDR]], align 4
// CHECK-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_0:%.*]], ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[LOGICAL_ADDR]], align 4
// CHECK-NEXT:    [[MUL:%.*]] = mul i32 1, [[TMP3]]
// CHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP2]], [[MUL]]
// CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[LOOPVAR_ADDR]], align 8
// CHECK-NEXT:    store i32 [[ADD]], ptr [[TMP4]], align 4
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@_Z14parallel_for_1Pfid
// CHECK-SAME: (ptr noundef [[R:%.*]], i32 noundef [[A:%.*]], double noundef [[B:%.*]]) #[[ATTR0]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[STRUCTARG17:%.*]] = alloca { ptr, ptr, ptr }, align 8
// CHECK-NEXT:    [[R_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[B_ADDR:%.*]] = alloca double, align 8
// CHECK-NEXT:    store ptr [[R]], ptr [[R_ADDR]], align 8
// CHECK-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK-NEXT:    store double [[B]], ptr [[B_ADDR]], align 8
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    br label [[OMP_PARALLEL:%.*]]
// CHECK:       omp_parallel:
// CHECK-NEXT:    [[GEP_A_ADDR18:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG17]], i32 0, i32 0
// CHECK-NEXT:    store ptr [[A_ADDR]], ptr [[GEP_A_ADDR18]], align 8
// CHECK-NEXT:    [[GEP_B_ADDR19:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG17]], i32 0, i32 1
// CHECK-NEXT:    store ptr [[B_ADDR]], ptr [[GEP_B_ADDR19]], align 8
// CHECK-NEXT:    [[GEP_R_ADDR20:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG17]], i32 0, i32 2
// CHECK-NEXT:    store ptr [[R_ADDR]], ptr [[GEP_R_ADDR20]], align 8
// CHECK-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 1, ptr @_Z14parallel_for_1Pfid..omp_par.4, ptr [[STRUCTARG17]])
// CHECK-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT16:%.*]]
// CHECK:       omp.par.outlined.exit16:
// CHECK-NEXT:    br label [[OMP_PAR_EXIT_SPLIT:%.*]]
// CHECK:       omp.par.exit.split:
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@_Z14parallel_for_1Pfid..omp_par.4
// CHECK-SAME: (ptr noalias [[TID_ADDR:%.*]], ptr noalias [[ZERO_ADDR:%.*]], ptr [[TMP0:%.*]]) #[[ATTR1]] {
// CHECK-NEXT:  omp.par.entry:
// CHECK-NEXT:    [[GEP_A_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[LOADGEP_A_ADDR:%.*]] = load ptr, ptr [[GEP_A_ADDR]], align 8
// CHECK-NEXT:    [[GEP_B_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 1
// CHECK-NEXT:    [[LOADGEP_B_ADDR:%.*]] = load ptr, ptr [[GEP_B_ADDR]], align 8
// CHECK-NEXT:    [[GEP_R_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 2
// CHECK-NEXT:    [[LOADGEP_R_ADDR:%.*]] = load ptr, ptr [[GEP_R_ADDR]], align 8
// CHECK-NEXT:    [[STRUCTARG:%.*]] = alloca { ptr, ptr, ptr }, align 8
// CHECK-NEXT:    [[TID_ADDR_LOCAL:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TID_ADDR]], align 4
// CHECK-NEXT:    store i32 [[TMP1]], ptr [[TID_ADDR_LOCAL]], align 4
// CHECK-NEXT:    [[TID:%.*]] = load i32, ptr [[TID_ADDR_LOCAL]], align 4
// CHECK-NEXT:    br label [[OMP_PAR_REGION:%.*]]
// CHECK:       omp.par.region:
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    br label [[OMP_PARALLEL:%.*]]
// CHECK:       omp_parallel:
// CHECK-NEXT:    [[GEP_A_ADDR1:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG]], i32 0, i32 0
// CHECK-NEXT:    store ptr [[LOADGEP_A_ADDR]], ptr [[GEP_A_ADDR1]], align 8
// CHECK-NEXT:    [[GEP_B_ADDR2:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG]], i32 0, i32 1
// CHECK-NEXT:    store ptr [[LOADGEP_B_ADDR]], ptr [[GEP_B_ADDR2]], align 8
// CHECK-NEXT:    [[GEP_R_ADDR3:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG]], i32 0, i32 2
// CHECK-NEXT:    store ptr [[LOADGEP_R_ADDR]], ptr [[GEP_R_ADDR3]], align 8
// CHECK-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 1, ptr @_Z14parallel_for_1Pfid..omp_par, ptr [[STRUCTARG]])
// CHECK-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT:%.*]]
// CHECK:       omp.par.outlined.exit:
// CHECK-NEXT:    br label [[OMP_PAR_EXIT7_SPLIT:%.*]]
// CHECK:       omp.par.exit7.split:
// CHECK-NEXT:    br label [[OMP_PAR_REGION_PARALLEL_AFTER:%.*]]
// CHECK:       omp.par.region.parallel.after:
// CHECK-NEXT:    br label [[OMP_PAR_PRE_FINALIZE:%.*]]
// CHECK:       omp.par.pre_finalize:
// CHECK-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT16_EXITSTUB:%.*]]
// CHECK:       omp.par.outlined.exit16.exitStub:
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@_Z14parallel_for_1Pfid..omp_par
// CHECK-SAME: (ptr noalias [[TID_ADDR2:%.*]], ptr noalias [[ZERO_ADDR3:%.*]], ptr [[TMP0:%.*]]) #[[ATTR1]] {
// CHECK-NEXT:  omp.par.entry4:
// CHECK-NEXT:    [[GEP_A_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[LOADGEP_A_ADDR:%.*]] = load ptr, ptr [[GEP_A_ADDR]], align 8
// CHECK-NEXT:    [[GEP_B_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 1
// CHECK-NEXT:    [[LOADGEP_B_ADDR:%.*]] = load ptr, ptr [[GEP_B_ADDR]], align 8
// CHECK-NEXT:    [[GEP_R_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 2
// CHECK-NEXT:    [[LOADGEP_R_ADDR:%.*]] = load ptr, ptr [[GEP_R_ADDR]], align 8
// CHECK-NEXT:    [[TID_ADDR_LOCAL8:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TID_ADDR2]], align 4
// CHECK-NEXT:    store i32 [[TMP1]], ptr [[TID_ADDR_LOCAL8]], align 4
// CHECK-NEXT:    [[TID9:%.*]] = load i32, ptr [[TID_ADDR_LOCAL8]], align 4
// CHECK-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[AGG_CAPTURED:%.*]] = alloca [[STRUCT_ANON_1:%.*]], align 8
// CHECK-NEXT:    [[AGG_CAPTURED12:%.*]] = alloca [[STRUCT_ANON_2:%.*]], align 4
// CHECK-NEXT:    [[DOTCOUNT_ADDR:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_LASTITER:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_LOWERBOUND:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_UPPERBOUND:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NEXT:    br label [[OMP_PAR_REGION5:%.*]]
// CHECK:       omp.par.region5:
// CHECK-NEXT:    store i32 0, ptr [[I]], align 4
// CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_ANON_1]], ptr [[AGG_CAPTURED]], i32 0, i32 0
// CHECK-NEXT:    store ptr [[I]], ptr [[TMP2]], align 8
// CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [[STRUCT_ANON_2]], ptr [[AGG_CAPTURED12]], i32 0, i32 0
// CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr [[I]], align 4
// CHECK-NEXT:    store i32 [[TMP4]], ptr [[TMP3]], align 4
// CHECK-NEXT:    call void @__captured_stmt.2(ptr [[DOTCOUNT_ADDR]], ptr [[AGG_CAPTURED]])
// CHECK-NEXT:    [[DOTCOUNT:%.*]] = load i32, ptr [[DOTCOUNT_ADDR]], align 4
// CHECK-NEXT:    br label [[OMP_LOOP_PREHEADER:%.*]]
// CHECK:       omp_loop.preheader:
// CHECK-NEXT:    store i32 0, ptr [[P_LOWERBOUND]], align 4
// CHECK-NEXT:    [[TMP5:%.*]] = sub i32 [[DOTCOUNT]], 1
// CHECK-NEXT:    store i32 [[TMP5]], ptr [[P_UPPERBOUND]], align 4
// CHECK-NEXT:    store i32 1, ptr [[P_STRIDE]], align 4
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM14:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    call void @__kmpc_for_static_init_4u(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM14]], i32 34, ptr [[P_LASTITER]], ptr [[P_LOWERBOUND]], ptr [[P_UPPERBOUND]], ptr [[P_STRIDE]], i32 1, i32 0)
// CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[P_LOWERBOUND]], align 4
// CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[P_UPPERBOUND]], align 4
// CHECK-NEXT:    [[TMP8:%.*]] = sub i32 [[TMP7]], [[TMP6]]
// CHECK-NEXT:    [[TMP9:%.*]] = add i32 [[TMP8]], 1
// CHECK-NEXT:    br label [[OMP_LOOP_HEADER:%.*]]
// CHECK:       omp_loop.header:
// CHECK-NEXT:    [[OMP_LOOP_IV:%.*]] = phi i32 [ 0, [[OMP_LOOP_PREHEADER]] ], [ [[OMP_LOOP_NEXT:%.*]], [[OMP_LOOP_INC:%.*]] ]
// CHECK-NEXT:    br label [[OMP_LOOP_COND:%.*]]
// CHECK:       omp_loop.cond:
// CHECK-NEXT:    [[OMP_LOOP_CMP:%.*]] = icmp ult i32 [[OMP_LOOP_IV]], [[TMP9]]
// CHECK-NEXT:    br i1 [[OMP_LOOP_CMP]], label [[OMP_LOOP_BODY:%.*]], label [[OMP_LOOP_EXIT:%.*]]
// CHECK:       omp_loop.exit:
// CHECK-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM14]])
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM15:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[OMP_GLOBAL_THREAD_NUM15]])
// CHECK-NEXT:    br label [[OMP_LOOP_AFTER:%.*]]
// CHECK:       omp_loop.after:
// CHECK-NEXT:    br label [[OMP_PAR_REGION5_PARALLEL_AFTER:%.*]]
// CHECK:       omp.par.region5.parallel.after:
// CHECK-NEXT:    br label [[OMP_PAR_PRE_FINALIZE6:%.*]]
// CHECK:       omp.par.pre_finalize6:
// CHECK-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT_EXITSTUB:%.*]]
// CHECK:       omp_loop.body:
// CHECK-NEXT:    [[TMP10:%.*]] = add i32 [[OMP_LOOP_IV]], [[TMP6]]
// CHECK-NEXT:    call void @__captured_stmt.3(ptr [[I]], i32 [[TMP10]], ptr [[AGG_CAPTURED12]])
// CHECK-NEXT:    [[TMP11:%.*]] = load i32, ptr [[LOADGEP_A_ADDR]], align 4
// CHECK-NEXT:    [[CONV:%.*]] = sitofp i32 [[TMP11]] to double
// CHECK-NEXT:    [[TMP12:%.*]] = load double, ptr [[LOADGEP_B_ADDR]], align 8
// CHECK-NEXT:    [[ADD:%.*]] = fadd double [[CONV]], [[TMP12]]
// CHECK-NEXT:    [[CONV13:%.*]] = fptrunc double [[ADD]] to float
// CHECK-NEXT:    [[TMP13:%.*]] = load ptr, ptr [[LOADGEP_R_ADDR]], align 8
// CHECK-NEXT:    store float [[CONV13]], ptr [[TMP13]], align 4
// CHECK-NEXT:    br label [[OMP_LOOP_INC]]
// CHECK:       omp_loop.inc:
// CHECK-NEXT:    [[OMP_LOOP_NEXT]] = add nuw i32 [[OMP_LOOP_IV]], 1
// CHECK-NEXT:    br label [[OMP_LOOP_HEADER]]
// CHECK:       omp.par.outlined.exit.exitStub:
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@__captured_stmt.2
// CHECK-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[DISTANCE:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[DISTANCE_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[DOTSTART:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[DOTSTOP:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[DOTSTEP:%.*]] = alloca i32, align 4
// CHECK-NEXT:    store ptr [[DISTANCE]], ptr [[DISTANCE_ADDR]], align 8
// CHECK-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_1:%.*]], ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP1]], align 8
// CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[TMP2]], align 4
// CHECK-NEXT:    store i32 [[TMP3]], ptr [[DOTSTART]], align 4
// CHECK-NEXT:    store i32 100, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    store i32 1, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTSTART]], align 4
// CHECK-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP4]], [[TMP5]]
// CHECK-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK:       cond.true:
// CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTSTART]], align 4
// CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], [[TMP7]]
// CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[SUB1:%.*]] = sub i32 [[TMP8]], 1
// CHECK-NEXT:    [[ADD:%.*]] = add i32 [[SUB]], [[SUB1]]
// CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], [[TMP9]]
// CHECK-NEXT:    br label [[COND_END:%.*]]
// CHECK:       cond.false:
// CHECK-NEXT:    br label [[COND_END]]
// CHECK:       cond.end:
// CHECK-NEXT:    [[COND:%.*]] = phi i32 [ [[DIV]], [[COND_TRUE]] ], [ 0, [[COND_FALSE]] ]
// CHECK-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DISTANCE_ADDR]], align 8
// CHECK-NEXT:    store i32 [[COND]], ptr [[TMP10]], align 4
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@__captured_stmt.3
// CHECK-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[LOOPVAR:%.*]], i32 noundef [[LOGICAL:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[LOOPVAR_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[LOGICAL_ADDR:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    store ptr [[LOOPVAR]], ptr [[LOOPVAR_ADDR]], align 8
// CHECK-NEXT:    store i32 [[LOGICAL]], ptr [[LOGICAL_ADDR]], align 4
// CHECK-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_2:%.*]], ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[LOGICAL_ADDR]], align 4
// CHECK-NEXT:    [[MUL:%.*]] = mul i32 1, [[TMP3]]
// CHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP2]], [[MUL]]
// CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[LOOPVAR_ADDR]], align 8
// CHECK-NEXT:    store i32 [[ADD]], ptr [[TMP4]], align 4
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@_Z14parallel_for_2Pfid
// CHECK-SAME: (ptr noundef [[R:%.*]], i32 noundef [[A:%.*]], double noundef [[B:%.*]]) #[[ATTR0]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[STRUCTARG:%.*]] = alloca { ptr, ptr, ptr }, align 8
// CHECK-NEXT:    [[R_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[B_ADDR:%.*]] = alloca double, align 8
// CHECK-NEXT:    [[I185:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[AGG_CAPTURED186:%.*]] = alloca [[STRUCT_ANON_17:%.*]], align 8
// CHECK-NEXT:    [[AGG_CAPTURED187:%.*]] = alloca [[STRUCT_ANON_18:%.*]], align 4
// CHECK-NEXT:    [[DOTCOUNT_ADDR188:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_LASTITER203:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_LOWERBOUND204:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_UPPERBOUND205:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_STRIDE206:%.*]] = alloca i32, align 4
// CHECK-NEXT:    store ptr [[R]], ptr [[R_ADDR]], align 8
// CHECK-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK-NEXT:    store double [[B]], ptr [[B_ADDR]], align 8
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    br label [[OMP_PARALLEL:%.*]]
// CHECK:       omp_parallel:
// CHECK-NEXT:    [[GEP_A_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG]], i32 0, i32 0
// CHECK-NEXT:    store ptr [[A_ADDR]], ptr [[GEP_A_ADDR]], align 8
// CHECK-NEXT:    [[GEP_B_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG]], i32 0, i32 1
// CHECK-NEXT:    store ptr [[B_ADDR]], ptr [[GEP_B_ADDR]], align 8
// CHECK-NEXT:    [[GEP_R_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG]], i32 0, i32 2
// CHECK-NEXT:    store ptr [[R_ADDR]], ptr [[GEP_R_ADDR]], align 8
// CHECK-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 1, ptr @_Z14parallel_for_2Pfid..omp_par.23, ptr [[STRUCTARG]])
// CHECK-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT184:%.*]]
// CHECK:       omp.par.outlined.exit184:
// CHECK-NEXT:    br label [[OMP_PAR_EXIT_SPLIT:%.*]]
// CHECK:       omp.par.exit.split:
// CHECK-NEXT:    store i32 0, ptr [[I185]], align 4
// CHECK-NEXT:    [[TMP0:%.*]] = getelementptr inbounds [[STRUCT_ANON_17]], ptr [[AGG_CAPTURED186]], i32 0, i32 0
// CHECK-NEXT:    store ptr [[I185]], ptr [[TMP0]], align 8
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_18]], ptr [[AGG_CAPTURED187]], i32 0, i32 0
// CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr [[I185]], align 4
// CHECK-NEXT:    store i32 [[TMP2]], ptr [[TMP1]], align 4
// CHECK-NEXT:    call void @__captured_stmt.19(ptr [[DOTCOUNT_ADDR188]], ptr [[AGG_CAPTURED186]])
// CHECK-NEXT:    [[DOTCOUNT189:%.*]] = load i32, ptr [[DOTCOUNT_ADDR188]], align 4
// CHECK-NEXT:    br label [[OMP_LOOP_PREHEADER190:%.*]]
// CHECK:       omp_loop.preheader190:
// CHECK-NEXT:    store i32 0, ptr [[P_LOWERBOUND204]], align 4
// CHECK-NEXT:    [[TMP3:%.*]] = sub i32 [[DOTCOUNT189]], 1
// CHECK-NEXT:    store i32 [[TMP3]], ptr [[P_UPPERBOUND205]], align 4
// CHECK-NEXT:    store i32 1, ptr [[P_STRIDE206]], align 4
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM207:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    call void @__kmpc_for_static_init_4u(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM207]], i32 34, ptr [[P_LASTITER203]], ptr [[P_LOWERBOUND204]], ptr [[P_UPPERBOUND205]], ptr [[P_STRIDE206]], i32 1, i32 0)
// CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr [[P_LOWERBOUND204]], align 4
// CHECK-NEXT:    [[TMP5:%.*]] = load i32, ptr [[P_UPPERBOUND205]], align 4
// CHECK-NEXT:    [[TMP6:%.*]] = sub i32 [[TMP5]], [[TMP4]]
// CHECK-NEXT:    [[TMP7:%.*]] = add i32 [[TMP6]], 1
// CHECK-NEXT:    br label [[OMP_LOOP_HEADER191:%.*]]
// CHECK:       omp_loop.header191:
// CHECK-NEXT:    [[OMP_LOOP_IV197:%.*]] = phi i32 [ 0, [[OMP_LOOP_PREHEADER190]] ], [ [[OMP_LOOP_NEXT199:%.*]], [[OMP_LOOP_INC194:%.*]] ]
// CHECK-NEXT:    br label [[OMP_LOOP_COND192:%.*]]
// CHECK:       omp_loop.cond192:
// CHECK-NEXT:    [[OMP_LOOP_CMP198:%.*]] = icmp ult i32 [[OMP_LOOP_IV197]], [[TMP7]]
// CHECK-NEXT:    br i1 [[OMP_LOOP_CMP198]], label [[OMP_LOOP_BODY193:%.*]], label [[OMP_LOOP_EXIT195:%.*]]
// CHECK:       omp_loop.body193:
// CHECK-NEXT:    [[TMP8:%.*]] = add i32 [[OMP_LOOP_IV197]], [[TMP4]]
// CHECK-NEXT:    call void @__captured_stmt.20(ptr [[I185]], i32 [[TMP8]], ptr [[AGG_CAPTURED187]])
// CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NEXT:    [[CONV200:%.*]] = sitofp i32 [[TMP9]] to double
// CHECK-NEXT:    [[TMP10:%.*]] = load double, ptr [[B_ADDR]], align 8
// CHECK-NEXT:    [[ADD201:%.*]] = fadd double [[CONV200]], [[TMP10]]
// CHECK-NEXT:    [[CONV202:%.*]] = fptrunc double [[ADD201]] to float
// CHECK-NEXT:    [[TMP11:%.*]] = load ptr, ptr [[R_ADDR]], align 8
// CHECK-NEXT:    store float [[CONV202]], ptr [[TMP11]], align 4
// CHECK-NEXT:    br label [[OMP_LOOP_INC194]]
// CHECK:       omp_loop.inc194:
// CHECK-NEXT:    [[OMP_LOOP_NEXT199]] = add nuw i32 [[OMP_LOOP_IV197]], 1
// CHECK-NEXT:    br label [[OMP_LOOP_HEADER191]]
// CHECK:       omp_loop.exit195:
// CHECK-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM207]])
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM208:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[OMP_GLOBAL_THREAD_NUM208]])
// CHECK-NEXT:    br label [[OMP_LOOP_AFTER196:%.*]]
// CHECK:       omp_loop.after196:
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@_Z14parallel_for_2Pfid..omp_par.23
// CHECK-SAME: (ptr noalias [[TID_ADDR:%.*]], ptr noalias [[ZERO_ADDR:%.*]], ptr [[TMP0:%.*]]) #[[ATTR1]] {
// CHECK-NEXT:  omp.par.entry:
// CHECK-NEXT:    [[GEP_A_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[LOADGEP_A_ADDR:%.*]] = load ptr, ptr [[GEP_A_ADDR]], align 8
// CHECK-NEXT:    [[GEP_B_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 1
// CHECK-NEXT:    [[LOADGEP_B_ADDR:%.*]] = load ptr, ptr [[GEP_B_ADDR]], align 8
// CHECK-NEXT:    [[GEP_R_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 2
// CHECK-NEXT:    [[LOADGEP_R_ADDR:%.*]] = load ptr, ptr [[GEP_R_ADDR]], align 8
// CHECK-NEXT:    [[STRUCTARG214:%.*]] = alloca { ptr, ptr, ptr }, align 8
// CHECK-NEXT:    [[TID_ADDR_LOCAL:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TID_ADDR]], align 4
// CHECK-NEXT:    store i32 [[TMP1]], ptr [[TID_ADDR_LOCAL]], align 4
// CHECK-NEXT:    [[TID:%.*]] = load i32, ptr [[TID_ADDR_LOCAL]], align 4
// CHECK-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[AGG_CAPTURED:%.*]] = alloca [[STRUCT_ANON_3:%.*]], align 8
// CHECK-NEXT:    [[AGG_CAPTURED1:%.*]] = alloca [[STRUCT_ANON_4:%.*]], align 4
// CHECK-NEXT:    [[DOTCOUNT_ADDR:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_LASTITER:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_LOWERBOUND:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_UPPERBOUND:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[I160:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[AGG_CAPTURED161:%.*]] = alloca [[STRUCT_ANON_15:%.*]], align 8
// CHECK-NEXT:    [[AGG_CAPTURED162:%.*]] = alloca [[STRUCT_ANON_16:%.*]], align 4
// CHECK-NEXT:    [[DOTCOUNT_ADDR163:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_LASTITER178:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_LOWERBOUND179:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_UPPERBOUND180:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_STRIDE181:%.*]] = alloca i32, align 4
// CHECK-NEXT:    br label [[OMP_PAR_REGION:%.*]]
// CHECK:       omp.par.region:
// CHECK-NEXT:    store i32 0, ptr [[I]], align 4
// CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_ANON_3]], ptr [[AGG_CAPTURED]], i32 0, i32 0
// CHECK-NEXT:    store ptr [[I]], ptr [[TMP2]], align 8
// CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [[STRUCT_ANON_4]], ptr [[AGG_CAPTURED1]], i32 0, i32 0
// CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr [[I]], align 4
// CHECK-NEXT:    store i32 [[TMP4]], ptr [[TMP3]], align 4
// CHECK-NEXT:    call void @__captured_stmt.5(ptr [[DOTCOUNT_ADDR]], ptr [[AGG_CAPTURED]])
// CHECK-NEXT:    [[DOTCOUNT:%.*]] = load i32, ptr [[DOTCOUNT_ADDR]], align 4
// CHECK-NEXT:    br label [[OMP_LOOP_PREHEADER:%.*]]
// CHECK:       omp_loop.preheader:
// CHECK-NEXT:    store i32 0, ptr [[P_LOWERBOUND]], align 4
// CHECK-NEXT:    [[TMP5:%.*]] = sub i32 [[DOTCOUNT]], 1
// CHECK-NEXT:    store i32 [[TMP5]], ptr [[P_UPPERBOUND]], align 4
// CHECK-NEXT:    store i32 1, ptr [[P_STRIDE]], align 4
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM3:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    call void @__kmpc_for_static_init_4u(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM3]], i32 34, ptr [[P_LASTITER]], ptr [[P_LOWERBOUND]], ptr [[P_UPPERBOUND]], ptr [[P_STRIDE]], i32 1, i32 0)
// CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[P_LOWERBOUND]], align 4
// CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[P_UPPERBOUND]], align 4
// CHECK-NEXT:    [[TMP8:%.*]] = sub i32 [[TMP7]], [[TMP6]]
// CHECK-NEXT:    [[TMP9:%.*]] = add i32 [[TMP8]], 1
// CHECK-NEXT:    br label [[OMP_LOOP_HEADER:%.*]]
// CHECK:       omp_loop.header:
// CHECK-NEXT:    [[OMP_LOOP_IV:%.*]] = phi i32 [ 0, [[OMP_LOOP_PREHEADER]] ], [ [[OMP_LOOP_NEXT:%.*]], [[OMP_LOOP_INC:%.*]] ]
// CHECK-NEXT:    br label [[OMP_LOOP_COND:%.*]]
// CHECK:       omp_loop.cond:
// CHECK-NEXT:    [[OMP_LOOP_CMP:%.*]] = icmp ult i32 [[OMP_LOOP_IV]], [[TMP9]]
// CHECK-NEXT:    br i1 [[OMP_LOOP_CMP]], label [[OMP_LOOP_BODY:%.*]], label [[OMP_LOOP_EXIT:%.*]]
// CHECK:       omp_loop.exit:
// CHECK-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM3]])
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM4:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[OMP_GLOBAL_THREAD_NUM4]])
// CHECK-NEXT:    br label [[OMP_LOOP_AFTER:%.*]]
// CHECK:       omp_loop.after:
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM5:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    br label [[OMP_PARALLEL:%.*]]
// CHECK:       omp_parallel:
// CHECK-NEXT:    [[GEP_A_ADDR215:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG214]], i32 0, i32 0
// CHECK-NEXT:    store ptr [[LOADGEP_A_ADDR]], ptr [[GEP_A_ADDR215]], align 8
// CHECK-NEXT:    [[GEP_B_ADDR216:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG214]], i32 0, i32 1
// CHECK-NEXT:    store ptr [[LOADGEP_B_ADDR]], ptr [[GEP_B_ADDR216]], align 8
// CHECK-NEXT:    [[GEP_R_ADDR217:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG214]], i32 0, i32 2
// CHECK-NEXT:    store ptr [[LOADGEP_R_ADDR]], ptr [[GEP_R_ADDR217]], align 8
// CHECK-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 1, ptr @_Z14parallel_for_2Pfid..omp_par.22, ptr [[STRUCTARG214]])
// CHECK-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT159:%.*]]
// CHECK:       omp.par.outlined.exit159:
// CHECK-NEXT:    br label [[OMP_PAR_EXIT11_SPLIT:%.*]]
// CHECK:       omp.par.exit11.split:
// CHECK-NEXT:    store i32 0, ptr [[I160]], align 4
// CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [[STRUCT_ANON_15]], ptr [[AGG_CAPTURED161]], i32 0, i32 0
// CHECK-NEXT:    store ptr [[I160]], ptr [[TMP10]], align 8
// CHECK-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_ANON_16]], ptr [[AGG_CAPTURED162]], i32 0, i32 0
// CHECK-NEXT:    [[TMP12:%.*]] = load i32, ptr [[I160]], align 4
// CHECK-NEXT:    store i32 [[TMP12]], ptr [[TMP11]], align 4
// CHECK-NEXT:    call void @__captured_stmt.17(ptr [[DOTCOUNT_ADDR163]], ptr [[AGG_CAPTURED161]])
// CHECK-NEXT:    [[DOTCOUNT164:%.*]] = load i32, ptr [[DOTCOUNT_ADDR163]], align 4
// CHECK-NEXT:    br label [[OMP_LOOP_PREHEADER165:%.*]]
// CHECK:       omp_loop.preheader165:
// CHECK-NEXT:    store i32 0, ptr [[P_LOWERBOUND179]], align 4
// CHECK-NEXT:    [[TMP13:%.*]] = sub i32 [[DOTCOUNT164]], 1
// CHECK-NEXT:    store i32 [[TMP13]], ptr [[P_UPPERBOUND180]], align 4
// CHECK-NEXT:    store i32 1, ptr [[P_STRIDE181]], align 4
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM182:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    call void @__kmpc_for_static_init_4u(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM182]], i32 34, ptr [[P_LASTITER178]], ptr [[P_LOWERBOUND179]], ptr [[P_UPPERBOUND180]], ptr [[P_STRIDE181]], i32 1, i32 0)
// CHECK-NEXT:    [[TMP14:%.*]] = load i32, ptr [[P_LOWERBOUND179]], align 4
// CHECK-NEXT:    [[TMP15:%.*]] = load i32, ptr [[P_UPPERBOUND180]], align 4
// CHECK-NEXT:    [[TMP16:%.*]] = sub i32 [[TMP15]], [[TMP14]]
// CHECK-NEXT:    [[TMP17:%.*]] = add i32 [[TMP16]], 1
// CHECK-NEXT:    br label [[OMP_LOOP_HEADER166:%.*]]
// CHECK:       omp_loop.header166:
// CHECK-NEXT:    [[OMP_LOOP_IV172:%.*]] = phi i32 [ 0, [[OMP_LOOP_PREHEADER165]] ], [ [[OMP_LOOP_NEXT174:%.*]], [[OMP_LOOP_INC169:%.*]] ]
// CHECK-NEXT:    br label [[OMP_LOOP_COND167:%.*]]
// CHECK:       omp_loop.cond167:
// CHECK-NEXT:    [[OMP_LOOP_CMP173:%.*]] = icmp ult i32 [[OMP_LOOP_IV172]], [[TMP17]]
// CHECK-NEXT:    br i1 [[OMP_LOOP_CMP173]], label [[OMP_LOOP_BODY168:%.*]], label [[OMP_LOOP_EXIT170:%.*]]
// CHECK:       omp_loop.exit170:
// CHECK-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM182]])
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM183:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[OMP_GLOBAL_THREAD_NUM183]])
// CHECK-NEXT:    br label [[OMP_LOOP_AFTER171:%.*]]
// CHECK:       omp_loop.after171:
// CHECK-NEXT:    br label [[OMP_PAR_REGION_PARALLEL_AFTER:%.*]]
// CHECK:       omp.par.region.parallel.after:
// CHECK-NEXT:    br label [[OMP_PAR_PRE_FINALIZE:%.*]]
// CHECK:       omp.par.pre_finalize:
// CHECK-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT184_EXITSTUB:%.*]]
// CHECK:       omp_loop.body168:
// CHECK-NEXT:    [[TMP18:%.*]] = add i32 [[OMP_LOOP_IV172]], [[TMP14]]
// CHECK-NEXT:    call void @__captured_stmt.18(ptr [[I160]], i32 [[TMP18]], ptr [[AGG_CAPTURED162]])
// CHECK-NEXT:    [[TMP19:%.*]] = load i32, ptr [[LOADGEP_A_ADDR]], align 4
// CHECK-NEXT:    [[CONV175:%.*]] = sitofp i32 [[TMP19]] to double
// CHECK-NEXT:    [[TMP20:%.*]] = load double, ptr [[LOADGEP_B_ADDR]], align 8
// CHECK-NEXT:    [[ADD176:%.*]] = fadd double [[CONV175]], [[TMP20]]
// CHECK-NEXT:    [[CONV177:%.*]] = fptrunc double [[ADD176]] to float
// CHECK-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[LOADGEP_R_ADDR]], align 8
// CHECK-NEXT:    store float [[CONV177]], ptr [[TMP21]], align 4
// CHECK-NEXT:    br label [[OMP_LOOP_INC169]]
// CHECK:       omp_loop.inc169:
// CHECK-NEXT:    [[OMP_LOOP_NEXT174]] = add nuw i32 [[OMP_LOOP_IV172]], 1
// CHECK-NEXT:    br label [[OMP_LOOP_HEADER166]]
// CHECK:       omp_loop.body:
// CHECK-NEXT:    [[TMP22:%.*]] = add i32 [[OMP_LOOP_IV]], [[TMP6]]
// CHECK-NEXT:    call void @__captured_stmt.6(ptr [[I]], i32 [[TMP22]], ptr [[AGG_CAPTURED1]])
// CHECK-NEXT:    [[TMP23:%.*]] = load i32, ptr [[LOADGEP_A_ADDR]], align 4
// CHECK-NEXT:    [[CONV:%.*]] = sitofp i32 [[TMP23]] to double
// CHECK-NEXT:    [[TMP24:%.*]] = load double, ptr [[LOADGEP_B_ADDR]], align 8
// CHECK-NEXT:    [[ADD:%.*]] = fadd double [[CONV]], [[TMP24]]
// CHECK-NEXT:    [[CONV2:%.*]] = fptrunc double [[ADD]] to float
// CHECK-NEXT:    [[TMP25:%.*]] = load ptr, ptr [[LOADGEP_R_ADDR]], align 8
// CHECK-NEXT:    store float [[CONV2]], ptr [[TMP25]], align 4
// CHECK-NEXT:    br label [[OMP_LOOP_INC]]
// CHECK:       omp_loop.inc:
// CHECK-NEXT:    [[OMP_LOOP_NEXT]] = add nuw i32 [[OMP_LOOP_IV]], 1
// CHECK-NEXT:    br label [[OMP_LOOP_HEADER]]
// CHECK:       omp.par.outlined.exit184.exitStub:
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@_Z14parallel_for_2Pfid..omp_par.22
// CHECK-SAME: (ptr noalias [[TID_ADDR6:%.*]], ptr noalias [[ZERO_ADDR7:%.*]], ptr [[TMP0:%.*]]) #[[ATTR1]] {
// CHECK-NEXT:  omp.par.entry8:
// CHECK-NEXT:    [[GEP_A_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[LOADGEP_A_ADDR:%.*]] = load ptr, ptr [[GEP_A_ADDR]], align 8
// CHECK-NEXT:    [[GEP_B_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 1
// CHECK-NEXT:    [[LOADGEP_B_ADDR:%.*]] = load ptr, ptr [[GEP_B_ADDR]], align 8
// CHECK-NEXT:    [[GEP_R_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 2
// CHECK-NEXT:    [[LOADGEP_R_ADDR:%.*]] = load ptr, ptr [[GEP_R_ADDR]], align 8
// CHECK-NEXT:    [[STRUCTARG209:%.*]] = alloca { ptr, ptr, ptr }, align 8
// CHECK-NEXT:    [[STRUCTARG:%.*]] = alloca { ptr, ptr, ptr }, align 8
// CHECK-NEXT:    [[TID_ADDR_LOCAL12:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TID_ADDR6]], align 4
// CHECK-NEXT:    store i32 [[TMP1]], ptr [[TID_ADDR_LOCAL12]], align 4
// CHECK-NEXT:    [[TID13:%.*]] = load i32, ptr [[TID_ADDR_LOCAL12]], align 4
// CHECK-NEXT:    [[I16:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[AGG_CAPTURED17:%.*]] = alloca [[STRUCT_ANON_5:%.*]], align 8
// CHECK-NEXT:    [[AGG_CAPTURED18:%.*]] = alloca [[STRUCT_ANON_6:%.*]], align 4
// CHECK-NEXT:    [[DOTCOUNT_ADDR19:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_LASTITER34:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_LOWERBOUND35:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_UPPERBOUND36:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_STRIDE37:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[I75:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[AGG_CAPTURED76:%.*]] = alloca [[STRUCT_ANON_9:%.*]], align 8
// CHECK-NEXT:    [[AGG_CAPTURED77:%.*]] = alloca [[STRUCT_ANON_10:%.*]], align 4
// CHECK-NEXT:    [[DOTCOUNT_ADDR78:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_LASTITER93:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_LOWERBOUND94:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_UPPERBOUND95:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_STRIDE96:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[I135:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[AGG_CAPTURED136:%.*]] = alloca [[STRUCT_ANON_13:%.*]], align 8
// CHECK-NEXT:    [[AGG_CAPTURED137:%.*]] = alloca [[STRUCT_ANON_14:%.*]], align 4
// CHECK-NEXT:    [[DOTCOUNT_ADDR138:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_LASTITER153:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_LOWERBOUND154:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_UPPERBOUND155:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_STRIDE156:%.*]] = alloca i32, align 4
// CHECK-NEXT:    br label [[OMP_PAR_REGION9:%.*]]
// CHECK:       omp.par.region9:
// CHECK-NEXT:    store i32 0, ptr [[I16]], align 4
// CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_ANON_5]], ptr [[AGG_CAPTURED17]], i32 0, i32 0
// CHECK-NEXT:    store ptr [[I16]], ptr [[TMP2]], align 8
// CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [[STRUCT_ANON_6]], ptr [[AGG_CAPTURED18]], i32 0, i32 0
// CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr [[I16]], align 4
// CHECK-NEXT:    store i32 [[TMP4]], ptr [[TMP3]], align 4
// CHECK-NEXT:    call void @__captured_stmt.7(ptr [[DOTCOUNT_ADDR19]], ptr [[AGG_CAPTURED17]])
// CHECK-NEXT:    [[DOTCOUNT20:%.*]] = load i32, ptr [[DOTCOUNT_ADDR19]], align 4
// CHECK-NEXT:    br label [[OMP_LOOP_PREHEADER21:%.*]]
// CHECK:       omp_loop.preheader21:
// CHECK-NEXT:    store i32 0, ptr [[P_LOWERBOUND35]], align 4
// CHECK-NEXT:    [[TMP5:%.*]] = sub i32 [[DOTCOUNT20]], 1
// CHECK-NEXT:    store i32 [[TMP5]], ptr [[P_UPPERBOUND36]], align 4
// CHECK-NEXT:    store i32 1, ptr [[P_STRIDE37]], align 4
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM38:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    call void @__kmpc_for_static_init_4u(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM38]], i32 34, ptr [[P_LASTITER34]], ptr [[P_LOWERBOUND35]], ptr [[P_UPPERBOUND36]], ptr [[P_STRIDE37]], i32 1, i32 0)
// CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[P_LOWERBOUND35]], align 4
// CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[P_UPPERBOUND36]], align 4
// CHECK-NEXT:    [[TMP8:%.*]] = sub i32 [[TMP7]], [[TMP6]]
// CHECK-NEXT:    [[TMP9:%.*]] = add i32 [[TMP8]], 1
// CHECK-NEXT:    br label [[OMP_LOOP_HEADER22:%.*]]
// CHECK:       omp_loop.header22:
// CHECK-NEXT:    [[OMP_LOOP_IV28:%.*]] = phi i32 [ 0, [[OMP_LOOP_PREHEADER21]] ], [ [[OMP_LOOP_NEXT30:%.*]], [[OMP_LOOP_INC25:%.*]] ]
// CHECK-NEXT:    br label [[OMP_LOOP_COND23:%.*]]
// CHECK:       omp_loop.cond23:
// CHECK-NEXT:    [[OMP_LOOP_CMP29:%.*]] = icmp ult i32 [[OMP_LOOP_IV28]], [[TMP9]]
// CHECK-NEXT:    br i1 [[OMP_LOOP_CMP29]], label [[OMP_LOOP_BODY24:%.*]], label [[OMP_LOOP_EXIT26:%.*]]
// CHECK:       omp_loop.exit26:
// CHECK-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM38]])
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM39:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[OMP_GLOBAL_THREAD_NUM39]])
// CHECK-NEXT:    br label [[OMP_LOOP_AFTER27:%.*]]
// CHECK:       omp_loop.after27:
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM40:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    br label [[OMP_PARALLEL:%.*]]
// CHECK:       omp_parallel:
// CHECK-NEXT:    [[GEP_A_ADDR1:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG]], i32 0, i32 0
// CHECK-NEXT:    store ptr [[LOADGEP_A_ADDR]], ptr [[GEP_A_ADDR1]], align 8
// CHECK-NEXT:    [[GEP_B_ADDR2:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG]], i32 0, i32 1
// CHECK-NEXT:    store ptr [[LOADGEP_B_ADDR]], ptr [[GEP_B_ADDR2]], align 8
// CHECK-NEXT:    [[GEP_R_ADDR3:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG]], i32 0, i32 2
// CHECK-NEXT:    store ptr [[LOADGEP_R_ADDR]], ptr [[GEP_R_ADDR3]], align 8
// CHECK-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 1, ptr @_Z14parallel_for_2Pfid..omp_par, ptr [[STRUCTARG]])
// CHECK-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT:%.*]]
// CHECK:       omp.par.outlined.exit:
// CHECK-NEXT:    br label [[OMP_PAR_EXIT46_SPLIT:%.*]]
// CHECK:       omp.par.exit46.split:
// CHECK-NEXT:    store i32 0, ptr [[I75]], align 4
// CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [[STRUCT_ANON_9]], ptr [[AGG_CAPTURED76]], i32 0, i32 0
// CHECK-NEXT:    store ptr [[I75]], ptr [[TMP10]], align 8
// CHECK-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_ANON_10]], ptr [[AGG_CAPTURED77]], i32 0, i32 0
// CHECK-NEXT:    [[TMP12:%.*]] = load i32, ptr [[I75]], align 4
// CHECK-NEXT:    store i32 [[TMP12]], ptr [[TMP11]], align 4
// CHECK-NEXT:    call void @__captured_stmt.11(ptr [[DOTCOUNT_ADDR78]], ptr [[AGG_CAPTURED76]])
// CHECK-NEXT:    [[DOTCOUNT79:%.*]] = load i32, ptr [[DOTCOUNT_ADDR78]], align 4
// CHECK-NEXT:    br label [[OMP_LOOP_PREHEADER80:%.*]]
// CHECK:       omp_loop.preheader80:
// CHECK-NEXT:    store i32 0, ptr [[P_LOWERBOUND94]], align 4
// CHECK-NEXT:    [[TMP13:%.*]] = sub i32 [[DOTCOUNT79]], 1
// CHECK-NEXT:    store i32 [[TMP13]], ptr [[P_UPPERBOUND95]], align 4
// CHECK-NEXT:    store i32 1, ptr [[P_STRIDE96]], align 4
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM97:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    call void @__kmpc_for_static_init_4u(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM97]], i32 34, ptr [[P_LASTITER93]], ptr [[P_LOWERBOUND94]], ptr [[P_UPPERBOUND95]], ptr [[P_STRIDE96]], i32 1, i32 0)
// CHECK-NEXT:    [[TMP14:%.*]] = load i32, ptr [[P_LOWERBOUND94]], align 4
// CHECK-NEXT:    [[TMP15:%.*]] = load i32, ptr [[P_UPPERBOUND95]], align 4
// CHECK-NEXT:    [[TMP16:%.*]] = sub i32 [[TMP15]], [[TMP14]]
// CHECK-NEXT:    [[TMP17:%.*]] = add i32 [[TMP16]], 1
// CHECK-NEXT:    br label [[OMP_LOOP_HEADER81:%.*]]
// CHECK:       omp_loop.header81:
// CHECK-NEXT:    [[OMP_LOOP_IV87:%.*]] = phi i32 [ 0, [[OMP_LOOP_PREHEADER80]] ], [ [[OMP_LOOP_NEXT89:%.*]], [[OMP_LOOP_INC84:%.*]] ]
// CHECK-NEXT:    br label [[OMP_LOOP_COND82:%.*]]
// CHECK:       omp_loop.cond82:
// CHECK-NEXT:    [[OMP_LOOP_CMP88:%.*]] = icmp ult i32 [[OMP_LOOP_IV87]], [[TMP17]]
// CHECK-NEXT:    br i1 [[OMP_LOOP_CMP88]], label [[OMP_LOOP_BODY83:%.*]], label [[OMP_LOOP_EXIT85:%.*]]
// CHECK:       omp_loop.exit85:
// CHECK-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM97]])
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM98:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[OMP_GLOBAL_THREAD_NUM98]])
// CHECK-NEXT:    br label [[OMP_LOOP_AFTER86:%.*]]
// CHECK:       omp_loop.after86:
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM99:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    br label [[OMP_PARALLEL213:%.*]]
// CHECK:       omp_parallel213:
// CHECK-NEXT:    [[GEP_A_ADDR210:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG209]], i32 0, i32 0
// CHECK-NEXT:    store ptr [[LOADGEP_A_ADDR]], ptr [[GEP_A_ADDR210]], align 8
// CHECK-NEXT:    [[GEP_B_ADDR211:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG209]], i32 0, i32 1
// CHECK-NEXT:    store ptr [[LOADGEP_B_ADDR]], ptr [[GEP_B_ADDR211]], align 8
// CHECK-NEXT:    [[GEP_R_ADDR212:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG209]], i32 0, i32 2
// CHECK-NEXT:    store ptr [[LOADGEP_R_ADDR]], ptr [[GEP_R_ADDR212]], align 8
// CHECK-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 1, ptr @_Z14parallel_for_2Pfid..omp_par.21, ptr [[STRUCTARG209]])
// CHECK-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT134:%.*]]
// CHECK:       omp.par.outlined.exit134:
// CHECK-NEXT:    br label [[OMP_PAR_EXIT105_SPLIT:%.*]]
// CHECK:       omp.par.exit105.split:
// CHECK-NEXT:    store i32 0, ptr [[I135]], align 4
// CHECK-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [[STRUCT_ANON_13]], ptr [[AGG_CAPTURED136]], i32 0, i32 0
// CHECK-NEXT:    store ptr [[I135]], ptr [[TMP18]], align 8
// CHECK-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [[STRUCT_ANON_14]], ptr [[AGG_CAPTURED137]], i32 0, i32 0
// CHECK-NEXT:    [[TMP20:%.*]] = load i32, ptr [[I135]], align 4
// CHECK-NEXT:    store i32 [[TMP20]], ptr [[TMP19]], align 4
// CHECK-NEXT:    call void @__captured_stmt.15(ptr [[DOTCOUNT_ADDR138]], ptr [[AGG_CAPTURED136]])
// CHECK-NEXT:    [[DOTCOUNT139:%.*]] = load i32, ptr [[DOTCOUNT_ADDR138]], align 4
// CHECK-NEXT:    br label [[OMP_LOOP_PREHEADER140:%.*]]
// CHECK:       omp_loop.preheader140:
// CHECK-NEXT:    store i32 0, ptr [[P_LOWERBOUND154]], align 4
// CHECK-NEXT:    [[TMP21:%.*]] = sub i32 [[DOTCOUNT139]], 1
// CHECK-NEXT:    store i32 [[TMP21]], ptr [[P_UPPERBOUND155]], align 4
// CHECK-NEXT:    store i32 1, ptr [[P_STRIDE156]], align 4
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM157:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    call void @__kmpc_for_static_init_4u(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM157]], i32 34, ptr [[P_LASTITER153]], ptr [[P_LOWERBOUND154]], ptr [[P_UPPERBOUND155]], ptr [[P_STRIDE156]], i32 1, i32 0)
// CHECK-NEXT:    [[TMP22:%.*]] = load i32, ptr [[P_LOWERBOUND154]], align 4
// CHECK-NEXT:    [[TMP23:%.*]] = load i32, ptr [[P_UPPERBOUND155]], align 4
// CHECK-NEXT:    [[TMP24:%.*]] = sub i32 [[TMP23]], [[TMP22]]
// CHECK-NEXT:    [[TMP25:%.*]] = add i32 [[TMP24]], 1
// CHECK-NEXT:    br label [[OMP_LOOP_HEADER141:%.*]]
// CHECK:       omp_loop.header141:
// CHECK-NEXT:    [[OMP_LOOP_IV147:%.*]] = phi i32 [ 0, [[OMP_LOOP_PREHEADER140]] ], [ [[OMP_LOOP_NEXT149:%.*]], [[OMP_LOOP_INC144:%.*]] ]
// CHECK-NEXT:    br label [[OMP_LOOP_COND142:%.*]]
// CHECK:       omp_loop.cond142:
// CHECK-NEXT:    [[OMP_LOOP_CMP148:%.*]] = icmp ult i32 [[OMP_LOOP_IV147]], [[TMP25]]
// CHECK-NEXT:    br i1 [[OMP_LOOP_CMP148]], label [[OMP_LOOP_BODY143:%.*]], label [[OMP_LOOP_EXIT145:%.*]]
// CHECK:       omp_loop.exit145:
// CHECK-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM157]])
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM158:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[OMP_GLOBAL_THREAD_NUM158]])
// CHECK-NEXT:    br label [[OMP_LOOP_AFTER146:%.*]]
// CHECK:       omp_loop.after146:
// CHECK-NEXT:    br label [[OMP_PAR_REGION9_PARALLEL_AFTER:%.*]]
// CHECK:       omp.par.region9.parallel.after:
// CHECK-NEXT:    br label [[OMP_PAR_PRE_FINALIZE10:%.*]]
// CHECK:       omp.par.pre_finalize10:
// CHECK-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT159_EXITSTUB:%.*]]
// CHECK:       omp_loop.body143:
// CHECK-NEXT:    [[TMP26:%.*]] = add i32 [[OMP_LOOP_IV147]], [[TMP22]]
// CHECK-NEXT:    call void @__captured_stmt.16(ptr [[I135]], i32 [[TMP26]], ptr [[AGG_CAPTURED137]])
// CHECK-NEXT:    [[TMP27:%.*]] = load i32, ptr [[LOADGEP_A_ADDR]], align 4
// CHECK-NEXT:    [[CONV150:%.*]] = sitofp i32 [[TMP27]] to double
// CHECK-NEXT:    [[TMP28:%.*]] = load double, ptr [[LOADGEP_B_ADDR]], align 8
// CHECK-NEXT:    [[ADD151:%.*]] = fadd double [[CONV150]], [[TMP28]]
// CHECK-NEXT:    [[CONV152:%.*]] = fptrunc double [[ADD151]] to float
// CHECK-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[LOADGEP_R_ADDR]], align 8
// CHECK-NEXT:    store float [[CONV152]], ptr [[TMP29]], align 4
// CHECK-NEXT:    br label [[OMP_LOOP_INC144]]
// CHECK:       omp_loop.inc144:
// CHECK-NEXT:    [[OMP_LOOP_NEXT149]] = add nuw i32 [[OMP_LOOP_IV147]], 1
// CHECK-NEXT:    br label [[OMP_LOOP_HEADER141]]
// CHECK:       omp_loop.body83:
// CHECK-NEXT:    [[TMP30:%.*]] = add i32 [[OMP_LOOP_IV87]], [[TMP14]]
// CHECK-NEXT:    call void @__captured_stmt.12(ptr [[I75]], i32 [[TMP30]], ptr [[AGG_CAPTURED77]])
// CHECK-NEXT:    [[TMP31:%.*]] = load i32, ptr [[LOADGEP_A_ADDR]], align 4
// CHECK-NEXT:    [[CONV90:%.*]] = sitofp i32 [[TMP31]] to double
// CHECK-NEXT:    [[TMP32:%.*]] = load double, ptr [[LOADGEP_B_ADDR]], align 8
// CHECK-NEXT:    [[ADD91:%.*]] = fadd double [[CONV90]], [[TMP32]]
// CHECK-NEXT:    [[CONV92:%.*]] = fptrunc double [[ADD91]] to float
// CHECK-NEXT:    [[TMP33:%.*]] = load ptr, ptr [[LOADGEP_R_ADDR]], align 8
// CHECK-NEXT:    store float [[CONV92]], ptr [[TMP33]], align 4
// CHECK-NEXT:    br label [[OMP_LOOP_INC84]]
// CHECK:       omp_loop.inc84:
// CHECK-NEXT:    [[OMP_LOOP_NEXT89]] = add nuw i32 [[OMP_LOOP_IV87]], 1
// CHECK-NEXT:    br label [[OMP_LOOP_HEADER81]]
// CHECK:       omp_loop.body24:
// CHECK-NEXT:    [[TMP34:%.*]] = add i32 [[OMP_LOOP_IV28]], [[TMP6]]
// CHECK-NEXT:    call void @__captured_stmt.8(ptr [[I16]], i32 [[TMP34]], ptr [[AGG_CAPTURED18]])
// CHECK-NEXT:    [[TMP35:%.*]] = load i32, ptr [[LOADGEP_A_ADDR]], align 4
// CHECK-NEXT:    [[CONV31:%.*]] = sitofp i32 [[TMP35]] to double
// CHECK-NEXT:    [[TMP36:%.*]] = load double, ptr [[LOADGEP_B_ADDR]], align 8
// CHECK-NEXT:    [[ADD32:%.*]] = fadd double [[CONV31]], [[TMP36]]
// CHECK-NEXT:    [[CONV33:%.*]] = fptrunc double [[ADD32]] to float
// CHECK-NEXT:    [[TMP37:%.*]] = load ptr, ptr [[LOADGEP_R_ADDR]], align 8
// CHECK-NEXT:    store float [[CONV33]], ptr [[TMP37]], align 4
// CHECK-NEXT:    br label [[OMP_LOOP_INC25]]
// CHECK:       omp_loop.inc25:
// CHECK-NEXT:    [[OMP_LOOP_NEXT30]] = add nuw i32 [[OMP_LOOP_IV28]], 1
// CHECK-NEXT:    br label [[OMP_LOOP_HEADER22]]
// CHECK:       omp.par.outlined.exit159.exitStub:
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@_Z14parallel_for_2Pfid..omp_par.21
// CHECK-SAME: (ptr noalias [[TID_ADDR100:%.*]], ptr noalias [[ZERO_ADDR101:%.*]], ptr [[TMP0:%.*]]) #[[ATTR1]] {
// CHECK-NEXT:  omp.par.entry102:
// CHECK-NEXT:    [[GEP_A_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[LOADGEP_A_ADDR:%.*]] = load ptr, ptr [[GEP_A_ADDR]], align 8
// CHECK-NEXT:    [[GEP_B_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 1
// CHECK-NEXT:    [[LOADGEP_B_ADDR:%.*]] = load ptr, ptr [[GEP_B_ADDR]], align 8
// CHECK-NEXT:    [[GEP_R_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 2
// CHECK-NEXT:    [[LOADGEP_R_ADDR:%.*]] = load ptr, ptr [[GEP_R_ADDR]], align 8
// CHECK-NEXT:    [[TID_ADDR_LOCAL106:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TID_ADDR100]], align 4
// CHECK-NEXT:    store i32 [[TMP1]], ptr [[TID_ADDR_LOCAL106]], align 4
// CHECK-NEXT:    [[TID107:%.*]] = load i32, ptr [[TID_ADDR_LOCAL106]], align 4
// CHECK-NEXT:    [[I110:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[AGG_CAPTURED111:%.*]] = alloca [[STRUCT_ANON_11:%.*]], align 8
// CHECK-NEXT:    [[AGG_CAPTURED112:%.*]] = alloca [[STRUCT_ANON_12:%.*]], align 4
// CHECK-NEXT:    [[DOTCOUNT_ADDR113:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_LASTITER128:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_LOWERBOUND129:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_UPPERBOUND130:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_STRIDE131:%.*]] = alloca i32, align 4
// CHECK-NEXT:    br label [[OMP_PAR_REGION103:%.*]]
// CHECK:       omp.par.region103:
// CHECK-NEXT:    store i32 0, ptr [[I110]], align 4
// CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_ANON_11]], ptr [[AGG_CAPTURED111]], i32 0, i32 0
// CHECK-NEXT:    store ptr [[I110]], ptr [[TMP2]], align 8
// CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [[STRUCT_ANON_12]], ptr [[AGG_CAPTURED112]], i32 0, i32 0
// CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr [[I110]], align 4
// CHECK-NEXT:    store i32 [[TMP4]], ptr [[TMP3]], align 4
// CHECK-NEXT:    call void @__captured_stmt.13(ptr [[DOTCOUNT_ADDR113]], ptr [[AGG_CAPTURED111]])
// CHECK-NEXT:    [[DOTCOUNT114:%.*]] = load i32, ptr [[DOTCOUNT_ADDR113]], align 4
// CHECK-NEXT:    br label [[OMP_LOOP_PREHEADER115:%.*]]
// CHECK:       omp_loop.preheader115:
// CHECK-NEXT:    store i32 0, ptr [[P_LOWERBOUND129]], align 4
// CHECK-NEXT:    [[TMP5:%.*]] = sub i32 [[DOTCOUNT114]], 1
// CHECK-NEXT:    store i32 [[TMP5]], ptr [[P_UPPERBOUND130]], align 4
// CHECK-NEXT:    store i32 1, ptr [[P_STRIDE131]], align 4
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM132:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    call void @__kmpc_for_static_init_4u(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM132]], i32 34, ptr [[P_LASTITER128]], ptr [[P_LOWERBOUND129]], ptr [[P_UPPERBOUND130]], ptr [[P_STRIDE131]], i32 1, i32 0)
// CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[P_LOWERBOUND129]], align 4
// CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[P_UPPERBOUND130]], align 4
// CHECK-NEXT:    [[TMP8:%.*]] = sub i32 [[TMP7]], [[TMP6]]
// CHECK-NEXT:    [[TMP9:%.*]] = add i32 [[TMP8]], 1
// CHECK-NEXT:    br label [[OMP_LOOP_HEADER116:%.*]]
// CHECK:       omp_loop.header116:
// CHECK-NEXT:    [[OMP_LOOP_IV122:%.*]] = phi i32 [ 0, [[OMP_LOOP_PREHEADER115]] ], [ [[OMP_LOOP_NEXT124:%.*]], [[OMP_LOOP_INC119:%.*]] ]
// CHECK-NEXT:    br label [[OMP_LOOP_COND117:%.*]]
// CHECK:       omp_loop.cond117:
// CHECK-NEXT:    [[OMP_LOOP_CMP123:%.*]] = icmp ult i32 [[OMP_LOOP_IV122]], [[TMP9]]
// CHECK-NEXT:    br i1 [[OMP_LOOP_CMP123]], label [[OMP_LOOP_BODY118:%.*]], label [[OMP_LOOP_EXIT120:%.*]]
// CHECK:       omp_loop.exit120:
// CHECK-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM132]])
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM133:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[OMP_GLOBAL_THREAD_NUM133]])
// CHECK-NEXT:    br label [[OMP_LOOP_AFTER121:%.*]]
// CHECK:       omp_loop.after121:
// CHECK-NEXT:    br label [[OMP_PAR_REGION103_PARALLEL_AFTER:%.*]]
// CHECK:       omp.par.region103.parallel.after:
// CHECK-NEXT:    br label [[OMP_PAR_PRE_FINALIZE104:%.*]]
// CHECK:       omp.par.pre_finalize104:
// CHECK-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT134_EXITSTUB:%.*]]
// CHECK:       omp_loop.body118:
// CHECK-NEXT:    [[TMP10:%.*]] = add i32 [[OMP_LOOP_IV122]], [[TMP6]]
// CHECK-NEXT:    call void @__captured_stmt.14(ptr [[I110]], i32 [[TMP10]], ptr [[AGG_CAPTURED112]])
// CHECK-NEXT:    [[TMP11:%.*]] = load i32, ptr [[LOADGEP_A_ADDR]], align 4
// CHECK-NEXT:    [[CONV125:%.*]] = sitofp i32 [[TMP11]] to double
// CHECK-NEXT:    [[TMP12:%.*]] = load double, ptr [[LOADGEP_B_ADDR]], align 8
// CHECK-NEXT:    [[ADD126:%.*]] = fadd double [[CONV125]], [[TMP12]]
// CHECK-NEXT:    [[CONV127:%.*]] = fptrunc double [[ADD126]] to float
// CHECK-NEXT:    [[TMP13:%.*]] = load ptr, ptr [[LOADGEP_R_ADDR]], align 8
// CHECK-NEXT:    store float [[CONV127]], ptr [[TMP13]], align 4
// CHECK-NEXT:    br label [[OMP_LOOP_INC119]]
// CHECK:       omp_loop.inc119:
// CHECK-NEXT:    [[OMP_LOOP_NEXT124]] = add nuw i32 [[OMP_LOOP_IV122]], 1
// CHECK-NEXT:    br label [[OMP_LOOP_HEADER116]]
// CHECK:       omp.par.outlined.exit134.exitStub:
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@_Z14parallel_for_2Pfid..omp_par
// CHECK-SAME: (ptr noalias [[TID_ADDR41:%.*]], ptr noalias [[ZERO_ADDR42:%.*]], ptr [[TMP0:%.*]]) #[[ATTR1]] {
// CHECK-NEXT:  omp.par.entry43:
// CHECK-NEXT:    [[GEP_A_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[LOADGEP_A_ADDR:%.*]] = load ptr, ptr [[GEP_A_ADDR]], align 8
// CHECK-NEXT:    [[GEP_B_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 1
// CHECK-NEXT:    [[LOADGEP_B_ADDR:%.*]] = load ptr, ptr [[GEP_B_ADDR]], align 8
// CHECK-NEXT:    [[GEP_R_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 2
// CHECK-NEXT:    [[LOADGEP_R_ADDR:%.*]] = load ptr, ptr [[GEP_R_ADDR]], align 8
// CHECK-NEXT:    [[TID_ADDR_LOCAL47:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TID_ADDR41]], align 4
// CHECK-NEXT:    store i32 [[TMP1]], ptr [[TID_ADDR_LOCAL47]], align 4
// CHECK-NEXT:    [[TID48:%.*]] = load i32, ptr [[TID_ADDR_LOCAL47]], align 4
// CHECK-NEXT:    [[I51:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[AGG_CAPTURED52:%.*]] = alloca [[STRUCT_ANON_7:%.*]], align 8
// CHECK-NEXT:    [[AGG_CAPTURED53:%.*]] = alloca [[STRUCT_ANON_8:%.*]], align 4
// CHECK-NEXT:    [[DOTCOUNT_ADDR54:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_LASTITER69:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_LOWERBOUND70:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_UPPERBOUND71:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[P_STRIDE72:%.*]] = alloca i32, align 4
// CHECK-NEXT:    br label [[OMP_PAR_REGION44:%.*]]
// CHECK:       omp.par.region44:
// CHECK-NEXT:    store i32 0, ptr [[I51]], align 4
// CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_ANON_7]], ptr [[AGG_CAPTURED52]], i32 0, i32 0
// CHECK-NEXT:    store ptr [[I51]], ptr [[TMP2]], align 8
// CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [[STRUCT_ANON_8]], ptr [[AGG_CAPTURED53]], i32 0, i32 0
// CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr [[I51]], align 4
// CHECK-NEXT:    store i32 [[TMP4]], ptr [[TMP3]], align 4
// CHECK-NEXT:    call void @__captured_stmt.9(ptr [[DOTCOUNT_ADDR54]], ptr [[AGG_CAPTURED52]])
// CHECK-NEXT:    [[DOTCOUNT55:%.*]] = load i32, ptr [[DOTCOUNT_ADDR54]], align 4
// CHECK-NEXT:    br label [[OMP_LOOP_PREHEADER56:%.*]]
// CHECK:       omp_loop.preheader56:
// CHECK-NEXT:    store i32 0, ptr [[P_LOWERBOUND70]], align 4
// CHECK-NEXT:    [[TMP5:%.*]] = sub i32 [[DOTCOUNT55]], 1
// CHECK-NEXT:    store i32 [[TMP5]], ptr [[P_UPPERBOUND71]], align 4
// CHECK-NEXT:    store i32 1, ptr [[P_STRIDE72]], align 4
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM73:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    call void @__kmpc_for_static_init_4u(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM73]], i32 34, ptr [[P_LASTITER69]], ptr [[P_LOWERBOUND70]], ptr [[P_UPPERBOUND71]], ptr [[P_STRIDE72]], i32 1, i32 0)
// CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[P_LOWERBOUND70]], align 4
// CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[P_UPPERBOUND71]], align 4
// CHECK-NEXT:    [[TMP8:%.*]] = sub i32 [[TMP7]], [[TMP6]]
// CHECK-NEXT:    [[TMP9:%.*]] = add i32 [[TMP8]], 1
// CHECK-NEXT:    br label [[OMP_LOOP_HEADER57:%.*]]
// CHECK:       omp_loop.header57:
// CHECK-NEXT:    [[OMP_LOOP_IV63:%.*]] = phi i32 [ 0, [[OMP_LOOP_PREHEADER56]] ], [ [[OMP_LOOP_NEXT65:%.*]], [[OMP_LOOP_INC60:%.*]] ]
// CHECK-NEXT:    br label [[OMP_LOOP_COND58:%.*]]
// CHECK:       omp_loop.cond58:
// CHECK-NEXT:    [[OMP_LOOP_CMP64:%.*]] = icmp ult i32 [[OMP_LOOP_IV63]], [[TMP9]]
// CHECK-NEXT:    br i1 [[OMP_LOOP_CMP64]], label [[OMP_LOOP_BODY59:%.*]], label [[OMP_LOOP_EXIT61:%.*]]
// CHECK:       omp_loop.exit61:
// CHECK-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM73]])
// CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM74:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[OMP_GLOBAL_THREAD_NUM74]])
// CHECK-NEXT:    br label [[OMP_LOOP_AFTER62:%.*]]
// CHECK:       omp_loop.after62:
// CHECK-NEXT:    br label [[OMP_PAR_REGION44_PARALLEL_AFTER:%.*]]
// CHECK:       omp.par.region44.parallel.after:
// CHECK-NEXT:    br label [[OMP_PAR_PRE_FINALIZE45:%.*]]
// CHECK:       omp.par.pre_finalize45:
// CHECK-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT_EXITSTUB:%.*]]
// CHECK:       omp_loop.body59:
// CHECK-NEXT:    [[TMP10:%.*]] = add i32 [[OMP_LOOP_IV63]], [[TMP6]]
// CHECK-NEXT:    call void @__captured_stmt.10(ptr [[I51]], i32 [[TMP10]], ptr [[AGG_CAPTURED53]])
// CHECK-NEXT:    [[TMP11:%.*]] = load i32, ptr [[LOADGEP_A_ADDR]], align 4
// CHECK-NEXT:    [[CONV66:%.*]] = sitofp i32 [[TMP11]] to double
// CHECK-NEXT:    [[TMP12:%.*]] = load double, ptr [[LOADGEP_B_ADDR]], align 8
// CHECK-NEXT:    [[ADD67:%.*]] = fadd double [[CONV66]], [[TMP12]]
// CHECK-NEXT:    [[CONV68:%.*]] = fptrunc double [[ADD67]] to float
// CHECK-NEXT:    [[TMP13:%.*]] = load ptr, ptr [[LOADGEP_R_ADDR]], align 8
// CHECK-NEXT:    store float [[CONV68]], ptr [[TMP13]], align 4
// CHECK-NEXT:    br label [[OMP_LOOP_INC60]]
// CHECK:       omp_loop.inc60:
// CHECK-NEXT:    [[OMP_LOOP_NEXT65]] = add nuw i32 [[OMP_LOOP_IV63]], 1
// CHECK-NEXT:    br label [[OMP_LOOP_HEADER57]]
// CHECK:       omp.par.outlined.exit.exitStub:
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@__captured_stmt.5
// CHECK-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[DISTANCE:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[DISTANCE_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[DOTSTART:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[DOTSTOP:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[DOTSTEP:%.*]] = alloca i32, align 4
// CHECK-NEXT:    store ptr [[DISTANCE]], ptr [[DISTANCE_ADDR]], align 8
// CHECK-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_3:%.*]], ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP1]], align 8
// CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[TMP2]], align 4
// CHECK-NEXT:    store i32 [[TMP3]], ptr [[DOTSTART]], align 4
// CHECK-NEXT:    store i32 100, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    store i32 1, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTSTART]], align 4
// CHECK-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP4]], [[TMP5]]
// CHECK-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK:       cond.true:
// CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTSTART]], align 4
// CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], [[TMP7]]
// CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[SUB1:%.*]] = sub i32 [[TMP8]], 1
// CHECK-NEXT:    [[ADD:%.*]] = add i32 [[SUB]], [[SUB1]]
// CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], [[TMP9]]
// CHECK-NEXT:    br label [[COND_END:%.*]]
// CHECK:       cond.false:
// CHECK-NEXT:    br label [[COND_END]]
// CHECK:       cond.end:
// CHECK-NEXT:    [[COND:%.*]] = phi i32 [ [[DIV]], [[COND_TRUE]] ], [ 0, [[COND_FALSE]] ]
// CHECK-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DISTANCE_ADDR]], align 8
// CHECK-NEXT:    store i32 [[COND]], ptr [[TMP10]], align 4
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@__captured_stmt.6
// CHECK-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[LOOPVAR:%.*]], i32 noundef [[LOGICAL:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[LOOPVAR_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[LOGICAL_ADDR:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    store ptr [[LOOPVAR]], ptr [[LOOPVAR_ADDR]], align 8
// CHECK-NEXT:    store i32 [[LOGICAL]], ptr [[LOGICAL_ADDR]], align 4
// CHECK-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_4:%.*]], ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[LOGICAL_ADDR]], align 4
// CHECK-NEXT:    [[MUL:%.*]] = mul i32 1, [[TMP3]]
// CHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP2]], [[MUL]]
// CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[LOOPVAR_ADDR]], align 8
// CHECK-NEXT:    store i32 [[ADD]], ptr [[TMP4]], align 4
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@__captured_stmt.7
// CHECK-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[DISTANCE:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[DISTANCE_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[DOTSTART:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[DOTSTOP:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[DOTSTEP:%.*]] = alloca i32, align 4
// CHECK-NEXT:    store ptr [[DISTANCE]], ptr [[DISTANCE_ADDR]], align 8
// CHECK-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_5:%.*]], ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP1]], align 8
// CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[TMP2]], align 4
// CHECK-NEXT:    store i32 [[TMP3]], ptr [[DOTSTART]], align 4
// CHECK-NEXT:    store i32 100, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    store i32 1, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTSTART]], align 4
// CHECK-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP4]], [[TMP5]]
// CHECK-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK:       cond.true:
// CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTSTART]], align 4
// CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], [[TMP7]]
// CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[SUB1:%.*]] = sub i32 [[TMP8]], 1
// CHECK-NEXT:    [[ADD:%.*]] = add i32 [[SUB]], [[SUB1]]
// CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], [[TMP9]]
// CHECK-NEXT:    br label [[COND_END:%.*]]
// CHECK:       cond.false:
// CHECK-NEXT:    br label [[COND_END]]
// CHECK:       cond.end:
// CHECK-NEXT:    [[COND:%.*]] = phi i32 [ [[DIV]], [[COND_TRUE]] ], [ 0, [[COND_FALSE]] ]
// CHECK-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DISTANCE_ADDR]], align 8
// CHECK-NEXT:    store i32 [[COND]], ptr [[TMP10]], align 4
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@__captured_stmt.8
// CHECK-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[LOOPVAR:%.*]], i32 noundef [[LOGICAL:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[LOOPVAR_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[LOGICAL_ADDR:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    store ptr [[LOOPVAR]], ptr [[LOOPVAR_ADDR]], align 8
// CHECK-NEXT:    store i32 [[LOGICAL]], ptr [[LOGICAL_ADDR]], align 4
// CHECK-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_6:%.*]], ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[LOGICAL_ADDR]], align 4
// CHECK-NEXT:    [[MUL:%.*]] = mul i32 1, [[TMP3]]
// CHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP2]], [[MUL]]
// CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[LOOPVAR_ADDR]], align 8
// CHECK-NEXT:    store i32 [[ADD]], ptr [[TMP4]], align 4
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@__captured_stmt.9
// CHECK-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[DISTANCE:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[DISTANCE_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[DOTSTART:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[DOTSTOP:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[DOTSTEP:%.*]] = alloca i32, align 4
// CHECK-NEXT:    store ptr [[DISTANCE]], ptr [[DISTANCE_ADDR]], align 8
// CHECK-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_7:%.*]], ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP1]], align 8
// CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[TMP2]], align 4
// CHECK-NEXT:    store i32 [[TMP3]], ptr [[DOTSTART]], align 4
// CHECK-NEXT:    store i32 100, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    store i32 1, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTSTART]], align 4
// CHECK-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP4]], [[TMP5]]
// CHECK-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK:       cond.true:
// CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTSTART]], align 4
// CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], [[TMP7]]
// CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[SUB1:%.*]] = sub i32 [[TMP8]], 1
// CHECK-NEXT:    [[ADD:%.*]] = add i32 [[SUB]], [[SUB1]]
// CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], [[TMP9]]
// CHECK-NEXT:    br label [[COND_END:%.*]]
// CHECK:       cond.false:
// CHECK-NEXT:    br label [[COND_END]]
// CHECK:       cond.end:
// CHECK-NEXT:    [[COND:%.*]] = phi i32 [ [[DIV]], [[COND_TRUE]] ], [ 0, [[COND_FALSE]] ]
// CHECK-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DISTANCE_ADDR]], align 8
// CHECK-NEXT:    store i32 [[COND]], ptr [[TMP10]], align 4
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@__captured_stmt.10
// CHECK-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[LOOPVAR:%.*]], i32 noundef [[LOGICAL:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[LOOPVAR_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[LOGICAL_ADDR:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    store ptr [[LOOPVAR]], ptr [[LOOPVAR_ADDR]], align 8
// CHECK-NEXT:    store i32 [[LOGICAL]], ptr [[LOGICAL_ADDR]], align 4
// CHECK-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_8:%.*]], ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[LOGICAL_ADDR]], align 4
// CHECK-NEXT:    [[MUL:%.*]] = mul i32 1, [[TMP3]]
// CHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP2]], [[MUL]]
// CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[LOOPVAR_ADDR]], align 8
// CHECK-NEXT:    store i32 [[ADD]], ptr [[TMP4]], align 4
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@__captured_stmt.11
// CHECK-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[DISTANCE:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[DISTANCE_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[DOTSTART:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[DOTSTOP:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[DOTSTEP:%.*]] = alloca i32, align 4
// CHECK-NEXT:    store ptr [[DISTANCE]], ptr [[DISTANCE_ADDR]], align 8
// CHECK-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_9:%.*]], ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP1]], align 8
// CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[TMP2]], align 4
// CHECK-NEXT:    store i32 [[TMP3]], ptr [[DOTSTART]], align 4
// CHECK-NEXT:    store i32 100, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    store i32 1, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTSTART]], align 4
// CHECK-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP4]], [[TMP5]]
// CHECK-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK:       cond.true:
// CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTSTART]], align 4
// CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], [[TMP7]]
// CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[SUB1:%.*]] = sub i32 [[TMP8]], 1
// CHECK-NEXT:    [[ADD:%.*]] = add i32 [[SUB]], [[SUB1]]
// CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], [[TMP9]]
// CHECK-NEXT:    br label [[COND_END:%.*]]
// CHECK:       cond.false:
// CHECK-NEXT:    br label [[COND_END]]
// CHECK:       cond.end:
// CHECK-NEXT:    [[COND:%.*]] = phi i32 [ [[DIV]], [[COND_TRUE]] ], [ 0, [[COND_FALSE]] ]
// CHECK-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DISTANCE_ADDR]], align 8
// CHECK-NEXT:    store i32 [[COND]], ptr [[TMP10]], align 4
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@__captured_stmt.12
// CHECK-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[LOOPVAR:%.*]], i32 noundef [[LOGICAL:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[LOOPVAR_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[LOGICAL_ADDR:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    store ptr [[LOOPVAR]], ptr [[LOOPVAR_ADDR]], align 8
// CHECK-NEXT:    store i32 [[LOGICAL]], ptr [[LOGICAL_ADDR]], align 4
// CHECK-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_10:%.*]], ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[LOGICAL_ADDR]], align 4
// CHECK-NEXT:    [[MUL:%.*]] = mul i32 1, [[TMP3]]
// CHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP2]], [[MUL]]
// CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[LOOPVAR_ADDR]], align 8
// CHECK-NEXT:    store i32 [[ADD]], ptr [[TMP4]], align 4
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@__captured_stmt.13
// CHECK-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[DISTANCE:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[DISTANCE_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[DOTSTART:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[DOTSTOP:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[DOTSTEP:%.*]] = alloca i32, align 4
// CHECK-NEXT:    store ptr [[DISTANCE]], ptr [[DISTANCE_ADDR]], align 8
// CHECK-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_11:%.*]], ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP1]], align 8
// CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[TMP2]], align 4
// CHECK-NEXT:    store i32 [[TMP3]], ptr [[DOTSTART]], align 4
// CHECK-NEXT:    store i32 100, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    store i32 1, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTSTART]], align 4
// CHECK-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP4]], [[TMP5]]
// CHECK-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK:       cond.true:
// CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTSTART]], align 4
// CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], [[TMP7]]
// CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[SUB1:%.*]] = sub i32 [[TMP8]], 1
// CHECK-NEXT:    [[ADD:%.*]] = add i32 [[SUB]], [[SUB1]]
// CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], [[TMP9]]
// CHECK-NEXT:    br label [[COND_END:%.*]]
// CHECK:       cond.false:
// CHECK-NEXT:    br label [[COND_END]]
// CHECK:       cond.end:
// CHECK-NEXT:    [[COND:%.*]] = phi i32 [ [[DIV]], [[COND_TRUE]] ], [ 0, [[COND_FALSE]] ]
// CHECK-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DISTANCE_ADDR]], align 8
// CHECK-NEXT:    store i32 [[COND]], ptr [[TMP10]], align 4
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@__captured_stmt.14
// CHECK-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[LOOPVAR:%.*]], i32 noundef [[LOGICAL:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[LOOPVAR_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[LOGICAL_ADDR:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    store ptr [[LOOPVAR]], ptr [[LOOPVAR_ADDR]], align 8
// CHECK-NEXT:    store i32 [[LOGICAL]], ptr [[LOGICAL_ADDR]], align 4
// CHECK-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_12:%.*]], ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[LOGICAL_ADDR]], align 4
// CHECK-NEXT:    [[MUL:%.*]] = mul i32 1, [[TMP3]]
// CHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP2]], [[MUL]]
// CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[LOOPVAR_ADDR]], align 8
// CHECK-NEXT:    store i32 [[ADD]], ptr [[TMP4]], align 4
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@__captured_stmt.15
// CHECK-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[DISTANCE:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[DISTANCE_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[DOTSTART:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[DOTSTOP:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[DOTSTEP:%.*]] = alloca i32, align 4
// CHECK-NEXT:    store ptr [[DISTANCE]], ptr [[DISTANCE_ADDR]], align 8
// CHECK-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_13:%.*]], ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP1]], align 8
// CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[TMP2]], align 4
// CHECK-NEXT:    store i32 [[TMP3]], ptr [[DOTSTART]], align 4
// CHECK-NEXT:    store i32 100, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    store i32 1, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTSTART]], align 4
// CHECK-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP4]], [[TMP5]]
// CHECK-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK:       cond.true:
// CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTSTART]], align 4
// CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], [[TMP7]]
// CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[SUB1:%.*]] = sub i32 [[TMP8]], 1
// CHECK-NEXT:    [[ADD:%.*]] = add i32 [[SUB]], [[SUB1]]
// CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], [[TMP9]]
// CHECK-NEXT:    br label [[COND_END:%.*]]
// CHECK:       cond.false:
// CHECK-NEXT:    br label [[COND_END]]
// CHECK:       cond.end:
// CHECK-NEXT:    [[COND:%.*]] = phi i32 [ [[DIV]], [[COND_TRUE]] ], [ 0, [[COND_FALSE]] ]
// CHECK-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DISTANCE_ADDR]], align 8
// CHECK-NEXT:    store i32 [[COND]], ptr [[TMP10]], align 4
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@__captured_stmt.16
// CHECK-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[LOOPVAR:%.*]], i32 noundef [[LOGICAL:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[LOOPVAR_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[LOGICAL_ADDR:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    store ptr [[LOOPVAR]], ptr [[LOOPVAR_ADDR]], align 8
// CHECK-NEXT:    store i32 [[LOGICAL]], ptr [[LOGICAL_ADDR]], align 4
// CHECK-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_14:%.*]], ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[LOGICAL_ADDR]], align 4
// CHECK-NEXT:    [[MUL:%.*]] = mul i32 1, [[TMP3]]
// CHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP2]], [[MUL]]
// CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[LOOPVAR_ADDR]], align 8
// CHECK-NEXT:    store i32 [[ADD]], ptr [[TMP4]], align 4
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@__captured_stmt.17
// CHECK-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[DISTANCE:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[DISTANCE_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[DOTSTART:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[DOTSTOP:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[DOTSTEP:%.*]] = alloca i32, align 4
// CHECK-NEXT:    store ptr [[DISTANCE]], ptr [[DISTANCE_ADDR]], align 8
// CHECK-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_15:%.*]], ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP1]], align 8
// CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[TMP2]], align 4
// CHECK-NEXT:    store i32 [[TMP3]], ptr [[DOTSTART]], align 4
// CHECK-NEXT:    store i32 100, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    store i32 1, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTSTART]], align 4
// CHECK-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP4]], [[TMP5]]
// CHECK-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK:       cond.true:
// CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTSTART]], align 4
// CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], [[TMP7]]
// CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[SUB1:%.*]] = sub i32 [[TMP8]], 1
// CHECK-NEXT:    [[ADD:%.*]] = add i32 [[SUB]], [[SUB1]]
// CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], [[TMP9]]
// CHECK-NEXT:    br label [[COND_END:%.*]]
// CHECK:       cond.false:
// CHECK-NEXT:    br label [[COND_END]]
// CHECK:       cond.end:
// CHECK-NEXT:    [[COND:%.*]] = phi i32 [ [[DIV]], [[COND_TRUE]] ], [ 0, [[COND_FALSE]] ]
// CHECK-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DISTANCE_ADDR]], align 8
// CHECK-NEXT:    store i32 [[COND]], ptr [[TMP10]], align 4
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@__captured_stmt.18
// CHECK-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[LOOPVAR:%.*]], i32 noundef [[LOGICAL:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[LOOPVAR_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[LOGICAL_ADDR:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    store ptr [[LOOPVAR]], ptr [[LOOPVAR_ADDR]], align 8
// CHECK-NEXT:    store i32 [[LOGICAL]], ptr [[LOGICAL_ADDR]], align 4
// CHECK-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_16:%.*]], ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[LOGICAL_ADDR]], align 4
// CHECK-NEXT:    [[MUL:%.*]] = mul i32 1, [[TMP3]]
// CHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP2]], [[MUL]]
// CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[LOOPVAR_ADDR]], align 8
// CHECK-NEXT:    store i32 [[ADD]], ptr [[TMP4]], align 4
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@__captured_stmt.19
// CHECK-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[DISTANCE:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[DISTANCE_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[DOTSTART:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[DOTSTOP:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[DOTSTEP:%.*]] = alloca i32, align 4
// CHECK-NEXT:    store ptr [[DISTANCE]], ptr [[DISTANCE_ADDR]], align 8
// CHECK-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_17:%.*]], ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP1]], align 8
// CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[TMP2]], align 4
// CHECK-NEXT:    store i32 [[TMP3]], ptr [[DOTSTART]], align 4
// CHECK-NEXT:    store i32 100, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    store i32 1, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTSTART]], align 4
// CHECK-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP4]], [[TMP5]]
// CHECK-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK:       cond.true:
// CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTSTOP]], align 4
// CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTSTART]], align 4
// CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], [[TMP7]]
// CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[SUB1:%.*]] = sub i32 [[TMP8]], 1
// CHECK-NEXT:    [[ADD:%.*]] = add i32 [[SUB]], [[SUB1]]
// CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTSTEP]], align 4
// CHECK-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], [[TMP9]]
// CHECK-NEXT:    br label [[COND_END:%.*]]
// CHECK:       cond.false:
// CHECK-NEXT:    br label [[COND_END]]
// CHECK:       cond.end:
// CHECK-NEXT:    [[COND:%.*]] = phi i32 [ [[DIV]], [[COND_TRUE]] ], [ 0, [[COND_FALSE]] ]
// CHECK-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DISTANCE_ADDR]], align 8
// CHECK-NEXT:    store i32 [[COND]], ptr [[TMP10]], align 4
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@__captured_stmt.20
// CHECK-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[LOOPVAR:%.*]], i32 noundef [[LOGICAL:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[LOOPVAR_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    [[LOGICAL_ADDR:%.*]] = alloca i32, align 4
// CHECK-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT:    store ptr [[LOOPVAR]], ptr [[LOOPVAR_ADDR]], align 8
// CHECK-NEXT:    store i32 [[LOGICAL]], ptr [[LOGICAL_ADDR]], align 4
// CHECK-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_18:%.*]], ptr [[TMP0]], i32 0, i32 0
// CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[LOGICAL_ADDR]], align 4
// CHECK-NEXT:    [[MUL:%.*]] = mul i32 1, [[TMP3]]
// CHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP2]], [[MUL]]
// CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[LOOPVAR_ADDR]], align 8
// CHECK-NEXT:    store i32 [[ADD]], ptr [[TMP4]], align 4
// CHECK-NEXT:    ret void
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@_Z14parallel_for_0v
// CHECK-DEBUG-SAME: () #[[ATTR0:[0-9]+]] !dbg [[DBG8:![0-9]+]] {
// CHECK-DEBUG-NEXT:  entry:
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]]), !dbg [[DBG13:![0-9]+]]
// CHECK-DEBUG-NEXT:    br label [[OMP_PARALLEL:%.*]]
// CHECK-DEBUG:       omp_parallel:
// CHECK-DEBUG-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 0, ptr @_Z14parallel_for_0v..omp_par), !dbg [[DBG14:![0-9]+]]
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT:%.*]]
// CHECK-DEBUG:       omp.par.outlined.exit:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_EXIT_SPLIT:%.*]]
// CHECK-DEBUG:       omp.par.exit.split:
// CHECK-DEBUG-NEXT:    ret void, !dbg [[DBG18:![0-9]+]]
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@_Z14parallel_for_0v..omp_par
// CHECK-DEBUG-SAME: (ptr noalias [[TID_ADDR:%.*]], ptr noalias [[ZERO_ADDR:%.*]]) #[[ATTR1:[0-9]+]] !dbg [[DBG19:![0-9]+]] {
// CHECK-DEBUG-NEXT:  omp.par.entry:
// CHECK-DEBUG-NEXT:    [[TID_ADDR_LOCAL:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[TMP0:%.*]] = load i32, ptr [[TID_ADDR]], align 4
// CHECK-DEBUG-NEXT:    store i32 [[TMP0]], ptr [[TID_ADDR_LOCAL]], align 4
// CHECK-DEBUG-NEXT:    [[TID:%.*]] = load i32, ptr [[TID_ADDR_LOCAL]], align 4
// CHECK-DEBUG-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[AGG_CAPTURED:%.*]] = alloca [[STRUCT_ANON:%.*]], align 8
// CHECK-DEBUG-NEXT:    [[AGG_CAPTURED1:%.*]] = alloca [[STRUCT_ANON_0:%.*]], align 4
// CHECK-DEBUG-NEXT:    [[DOTCOUNT_ADDR:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_LASTITER:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_LOWERBOUND:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_UPPERBOUND:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_STRIDE:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_REGION:%.*]]
// CHECK-DEBUG:       omp.par.region:
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[I]], metadata [[META21:![0-9]+]], metadata !DIExpression()), !dbg [[DBG26:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 0, ptr [[I]], align 4, !dbg [[DBG26]]
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 0, !dbg [[DBG27:![0-9]+]]
// CHECK-DEBUG-NEXT:    store ptr [[I]], ptr [[TMP1]], align 8, !dbg [[DBG27]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_ANON_0]], ptr [[AGG_CAPTURED1]], i32 0, i32 0, !dbg [[DBG27]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = load i32, ptr [[I]], align 4, !dbg [[DBG28:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP3]], ptr [[TMP2]], align 4, !dbg [[DBG27]]
// CHECK-DEBUG-NEXT:    call void @__captured_stmt(ptr [[DOTCOUNT_ADDR]], ptr [[AGG_CAPTURED]]), !dbg [[DBG27]]
// CHECK-DEBUG-NEXT:    [[DOTCOUNT:%.*]] = load i32, ptr [[DOTCOUNT_ADDR]], align 4, !dbg [[DBG27]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_PREHEADER:%.*]], !dbg [[DBG27]]
// CHECK-DEBUG:       omp_loop.preheader:
// CHECK-DEBUG-NEXT:    store i32 0, ptr [[P_LOWERBOUND]], align 4, !dbg [[DBG27]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = sub i32 [[DOTCOUNT]], 1, !dbg [[DBG27]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP4]], ptr [[P_UPPERBOUND]], align 4, !dbg [[DBG27]]
// CHECK-DEBUG-NEXT:    store i32 1, ptr [[P_STRIDE]], align 4, !dbg [[DBG27]]
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB3:[0-9]+]]), !dbg [[DBG27]]
// CHECK-DEBUG-NEXT:    call void @__kmpc_for_static_init_4u(ptr @[[GLOB3]], i32 [[OMP_GLOBAL_THREAD_NUM2]], i32 34, ptr [[P_LASTITER]], ptr [[P_LOWERBOUND]], ptr [[P_UPPERBOUND]], ptr [[P_STRIDE]], i32 1, i32 0), !dbg [[DBG27]]
// CHECK-DEBUG-NEXT:    [[TMP5:%.*]] = load i32, ptr [[P_LOWERBOUND]], align 4, !dbg [[DBG27]]
// CHECK-DEBUG-NEXT:    [[TMP6:%.*]] = load i32, ptr [[P_UPPERBOUND]], align 4, !dbg [[DBG27]]
// CHECK-DEBUG-NEXT:    [[TMP7:%.*]] = sub i32 [[TMP6]], [[TMP5]], !dbg [[DBG27]]
// CHECK-DEBUG-NEXT:    [[TMP8:%.*]] = add i32 [[TMP7]], 1, !dbg [[DBG27]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_HEADER:%.*]], !dbg [[DBG27]]
// CHECK-DEBUG:       omp_loop.header:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_IV:%.*]] = phi i32 [ 0, [[OMP_LOOP_PREHEADER]] ], [ [[OMP_LOOP_NEXT:%.*]], [[OMP_LOOP_INC:%.*]] ], !dbg [[DBG27]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_COND:%.*]], !dbg [[DBG27]]
// CHECK-DEBUG:       omp_loop.cond:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_CMP:%.*]] = icmp ult i32 [[OMP_LOOP_IV]], [[TMP8]], !dbg [[DBG27]]
// CHECK-DEBUG-NEXT:    br i1 [[OMP_LOOP_CMP]], label [[OMP_LOOP_BODY:%.*]], label [[OMP_LOOP_EXIT:%.*]], !dbg [[DBG27]]
// CHECK-DEBUG:       omp_loop.exit:
// CHECK-DEBUG-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB3]], i32 [[OMP_GLOBAL_THREAD_NUM2]]), !dbg [[DBG27]]
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM3:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB3]]), !dbg [[DBG29:![0-9]+]]
// CHECK-DEBUG-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB4:[0-9]+]], i32 [[OMP_GLOBAL_THREAD_NUM3]]), !dbg [[DBG29]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_AFTER:%.*]], !dbg [[DBG27]]
// CHECK-DEBUG:       omp_loop.after:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_REGION_PARALLEL_AFTER:%.*]], !dbg [[DBG30:![0-9]+]]
// CHECK-DEBUG:       omp.par.region.parallel.after:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_PRE_FINALIZE:%.*]]
// CHECK-DEBUG:       omp.par.pre_finalize:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT_EXITSTUB:%.*]], !dbg [[DBG30]]
// CHECK-DEBUG:       omp_loop.body:
// CHECK-DEBUG-NEXT:    [[TMP9:%.*]] = add i32 [[OMP_LOOP_IV]], [[TMP5]], !dbg [[DBG29]]
// CHECK-DEBUG-NEXT:    call void @__captured_stmt.1(ptr [[I]], i32 [[TMP9]], ptr [[AGG_CAPTURED1]]), !dbg [[DBG27]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_INC]], !dbg [[DBG27]]
// CHECK-DEBUG:       omp_loop.inc:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_NEXT]] = add nuw i32 [[OMP_LOOP_IV]], 1, !dbg [[DBG27]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_HEADER]], !dbg [[DBG27]]
// CHECK-DEBUG:       omp.par.outlined.exit.exitStub:
// CHECK-DEBUG-NEXT:    ret void
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@__captured_stmt
// CHECK-DEBUG-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[DISTANCE:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR4:[0-9]+]] !dbg [[DBG31:![0-9]+]] {
// CHECK-DEBUG-NEXT:  entry:
// CHECK-DEBUG-NEXT:    [[DISTANCE_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[DOTSTART:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[DOTSTOP:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[DOTSTEP:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    store ptr [[DISTANCE]], ptr [[DISTANCE_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DISTANCE_ADDR]], metadata [[META39:![0-9]+]], metadata !DIExpression()), !dbg [[DBG40:![0-9]+]]
// CHECK-DEBUG-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[__CONTEXT_ADDR]], metadata [[META41:![0-9]+]], metadata !DIExpression()), !dbg [[DBG40]]
// CHECK-DEBUG-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTART]], metadata [[META42:![0-9]+]], metadata !DIExpression()), !dbg [[DBG44:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON:%.*]], ptr [[TMP0]], i32 0, i32 0, !dbg [[DBG45:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP1]], align 8, !dbg [[DBG45]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = load i32, ptr [[TMP2]], align 4, !dbg [[DBG45]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP3]], ptr [[DOTSTART]], align 4, !dbg [[DBG44]]
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTOP]], metadata [[META47:![0-9]+]], metadata !DIExpression()), !dbg [[DBG48:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 100, ptr [[DOTSTOP]], align 4, !dbg [[DBG48]]
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTEP]], metadata [[META49:![0-9]+]], metadata !DIExpression()), !dbg [[DBG48]]
// CHECK-DEBUG-NEXT:    store i32 1, ptr [[DOTSTEP]], align 4, !dbg [[DBG48]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTSTART]], align 4, !dbg [[DBG48]]
// CHECK-DEBUG-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTSTOP]], align 4, !dbg [[DBG48]]
// CHECK-DEBUG-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP4]], [[TMP5]], !dbg [[DBG48]]
// CHECK-DEBUG-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]], !dbg [[DBG48]]
// CHECK-DEBUG:       cond.true:
// CHECK-DEBUG-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTSTOP]], align 4, !dbg [[DBG48]]
// CHECK-DEBUG-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTSTART]], align 4, !dbg [[DBG48]]
// CHECK-DEBUG-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], [[TMP7]], !dbg [[DBG48]]
// CHECK-DEBUG-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTSTEP]], align 4, !dbg [[DBG48]]
// CHECK-DEBUG-NEXT:    [[SUB1:%.*]] = sub i32 [[TMP8]], 1, !dbg [[DBG48]]
// CHECK-DEBUG-NEXT:    [[ADD:%.*]] = add i32 [[SUB]], [[SUB1]], !dbg [[DBG48]]
// CHECK-DEBUG-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTSTEP]], align 4, !dbg [[DBG48]]
// CHECK-DEBUG-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], [[TMP9]], !dbg [[DBG48]]
// CHECK-DEBUG-NEXT:    br label [[COND_END:%.*]], !dbg [[DBG48]]
// CHECK-DEBUG:       cond.false:
// CHECK-DEBUG-NEXT:    br label [[COND_END]], !dbg [[DBG48]]
// CHECK-DEBUG:       cond.end:
// CHECK-DEBUG-NEXT:    [[COND:%.*]] = phi i32 [ [[DIV]], [[COND_TRUE]] ], [ 0, [[COND_FALSE]] ], !dbg [[DBG48]]
// CHECK-DEBUG-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DISTANCE_ADDR]], align 8, !dbg [[DBG48]]
// CHECK-DEBUG-NEXT:    store i32 [[COND]], ptr [[TMP10]], align 4, !dbg [[DBG48]]
// CHECK-DEBUG-NEXT:    ret void, !dbg [[DBG50:![0-9]+]]
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@__captured_stmt.1
// CHECK-DEBUG-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[LOOPVAR:%.*]], i32 noundef [[LOGICAL:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR4]] !dbg [[DBG52:![0-9]+]] {
// CHECK-DEBUG-NEXT:  entry:
// CHECK-DEBUG-NEXT:    [[LOOPVAR_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[LOGICAL_ADDR:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    store ptr [[LOOPVAR]], ptr [[LOOPVAR_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[LOOPVAR_ADDR]], metadata [[META60:![0-9]+]], metadata !DIExpression()), !dbg [[DBG61:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 [[LOGICAL]], ptr [[LOGICAL_ADDR]], align 4
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[LOGICAL_ADDR]], metadata [[META62:![0-9]+]], metadata !DIExpression()), !dbg [[DBG61]]
// CHECK-DEBUG-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[__CONTEXT_ADDR]], metadata [[META63:![0-9]+]], metadata !DIExpression()), !dbg [[DBG61]]
// CHECK-DEBUG-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_0:%.*]], ptr [[TMP0]], i32 0, i32 0, !dbg [[DBG64:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4, !dbg [[DBG64]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = load i32, ptr [[LOGICAL_ADDR]], align 4, !dbg [[DBG66:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[MUL:%.*]] = mul i32 1, [[TMP3]], !dbg [[DBG66]]
// CHECK-DEBUG-NEXT:    [[ADD:%.*]] = add i32 [[TMP2]], [[MUL]], !dbg [[DBG66]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[LOOPVAR_ADDR]], align 8, !dbg [[DBG66]]
// CHECK-DEBUG-NEXT:    store i32 [[ADD]], ptr [[TMP4]], align 4, !dbg [[DBG61]]
// CHECK-DEBUG-NEXT:    ret void, !dbg [[DBG64]]
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@_Z14parallel_for_1Pfid
// CHECK-DEBUG-SAME: (ptr noundef [[R:%.*]], i32 noundef [[A:%.*]], double noundef [[B:%.*]]) #[[ATTR0]] !dbg [[DBG69:![0-9]+]] {
// CHECK-DEBUG-NEXT:  entry:
// CHECK-DEBUG-NEXT:    [[STRUCTARG17:%.*]] = alloca { ptr, ptr, ptr }, align 8
// CHECK-DEBUG-NEXT:    [[R_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[B_ADDR:%.*]] = alloca double, align 8
// CHECK-DEBUG-NEXT:    store ptr [[R]], ptr [[R_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[R_ADDR]], metadata [[META75:![0-9]+]], metadata !DIExpression()), !dbg [[DBG76:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[A_ADDR]], metadata [[META77:![0-9]+]], metadata !DIExpression()), !dbg [[DBG78:![0-9]+]]
// CHECK-DEBUG-NEXT:    store double [[B]], ptr [[B_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[B_ADDR]], metadata [[META79:![0-9]+]], metadata !DIExpression()), !dbg [[DBG80:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB6:[0-9]+]]), !dbg [[DBG81:![0-9]+]]
// CHECK-DEBUG-NEXT:    br label [[OMP_PARALLEL:%.*]]
// CHECK-DEBUG:       omp_parallel:
// CHECK-DEBUG-NEXT:    [[GEP_A_ADDR18:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG17]], i32 0, i32 0
// CHECK-DEBUG-NEXT:    store ptr [[A_ADDR]], ptr [[GEP_A_ADDR18]], align 8
// CHECK-DEBUG-NEXT:    [[GEP_B_ADDR19:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG17]], i32 0, i32 1
// CHECK-DEBUG-NEXT:    store ptr [[B_ADDR]], ptr [[GEP_B_ADDR19]], align 8
// CHECK-DEBUG-NEXT:    [[GEP_R_ADDR20:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG17]], i32 0, i32 2
// CHECK-DEBUG-NEXT:    store ptr [[R_ADDR]], ptr [[GEP_R_ADDR20]], align 8
// CHECK-DEBUG-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB6]], i32 1, ptr @_Z14parallel_for_1Pfid..omp_par.4, ptr [[STRUCTARG17]]), !dbg [[DBG82:![0-9]+]]
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT16:%.*]]
// CHECK-DEBUG:       omp.par.outlined.exit16:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_EXIT_SPLIT:%.*]]
// CHECK-DEBUG:       omp.par.exit.split:
// CHECK-DEBUG-NEXT:    ret void, !dbg [[DBG84:![0-9]+]]
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@_Z14parallel_for_1Pfid..omp_par.4
// CHECK-DEBUG-SAME: (ptr noalias [[TID_ADDR:%.*]], ptr noalias [[ZERO_ADDR:%.*]], ptr [[TMP0:%.*]]) #[[ATTR1]] !dbg [[DBG85:![0-9]+]] {
// CHECK-DEBUG-NEXT:  omp.par.entry:
// CHECK-DEBUG-NEXT:    [[GEP_A_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 0
// CHECK-DEBUG-NEXT:    [[LOADGEP_A_ADDR:%.*]] = load ptr, ptr [[GEP_A_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[GEP_B_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 1
// CHECK-DEBUG-NEXT:    [[LOADGEP_B_ADDR:%.*]] = load ptr, ptr [[GEP_B_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[GEP_R_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 2
// CHECK-DEBUG-NEXT:    [[LOADGEP_R_ADDR:%.*]] = load ptr, ptr [[GEP_R_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[STRUCTARG:%.*]] = alloca { ptr, ptr, ptr }, align 8
// CHECK-DEBUG-NEXT:    [[TID_ADDR_LOCAL:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TID_ADDR]], align 4
// CHECK-DEBUG-NEXT:    store i32 [[TMP1]], ptr [[TID_ADDR_LOCAL]], align 4
// CHECK-DEBUG-NEXT:    [[TID:%.*]] = load i32, ptr [[TID_ADDR_LOCAL]], align 4
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_REGION:%.*]]
// CHECK-DEBUG:       omp.par.region:
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB8:[0-9]+]]), !dbg [[DBG86:![0-9]+]]
// CHECK-DEBUG-NEXT:    br label [[OMP_PARALLEL:%.*]]
// CHECK-DEBUG:       omp_parallel:
// CHECK-DEBUG-NEXT:    [[GEP_A_ADDR1:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG]], i32 0, i32 0
// CHECK-DEBUG-NEXT:    store ptr [[LOADGEP_A_ADDR]], ptr [[GEP_A_ADDR1]], align 8
// CHECK-DEBUG-NEXT:    [[GEP_B_ADDR2:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG]], i32 0, i32 1
// CHECK-DEBUG-NEXT:    store ptr [[LOADGEP_B_ADDR]], ptr [[GEP_B_ADDR2]], align 8
// CHECK-DEBUG-NEXT:    [[GEP_R_ADDR3:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG]], i32 0, i32 2
// CHECK-DEBUG-NEXT:    store ptr [[LOADGEP_R_ADDR]], ptr [[GEP_R_ADDR3]], align 8
// CHECK-DEBUG-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB8]], i32 1, ptr @_Z14parallel_for_1Pfid..omp_par, ptr [[STRUCTARG]]), !dbg [[DBG88:![0-9]+]]
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT:%.*]]
// CHECK-DEBUG:       omp.par.outlined.exit:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_EXIT7_SPLIT:%.*]]
// CHECK-DEBUG:       omp.par.exit7.split:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_REGION_PARALLEL_AFTER:%.*]], !dbg [[DBG92:![0-9]+]]
// CHECK-DEBUG:       omp.par.region.parallel.after:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_PRE_FINALIZE:%.*]]
// CHECK-DEBUG:       omp.par.pre_finalize:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT16_EXITSTUB:%.*]], !dbg [[DBG92]]
// CHECK-DEBUG:       omp.par.outlined.exit16.exitStub:
// CHECK-DEBUG-NEXT:    ret void
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@_Z14parallel_for_1Pfid..omp_par
// CHECK-DEBUG-SAME: (ptr noalias [[TID_ADDR2:%.*]], ptr noalias [[ZERO_ADDR3:%.*]], ptr [[TMP0:%.*]]) #[[ATTR1]] !dbg [[DBG93:![0-9]+]] {
// CHECK-DEBUG-NEXT:  omp.par.entry4:
// CHECK-DEBUG-NEXT:    [[GEP_A_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 0
// CHECK-DEBUG-NEXT:    [[LOADGEP_A_ADDR:%.*]] = load ptr, ptr [[GEP_A_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[GEP_B_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 1
// CHECK-DEBUG-NEXT:    [[LOADGEP_B_ADDR:%.*]] = load ptr, ptr [[GEP_B_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[GEP_R_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 2
// CHECK-DEBUG-NEXT:    [[LOADGEP_R_ADDR:%.*]] = load ptr, ptr [[GEP_R_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[TID_ADDR_LOCAL8:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TID_ADDR2]], align 4
// CHECK-DEBUG-NEXT:    store i32 [[TMP1]], ptr [[TID_ADDR_LOCAL8]], align 4
// CHECK-DEBUG-NEXT:    [[TID9:%.*]] = load i32, ptr [[TID_ADDR_LOCAL8]], align 4
// CHECK-DEBUG-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[AGG_CAPTURED:%.*]] = alloca [[STRUCT_ANON_1:%.*]], align 8
// CHECK-DEBUG-NEXT:    [[AGG_CAPTURED12:%.*]] = alloca [[STRUCT_ANON_2:%.*]], align 4
// CHECK-DEBUG-NEXT:    [[DOTCOUNT_ADDR:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_LASTITER:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_LOWERBOUND:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_UPPERBOUND:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_STRIDE:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_REGION5:%.*]]
// CHECK-DEBUG:       omp.par.region5:
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[I]], metadata [[META94:![0-9]+]], metadata !DIExpression()), !dbg [[DBG99:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 0, ptr [[I]], align 4, !dbg [[DBG99]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_ANON_1]], ptr [[AGG_CAPTURED]], i32 0, i32 0, !dbg [[DBG100:![0-9]+]]
// CHECK-DEBUG-NEXT:    store ptr [[I]], ptr [[TMP2]], align 8, !dbg [[DBG100]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [[STRUCT_ANON_2]], ptr [[AGG_CAPTURED12]], i32 0, i32 0, !dbg [[DBG100]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load i32, ptr [[I]], align 4, !dbg [[DBG101:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP4]], ptr [[TMP3]], align 4, !dbg [[DBG100]]
// CHECK-DEBUG-NEXT:    call void @__captured_stmt.2(ptr [[DOTCOUNT_ADDR]], ptr [[AGG_CAPTURED]]), !dbg [[DBG100]]
// CHECK-DEBUG-NEXT:    [[DOTCOUNT:%.*]] = load i32, ptr [[DOTCOUNT_ADDR]], align 4, !dbg [[DBG100]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_PREHEADER:%.*]], !dbg [[DBG100]]
// CHECK-DEBUG:       omp_loop.preheader:
// CHECK-DEBUG-NEXT:    store i32 0, ptr [[P_LOWERBOUND]], align 4, !dbg [[DBG100]]
// CHECK-DEBUG-NEXT:    [[TMP5:%.*]] = sub i32 [[DOTCOUNT]], 1, !dbg [[DBG100]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP5]], ptr [[P_UPPERBOUND]], align 4, !dbg [[DBG100]]
// CHECK-DEBUG-NEXT:    store i32 1, ptr [[P_STRIDE]], align 4, !dbg [[DBG100]]
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM14:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB10:[0-9]+]]), !dbg [[DBG100]]
// CHECK-DEBUG-NEXT:    call void @__kmpc_for_static_init_4u(ptr @[[GLOB10]], i32 [[OMP_GLOBAL_THREAD_NUM14]], i32 34, ptr [[P_LASTITER]], ptr [[P_LOWERBOUND]], ptr [[P_UPPERBOUND]], ptr [[P_STRIDE]], i32 1, i32 0), !dbg [[DBG100]]
// CHECK-DEBUG-NEXT:    [[TMP6:%.*]] = load i32, ptr [[P_LOWERBOUND]], align 4, !dbg [[DBG100]]
// CHECK-DEBUG-NEXT:    [[TMP7:%.*]] = load i32, ptr [[P_UPPERBOUND]], align 4, !dbg [[DBG100]]
// CHECK-DEBUG-NEXT:    [[TMP8:%.*]] = sub i32 [[TMP7]], [[TMP6]], !dbg [[DBG100]]
// CHECK-DEBUG-NEXT:    [[TMP9:%.*]] = add i32 [[TMP8]], 1, !dbg [[DBG100]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_HEADER:%.*]], !dbg [[DBG100]]
// CHECK-DEBUG:       omp_loop.header:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_IV:%.*]] = phi i32 [ 0, [[OMP_LOOP_PREHEADER]] ], [ [[OMP_LOOP_NEXT:%.*]], [[OMP_LOOP_INC:%.*]] ], !dbg [[DBG100]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_COND:%.*]], !dbg [[DBG100]]
// CHECK-DEBUG:       omp_loop.cond:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_CMP:%.*]] = icmp ult i32 [[OMP_LOOP_IV]], [[TMP9]], !dbg [[DBG100]]
// CHECK-DEBUG-NEXT:    br i1 [[OMP_LOOP_CMP]], label [[OMP_LOOP_BODY:%.*]], label [[OMP_LOOP_EXIT:%.*]], !dbg [[DBG100]]
// CHECK-DEBUG:       omp_loop.exit:
// CHECK-DEBUG-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB10]], i32 [[OMP_GLOBAL_THREAD_NUM14]]), !dbg [[DBG100]]
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM15:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB10]]), !dbg [[DBG102:![0-9]+]]
// CHECK-DEBUG-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB11:[0-9]+]], i32 [[OMP_GLOBAL_THREAD_NUM15]]), !dbg [[DBG102]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_AFTER:%.*]], !dbg [[DBG100]]
// CHECK-DEBUG:       omp_loop.after:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_REGION5_PARALLEL_AFTER:%.*]], !dbg [[DBG103:![0-9]+]]
// CHECK-DEBUG:       omp.par.region5.parallel.after:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_PRE_FINALIZE6:%.*]]
// CHECK-DEBUG:       omp.par.pre_finalize6:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT_EXITSTUB:%.*]], !dbg [[DBG103]]
// CHECK-DEBUG:       omp_loop.body:
// CHECK-DEBUG-NEXT:    [[TMP10:%.*]] = add i32 [[OMP_LOOP_IV]], [[TMP6]], !dbg [[DBG102]]
// CHECK-DEBUG-NEXT:    call void @__captured_stmt.3(ptr [[I]], i32 [[TMP10]], ptr [[AGG_CAPTURED12]]), !dbg [[DBG100]]
// CHECK-DEBUG-NEXT:    [[TMP11:%.*]] = load i32, ptr [[LOADGEP_A_ADDR]], align 4, !dbg [[DBG104:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[CONV:%.*]] = sitofp i32 [[TMP11]] to double, !dbg [[DBG104]]
// CHECK-DEBUG-NEXT:    [[TMP12:%.*]] = load double, ptr [[LOADGEP_B_ADDR]], align 8, !dbg [[DBG106:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[ADD:%.*]] = fadd double [[CONV]], [[TMP12]], !dbg [[DBG107:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[CONV13:%.*]] = fptrunc double [[ADD]] to float, !dbg [[DBG104]]
// CHECK-DEBUG-NEXT:    [[TMP13:%.*]] = load ptr, ptr [[LOADGEP_R_ADDR]], align 8, !dbg [[DBG108:![0-9]+]]
// CHECK-DEBUG-NEXT:    store float [[CONV13]], ptr [[TMP13]], align 4, !dbg [[DBG109:![0-9]+]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_INC]], !dbg [[DBG100]]
// CHECK-DEBUG:       omp_loop.inc:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_NEXT]] = add nuw i32 [[OMP_LOOP_IV]], 1, !dbg [[DBG100]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_HEADER]], !dbg [[DBG100]]
// CHECK-DEBUG:       omp.par.outlined.exit.exitStub:
// CHECK-DEBUG-NEXT:    ret void
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@__captured_stmt.2
// CHECK-DEBUG-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[DISTANCE:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR4]] !dbg [[DBG110:![0-9]+]] {
// CHECK-DEBUG-NEXT:  entry:
// CHECK-DEBUG-NEXT:    [[DISTANCE_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[DOTSTART:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[DOTSTOP:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[DOTSTEP:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    store ptr [[DISTANCE]], ptr [[DISTANCE_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DISTANCE_ADDR]], metadata [[META111:![0-9]+]], metadata !DIExpression()), !dbg [[DBG112:![0-9]+]]
// CHECK-DEBUG-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[__CONTEXT_ADDR]], metadata [[META113:![0-9]+]], metadata !DIExpression()), !dbg [[DBG112]]
// CHECK-DEBUG-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTART]], metadata [[META114:![0-9]+]], metadata !DIExpression()), !dbg [[DBG116:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_1:%.*]], ptr [[TMP0]], i32 0, i32 0, !dbg [[DBG117:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP1]], align 8, !dbg [[DBG117]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = load i32, ptr [[TMP2]], align 4, !dbg [[DBG117]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP3]], ptr [[DOTSTART]], align 4, !dbg [[DBG116]]
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTOP]], metadata [[META119:![0-9]+]], metadata !DIExpression()), !dbg [[DBG120:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 100, ptr [[DOTSTOP]], align 4, !dbg [[DBG120]]
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTEP]], metadata [[META121:![0-9]+]], metadata !DIExpression()), !dbg [[DBG120]]
// CHECK-DEBUG-NEXT:    store i32 1, ptr [[DOTSTEP]], align 4, !dbg [[DBG120]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTSTART]], align 4, !dbg [[DBG120]]
// CHECK-DEBUG-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTSTOP]], align 4, !dbg [[DBG120]]
// CHECK-DEBUG-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP4]], [[TMP5]], !dbg [[DBG120]]
// CHECK-DEBUG-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]], !dbg [[DBG120]]
// CHECK-DEBUG:       cond.true:
// CHECK-DEBUG-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTSTOP]], align 4, !dbg [[DBG120]]
// CHECK-DEBUG-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTSTART]], align 4, !dbg [[DBG120]]
// CHECK-DEBUG-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], [[TMP7]], !dbg [[DBG120]]
// CHECK-DEBUG-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTSTEP]], align 4, !dbg [[DBG120]]
// CHECK-DEBUG-NEXT:    [[SUB1:%.*]] = sub i32 [[TMP8]], 1, !dbg [[DBG120]]
// CHECK-DEBUG-NEXT:    [[ADD:%.*]] = add i32 [[SUB]], [[SUB1]], !dbg [[DBG120]]
// CHECK-DEBUG-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTSTEP]], align 4, !dbg [[DBG120]]
// CHECK-DEBUG-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], [[TMP9]], !dbg [[DBG120]]
// CHECK-DEBUG-NEXT:    br label [[COND_END:%.*]], !dbg [[DBG120]]
// CHECK-DEBUG:       cond.false:
// CHECK-DEBUG-NEXT:    br label [[COND_END]], !dbg [[DBG120]]
// CHECK-DEBUG:       cond.end:
// CHECK-DEBUG-NEXT:    [[COND:%.*]] = phi i32 [ [[DIV]], [[COND_TRUE]] ], [ 0, [[COND_FALSE]] ], !dbg [[DBG120]]
// CHECK-DEBUG-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DISTANCE_ADDR]], align 8, !dbg [[DBG120]]
// CHECK-DEBUG-NEXT:    store i32 [[COND]], ptr [[TMP10]], align 4, !dbg [[DBG120]]
// CHECK-DEBUG-NEXT:    ret void, !dbg [[DBG122:![0-9]+]]
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@__captured_stmt.3
// CHECK-DEBUG-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[LOOPVAR:%.*]], i32 noundef [[LOGICAL:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR4]] !dbg [[DBG124:![0-9]+]] {
// CHECK-DEBUG-NEXT:  entry:
// CHECK-DEBUG-NEXT:    [[LOOPVAR_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[LOGICAL_ADDR:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    store ptr [[LOOPVAR]], ptr [[LOOPVAR_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[LOOPVAR_ADDR]], metadata [[META125:![0-9]+]], metadata !DIExpression()), !dbg [[DBG126:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 [[LOGICAL]], ptr [[LOGICAL_ADDR]], align 4
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[LOGICAL_ADDR]], metadata [[META127:![0-9]+]], metadata !DIExpression()), !dbg [[DBG126]]
// CHECK-DEBUG-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[__CONTEXT_ADDR]], metadata [[META128:![0-9]+]], metadata !DIExpression()), !dbg [[DBG126]]
// CHECK-DEBUG-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_2:%.*]], ptr [[TMP0]], i32 0, i32 0, !dbg [[DBG129:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4, !dbg [[DBG129]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = load i32, ptr [[LOGICAL_ADDR]], align 4, !dbg [[DBG131:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[MUL:%.*]] = mul i32 1, [[TMP3]], !dbg [[DBG131]]
// CHECK-DEBUG-NEXT:    [[ADD:%.*]] = add i32 [[TMP2]], [[MUL]], !dbg [[DBG131]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[LOOPVAR_ADDR]], align 8, !dbg [[DBG131]]
// CHECK-DEBUG-NEXT:    store i32 [[ADD]], ptr [[TMP4]], align 4, !dbg [[DBG126]]
// CHECK-DEBUG-NEXT:    ret void, !dbg [[DBG129]]
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@_Z14parallel_for_2Pfid
// CHECK-DEBUG-SAME: (ptr noundef [[R:%.*]], i32 noundef [[A:%.*]], double noundef [[B:%.*]]) #[[ATTR0]] !dbg [[DBG132:![0-9]+]] {
// CHECK-DEBUG-NEXT:  entry:
// CHECK-DEBUG-NEXT:    [[STRUCTARG:%.*]] = alloca { ptr, ptr, ptr }, align 8
// CHECK-DEBUG-NEXT:    [[R_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[B_ADDR:%.*]] = alloca double, align 8
// CHECK-DEBUG-NEXT:    [[I185:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[AGG_CAPTURED186:%.*]] = alloca [[STRUCT_ANON_17:%.*]], align 8
// CHECK-DEBUG-NEXT:    [[AGG_CAPTURED187:%.*]] = alloca [[STRUCT_ANON_18:%.*]], align 4
// CHECK-DEBUG-NEXT:    [[DOTCOUNT_ADDR188:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_LASTITER203:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_LOWERBOUND204:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_UPPERBOUND205:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_STRIDE206:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    store ptr [[R]], ptr [[R_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[R_ADDR]], metadata [[META133:![0-9]+]], metadata !DIExpression()), !dbg [[DBG134:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[A_ADDR]], metadata [[META135:![0-9]+]], metadata !DIExpression()), !dbg [[DBG136:![0-9]+]]
// CHECK-DEBUG-NEXT:    store double [[B]], ptr [[B_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[B_ADDR]], metadata [[META137:![0-9]+]], metadata !DIExpression()), !dbg [[DBG138:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB13:[0-9]+]]), !dbg [[DBG139:![0-9]+]]
// CHECK-DEBUG-NEXT:    br label [[OMP_PARALLEL:%.*]]
// CHECK-DEBUG:       omp_parallel:
// CHECK-DEBUG-NEXT:    [[GEP_A_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG]], i32 0, i32 0
// CHECK-DEBUG-NEXT:    store ptr [[A_ADDR]], ptr [[GEP_A_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[GEP_B_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG]], i32 0, i32 1
// CHECK-DEBUG-NEXT:    store ptr [[B_ADDR]], ptr [[GEP_B_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[GEP_R_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG]], i32 0, i32 2
// CHECK-DEBUG-NEXT:    store ptr [[R_ADDR]], ptr [[GEP_R_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB13]], i32 1, ptr @_Z14parallel_for_2Pfid..omp_par.23, ptr [[STRUCTARG]]), !dbg [[DBG140:![0-9]+]]
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT184:%.*]]
// CHECK-DEBUG:       omp.par.outlined.exit184:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_EXIT_SPLIT:%.*]]
// CHECK-DEBUG:       omp.par.exit.split:
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[I185]], metadata [[META144:![0-9]+]], metadata !DIExpression()), !dbg [[DBG147:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 0, ptr [[I185]], align 4, !dbg [[DBG147]]
// CHECK-DEBUG-NEXT:    [[TMP0:%.*]] = getelementptr inbounds [[STRUCT_ANON_17]], ptr [[AGG_CAPTURED186]], i32 0, i32 0, !dbg [[DBG148:![0-9]+]]
// CHECK-DEBUG-NEXT:    store ptr [[I185]], ptr [[TMP0]], align 8, !dbg [[DBG148]]
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_18]], ptr [[AGG_CAPTURED187]], i32 0, i32 0, !dbg [[DBG148]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = load i32, ptr [[I185]], align 4, !dbg [[DBG149:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP2]], ptr [[TMP1]], align 4, !dbg [[DBG148]]
// CHECK-DEBUG-NEXT:    call void @__captured_stmt.19(ptr [[DOTCOUNT_ADDR188]], ptr [[AGG_CAPTURED186]]), !dbg [[DBG148]]
// CHECK-DEBUG-NEXT:    [[DOTCOUNT189:%.*]] = load i32, ptr [[DOTCOUNT_ADDR188]], align 4, !dbg [[DBG148]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_PREHEADER190:%.*]], !dbg [[DBG148]]
// CHECK-DEBUG:       omp_loop.preheader190:
// CHECK-DEBUG-NEXT:    store i32 0, ptr [[P_LOWERBOUND204]], align 4, !dbg [[DBG148]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = sub i32 [[DOTCOUNT189]], 1, !dbg [[DBG148]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP3]], ptr [[P_UPPERBOUND205]], align 4, !dbg [[DBG148]]
// CHECK-DEBUG-NEXT:    store i32 1, ptr [[P_STRIDE206]], align 4, !dbg [[DBG148]]
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM207:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB42:[0-9]+]]), !dbg [[DBG148]]
// CHECK-DEBUG-NEXT:    call void @__kmpc_for_static_init_4u(ptr @[[GLOB42]], i32 [[OMP_GLOBAL_THREAD_NUM207]], i32 34, ptr [[P_LASTITER203]], ptr [[P_LOWERBOUND204]], ptr [[P_UPPERBOUND205]], ptr [[P_STRIDE206]], i32 1, i32 0), !dbg [[DBG148]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load i32, ptr [[P_LOWERBOUND204]], align 4, !dbg [[DBG148]]
// CHECK-DEBUG-NEXT:    [[TMP5:%.*]] = load i32, ptr [[P_UPPERBOUND205]], align 4, !dbg [[DBG148]]
// CHECK-DEBUG-NEXT:    [[TMP6:%.*]] = sub i32 [[TMP5]], [[TMP4]], !dbg [[DBG148]]
// CHECK-DEBUG-NEXT:    [[TMP7:%.*]] = add i32 [[TMP6]], 1, !dbg [[DBG148]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_HEADER191:%.*]], !dbg [[DBG148]]
// CHECK-DEBUG:       omp_loop.header191:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_IV197:%.*]] = phi i32 [ 0, [[OMP_LOOP_PREHEADER190]] ], [ [[OMP_LOOP_NEXT199:%.*]], [[OMP_LOOP_INC194:%.*]] ], !dbg [[DBG148]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_COND192:%.*]], !dbg [[DBG148]]
// CHECK-DEBUG:       omp_loop.cond192:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_CMP198:%.*]] = icmp ult i32 [[OMP_LOOP_IV197]], [[TMP7]], !dbg [[DBG148]]
// CHECK-DEBUG-NEXT:    br i1 [[OMP_LOOP_CMP198]], label [[OMP_LOOP_BODY193:%.*]], label [[OMP_LOOP_EXIT195:%.*]], !dbg [[DBG148]]
// CHECK-DEBUG:       omp_loop.body193:
// CHECK-DEBUG-NEXT:    [[TMP8:%.*]] = add i32 [[OMP_LOOP_IV197]], [[TMP4]], !dbg [[DBG150:![0-9]+]]
// CHECK-DEBUG-NEXT:    call void @__captured_stmt.20(ptr [[I185]], i32 [[TMP8]], ptr [[AGG_CAPTURED187]]), !dbg [[DBG148]]
// CHECK-DEBUG-NEXT:    [[TMP9:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG151:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[CONV200:%.*]] = sitofp i32 [[TMP9]] to double, !dbg [[DBG151]]
// CHECK-DEBUG-NEXT:    [[TMP10:%.*]] = load double, ptr [[B_ADDR]], align 8, !dbg [[DBG150]]
// CHECK-DEBUG-NEXT:    [[ADD201:%.*]] = fadd double [[CONV200]], [[TMP10]], !dbg [[DBG152:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[CONV202:%.*]] = fptrunc double [[ADD201]] to float, !dbg [[DBG151]]
// CHECK-DEBUG-NEXT:    [[TMP11:%.*]] = load ptr, ptr [[R_ADDR]], align 8, !dbg [[DBG153:![0-9]+]]
// CHECK-DEBUG-NEXT:    store float [[CONV202]], ptr [[TMP11]], align 4, !dbg [[DBG154:![0-9]+]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_INC194]], !dbg [[DBG148]]
// CHECK-DEBUG:       omp_loop.inc194:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_NEXT199]] = add nuw i32 [[OMP_LOOP_IV197]], 1, !dbg [[DBG148]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_HEADER191]], !dbg [[DBG148]]
// CHECK-DEBUG:       omp_loop.exit195:
// CHECK-DEBUG-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB42]], i32 [[OMP_GLOBAL_THREAD_NUM207]]), !dbg [[DBG148]]
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM208:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB42]]), !dbg [[DBG150]]
// CHECK-DEBUG-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB43:[0-9]+]], i32 [[OMP_GLOBAL_THREAD_NUM208]]), !dbg [[DBG150]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_AFTER196:%.*]], !dbg [[DBG148]]
// CHECK-DEBUG:       omp_loop.after196:
// CHECK-DEBUG-NEXT:    ret void, !dbg [[DBG155:![0-9]+]]
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@_Z14parallel_for_2Pfid..omp_par.23
// CHECK-DEBUG-SAME: (ptr noalias [[TID_ADDR:%.*]], ptr noalias [[ZERO_ADDR:%.*]], ptr [[TMP0:%.*]]) #[[ATTR1]] !dbg [[DBG156:![0-9]+]] {
// CHECK-DEBUG-NEXT:  omp.par.entry:
// CHECK-DEBUG-NEXT:    [[GEP_A_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 0
// CHECK-DEBUG-NEXT:    [[LOADGEP_A_ADDR:%.*]] = load ptr, ptr [[GEP_A_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[GEP_B_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 1
// CHECK-DEBUG-NEXT:    [[LOADGEP_B_ADDR:%.*]] = load ptr, ptr [[GEP_B_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[GEP_R_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 2
// CHECK-DEBUG-NEXT:    [[LOADGEP_R_ADDR:%.*]] = load ptr, ptr [[GEP_R_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[STRUCTARG214:%.*]] = alloca { ptr, ptr, ptr }, align 8
// CHECK-DEBUG-NEXT:    [[TID_ADDR_LOCAL:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TID_ADDR]], align 4
// CHECK-DEBUG-NEXT:    store i32 [[TMP1]], ptr [[TID_ADDR_LOCAL]], align 4
// CHECK-DEBUG-NEXT:    [[TID:%.*]] = load i32, ptr [[TID_ADDR_LOCAL]], align 4
// CHECK-DEBUG-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[AGG_CAPTURED:%.*]] = alloca [[STRUCT_ANON_3:%.*]], align 8
// CHECK-DEBUG-NEXT:    [[AGG_CAPTURED1:%.*]] = alloca [[STRUCT_ANON_4:%.*]], align 4
// CHECK-DEBUG-NEXT:    [[DOTCOUNT_ADDR:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_LASTITER:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_LOWERBOUND:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_UPPERBOUND:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_STRIDE:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[I160:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[AGG_CAPTURED161:%.*]] = alloca [[STRUCT_ANON_15:%.*]], align 8
// CHECK-DEBUG-NEXT:    [[AGG_CAPTURED162:%.*]] = alloca [[STRUCT_ANON_16:%.*]], align 4
// CHECK-DEBUG-NEXT:    [[DOTCOUNT_ADDR163:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_LASTITER178:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_LOWERBOUND179:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_UPPERBOUND180:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_STRIDE181:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_REGION:%.*]]
// CHECK-DEBUG:       omp.par.region:
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[I]], metadata [[META157:![0-9]+]], metadata !DIExpression()), !dbg [[DBG161:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 0, ptr [[I]], align 4, !dbg [[DBG161]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_ANON_3]], ptr [[AGG_CAPTURED]], i32 0, i32 0, !dbg [[DBG162:![0-9]+]]
// CHECK-DEBUG-NEXT:    store ptr [[I]], ptr [[TMP2]], align 8, !dbg [[DBG162]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [[STRUCT_ANON_4]], ptr [[AGG_CAPTURED1]], i32 0, i32 0, !dbg [[DBG162]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load i32, ptr [[I]], align 4, !dbg [[DBG163:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP4]], ptr [[TMP3]], align 4, !dbg [[DBG162]]
// CHECK-DEBUG-NEXT:    call void @__captured_stmt.5(ptr [[DOTCOUNT_ADDR]], ptr [[AGG_CAPTURED]]), !dbg [[DBG162]]
// CHECK-DEBUG-NEXT:    [[DOTCOUNT:%.*]] = load i32, ptr [[DOTCOUNT_ADDR]], align 4, !dbg [[DBG162]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_PREHEADER:%.*]], !dbg [[DBG162]]
// CHECK-DEBUG:       omp_loop.preheader:
// CHECK-DEBUG-NEXT:    store i32 0, ptr [[P_LOWERBOUND]], align 4, !dbg [[DBG162]]
// CHECK-DEBUG-NEXT:    [[TMP5:%.*]] = sub i32 [[DOTCOUNT]], 1, !dbg [[DBG162]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP5]], ptr [[P_UPPERBOUND]], align 4, !dbg [[DBG162]]
// CHECK-DEBUG-NEXT:    store i32 1, ptr [[P_STRIDE]], align 4, !dbg [[DBG162]]
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM3:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB15:[0-9]+]]), !dbg [[DBG162]]
// CHECK-DEBUG-NEXT:    call void @__kmpc_for_static_init_4u(ptr @[[GLOB15]], i32 [[OMP_GLOBAL_THREAD_NUM3]], i32 34, ptr [[P_LASTITER]], ptr [[P_LOWERBOUND]], ptr [[P_UPPERBOUND]], ptr [[P_STRIDE]], i32 1, i32 0), !dbg [[DBG162]]
// CHECK-DEBUG-NEXT:    [[TMP6:%.*]] = load i32, ptr [[P_LOWERBOUND]], align 4, !dbg [[DBG162]]
// CHECK-DEBUG-NEXT:    [[TMP7:%.*]] = load i32, ptr [[P_UPPERBOUND]], align 4, !dbg [[DBG162]]
// CHECK-DEBUG-NEXT:    [[TMP8:%.*]] = sub i32 [[TMP7]], [[TMP6]], !dbg [[DBG162]]
// CHECK-DEBUG-NEXT:    [[TMP9:%.*]] = add i32 [[TMP8]], 1, !dbg [[DBG162]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_HEADER:%.*]], !dbg [[DBG162]]
// CHECK-DEBUG:       omp_loop.header:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_IV:%.*]] = phi i32 [ 0, [[OMP_LOOP_PREHEADER]] ], [ [[OMP_LOOP_NEXT:%.*]], [[OMP_LOOP_INC:%.*]] ], !dbg [[DBG162]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_COND:%.*]], !dbg [[DBG162]]
// CHECK-DEBUG:       omp_loop.cond:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_CMP:%.*]] = icmp ult i32 [[OMP_LOOP_IV]], [[TMP9]], !dbg [[DBG162]]
// CHECK-DEBUG-NEXT:    br i1 [[OMP_LOOP_CMP]], label [[OMP_LOOP_BODY:%.*]], label [[OMP_LOOP_EXIT:%.*]], !dbg [[DBG162]]
// CHECK-DEBUG:       omp_loop.exit:
// CHECK-DEBUG-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB15]], i32 [[OMP_GLOBAL_THREAD_NUM3]]), !dbg [[DBG162]]
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM4:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB15]]), !dbg [[DBG164:![0-9]+]]
// CHECK-DEBUG-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB16:[0-9]+]], i32 [[OMP_GLOBAL_THREAD_NUM4]]), !dbg [[DBG164]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_AFTER:%.*]], !dbg [[DBG162]]
// CHECK-DEBUG:       omp_loop.after:
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM5:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB18:[0-9]+]]), !dbg [[DBG165:![0-9]+]]
// CHECK-DEBUG-NEXT:    br label [[OMP_PARALLEL:%.*]]
// CHECK-DEBUG:       omp_parallel:
// CHECK-DEBUG-NEXT:    [[GEP_A_ADDR215:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG214]], i32 0, i32 0
// CHECK-DEBUG-NEXT:    store ptr [[LOADGEP_A_ADDR]], ptr [[GEP_A_ADDR215]], align 8
// CHECK-DEBUG-NEXT:    [[GEP_B_ADDR216:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG214]], i32 0, i32 1
// CHECK-DEBUG-NEXT:    store ptr [[LOADGEP_B_ADDR]], ptr [[GEP_B_ADDR216]], align 8
// CHECK-DEBUG-NEXT:    [[GEP_R_ADDR217:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG214]], i32 0, i32 2
// CHECK-DEBUG-NEXT:    store ptr [[LOADGEP_R_ADDR]], ptr [[GEP_R_ADDR217]], align 8
// CHECK-DEBUG-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB18]], i32 1, ptr @_Z14parallel_for_2Pfid..omp_par.22, ptr [[STRUCTARG214]]), !dbg [[DBG166:![0-9]+]]
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT159:%.*]]
// CHECK-DEBUG:       omp.par.outlined.exit159:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_EXIT11_SPLIT:%.*]]
// CHECK-DEBUG:       omp.par.exit11.split:
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[I160]], metadata [[META170:![0-9]+]], metadata !DIExpression()), !dbg [[DBG173:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 0, ptr [[I160]], align 4, !dbg [[DBG173]]
// CHECK-DEBUG-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [[STRUCT_ANON_15]], ptr [[AGG_CAPTURED161]], i32 0, i32 0, !dbg [[DBG174:![0-9]+]]
// CHECK-DEBUG-NEXT:    store ptr [[I160]], ptr [[TMP10]], align 8, !dbg [[DBG174]]
// CHECK-DEBUG-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_ANON_16]], ptr [[AGG_CAPTURED162]], i32 0, i32 0, !dbg [[DBG174]]
// CHECK-DEBUG-NEXT:    [[TMP12:%.*]] = load i32, ptr [[I160]], align 4, !dbg [[DBG175:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP12]], ptr [[TMP11]], align 4, !dbg [[DBG174]]
// CHECK-DEBUG-NEXT:    call void @__captured_stmt.17(ptr [[DOTCOUNT_ADDR163]], ptr [[AGG_CAPTURED161]]), !dbg [[DBG174]]
// CHECK-DEBUG-NEXT:    [[DOTCOUNT164:%.*]] = load i32, ptr [[DOTCOUNT_ADDR163]], align 4, !dbg [[DBG174]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_PREHEADER165:%.*]], !dbg [[DBG174]]
// CHECK-DEBUG:       omp_loop.preheader165:
// CHECK-DEBUG-NEXT:    store i32 0, ptr [[P_LOWERBOUND179]], align 4, !dbg [[DBG174]]
// CHECK-DEBUG-NEXT:    [[TMP13:%.*]] = sub i32 [[DOTCOUNT164]], 1, !dbg [[DBG174]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP13]], ptr [[P_UPPERBOUND180]], align 4, !dbg [[DBG174]]
// CHECK-DEBUG-NEXT:    store i32 1, ptr [[P_STRIDE181]], align 4, !dbg [[DBG174]]
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM182:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB39:[0-9]+]]), !dbg [[DBG174]]
// CHECK-DEBUG-NEXT:    call void @__kmpc_for_static_init_4u(ptr @[[GLOB39]], i32 [[OMP_GLOBAL_THREAD_NUM182]], i32 34, ptr [[P_LASTITER178]], ptr [[P_LOWERBOUND179]], ptr [[P_UPPERBOUND180]], ptr [[P_STRIDE181]], i32 1, i32 0), !dbg [[DBG174]]
// CHECK-DEBUG-NEXT:    [[TMP14:%.*]] = load i32, ptr [[P_LOWERBOUND179]], align 4, !dbg [[DBG174]]
// CHECK-DEBUG-NEXT:    [[TMP15:%.*]] = load i32, ptr [[P_UPPERBOUND180]], align 4, !dbg [[DBG174]]
// CHECK-DEBUG-NEXT:    [[TMP16:%.*]] = sub i32 [[TMP15]], [[TMP14]], !dbg [[DBG174]]
// CHECK-DEBUG-NEXT:    [[TMP17:%.*]] = add i32 [[TMP16]], 1, !dbg [[DBG174]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_HEADER166:%.*]], !dbg [[DBG174]]
// CHECK-DEBUG:       omp_loop.header166:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_IV172:%.*]] = phi i32 [ 0, [[OMP_LOOP_PREHEADER165]] ], [ [[OMP_LOOP_NEXT174:%.*]], [[OMP_LOOP_INC169:%.*]] ], !dbg [[DBG174]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_COND167:%.*]], !dbg [[DBG174]]
// CHECK-DEBUG:       omp_loop.cond167:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_CMP173:%.*]] = icmp ult i32 [[OMP_LOOP_IV172]], [[TMP17]], !dbg [[DBG174]]
// CHECK-DEBUG-NEXT:    br i1 [[OMP_LOOP_CMP173]], label [[OMP_LOOP_BODY168:%.*]], label [[OMP_LOOP_EXIT170:%.*]], !dbg [[DBG174]]
// CHECK-DEBUG:       omp_loop.exit170:
// CHECK-DEBUG-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB39]], i32 [[OMP_GLOBAL_THREAD_NUM182]]), !dbg [[DBG174]]
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM183:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB39]]), !dbg [[DBG176:![0-9]+]]
// CHECK-DEBUG-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB40:[0-9]+]], i32 [[OMP_GLOBAL_THREAD_NUM183]]), !dbg [[DBG176]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_AFTER171:%.*]], !dbg [[DBG174]]
// CHECK-DEBUG:       omp_loop.after171:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_REGION_PARALLEL_AFTER:%.*]], !dbg [[DBG177:![0-9]+]]
// CHECK-DEBUG:       omp.par.region.parallel.after:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_PRE_FINALIZE:%.*]]
// CHECK-DEBUG:       omp.par.pre_finalize:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT184_EXITSTUB:%.*]], !dbg [[DBG177]]
// CHECK-DEBUG:       omp_loop.body168:
// CHECK-DEBUG-NEXT:    [[TMP18:%.*]] = add i32 [[OMP_LOOP_IV172]], [[TMP14]], !dbg [[DBG176]]
// CHECK-DEBUG-NEXT:    call void @__captured_stmt.18(ptr [[I160]], i32 [[TMP18]], ptr [[AGG_CAPTURED162]]), !dbg [[DBG174]]
// CHECK-DEBUG-NEXT:    [[TMP19:%.*]] = load i32, ptr [[LOADGEP_A_ADDR]], align 4, !dbg [[DBG178:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[CONV175:%.*]] = sitofp i32 [[TMP19]] to double, !dbg [[DBG178]]
// CHECK-DEBUG-NEXT:    [[TMP20:%.*]] = load double, ptr [[LOADGEP_B_ADDR]], align 8, !dbg [[DBG176]]
// CHECK-DEBUG-NEXT:    [[ADD176:%.*]] = fadd double [[CONV175]], [[TMP20]], !dbg [[DBG179:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[CONV177:%.*]] = fptrunc double [[ADD176]] to float, !dbg [[DBG178]]
// CHECK-DEBUG-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[LOADGEP_R_ADDR]], align 8, !dbg [[DBG180:![0-9]+]]
// CHECK-DEBUG-NEXT:    store float [[CONV177]], ptr [[TMP21]], align 4, !dbg [[DBG181:![0-9]+]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_INC169]], !dbg [[DBG174]]
// CHECK-DEBUG:       omp_loop.inc169:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_NEXT174]] = add nuw i32 [[OMP_LOOP_IV172]], 1, !dbg [[DBG174]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_HEADER166]], !dbg [[DBG174]]
// CHECK-DEBUG:       omp_loop.body:
// CHECK-DEBUG-NEXT:    [[TMP22:%.*]] = add i32 [[OMP_LOOP_IV]], [[TMP6]], !dbg [[DBG164]]
// CHECK-DEBUG-NEXT:    call void @__captured_stmt.6(ptr [[I]], i32 [[TMP22]], ptr [[AGG_CAPTURED1]]), !dbg [[DBG162]]
// CHECK-DEBUG-NEXT:    [[TMP23:%.*]] = load i32, ptr [[LOADGEP_A_ADDR]], align 4, !dbg [[DBG182:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[CONV:%.*]] = sitofp i32 [[TMP23]] to double, !dbg [[DBG182]]
// CHECK-DEBUG-NEXT:    [[TMP24:%.*]] = load double, ptr [[LOADGEP_B_ADDR]], align 8, !dbg [[DBG164]]
// CHECK-DEBUG-NEXT:    [[ADD:%.*]] = fadd double [[CONV]], [[TMP24]], !dbg [[DBG183:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[CONV2:%.*]] = fptrunc double [[ADD]] to float, !dbg [[DBG182]]
// CHECK-DEBUG-NEXT:    [[TMP25:%.*]] = load ptr, ptr [[LOADGEP_R_ADDR]], align 8, !dbg [[DBG184:![0-9]+]]
// CHECK-DEBUG-NEXT:    store float [[CONV2]], ptr [[TMP25]], align 4, !dbg [[DBG185:![0-9]+]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_INC]], !dbg [[DBG162]]
// CHECK-DEBUG:       omp_loop.inc:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_NEXT]] = add nuw i32 [[OMP_LOOP_IV]], 1, !dbg [[DBG162]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_HEADER]], !dbg [[DBG162]]
// CHECK-DEBUG:       omp.par.outlined.exit184.exitStub:
// CHECK-DEBUG-NEXT:    ret void
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@_Z14parallel_for_2Pfid..omp_par.22
// CHECK-DEBUG-SAME: (ptr noalias [[TID_ADDR6:%.*]], ptr noalias [[ZERO_ADDR7:%.*]], ptr [[TMP0:%.*]]) #[[ATTR1]] !dbg [[DBG186:![0-9]+]] {
// CHECK-DEBUG-NEXT:  omp.par.entry8:
// CHECK-DEBUG-NEXT:    [[GEP_A_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 0
// CHECK-DEBUG-NEXT:    [[LOADGEP_A_ADDR:%.*]] = load ptr, ptr [[GEP_A_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[GEP_B_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 1
// CHECK-DEBUG-NEXT:    [[LOADGEP_B_ADDR:%.*]] = load ptr, ptr [[GEP_B_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[GEP_R_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 2
// CHECK-DEBUG-NEXT:    [[LOADGEP_R_ADDR:%.*]] = load ptr, ptr [[GEP_R_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[STRUCTARG209:%.*]] = alloca { ptr, ptr, ptr }, align 8
// CHECK-DEBUG-NEXT:    [[STRUCTARG:%.*]] = alloca { ptr, ptr, ptr }, align 8
// CHECK-DEBUG-NEXT:    [[TID_ADDR_LOCAL12:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TID_ADDR6]], align 4
// CHECK-DEBUG-NEXT:    store i32 [[TMP1]], ptr [[TID_ADDR_LOCAL12]], align 4
// CHECK-DEBUG-NEXT:    [[TID13:%.*]] = load i32, ptr [[TID_ADDR_LOCAL12]], align 4
// CHECK-DEBUG-NEXT:    [[I16:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[AGG_CAPTURED17:%.*]] = alloca [[STRUCT_ANON_5:%.*]], align 8
// CHECK-DEBUG-NEXT:    [[AGG_CAPTURED18:%.*]] = alloca [[STRUCT_ANON_6:%.*]], align 4
// CHECK-DEBUG-NEXT:    [[DOTCOUNT_ADDR19:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_LASTITER34:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_LOWERBOUND35:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_UPPERBOUND36:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_STRIDE37:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[I75:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[AGG_CAPTURED76:%.*]] = alloca [[STRUCT_ANON_9:%.*]], align 8
// CHECK-DEBUG-NEXT:    [[AGG_CAPTURED77:%.*]] = alloca [[STRUCT_ANON_10:%.*]], align 4
// CHECK-DEBUG-NEXT:    [[DOTCOUNT_ADDR78:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_LASTITER93:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_LOWERBOUND94:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_UPPERBOUND95:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_STRIDE96:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[I135:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[AGG_CAPTURED136:%.*]] = alloca [[STRUCT_ANON_13:%.*]], align 8
// CHECK-DEBUG-NEXT:    [[AGG_CAPTURED137:%.*]] = alloca [[STRUCT_ANON_14:%.*]], align 4
// CHECK-DEBUG-NEXT:    [[DOTCOUNT_ADDR138:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_LASTITER153:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_LOWERBOUND154:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_UPPERBOUND155:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_STRIDE156:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_REGION9:%.*]]
// CHECK-DEBUG:       omp.par.region9:
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[I16]], metadata [[META187:![0-9]+]], metadata !DIExpression()), !dbg [[DBG192:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 0, ptr [[I16]], align 4, !dbg [[DBG192]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_ANON_5]], ptr [[AGG_CAPTURED17]], i32 0, i32 0, !dbg [[DBG193:![0-9]+]]
// CHECK-DEBUG-NEXT:    store ptr [[I16]], ptr [[TMP2]], align 8, !dbg [[DBG193]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [[STRUCT_ANON_6]], ptr [[AGG_CAPTURED18]], i32 0, i32 0, !dbg [[DBG193]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load i32, ptr [[I16]], align 4, !dbg [[DBG194:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP4]], ptr [[TMP3]], align 4, !dbg [[DBG193]]
// CHECK-DEBUG-NEXT:    call void @__captured_stmt.7(ptr [[DOTCOUNT_ADDR19]], ptr [[AGG_CAPTURED17]]), !dbg [[DBG193]]
// CHECK-DEBUG-NEXT:    [[DOTCOUNT20:%.*]] = load i32, ptr [[DOTCOUNT_ADDR19]], align 4, !dbg [[DBG193]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_PREHEADER21:%.*]], !dbg [[DBG193]]
// CHECK-DEBUG:       omp_loop.preheader21:
// CHECK-DEBUG-NEXT:    store i32 0, ptr [[P_LOWERBOUND35]], align 4, !dbg [[DBG193]]
// CHECK-DEBUG-NEXT:    [[TMP5:%.*]] = sub i32 [[DOTCOUNT20]], 1, !dbg [[DBG193]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP5]], ptr [[P_UPPERBOUND36]], align 4, !dbg [[DBG193]]
// CHECK-DEBUG-NEXT:    store i32 1, ptr [[P_STRIDE37]], align 4, !dbg [[DBG193]]
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM38:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB20:[0-9]+]]), !dbg [[DBG193]]
// CHECK-DEBUG-NEXT:    call void @__kmpc_for_static_init_4u(ptr @[[GLOB20]], i32 [[OMP_GLOBAL_THREAD_NUM38]], i32 34, ptr [[P_LASTITER34]], ptr [[P_LOWERBOUND35]], ptr [[P_UPPERBOUND36]], ptr [[P_STRIDE37]], i32 1, i32 0), !dbg [[DBG193]]
// CHECK-DEBUG-NEXT:    [[TMP6:%.*]] = load i32, ptr [[P_LOWERBOUND35]], align 4, !dbg [[DBG193]]
// CHECK-DEBUG-NEXT:    [[TMP7:%.*]] = load i32, ptr [[P_UPPERBOUND36]], align 4, !dbg [[DBG193]]
// CHECK-DEBUG-NEXT:    [[TMP8:%.*]] = sub i32 [[TMP7]], [[TMP6]], !dbg [[DBG193]]
// CHECK-DEBUG-NEXT:    [[TMP9:%.*]] = add i32 [[TMP8]], 1, !dbg [[DBG193]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_HEADER22:%.*]], !dbg [[DBG193]]
// CHECK-DEBUG:       omp_loop.header22:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_IV28:%.*]] = phi i32 [ 0, [[OMP_LOOP_PREHEADER21]] ], [ [[OMP_LOOP_NEXT30:%.*]], [[OMP_LOOP_INC25:%.*]] ], !dbg [[DBG193]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_COND23:%.*]], !dbg [[DBG193]]
// CHECK-DEBUG:       omp_loop.cond23:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_CMP29:%.*]] = icmp ult i32 [[OMP_LOOP_IV28]], [[TMP9]], !dbg [[DBG193]]
// CHECK-DEBUG-NEXT:    br i1 [[OMP_LOOP_CMP29]], label [[OMP_LOOP_BODY24:%.*]], label [[OMP_LOOP_EXIT26:%.*]], !dbg [[DBG193]]
// CHECK-DEBUG:       omp_loop.exit26:
// CHECK-DEBUG-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB20]], i32 [[OMP_GLOBAL_THREAD_NUM38]]), !dbg [[DBG193]]
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM39:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB20]]), !dbg [[DBG195:![0-9]+]]
// CHECK-DEBUG-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB21:[0-9]+]], i32 [[OMP_GLOBAL_THREAD_NUM39]]), !dbg [[DBG195]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_AFTER27:%.*]], !dbg [[DBG193]]
// CHECK-DEBUG:       omp_loop.after27:
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM40:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB23:[0-9]+]]), !dbg [[DBG196:![0-9]+]]
// CHECK-DEBUG-NEXT:    br label [[OMP_PARALLEL:%.*]]
// CHECK-DEBUG:       omp_parallel:
// CHECK-DEBUG-NEXT:    [[GEP_A_ADDR1:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG]], i32 0, i32 0
// CHECK-DEBUG-NEXT:    store ptr [[LOADGEP_A_ADDR]], ptr [[GEP_A_ADDR1]], align 8
// CHECK-DEBUG-NEXT:    [[GEP_B_ADDR2:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG]], i32 0, i32 1
// CHECK-DEBUG-NEXT:    store ptr [[LOADGEP_B_ADDR]], ptr [[GEP_B_ADDR2]], align 8
// CHECK-DEBUG-NEXT:    [[GEP_R_ADDR3:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG]], i32 0, i32 2
// CHECK-DEBUG-NEXT:    store ptr [[LOADGEP_R_ADDR]], ptr [[GEP_R_ADDR3]], align 8
// CHECK-DEBUG-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB23]], i32 1, ptr @_Z14parallel_for_2Pfid..omp_par, ptr [[STRUCTARG]]), !dbg [[DBG197:![0-9]+]]
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT:%.*]]
// CHECK-DEBUG:       omp.par.outlined.exit:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_EXIT46_SPLIT:%.*]]
// CHECK-DEBUG:       omp.par.exit46.split:
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[I75]], metadata [[META201:![0-9]+]], metadata !DIExpression()), !dbg [[DBG204:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 0, ptr [[I75]], align 4, !dbg [[DBG204]]
// CHECK-DEBUG-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [[STRUCT_ANON_9]], ptr [[AGG_CAPTURED76]], i32 0, i32 0, !dbg [[DBG205:![0-9]+]]
// CHECK-DEBUG-NEXT:    store ptr [[I75]], ptr [[TMP10]], align 8, !dbg [[DBG205]]
// CHECK-DEBUG-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_ANON_10]], ptr [[AGG_CAPTURED77]], i32 0, i32 0, !dbg [[DBG205]]
// CHECK-DEBUG-NEXT:    [[TMP12:%.*]] = load i32, ptr [[I75]], align 4, !dbg [[DBG206:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP12]], ptr [[TMP11]], align 4, !dbg [[DBG205]]
// CHECK-DEBUG-NEXT:    call void @__captured_stmt.11(ptr [[DOTCOUNT_ADDR78]], ptr [[AGG_CAPTURED76]]), !dbg [[DBG205]]
// CHECK-DEBUG-NEXT:    [[DOTCOUNT79:%.*]] = load i32, ptr [[DOTCOUNT_ADDR78]], align 4, !dbg [[DBG205]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_PREHEADER80:%.*]], !dbg [[DBG205]]
// CHECK-DEBUG:       omp_loop.preheader80:
// CHECK-DEBUG-NEXT:    store i32 0, ptr [[P_LOWERBOUND94]], align 4, !dbg [[DBG205]]
// CHECK-DEBUG-NEXT:    [[TMP13:%.*]] = sub i32 [[DOTCOUNT79]], 1, !dbg [[DBG205]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP13]], ptr [[P_UPPERBOUND95]], align 4, !dbg [[DBG205]]
// CHECK-DEBUG-NEXT:    store i32 1, ptr [[P_STRIDE96]], align 4, !dbg [[DBG205]]
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM97:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB28:[0-9]+]]), !dbg [[DBG205]]
// CHECK-DEBUG-NEXT:    call void @__kmpc_for_static_init_4u(ptr @[[GLOB28]], i32 [[OMP_GLOBAL_THREAD_NUM97]], i32 34, ptr [[P_LASTITER93]], ptr [[P_LOWERBOUND94]], ptr [[P_UPPERBOUND95]], ptr [[P_STRIDE96]], i32 1, i32 0), !dbg [[DBG205]]
// CHECK-DEBUG-NEXT:    [[TMP14:%.*]] = load i32, ptr [[P_LOWERBOUND94]], align 4, !dbg [[DBG205]]
// CHECK-DEBUG-NEXT:    [[TMP15:%.*]] = load i32, ptr [[P_UPPERBOUND95]], align 4, !dbg [[DBG205]]
// CHECK-DEBUG-NEXT:    [[TMP16:%.*]] = sub i32 [[TMP15]], [[TMP14]], !dbg [[DBG205]]
// CHECK-DEBUG-NEXT:    [[TMP17:%.*]] = add i32 [[TMP16]], 1, !dbg [[DBG205]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_HEADER81:%.*]], !dbg [[DBG205]]
// CHECK-DEBUG:       omp_loop.header81:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_IV87:%.*]] = phi i32 [ 0, [[OMP_LOOP_PREHEADER80]] ], [ [[OMP_LOOP_NEXT89:%.*]], [[OMP_LOOP_INC84:%.*]] ], !dbg [[DBG205]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_COND82:%.*]], !dbg [[DBG205]]
// CHECK-DEBUG:       omp_loop.cond82:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_CMP88:%.*]] = icmp ult i32 [[OMP_LOOP_IV87]], [[TMP17]], !dbg [[DBG205]]
// CHECK-DEBUG-NEXT:    br i1 [[OMP_LOOP_CMP88]], label [[OMP_LOOP_BODY83:%.*]], label [[OMP_LOOP_EXIT85:%.*]], !dbg [[DBG205]]
// CHECK-DEBUG:       omp_loop.exit85:
// CHECK-DEBUG-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB28]], i32 [[OMP_GLOBAL_THREAD_NUM97]]), !dbg [[DBG205]]
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM98:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB28]]), !dbg [[DBG207:![0-9]+]]
// CHECK-DEBUG-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB29:[0-9]+]], i32 [[OMP_GLOBAL_THREAD_NUM98]]), !dbg [[DBG207]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_AFTER86:%.*]], !dbg [[DBG205]]
// CHECK-DEBUG:       omp_loop.after86:
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM99:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB31:[0-9]+]]), !dbg [[DBG208:![0-9]+]]
// CHECK-DEBUG-NEXT:    br label [[OMP_PARALLEL213:%.*]]
// CHECK-DEBUG:       omp_parallel213:
// CHECK-DEBUG-NEXT:    [[GEP_A_ADDR210:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG209]], i32 0, i32 0
// CHECK-DEBUG-NEXT:    store ptr [[LOADGEP_A_ADDR]], ptr [[GEP_A_ADDR210]], align 8
// CHECK-DEBUG-NEXT:    [[GEP_B_ADDR211:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG209]], i32 0, i32 1
// CHECK-DEBUG-NEXT:    store ptr [[LOADGEP_B_ADDR]], ptr [[GEP_B_ADDR211]], align 8
// CHECK-DEBUG-NEXT:    [[GEP_R_ADDR212:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[STRUCTARG209]], i32 0, i32 2
// CHECK-DEBUG-NEXT:    store ptr [[LOADGEP_R_ADDR]], ptr [[GEP_R_ADDR212]], align 8
// CHECK-DEBUG-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB31]], i32 1, ptr @_Z14parallel_for_2Pfid..omp_par.21, ptr [[STRUCTARG209]]), !dbg [[DBG209:![0-9]+]]
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT134:%.*]]
// CHECK-DEBUG:       omp.par.outlined.exit134:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_EXIT105_SPLIT:%.*]]
// CHECK-DEBUG:       omp.par.exit105.split:
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[I135]], metadata [[META213:![0-9]+]], metadata !DIExpression()), !dbg [[DBG216:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 0, ptr [[I135]], align 4, !dbg [[DBG216]]
// CHECK-DEBUG-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [[STRUCT_ANON_13]], ptr [[AGG_CAPTURED136]], i32 0, i32 0, !dbg [[DBG217:![0-9]+]]
// CHECK-DEBUG-NEXT:    store ptr [[I135]], ptr [[TMP18]], align 8, !dbg [[DBG217]]
// CHECK-DEBUG-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [[STRUCT_ANON_14]], ptr [[AGG_CAPTURED137]], i32 0, i32 0, !dbg [[DBG217]]
// CHECK-DEBUG-NEXT:    [[TMP20:%.*]] = load i32, ptr [[I135]], align 4, !dbg [[DBG218:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP20]], ptr [[TMP19]], align 4, !dbg [[DBG217]]
// CHECK-DEBUG-NEXT:    call void @__captured_stmt.15(ptr [[DOTCOUNT_ADDR138]], ptr [[AGG_CAPTURED136]]), !dbg [[DBG217]]
// CHECK-DEBUG-NEXT:    [[DOTCOUNT139:%.*]] = load i32, ptr [[DOTCOUNT_ADDR138]], align 4, !dbg [[DBG217]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_PREHEADER140:%.*]], !dbg [[DBG217]]
// CHECK-DEBUG:       omp_loop.preheader140:
// CHECK-DEBUG-NEXT:    store i32 0, ptr [[P_LOWERBOUND154]], align 4, !dbg [[DBG217]]
// CHECK-DEBUG-NEXT:    [[TMP21:%.*]] = sub i32 [[DOTCOUNT139]], 1, !dbg [[DBG217]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP21]], ptr [[P_UPPERBOUND155]], align 4, !dbg [[DBG217]]
// CHECK-DEBUG-NEXT:    store i32 1, ptr [[P_STRIDE156]], align 4, !dbg [[DBG217]]
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM157:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB36:[0-9]+]]), !dbg [[DBG217]]
// CHECK-DEBUG-NEXT:    call void @__kmpc_for_static_init_4u(ptr @[[GLOB36]], i32 [[OMP_GLOBAL_THREAD_NUM157]], i32 34, ptr [[P_LASTITER153]], ptr [[P_LOWERBOUND154]], ptr [[P_UPPERBOUND155]], ptr [[P_STRIDE156]], i32 1, i32 0), !dbg [[DBG217]]
// CHECK-DEBUG-NEXT:    [[TMP22:%.*]] = load i32, ptr [[P_LOWERBOUND154]], align 4, !dbg [[DBG217]]
// CHECK-DEBUG-NEXT:    [[TMP23:%.*]] = load i32, ptr [[P_UPPERBOUND155]], align 4, !dbg [[DBG217]]
// CHECK-DEBUG-NEXT:    [[TMP24:%.*]] = sub i32 [[TMP23]], [[TMP22]], !dbg [[DBG217]]
// CHECK-DEBUG-NEXT:    [[TMP25:%.*]] = add i32 [[TMP24]], 1, !dbg [[DBG217]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_HEADER141:%.*]], !dbg [[DBG217]]
// CHECK-DEBUG:       omp_loop.header141:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_IV147:%.*]] = phi i32 [ 0, [[OMP_LOOP_PREHEADER140]] ], [ [[OMP_LOOP_NEXT149:%.*]], [[OMP_LOOP_INC144:%.*]] ], !dbg [[DBG217]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_COND142:%.*]], !dbg [[DBG217]]
// CHECK-DEBUG:       omp_loop.cond142:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_CMP148:%.*]] = icmp ult i32 [[OMP_LOOP_IV147]], [[TMP25]], !dbg [[DBG217]]
// CHECK-DEBUG-NEXT:    br i1 [[OMP_LOOP_CMP148]], label [[OMP_LOOP_BODY143:%.*]], label [[OMP_LOOP_EXIT145:%.*]], !dbg [[DBG217]]
// CHECK-DEBUG:       omp_loop.exit145:
// CHECK-DEBUG-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB36]], i32 [[OMP_GLOBAL_THREAD_NUM157]]), !dbg [[DBG217]]
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM158:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB36]]), !dbg [[DBG219:![0-9]+]]
// CHECK-DEBUG-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB37:[0-9]+]], i32 [[OMP_GLOBAL_THREAD_NUM158]]), !dbg [[DBG219]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_AFTER146:%.*]], !dbg [[DBG217]]
// CHECK-DEBUG:       omp_loop.after146:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_REGION9_PARALLEL_AFTER:%.*]], !dbg [[DBG220:![0-9]+]]
// CHECK-DEBUG:       omp.par.region9.parallel.after:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_PRE_FINALIZE10:%.*]]
// CHECK-DEBUG:       omp.par.pre_finalize10:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT159_EXITSTUB:%.*]], !dbg [[DBG220]]
// CHECK-DEBUG:       omp_loop.body143:
// CHECK-DEBUG-NEXT:    [[TMP26:%.*]] = add i32 [[OMP_LOOP_IV147]], [[TMP22]], !dbg [[DBG219]]
// CHECK-DEBUG-NEXT:    call void @__captured_stmt.16(ptr [[I135]], i32 [[TMP26]], ptr [[AGG_CAPTURED137]]), !dbg [[DBG217]]
// CHECK-DEBUG-NEXT:    [[TMP27:%.*]] = load i32, ptr [[LOADGEP_A_ADDR]], align 4, !dbg [[DBG221:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[CONV150:%.*]] = sitofp i32 [[TMP27]] to double, !dbg [[DBG221]]
// CHECK-DEBUG-NEXT:    [[TMP28:%.*]] = load double, ptr [[LOADGEP_B_ADDR]], align 8, !dbg [[DBG219]]
// CHECK-DEBUG-NEXT:    [[ADD151:%.*]] = fadd double [[CONV150]], [[TMP28]], !dbg [[DBG222:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[CONV152:%.*]] = fptrunc double [[ADD151]] to float, !dbg [[DBG221]]
// CHECK-DEBUG-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[LOADGEP_R_ADDR]], align 8, !dbg [[DBG223:![0-9]+]]
// CHECK-DEBUG-NEXT:    store float [[CONV152]], ptr [[TMP29]], align 4, !dbg [[DBG224:![0-9]+]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_INC144]], !dbg [[DBG217]]
// CHECK-DEBUG:       omp_loop.inc144:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_NEXT149]] = add nuw i32 [[OMP_LOOP_IV147]], 1, !dbg [[DBG217]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_HEADER141]], !dbg [[DBG217]]
// CHECK-DEBUG:       omp_loop.body83:
// CHECK-DEBUG-NEXT:    [[TMP30:%.*]] = add i32 [[OMP_LOOP_IV87]], [[TMP14]], !dbg [[DBG207]]
// CHECK-DEBUG-NEXT:    call void @__captured_stmt.12(ptr [[I75]], i32 [[TMP30]], ptr [[AGG_CAPTURED77]]), !dbg [[DBG205]]
// CHECK-DEBUG-NEXT:    [[TMP31:%.*]] = load i32, ptr [[LOADGEP_A_ADDR]], align 4, !dbg [[DBG225:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[CONV90:%.*]] = sitofp i32 [[TMP31]] to double, !dbg [[DBG225]]
// CHECK-DEBUG-NEXT:    [[TMP32:%.*]] = load double, ptr [[LOADGEP_B_ADDR]], align 8, !dbg [[DBG207]]
// CHECK-DEBUG-NEXT:    [[ADD91:%.*]] = fadd double [[CONV90]], [[TMP32]], !dbg [[DBG226:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[CONV92:%.*]] = fptrunc double [[ADD91]] to float, !dbg [[DBG225]]
// CHECK-DEBUG-NEXT:    [[TMP33:%.*]] = load ptr, ptr [[LOADGEP_R_ADDR]], align 8, !dbg [[DBG227:![0-9]+]]
// CHECK-DEBUG-NEXT:    store float [[CONV92]], ptr [[TMP33]], align 4, !dbg [[DBG228:![0-9]+]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_INC84]], !dbg [[DBG205]]
// CHECK-DEBUG:       omp_loop.inc84:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_NEXT89]] = add nuw i32 [[OMP_LOOP_IV87]], 1, !dbg [[DBG205]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_HEADER81]], !dbg [[DBG205]]
// CHECK-DEBUG:       omp_loop.body24:
// CHECK-DEBUG-NEXT:    [[TMP34:%.*]] = add i32 [[OMP_LOOP_IV28]], [[TMP6]], !dbg [[DBG195]]
// CHECK-DEBUG-NEXT:    call void @__captured_stmt.8(ptr [[I16]], i32 [[TMP34]], ptr [[AGG_CAPTURED18]]), !dbg [[DBG193]]
// CHECK-DEBUG-NEXT:    [[TMP35:%.*]] = load i32, ptr [[LOADGEP_A_ADDR]], align 4, !dbg [[DBG229:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[CONV31:%.*]] = sitofp i32 [[TMP35]] to double, !dbg [[DBG229]]
// CHECK-DEBUG-NEXT:    [[TMP36:%.*]] = load double, ptr [[LOADGEP_B_ADDR]], align 8, !dbg [[DBG195]]
// CHECK-DEBUG-NEXT:    [[ADD32:%.*]] = fadd double [[CONV31]], [[TMP36]], !dbg [[DBG230:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[CONV33:%.*]] = fptrunc double [[ADD32]] to float, !dbg [[DBG229]]
// CHECK-DEBUG-NEXT:    [[TMP37:%.*]] = load ptr, ptr [[LOADGEP_R_ADDR]], align 8, !dbg [[DBG231:![0-9]+]]
// CHECK-DEBUG-NEXT:    store float [[CONV33]], ptr [[TMP37]], align 4, !dbg [[DBG232:![0-9]+]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_INC25]], !dbg [[DBG193]]
// CHECK-DEBUG:       omp_loop.inc25:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_NEXT30]] = add nuw i32 [[OMP_LOOP_IV28]], 1, !dbg [[DBG193]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_HEADER22]], !dbg [[DBG193]]
// CHECK-DEBUG:       omp.par.outlined.exit159.exitStub:
// CHECK-DEBUG-NEXT:    ret void
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@_Z14parallel_for_2Pfid..omp_par.21
// CHECK-DEBUG-SAME: (ptr noalias [[TID_ADDR100:%.*]], ptr noalias [[ZERO_ADDR101:%.*]], ptr [[TMP0:%.*]]) #[[ATTR1]] !dbg [[DBG233:![0-9]+]] {
// CHECK-DEBUG-NEXT:  omp.par.entry102:
// CHECK-DEBUG-NEXT:    [[GEP_A_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 0
// CHECK-DEBUG-NEXT:    [[LOADGEP_A_ADDR:%.*]] = load ptr, ptr [[GEP_A_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[GEP_B_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 1
// CHECK-DEBUG-NEXT:    [[LOADGEP_B_ADDR:%.*]] = load ptr, ptr [[GEP_B_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[GEP_R_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 2
// CHECK-DEBUG-NEXT:    [[LOADGEP_R_ADDR:%.*]] = load ptr, ptr [[GEP_R_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[TID_ADDR_LOCAL106:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TID_ADDR100]], align 4
// CHECK-DEBUG-NEXT:    store i32 [[TMP1]], ptr [[TID_ADDR_LOCAL106]], align 4
// CHECK-DEBUG-NEXT:    [[TID107:%.*]] = load i32, ptr [[TID_ADDR_LOCAL106]], align 4
// CHECK-DEBUG-NEXT:    [[I110:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[AGG_CAPTURED111:%.*]] = alloca [[STRUCT_ANON_11:%.*]], align 8
// CHECK-DEBUG-NEXT:    [[AGG_CAPTURED112:%.*]] = alloca [[STRUCT_ANON_12:%.*]], align 4
// CHECK-DEBUG-NEXT:    [[DOTCOUNT_ADDR113:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_LASTITER128:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_LOWERBOUND129:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_UPPERBOUND130:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_STRIDE131:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_REGION103:%.*]]
// CHECK-DEBUG:       omp.par.region103:
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[I110]], metadata [[META234:![0-9]+]], metadata !DIExpression()), !dbg [[DBG240:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 0, ptr [[I110]], align 4, !dbg [[DBG240]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_ANON_11]], ptr [[AGG_CAPTURED111]], i32 0, i32 0, !dbg [[DBG241:![0-9]+]]
// CHECK-DEBUG-NEXT:    store ptr [[I110]], ptr [[TMP2]], align 8, !dbg [[DBG241]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [[STRUCT_ANON_12]], ptr [[AGG_CAPTURED112]], i32 0, i32 0, !dbg [[DBG241]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load i32, ptr [[I110]], align 4, !dbg [[DBG242:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP4]], ptr [[TMP3]], align 4, !dbg [[DBG241]]
// CHECK-DEBUG-NEXT:    call void @__captured_stmt.13(ptr [[DOTCOUNT_ADDR113]], ptr [[AGG_CAPTURED111]]), !dbg [[DBG241]]
// CHECK-DEBUG-NEXT:    [[DOTCOUNT114:%.*]] = load i32, ptr [[DOTCOUNT_ADDR113]], align 4, !dbg [[DBG241]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_PREHEADER115:%.*]], !dbg [[DBG241]]
// CHECK-DEBUG:       omp_loop.preheader115:
// CHECK-DEBUG-NEXT:    store i32 0, ptr [[P_LOWERBOUND129]], align 4, !dbg [[DBG241]]
// CHECK-DEBUG-NEXT:    [[TMP5:%.*]] = sub i32 [[DOTCOUNT114]], 1, !dbg [[DBG241]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP5]], ptr [[P_UPPERBOUND130]], align 4, !dbg [[DBG241]]
// CHECK-DEBUG-NEXT:    store i32 1, ptr [[P_STRIDE131]], align 4, !dbg [[DBG241]]
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM132:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB33:[0-9]+]]), !dbg [[DBG241]]
// CHECK-DEBUG-NEXT:    call void @__kmpc_for_static_init_4u(ptr @[[GLOB33]], i32 [[OMP_GLOBAL_THREAD_NUM132]], i32 34, ptr [[P_LASTITER128]], ptr [[P_LOWERBOUND129]], ptr [[P_UPPERBOUND130]], ptr [[P_STRIDE131]], i32 1, i32 0), !dbg [[DBG241]]
// CHECK-DEBUG-NEXT:    [[TMP6:%.*]] = load i32, ptr [[P_LOWERBOUND129]], align 4, !dbg [[DBG241]]
// CHECK-DEBUG-NEXT:    [[TMP7:%.*]] = load i32, ptr [[P_UPPERBOUND130]], align 4, !dbg [[DBG241]]
// CHECK-DEBUG-NEXT:    [[TMP8:%.*]] = sub i32 [[TMP7]], [[TMP6]], !dbg [[DBG241]]
// CHECK-DEBUG-NEXT:    [[TMP9:%.*]] = add i32 [[TMP8]], 1, !dbg [[DBG241]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_HEADER116:%.*]], !dbg [[DBG241]]
// CHECK-DEBUG:       omp_loop.header116:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_IV122:%.*]] = phi i32 [ 0, [[OMP_LOOP_PREHEADER115]] ], [ [[OMP_LOOP_NEXT124:%.*]], [[OMP_LOOP_INC119:%.*]] ], !dbg [[DBG241]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_COND117:%.*]], !dbg [[DBG241]]
// CHECK-DEBUG:       omp_loop.cond117:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_CMP123:%.*]] = icmp ult i32 [[OMP_LOOP_IV122]], [[TMP9]], !dbg [[DBG241]]
// CHECK-DEBUG-NEXT:    br i1 [[OMP_LOOP_CMP123]], label [[OMP_LOOP_BODY118:%.*]], label [[OMP_LOOP_EXIT120:%.*]], !dbg [[DBG241]]
// CHECK-DEBUG:       omp_loop.exit120:
// CHECK-DEBUG-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB33]], i32 [[OMP_GLOBAL_THREAD_NUM132]]), !dbg [[DBG241]]
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM133:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB33]]), !dbg [[DBG243:![0-9]+]]
// CHECK-DEBUG-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB34:[0-9]+]], i32 [[OMP_GLOBAL_THREAD_NUM133]]), !dbg [[DBG243]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_AFTER121:%.*]], !dbg [[DBG241]]
// CHECK-DEBUG:       omp_loop.after121:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_REGION103_PARALLEL_AFTER:%.*]], !dbg [[DBG244:![0-9]+]]
// CHECK-DEBUG:       omp.par.region103.parallel.after:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_PRE_FINALIZE104:%.*]]
// CHECK-DEBUG:       omp.par.pre_finalize104:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT134_EXITSTUB:%.*]], !dbg [[DBG244]]
// CHECK-DEBUG:       omp_loop.body118:
// CHECK-DEBUG-NEXT:    [[TMP10:%.*]] = add i32 [[OMP_LOOP_IV122]], [[TMP6]], !dbg [[DBG243]]
// CHECK-DEBUG-NEXT:    call void @__captured_stmt.14(ptr [[I110]], i32 [[TMP10]], ptr [[AGG_CAPTURED112]]), !dbg [[DBG241]]
// CHECK-DEBUG-NEXT:    [[TMP11:%.*]] = load i32, ptr [[LOADGEP_A_ADDR]], align 4, !dbg [[DBG245:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[CONV125:%.*]] = sitofp i32 [[TMP11]] to double, !dbg [[DBG245]]
// CHECK-DEBUG-NEXT:    [[TMP12:%.*]] = load double, ptr [[LOADGEP_B_ADDR]], align 8, !dbg [[DBG243]]
// CHECK-DEBUG-NEXT:    [[ADD126:%.*]] = fadd double [[CONV125]], [[TMP12]], !dbg [[DBG246:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[CONV127:%.*]] = fptrunc double [[ADD126]] to float, !dbg [[DBG245]]
// CHECK-DEBUG-NEXT:    [[TMP13:%.*]] = load ptr, ptr [[LOADGEP_R_ADDR]], align 8, !dbg [[DBG247:![0-9]+]]
// CHECK-DEBUG-NEXT:    store float [[CONV127]], ptr [[TMP13]], align 4, !dbg [[DBG248:![0-9]+]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_INC119]], !dbg [[DBG241]]
// CHECK-DEBUG:       omp_loop.inc119:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_NEXT124]] = add nuw i32 [[OMP_LOOP_IV122]], 1, !dbg [[DBG241]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_HEADER116]], !dbg [[DBG241]]
// CHECK-DEBUG:       omp.par.outlined.exit134.exitStub:
// CHECK-DEBUG-NEXT:    ret void
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@_Z14parallel_for_2Pfid..omp_par
// CHECK-DEBUG-SAME: (ptr noalias [[TID_ADDR41:%.*]], ptr noalias [[ZERO_ADDR42:%.*]], ptr [[TMP0:%.*]]) #[[ATTR1]] !dbg [[DBG249:![0-9]+]] {
// CHECK-DEBUG-NEXT:  omp.par.entry43:
// CHECK-DEBUG-NEXT:    [[GEP_A_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 0
// CHECK-DEBUG-NEXT:    [[LOADGEP_A_ADDR:%.*]] = load ptr, ptr [[GEP_A_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[GEP_B_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 1
// CHECK-DEBUG-NEXT:    [[LOADGEP_B_ADDR:%.*]] = load ptr, ptr [[GEP_B_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[GEP_R_ADDR:%.*]] = getelementptr { ptr, ptr, ptr }, ptr [[TMP0]], i32 0, i32 2
// CHECK-DEBUG-NEXT:    [[LOADGEP_R_ADDR:%.*]] = load ptr, ptr [[GEP_R_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[TID_ADDR_LOCAL47:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TID_ADDR41]], align 4
// CHECK-DEBUG-NEXT:    store i32 [[TMP1]], ptr [[TID_ADDR_LOCAL47]], align 4
// CHECK-DEBUG-NEXT:    [[TID48:%.*]] = load i32, ptr [[TID_ADDR_LOCAL47]], align 4
// CHECK-DEBUG-NEXT:    [[I51:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[AGG_CAPTURED52:%.*]] = alloca [[STRUCT_ANON_7:%.*]], align 8
// CHECK-DEBUG-NEXT:    [[AGG_CAPTURED53:%.*]] = alloca [[STRUCT_ANON_8:%.*]], align 4
// CHECK-DEBUG-NEXT:    [[DOTCOUNT_ADDR54:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_LASTITER69:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_LOWERBOUND70:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_UPPERBOUND71:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[P_STRIDE72:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_REGION44:%.*]]
// CHECK-DEBUG:       omp.par.region44:
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[I51]], metadata [[META250:![0-9]+]], metadata !DIExpression()), !dbg [[DBG256:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 0, ptr [[I51]], align 4, !dbg [[DBG256]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_ANON_7]], ptr [[AGG_CAPTURED52]], i32 0, i32 0, !dbg [[DBG257:![0-9]+]]
// CHECK-DEBUG-NEXT:    store ptr [[I51]], ptr [[TMP2]], align 8, !dbg [[DBG257]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [[STRUCT_ANON_8]], ptr [[AGG_CAPTURED53]], i32 0, i32 0, !dbg [[DBG257]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load i32, ptr [[I51]], align 4, !dbg [[DBG258:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP4]], ptr [[TMP3]], align 4, !dbg [[DBG257]]
// CHECK-DEBUG-NEXT:    call void @__captured_stmt.9(ptr [[DOTCOUNT_ADDR54]], ptr [[AGG_CAPTURED52]]), !dbg [[DBG257]]
// CHECK-DEBUG-NEXT:    [[DOTCOUNT55:%.*]] = load i32, ptr [[DOTCOUNT_ADDR54]], align 4, !dbg [[DBG257]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_PREHEADER56:%.*]], !dbg [[DBG257]]
// CHECK-DEBUG:       omp_loop.preheader56:
// CHECK-DEBUG-NEXT:    store i32 0, ptr [[P_LOWERBOUND70]], align 4, !dbg [[DBG257]]
// CHECK-DEBUG-NEXT:    [[TMP5:%.*]] = sub i32 [[DOTCOUNT55]], 1, !dbg [[DBG257]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP5]], ptr [[P_UPPERBOUND71]], align 4, !dbg [[DBG257]]
// CHECK-DEBUG-NEXT:    store i32 1, ptr [[P_STRIDE72]], align 4, !dbg [[DBG257]]
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM73:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB25:[0-9]+]]), !dbg [[DBG257]]
// CHECK-DEBUG-NEXT:    call void @__kmpc_for_static_init_4u(ptr @[[GLOB25]], i32 [[OMP_GLOBAL_THREAD_NUM73]], i32 34, ptr [[P_LASTITER69]], ptr [[P_LOWERBOUND70]], ptr [[P_UPPERBOUND71]], ptr [[P_STRIDE72]], i32 1, i32 0), !dbg [[DBG257]]
// CHECK-DEBUG-NEXT:    [[TMP6:%.*]] = load i32, ptr [[P_LOWERBOUND70]], align 4, !dbg [[DBG257]]
// CHECK-DEBUG-NEXT:    [[TMP7:%.*]] = load i32, ptr [[P_UPPERBOUND71]], align 4, !dbg [[DBG257]]
// CHECK-DEBUG-NEXT:    [[TMP8:%.*]] = sub i32 [[TMP7]], [[TMP6]], !dbg [[DBG257]]
// CHECK-DEBUG-NEXT:    [[TMP9:%.*]] = add i32 [[TMP8]], 1, !dbg [[DBG257]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_HEADER57:%.*]], !dbg [[DBG257]]
// CHECK-DEBUG:       omp_loop.header57:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_IV63:%.*]] = phi i32 [ 0, [[OMP_LOOP_PREHEADER56]] ], [ [[OMP_LOOP_NEXT65:%.*]], [[OMP_LOOP_INC60:%.*]] ], !dbg [[DBG257]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_COND58:%.*]], !dbg [[DBG257]]
// CHECK-DEBUG:       omp_loop.cond58:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_CMP64:%.*]] = icmp ult i32 [[OMP_LOOP_IV63]], [[TMP9]], !dbg [[DBG257]]
// CHECK-DEBUG-NEXT:    br i1 [[OMP_LOOP_CMP64]], label [[OMP_LOOP_BODY59:%.*]], label [[OMP_LOOP_EXIT61:%.*]], !dbg [[DBG257]]
// CHECK-DEBUG:       omp_loop.exit61:
// CHECK-DEBUG-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB25]], i32 [[OMP_GLOBAL_THREAD_NUM73]]), !dbg [[DBG257]]
// CHECK-DEBUG-NEXT:    [[OMP_GLOBAL_THREAD_NUM74:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB25]]), !dbg [[DBG259:![0-9]+]]
// CHECK-DEBUG-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB26:[0-9]+]], i32 [[OMP_GLOBAL_THREAD_NUM74]]), !dbg [[DBG259]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_AFTER62:%.*]], !dbg [[DBG257]]
// CHECK-DEBUG:       omp_loop.after62:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_REGION44_PARALLEL_AFTER:%.*]], !dbg [[DBG260:![0-9]+]]
// CHECK-DEBUG:       omp.par.region44.parallel.after:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_PRE_FINALIZE45:%.*]]
// CHECK-DEBUG:       omp.par.pre_finalize45:
// CHECK-DEBUG-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT_EXITSTUB:%.*]], !dbg [[DBG260]]
// CHECK-DEBUG:       omp_loop.body59:
// CHECK-DEBUG-NEXT:    [[TMP10:%.*]] = add i32 [[OMP_LOOP_IV63]], [[TMP6]], !dbg [[DBG259]]
// CHECK-DEBUG-NEXT:    call void @__captured_stmt.10(ptr [[I51]], i32 [[TMP10]], ptr [[AGG_CAPTURED53]]), !dbg [[DBG257]]
// CHECK-DEBUG-NEXT:    [[TMP11:%.*]] = load i32, ptr [[LOADGEP_A_ADDR]], align 4, !dbg [[DBG261:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[CONV66:%.*]] = sitofp i32 [[TMP11]] to double, !dbg [[DBG261]]
// CHECK-DEBUG-NEXT:    [[TMP12:%.*]] = load double, ptr [[LOADGEP_B_ADDR]], align 8, !dbg [[DBG259]]
// CHECK-DEBUG-NEXT:    [[ADD67:%.*]] = fadd double [[CONV66]], [[TMP12]], !dbg [[DBG262:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[CONV68:%.*]] = fptrunc double [[ADD67]] to float, !dbg [[DBG261]]
// CHECK-DEBUG-NEXT:    [[TMP13:%.*]] = load ptr, ptr [[LOADGEP_R_ADDR]], align 8, !dbg [[DBG263:![0-9]+]]
// CHECK-DEBUG-NEXT:    store float [[CONV68]], ptr [[TMP13]], align 4, !dbg [[DBG264:![0-9]+]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_INC60]], !dbg [[DBG257]]
// CHECK-DEBUG:       omp_loop.inc60:
// CHECK-DEBUG-NEXT:    [[OMP_LOOP_NEXT65]] = add nuw i32 [[OMP_LOOP_IV63]], 1, !dbg [[DBG257]]
// CHECK-DEBUG-NEXT:    br label [[OMP_LOOP_HEADER57]], !dbg [[DBG257]]
// CHECK-DEBUG:       omp.par.outlined.exit.exitStub:
// CHECK-DEBUG-NEXT:    ret void
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@__captured_stmt.5
// CHECK-DEBUG-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[DISTANCE:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR4]] !dbg [[DBG265:![0-9]+]] {
// CHECK-DEBUG-NEXT:  entry:
// CHECK-DEBUG-NEXT:    [[DISTANCE_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[DOTSTART:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[DOTSTOP:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[DOTSTEP:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    store ptr [[DISTANCE]], ptr [[DISTANCE_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DISTANCE_ADDR]], metadata [[META266:![0-9]+]], metadata !DIExpression()), !dbg [[DBG267:![0-9]+]]
// CHECK-DEBUG-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[__CONTEXT_ADDR]], metadata [[META268:![0-9]+]], metadata !DIExpression()), !dbg [[DBG267]]
// CHECK-DEBUG-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTART]], metadata [[META269:![0-9]+]], metadata !DIExpression()), !dbg [[DBG271:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_3:%.*]], ptr [[TMP0]], i32 0, i32 0, !dbg [[DBG272:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP1]], align 8, !dbg [[DBG272]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = load i32, ptr [[TMP2]], align 4, !dbg [[DBG272]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP3]], ptr [[DOTSTART]], align 4, !dbg [[DBG271]]
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTOP]], metadata [[META274:![0-9]+]], metadata !DIExpression()), !dbg [[DBG275:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 100, ptr [[DOTSTOP]], align 4, !dbg [[DBG275]]
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTEP]], metadata [[META276:![0-9]+]], metadata !DIExpression()), !dbg [[DBG275]]
// CHECK-DEBUG-NEXT:    store i32 1, ptr [[DOTSTEP]], align 4, !dbg [[DBG275]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTSTART]], align 4, !dbg [[DBG275]]
// CHECK-DEBUG-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTSTOP]], align 4, !dbg [[DBG275]]
// CHECK-DEBUG-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP4]], [[TMP5]], !dbg [[DBG275]]
// CHECK-DEBUG-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]], !dbg [[DBG275]]
// CHECK-DEBUG:       cond.true:
// CHECK-DEBUG-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTSTOP]], align 4, !dbg [[DBG275]]
// CHECK-DEBUG-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTSTART]], align 4, !dbg [[DBG275]]
// CHECK-DEBUG-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], [[TMP7]], !dbg [[DBG275]]
// CHECK-DEBUG-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTSTEP]], align 4, !dbg [[DBG275]]
// CHECK-DEBUG-NEXT:    [[SUB1:%.*]] = sub i32 [[TMP8]], 1, !dbg [[DBG275]]
// CHECK-DEBUG-NEXT:    [[ADD:%.*]] = add i32 [[SUB]], [[SUB1]], !dbg [[DBG275]]
// CHECK-DEBUG-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTSTEP]], align 4, !dbg [[DBG275]]
// CHECK-DEBUG-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], [[TMP9]], !dbg [[DBG275]]
// CHECK-DEBUG-NEXT:    br label [[COND_END:%.*]], !dbg [[DBG275]]
// CHECK-DEBUG:       cond.false:
// CHECK-DEBUG-NEXT:    br label [[COND_END]], !dbg [[DBG275]]
// CHECK-DEBUG:       cond.end:
// CHECK-DEBUG-NEXT:    [[COND:%.*]] = phi i32 [ [[DIV]], [[COND_TRUE]] ], [ 0, [[COND_FALSE]] ], !dbg [[DBG275]]
// CHECK-DEBUG-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DISTANCE_ADDR]], align 8, !dbg [[DBG275]]
// CHECK-DEBUG-NEXT:    store i32 [[COND]], ptr [[TMP10]], align 4, !dbg [[DBG275]]
// CHECK-DEBUG-NEXT:    ret void, !dbg [[DBG277:![0-9]+]]
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@__captured_stmt.6
// CHECK-DEBUG-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[LOOPVAR:%.*]], i32 noundef [[LOGICAL:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR4]] !dbg [[DBG279:![0-9]+]] {
// CHECK-DEBUG-NEXT:  entry:
// CHECK-DEBUG-NEXT:    [[LOOPVAR_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[LOGICAL_ADDR:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    store ptr [[LOOPVAR]], ptr [[LOOPVAR_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[LOOPVAR_ADDR]], metadata [[META280:![0-9]+]], metadata !DIExpression()), !dbg [[DBG281:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 [[LOGICAL]], ptr [[LOGICAL_ADDR]], align 4
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[LOGICAL_ADDR]], metadata [[META282:![0-9]+]], metadata !DIExpression()), !dbg [[DBG281]]
// CHECK-DEBUG-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[__CONTEXT_ADDR]], metadata [[META283:![0-9]+]], metadata !DIExpression()), !dbg [[DBG281]]
// CHECK-DEBUG-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_4:%.*]], ptr [[TMP0]], i32 0, i32 0, !dbg [[DBG284:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4, !dbg [[DBG284]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = load i32, ptr [[LOGICAL_ADDR]], align 4, !dbg [[DBG286:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[MUL:%.*]] = mul i32 1, [[TMP3]], !dbg [[DBG286]]
// CHECK-DEBUG-NEXT:    [[ADD:%.*]] = add i32 [[TMP2]], [[MUL]], !dbg [[DBG286]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[LOOPVAR_ADDR]], align 8, !dbg [[DBG286]]
// CHECK-DEBUG-NEXT:    store i32 [[ADD]], ptr [[TMP4]], align 4, !dbg [[DBG281]]
// CHECK-DEBUG-NEXT:    ret void, !dbg [[DBG284]]
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@__captured_stmt.7
// CHECK-DEBUG-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[DISTANCE:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR4]] !dbg [[DBG287:![0-9]+]] {
// CHECK-DEBUG-NEXT:  entry:
// CHECK-DEBUG-NEXT:    [[DISTANCE_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[DOTSTART:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[DOTSTOP:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[DOTSTEP:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    store ptr [[DISTANCE]], ptr [[DISTANCE_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DISTANCE_ADDR]], metadata [[META288:![0-9]+]], metadata !DIExpression()), !dbg [[DBG289:![0-9]+]]
// CHECK-DEBUG-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[__CONTEXT_ADDR]], metadata [[META290:![0-9]+]], metadata !DIExpression()), !dbg [[DBG289]]
// CHECK-DEBUG-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTART]], metadata [[META291:![0-9]+]], metadata !DIExpression()), !dbg [[DBG293:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_5:%.*]], ptr [[TMP0]], i32 0, i32 0, !dbg [[DBG294:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP1]], align 8, !dbg [[DBG294]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = load i32, ptr [[TMP2]], align 4, !dbg [[DBG294]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP3]], ptr [[DOTSTART]], align 4, !dbg [[DBG293]]
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTOP]], metadata [[META296:![0-9]+]], metadata !DIExpression()), !dbg [[DBG297:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 100, ptr [[DOTSTOP]], align 4, !dbg [[DBG297]]
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTEP]], metadata [[META298:![0-9]+]], metadata !DIExpression()), !dbg [[DBG297]]
// CHECK-DEBUG-NEXT:    store i32 1, ptr [[DOTSTEP]], align 4, !dbg [[DBG297]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTSTART]], align 4, !dbg [[DBG297]]
// CHECK-DEBUG-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTSTOP]], align 4, !dbg [[DBG297]]
// CHECK-DEBUG-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP4]], [[TMP5]], !dbg [[DBG297]]
// CHECK-DEBUG-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]], !dbg [[DBG297]]
// CHECK-DEBUG:       cond.true:
// CHECK-DEBUG-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTSTOP]], align 4, !dbg [[DBG297]]
// CHECK-DEBUG-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTSTART]], align 4, !dbg [[DBG297]]
// CHECK-DEBUG-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], [[TMP7]], !dbg [[DBG297]]
// CHECK-DEBUG-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTSTEP]], align 4, !dbg [[DBG297]]
// CHECK-DEBUG-NEXT:    [[SUB1:%.*]] = sub i32 [[TMP8]], 1, !dbg [[DBG297]]
// CHECK-DEBUG-NEXT:    [[ADD:%.*]] = add i32 [[SUB]], [[SUB1]], !dbg [[DBG297]]
// CHECK-DEBUG-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTSTEP]], align 4, !dbg [[DBG297]]
// CHECK-DEBUG-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], [[TMP9]], !dbg [[DBG297]]
// CHECK-DEBUG-NEXT:    br label [[COND_END:%.*]], !dbg [[DBG297]]
// CHECK-DEBUG:       cond.false:
// CHECK-DEBUG-NEXT:    br label [[COND_END]], !dbg [[DBG297]]
// CHECK-DEBUG:       cond.end:
// CHECK-DEBUG-NEXT:    [[COND:%.*]] = phi i32 [ [[DIV]], [[COND_TRUE]] ], [ 0, [[COND_FALSE]] ], !dbg [[DBG297]]
// CHECK-DEBUG-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DISTANCE_ADDR]], align 8, !dbg [[DBG297]]
// CHECK-DEBUG-NEXT:    store i32 [[COND]], ptr [[TMP10]], align 4, !dbg [[DBG297]]
// CHECK-DEBUG-NEXT:    ret void, !dbg [[DBG299:![0-9]+]]
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@__captured_stmt.8
// CHECK-DEBUG-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[LOOPVAR:%.*]], i32 noundef [[LOGICAL:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR4]] !dbg [[DBG301:![0-9]+]] {
// CHECK-DEBUG-NEXT:  entry:
// CHECK-DEBUG-NEXT:    [[LOOPVAR_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[LOGICAL_ADDR:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    store ptr [[LOOPVAR]], ptr [[LOOPVAR_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[LOOPVAR_ADDR]], metadata [[META302:![0-9]+]], metadata !DIExpression()), !dbg [[DBG303:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 [[LOGICAL]], ptr [[LOGICAL_ADDR]], align 4
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[LOGICAL_ADDR]], metadata [[META304:![0-9]+]], metadata !DIExpression()), !dbg [[DBG303]]
// CHECK-DEBUG-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[__CONTEXT_ADDR]], metadata [[META305:![0-9]+]], metadata !DIExpression()), !dbg [[DBG303]]
// CHECK-DEBUG-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_6:%.*]], ptr [[TMP0]], i32 0, i32 0, !dbg [[DBG306:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4, !dbg [[DBG306]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = load i32, ptr [[LOGICAL_ADDR]], align 4, !dbg [[DBG308:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[MUL:%.*]] = mul i32 1, [[TMP3]], !dbg [[DBG308]]
// CHECK-DEBUG-NEXT:    [[ADD:%.*]] = add i32 [[TMP2]], [[MUL]], !dbg [[DBG308]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[LOOPVAR_ADDR]], align 8, !dbg [[DBG308]]
// CHECK-DEBUG-NEXT:    store i32 [[ADD]], ptr [[TMP4]], align 4, !dbg [[DBG303]]
// CHECK-DEBUG-NEXT:    ret void, !dbg [[DBG306]]
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@__captured_stmt.9
// CHECK-DEBUG-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[DISTANCE:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR4]] !dbg [[DBG309:![0-9]+]] {
// CHECK-DEBUG-NEXT:  entry:
// CHECK-DEBUG-NEXT:    [[DISTANCE_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[DOTSTART:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[DOTSTOP:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[DOTSTEP:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    store ptr [[DISTANCE]], ptr [[DISTANCE_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DISTANCE_ADDR]], metadata [[META310:![0-9]+]], metadata !DIExpression()), !dbg [[DBG311:![0-9]+]]
// CHECK-DEBUG-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[__CONTEXT_ADDR]], metadata [[META312:![0-9]+]], metadata !DIExpression()), !dbg [[DBG311]]
// CHECK-DEBUG-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTART]], metadata [[META313:![0-9]+]], metadata !DIExpression()), !dbg [[DBG315:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_7:%.*]], ptr [[TMP0]], i32 0, i32 0, !dbg [[DBG316:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP1]], align 8, !dbg [[DBG316]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = load i32, ptr [[TMP2]], align 4, !dbg [[DBG316]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP3]], ptr [[DOTSTART]], align 4, !dbg [[DBG315]]
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTOP]], metadata [[META318:![0-9]+]], metadata !DIExpression()), !dbg [[DBG319:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 100, ptr [[DOTSTOP]], align 4, !dbg [[DBG319]]
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTEP]], metadata [[META320:![0-9]+]], metadata !DIExpression()), !dbg [[DBG319]]
// CHECK-DEBUG-NEXT:    store i32 1, ptr [[DOTSTEP]], align 4, !dbg [[DBG319]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTSTART]], align 4, !dbg [[DBG319]]
// CHECK-DEBUG-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTSTOP]], align 4, !dbg [[DBG319]]
// CHECK-DEBUG-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP4]], [[TMP5]], !dbg [[DBG319]]
// CHECK-DEBUG-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]], !dbg [[DBG319]]
// CHECK-DEBUG:       cond.true:
// CHECK-DEBUG-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTSTOP]], align 4, !dbg [[DBG319]]
// CHECK-DEBUG-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTSTART]], align 4, !dbg [[DBG319]]
// CHECK-DEBUG-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], [[TMP7]], !dbg [[DBG319]]
// CHECK-DEBUG-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTSTEP]], align 4, !dbg [[DBG319]]
// CHECK-DEBUG-NEXT:    [[SUB1:%.*]] = sub i32 [[TMP8]], 1, !dbg [[DBG319]]
// CHECK-DEBUG-NEXT:    [[ADD:%.*]] = add i32 [[SUB]], [[SUB1]], !dbg [[DBG319]]
// CHECK-DEBUG-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTSTEP]], align 4, !dbg [[DBG319]]
// CHECK-DEBUG-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], [[TMP9]], !dbg [[DBG319]]
// CHECK-DEBUG-NEXT:    br label [[COND_END:%.*]], !dbg [[DBG319]]
// CHECK-DEBUG:       cond.false:
// CHECK-DEBUG-NEXT:    br label [[COND_END]], !dbg [[DBG319]]
// CHECK-DEBUG:       cond.end:
// CHECK-DEBUG-NEXT:    [[COND:%.*]] = phi i32 [ [[DIV]], [[COND_TRUE]] ], [ 0, [[COND_FALSE]] ], !dbg [[DBG319]]
// CHECK-DEBUG-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DISTANCE_ADDR]], align 8, !dbg [[DBG319]]
// CHECK-DEBUG-NEXT:    store i32 [[COND]], ptr [[TMP10]], align 4, !dbg [[DBG319]]
// CHECK-DEBUG-NEXT:    ret void, !dbg [[DBG321:![0-9]+]]
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@__captured_stmt.10
// CHECK-DEBUG-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[LOOPVAR:%.*]], i32 noundef [[LOGICAL:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR4]] !dbg [[DBG323:![0-9]+]] {
// CHECK-DEBUG-NEXT:  entry:
// CHECK-DEBUG-NEXT:    [[LOOPVAR_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[LOGICAL_ADDR:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    store ptr [[LOOPVAR]], ptr [[LOOPVAR_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[LOOPVAR_ADDR]], metadata [[META324:![0-9]+]], metadata !DIExpression()), !dbg [[DBG325:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 [[LOGICAL]], ptr [[LOGICAL_ADDR]], align 4
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[LOGICAL_ADDR]], metadata [[META326:![0-9]+]], metadata !DIExpression()), !dbg [[DBG325]]
// CHECK-DEBUG-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[__CONTEXT_ADDR]], metadata [[META327:![0-9]+]], metadata !DIExpression()), !dbg [[DBG325]]
// CHECK-DEBUG-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_8:%.*]], ptr [[TMP0]], i32 0, i32 0, !dbg [[DBG328:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4, !dbg [[DBG328]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = load i32, ptr [[LOGICAL_ADDR]], align 4, !dbg [[DBG330:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[MUL:%.*]] = mul i32 1, [[TMP3]], !dbg [[DBG330]]
// CHECK-DEBUG-NEXT:    [[ADD:%.*]] = add i32 [[TMP2]], [[MUL]], !dbg [[DBG330]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[LOOPVAR_ADDR]], align 8, !dbg [[DBG330]]
// CHECK-DEBUG-NEXT:    store i32 [[ADD]], ptr [[TMP4]], align 4, !dbg [[DBG325]]
// CHECK-DEBUG-NEXT:    ret void, !dbg [[DBG328]]
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@__captured_stmt.11
// CHECK-DEBUG-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[DISTANCE:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR4]] !dbg [[DBG331:![0-9]+]] {
// CHECK-DEBUG-NEXT:  entry:
// CHECK-DEBUG-NEXT:    [[DISTANCE_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[DOTSTART:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[DOTSTOP:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[DOTSTEP:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    store ptr [[DISTANCE]], ptr [[DISTANCE_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DISTANCE_ADDR]], metadata [[META332:![0-9]+]], metadata !DIExpression()), !dbg [[DBG333:![0-9]+]]
// CHECK-DEBUG-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[__CONTEXT_ADDR]], metadata [[META334:![0-9]+]], metadata !DIExpression()), !dbg [[DBG333]]
// CHECK-DEBUG-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTART]], metadata [[META335:![0-9]+]], metadata !DIExpression()), !dbg [[DBG337:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_9:%.*]], ptr [[TMP0]], i32 0, i32 0, !dbg [[DBG338:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP1]], align 8, !dbg [[DBG338]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = load i32, ptr [[TMP2]], align 4, !dbg [[DBG338]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP3]], ptr [[DOTSTART]], align 4, !dbg [[DBG337]]
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTOP]], metadata [[META340:![0-9]+]], metadata !DIExpression()), !dbg [[DBG341:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 100, ptr [[DOTSTOP]], align 4, !dbg [[DBG341]]
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTEP]], metadata [[META342:![0-9]+]], metadata !DIExpression()), !dbg [[DBG341]]
// CHECK-DEBUG-NEXT:    store i32 1, ptr [[DOTSTEP]], align 4, !dbg [[DBG341]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTSTART]], align 4, !dbg [[DBG341]]
// CHECK-DEBUG-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTSTOP]], align 4, !dbg [[DBG341]]
// CHECK-DEBUG-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP4]], [[TMP5]], !dbg [[DBG341]]
// CHECK-DEBUG-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]], !dbg [[DBG341]]
// CHECK-DEBUG:       cond.true:
// CHECK-DEBUG-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTSTOP]], align 4, !dbg [[DBG341]]
// CHECK-DEBUG-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTSTART]], align 4, !dbg [[DBG341]]
// CHECK-DEBUG-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], [[TMP7]], !dbg [[DBG341]]
// CHECK-DEBUG-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTSTEP]], align 4, !dbg [[DBG341]]
// CHECK-DEBUG-NEXT:    [[SUB1:%.*]] = sub i32 [[TMP8]], 1, !dbg [[DBG341]]
// CHECK-DEBUG-NEXT:    [[ADD:%.*]] = add i32 [[SUB]], [[SUB1]], !dbg [[DBG341]]
// CHECK-DEBUG-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTSTEP]], align 4, !dbg [[DBG341]]
// CHECK-DEBUG-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], [[TMP9]], !dbg [[DBG341]]
// CHECK-DEBUG-NEXT:    br label [[COND_END:%.*]], !dbg [[DBG341]]
// CHECK-DEBUG:       cond.false:
// CHECK-DEBUG-NEXT:    br label [[COND_END]], !dbg [[DBG341]]
// CHECK-DEBUG:       cond.end:
// CHECK-DEBUG-NEXT:    [[COND:%.*]] = phi i32 [ [[DIV]], [[COND_TRUE]] ], [ 0, [[COND_FALSE]] ], !dbg [[DBG341]]
// CHECK-DEBUG-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DISTANCE_ADDR]], align 8, !dbg [[DBG341]]
// CHECK-DEBUG-NEXT:    store i32 [[COND]], ptr [[TMP10]], align 4, !dbg [[DBG341]]
// CHECK-DEBUG-NEXT:    ret void, !dbg [[DBG343:![0-9]+]]
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@__captured_stmt.12
// CHECK-DEBUG-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[LOOPVAR:%.*]], i32 noundef [[LOGICAL:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR4]] !dbg [[DBG345:![0-9]+]] {
// CHECK-DEBUG-NEXT:  entry:
// CHECK-DEBUG-NEXT:    [[LOOPVAR_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[LOGICAL_ADDR:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    store ptr [[LOOPVAR]], ptr [[LOOPVAR_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[LOOPVAR_ADDR]], metadata [[META346:![0-9]+]], metadata !DIExpression()), !dbg [[DBG347:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 [[LOGICAL]], ptr [[LOGICAL_ADDR]], align 4
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[LOGICAL_ADDR]], metadata [[META348:![0-9]+]], metadata !DIExpression()), !dbg [[DBG347]]
// CHECK-DEBUG-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[__CONTEXT_ADDR]], metadata [[META349:![0-9]+]], metadata !DIExpression()), !dbg [[DBG347]]
// CHECK-DEBUG-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_10:%.*]], ptr [[TMP0]], i32 0, i32 0, !dbg [[DBG350:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4, !dbg [[DBG350]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = load i32, ptr [[LOGICAL_ADDR]], align 4, !dbg [[DBG352:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[MUL:%.*]] = mul i32 1, [[TMP3]], !dbg [[DBG352]]
// CHECK-DEBUG-NEXT:    [[ADD:%.*]] = add i32 [[TMP2]], [[MUL]], !dbg [[DBG352]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[LOOPVAR_ADDR]], align 8, !dbg [[DBG352]]
// CHECK-DEBUG-NEXT:    store i32 [[ADD]], ptr [[TMP4]], align 4, !dbg [[DBG347]]
// CHECK-DEBUG-NEXT:    ret void, !dbg [[DBG350]]
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@__captured_stmt.13
// CHECK-DEBUG-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[DISTANCE:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR4]] !dbg [[DBG353:![0-9]+]] {
// CHECK-DEBUG-NEXT:  entry:
// CHECK-DEBUG-NEXT:    [[DISTANCE_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[DOTSTART:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[DOTSTOP:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[DOTSTEP:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    store ptr [[DISTANCE]], ptr [[DISTANCE_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DISTANCE_ADDR]], metadata [[META354:![0-9]+]], metadata !DIExpression()), !dbg [[DBG355:![0-9]+]]
// CHECK-DEBUG-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[__CONTEXT_ADDR]], metadata [[META356:![0-9]+]], metadata !DIExpression()), !dbg [[DBG355]]
// CHECK-DEBUG-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTART]], metadata [[META357:![0-9]+]], metadata !DIExpression()), !dbg [[DBG359:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_11:%.*]], ptr [[TMP0]], i32 0, i32 0, !dbg [[DBG360:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP1]], align 8, !dbg [[DBG360]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = load i32, ptr [[TMP2]], align 4, !dbg [[DBG360]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP3]], ptr [[DOTSTART]], align 4, !dbg [[DBG359]]
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTOP]], metadata [[META362:![0-9]+]], metadata !DIExpression()), !dbg [[DBG363:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 100, ptr [[DOTSTOP]], align 4, !dbg [[DBG363]]
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTEP]], metadata [[META364:![0-9]+]], metadata !DIExpression()), !dbg [[DBG363]]
// CHECK-DEBUG-NEXT:    store i32 1, ptr [[DOTSTEP]], align 4, !dbg [[DBG363]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTSTART]], align 4, !dbg [[DBG363]]
// CHECK-DEBUG-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTSTOP]], align 4, !dbg [[DBG363]]
// CHECK-DEBUG-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP4]], [[TMP5]], !dbg [[DBG363]]
// CHECK-DEBUG-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]], !dbg [[DBG363]]
// CHECK-DEBUG:       cond.true:
// CHECK-DEBUG-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTSTOP]], align 4, !dbg [[DBG363]]
// CHECK-DEBUG-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTSTART]], align 4, !dbg [[DBG363]]
// CHECK-DEBUG-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], [[TMP7]], !dbg [[DBG363]]
// CHECK-DEBUG-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTSTEP]], align 4, !dbg [[DBG363]]
// CHECK-DEBUG-NEXT:    [[SUB1:%.*]] = sub i32 [[TMP8]], 1, !dbg [[DBG363]]
// CHECK-DEBUG-NEXT:    [[ADD:%.*]] = add i32 [[SUB]], [[SUB1]], !dbg [[DBG363]]
// CHECK-DEBUG-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTSTEP]], align 4, !dbg [[DBG363]]
// CHECK-DEBUG-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], [[TMP9]], !dbg [[DBG363]]
// CHECK-DEBUG-NEXT:    br label [[COND_END:%.*]], !dbg [[DBG363]]
// CHECK-DEBUG:       cond.false:
// CHECK-DEBUG-NEXT:    br label [[COND_END]], !dbg [[DBG363]]
// CHECK-DEBUG:       cond.end:
// CHECK-DEBUG-NEXT:    [[COND:%.*]] = phi i32 [ [[DIV]], [[COND_TRUE]] ], [ 0, [[COND_FALSE]] ], !dbg [[DBG363]]
// CHECK-DEBUG-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DISTANCE_ADDR]], align 8, !dbg [[DBG363]]
// CHECK-DEBUG-NEXT:    store i32 [[COND]], ptr [[TMP10]], align 4, !dbg [[DBG363]]
// CHECK-DEBUG-NEXT:    ret void, !dbg [[DBG365:![0-9]+]]
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@__captured_stmt.14
// CHECK-DEBUG-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[LOOPVAR:%.*]], i32 noundef [[LOGICAL:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR4]] !dbg [[DBG367:![0-9]+]] {
// CHECK-DEBUG-NEXT:  entry:
// CHECK-DEBUG-NEXT:    [[LOOPVAR_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[LOGICAL_ADDR:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    store ptr [[LOOPVAR]], ptr [[LOOPVAR_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[LOOPVAR_ADDR]], metadata [[META368:![0-9]+]], metadata !DIExpression()), !dbg [[DBG369:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 [[LOGICAL]], ptr [[LOGICAL_ADDR]], align 4
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[LOGICAL_ADDR]], metadata [[META370:![0-9]+]], metadata !DIExpression()), !dbg [[DBG369]]
// CHECK-DEBUG-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[__CONTEXT_ADDR]], metadata [[META371:![0-9]+]], metadata !DIExpression()), !dbg [[DBG369]]
// CHECK-DEBUG-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_12:%.*]], ptr [[TMP0]], i32 0, i32 0, !dbg [[DBG372:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4, !dbg [[DBG372]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = load i32, ptr [[LOGICAL_ADDR]], align 4, !dbg [[DBG374:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[MUL:%.*]] = mul i32 1, [[TMP3]], !dbg [[DBG374]]
// CHECK-DEBUG-NEXT:    [[ADD:%.*]] = add i32 [[TMP2]], [[MUL]], !dbg [[DBG374]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[LOOPVAR_ADDR]], align 8, !dbg [[DBG374]]
// CHECK-DEBUG-NEXT:    store i32 [[ADD]], ptr [[TMP4]], align 4, !dbg [[DBG369]]
// CHECK-DEBUG-NEXT:    ret void, !dbg [[DBG372]]
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@__captured_stmt.15
// CHECK-DEBUG-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[DISTANCE:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR4]] !dbg [[DBG375:![0-9]+]] {
// CHECK-DEBUG-NEXT:  entry:
// CHECK-DEBUG-NEXT:    [[DISTANCE_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[DOTSTART:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[DOTSTOP:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[DOTSTEP:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    store ptr [[DISTANCE]], ptr [[DISTANCE_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DISTANCE_ADDR]], metadata [[META376:![0-9]+]], metadata !DIExpression()), !dbg [[DBG377:![0-9]+]]
// CHECK-DEBUG-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[__CONTEXT_ADDR]], metadata [[META378:![0-9]+]], metadata !DIExpression()), !dbg [[DBG377]]
// CHECK-DEBUG-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTART]], metadata [[META379:![0-9]+]], metadata !DIExpression()), !dbg [[DBG381:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_13:%.*]], ptr [[TMP0]], i32 0, i32 0, !dbg [[DBG382:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP1]], align 8, !dbg [[DBG382]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = load i32, ptr [[TMP2]], align 4, !dbg [[DBG382]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP3]], ptr [[DOTSTART]], align 4, !dbg [[DBG381]]
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTOP]], metadata [[META384:![0-9]+]], metadata !DIExpression()), !dbg [[DBG385:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 100, ptr [[DOTSTOP]], align 4, !dbg [[DBG385]]
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTEP]], metadata [[META386:![0-9]+]], metadata !DIExpression()), !dbg [[DBG385]]
// CHECK-DEBUG-NEXT:    store i32 1, ptr [[DOTSTEP]], align 4, !dbg [[DBG385]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTSTART]], align 4, !dbg [[DBG385]]
// CHECK-DEBUG-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTSTOP]], align 4, !dbg [[DBG385]]
// CHECK-DEBUG-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP4]], [[TMP5]], !dbg [[DBG385]]
// CHECK-DEBUG-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]], !dbg [[DBG385]]
// CHECK-DEBUG:       cond.true:
// CHECK-DEBUG-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTSTOP]], align 4, !dbg [[DBG385]]
// CHECK-DEBUG-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTSTART]], align 4, !dbg [[DBG385]]
// CHECK-DEBUG-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], [[TMP7]], !dbg [[DBG385]]
// CHECK-DEBUG-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTSTEP]], align 4, !dbg [[DBG385]]
// CHECK-DEBUG-NEXT:    [[SUB1:%.*]] = sub i32 [[TMP8]], 1, !dbg [[DBG385]]
// CHECK-DEBUG-NEXT:    [[ADD:%.*]] = add i32 [[SUB]], [[SUB1]], !dbg [[DBG385]]
// CHECK-DEBUG-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTSTEP]], align 4, !dbg [[DBG385]]
// CHECK-DEBUG-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], [[TMP9]], !dbg [[DBG385]]
// CHECK-DEBUG-NEXT:    br label [[COND_END:%.*]], !dbg [[DBG385]]
// CHECK-DEBUG:       cond.false:
// CHECK-DEBUG-NEXT:    br label [[COND_END]], !dbg [[DBG385]]
// CHECK-DEBUG:       cond.end:
// CHECK-DEBUG-NEXT:    [[COND:%.*]] = phi i32 [ [[DIV]], [[COND_TRUE]] ], [ 0, [[COND_FALSE]] ], !dbg [[DBG385]]
// CHECK-DEBUG-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DISTANCE_ADDR]], align 8, !dbg [[DBG385]]
// CHECK-DEBUG-NEXT:    store i32 [[COND]], ptr [[TMP10]], align 4, !dbg [[DBG385]]
// CHECK-DEBUG-NEXT:    ret void, !dbg [[DBG387:![0-9]+]]
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@__captured_stmt.16
// CHECK-DEBUG-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[LOOPVAR:%.*]], i32 noundef [[LOGICAL:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR4]] !dbg [[DBG389:![0-9]+]] {
// CHECK-DEBUG-NEXT:  entry:
// CHECK-DEBUG-NEXT:    [[LOOPVAR_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[LOGICAL_ADDR:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    store ptr [[LOOPVAR]], ptr [[LOOPVAR_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[LOOPVAR_ADDR]], metadata [[META390:![0-9]+]], metadata !DIExpression()), !dbg [[DBG391:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 [[LOGICAL]], ptr [[LOGICAL_ADDR]], align 4
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[LOGICAL_ADDR]], metadata [[META392:![0-9]+]], metadata !DIExpression()), !dbg [[DBG391]]
// CHECK-DEBUG-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[__CONTEXT_ADDR]], metadata [[META393:![0-9]+]], metadata !DIExpression()), !dbg [[DBG391]]
// CHECK-DEBUG-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_14:%.*]], ptr [[TMP0]], i32 0, i32 0, !dbg [[DBG394:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4, !dbg [[DBG394]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = load i32, ptr [[LOGICAL_ADDR]], align 4, !dbg [[DBG396:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[MUL:%.*]] = mul i32 1, [[TMP3]], !dbg [[DBG396]]
// CHECK-DEBUG-NEXT:    [[ADD:%.*]] = add i32 [[TMP2]], [[MUL]], !dbg [[DBG396]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[LOOPVAR_ADDR]], align 8, !dbg [[DBG396]]
// CHECK-DEBUG-NEXT:    store i32 [[ADD]], ptr [[TMP4]], align 4, !dbg [[DBG391]]
// CHECK-DEBUG-NEXT:    ret void, !dbg [[DBG394]]
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@__captured_stmt.17
// CHECK-DEBUG-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[DISTANCE:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR4]] !dbg [[DBG397:![0-9]+]] {
// CHECK-DEBUG-NEXT:  entry:
// CHECK-DEBUG-NEXT:    [[DISTANCE_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[DOTSTART:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[DOTSTOP:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[DOTSTEP:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    store ptr [[DISTANCE]], ptr [[DISTANCE_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DISTANCE_ADDR]], metadata [[META398:![0-9]+]], metadata !DIExpression()), !dbg [[DBG399:![0-9]+]]
// CHECK-DEBUG-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[__CONTEXT_ADDR]], metadata [[META400:![0-9]+]], metadata !DIExpression()), !dbg [[DBG399]]
// CHECK-DEBUG-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTART]], metadata [[META401:![0-9]+]], metadata !DIExpression()), !dbg [[DBG403:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_15:%.*]], ptr [[TMP0]], i32 0, i32 0, !dbg [[DBG404:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP1]], align 8, !dbg [[DBG404]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = load i32, ptr [[TMP2]], align 4, !dbg [[DBG404]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP3]], ptr [[DOTSTART]], align 4, !dbg [[DBG403]]
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTOP]], metadata [[META406:![0-9]+]], metadata !DIExpression()), !dbg [[DBG407:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 100, ptr [[DOTSTOP]], align 4, !dbg [[DBG407]]
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTEP]], metadata [[META408:![0-9]+]], metadata !DIExpression()), !dbg [[DBG407]]
// CHECK-DEBUG-NEXT:    store i32 1, ptr [[DOTSTEP]], align 4, !dbg [[DBG407]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTSTART]], align 4, !dbg [[DBG407]]
// CHECK-DEBUG-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTSTOP]], align 4, !dbg [[DBG407]]
// CHECK-DEBUG-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP4]], [[TMP5]], !dbg [[DBG407]]
// CHECK-DEBUG-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]], !dbg [[DBG407]]
// CHECK-DEBUG:       cond.true:
// CHECK-DEBUG-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTSTOP]], align 4, !dbg [[DBG407]]
// CHECK-DEBUG-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTSTART]], align 4, !dbg [[DBG407]]
// CHECK-DEBUG-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], [[TMP7]], !dbg [[DBG407]]
// CHECK-DEBUG-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTSTEP]], align 4, !dbg [[DBG407]]
// CHECK-DEBUG-NEXT:    [[SUB1:%.*]] = sub i32 [[TMP8]], 1, !dbg [[DBG407]]
// CHECK-DEBUG-NEXT:    [[ADD:%.*]] = add i32 [[SUB]], [[SUB1]], !dbg [[DBG407]]
// CHECK-DEBUG-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTSTEP]], align 4, !dbg [[DBG407]]
// CHECK-DEBUG-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], [[TMP9]], !dbg [[DBG407]]
// CHECK-DEBUG-NEXT:    br label [[COND_END:%.*]], !dbg [[DBG407]]
// CHECK-DEBUG:       cond.false:
// CHECK-DEBUG-NEXT:    br label [[COND_END]], !dbg [[DBG407]]
// CHECK-DEBUG:       cond.end:
// CHECK-DEBUG-NEXT:    [[COND:%.*]] = phi i32 [ [[DIV]], [[COND_TRUE]] ], [ 0, [[COND_FALSE]] ], !dbg [[DBG407]]
// CHECK-DEBUG-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DISTANCE_ADDR]], align 8, !dbg [[DBG407]]
// CHECK-DEBUG-NEXT:    store i32 [[COND]], ptr [[TMP10]], align 4, !dbg [[DBG407]]
// CHECK-DEBUG-NEXT:    ret void, !dbg [[DBG409:![0-9]+]]
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@__captured_stmt.18
// CHECK-DEBUG-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[LOOPVAR:%.*]], i32 noundef [[LOGICAL:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR4]] !dbg [[DBG411:![0-9]+]] {
// CHECK-DEBUG-NEXT:  entry:
// CHECK-DEBUG-NEXT:    [[LOOPVAR_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[LOGICAL_ADDR:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    store ptr [[LOOPVAR]], ptr [[LOOPVAR_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[LOOPVAR_ADDR]], metadata [[META412:![0-9]+]], metadata !DIExpression()), !dbg [[DBG413:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 [[LOGICAL]], ptr [[LOGICAL_ADDR]], align 4
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[LOGICAL_ADDR]], metadata [[META414:![0-9]+]], metadata !DIExpression()), !dbg [[DBG413]]
// CHECK-DEBUG-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[__CONTEXT_ADDR]], metadata [[META415:![0-9]+]], metadata !DIExpression()), !dbg [[DBG413]]
// CHECK-DEBUG-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_16:%.*]], ptr [[TMP0]], i32 0, i32 0, !dbg [[DBG416:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4, !dbg [[DBG416]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = load i32, ptr [[LOGICAL_ADDR]], align 4, !dbg [[DBG418:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[MUL:%.*]] = mul i32 1, [[TMP3]], !dbg [[DBG418]]
// CHECK-DEBUG-NEXT:    [[ADD:%.*]] = add i32 [[TMP2]], [[MUL]], !dbg [[DBG418]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[LOOPVAR_ADDR]], align 8, !dbg [[DBG418]]
// CHECK-DEBUG-NEXT:    store i32 [[ADD]], ptr [[TMP4]], align 4, !dbg [[DBG413]]
// CHECK-DEBUG-NEXT:    ret void, !dbg [[DBG416]]
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@__captured_stmt.19
// CHECK-DEBUG-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[DISTANCE:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR4]] !dbg [[DBG419:![0-9]+]] {
// CHECK-DEBUG-NEXT:  entry:
// CHECK-DEBUG-NEXT:    [[DISTANCE_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[DOTSTART:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[DOTSTOP:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[DOTSTEP:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    store ptr [[DISTANCE]], ptr [[DISTANCE_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DISTANCE_ADDR]], metadata [[META420:![0-9]+]], metadata !DIExpression()), !dbg [[DBG421:![0-9]+]]
// CHECK-DEBUG-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[__CONTEXT_ADDR]], metadata [[META422:![0-9]+]], metadata !DIExpression()), !dbg [[DBG421]]
// CHECK-DEBUG-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTART]], metadata [[META423:![0-9]+]], metadata !DIExpression()), !dbg [[DBG425:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_17:%.*]], ptr [[TMP0]], i32 0, i32 0, !dbg [[DBG426:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP1]], align 8, !dbg [[DBG426]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = load i32, ptr [[TMP2]], align 4, !dbg [[DBG426]]
// CHECK-DEBUG-NEXT:    store i32 [[TMP3]], ptr [[DOTSTART]], align 4, !dbg [[DBG425]]
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTOP]], metadata [[META428:![0-9]+]], metadata !DIExpression()), !dbg [[DBG429:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 100, ptr [[DOTSTOP]], align 4, !dbg [[DBG429]]
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[DOTSTEP]], metadata [[META430:![0-9]+]], metadata !DIExpression()), !dbg [[DBG429]]
// CHECK-DEBUG-NEXT:    store i32 1, ptr [[DOTSTEP]], align 4, !dbg [[DBG429]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTSTART]], align 4, !dbg [[DBG429]]
// CHECK-DEBUG-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTSTOP]], align 4, !dbg [[DBG429]]
// CHECK-DEBUG-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP4]], [[TMP5]], !dbg [[DBG429]]
// CHECK-DEBUG-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]], !dbg [[DBG429]]
// CHECK-DEBUG:       cond.true:
// CHECK-DEBUG-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTSTOP]], align 4, !dbg [[DBG429]]
// CHECK-DEBUG-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTSTART]], align 4, !dbg [[DBG429]]
// CHECK-DEBUG-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], [[TMP7]], !dbg [[DBG429]]
// CHECK-DEBUG-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTSTEP]], align 4, !dbg [[DBG429]]
// CHECK-DEBUG-NEXT:    [[SUB1:%.*]] = sub i32 [[TMP8]], 1, !dbg [[DBG429]]
// CHECK-DEBUG-NEXT:    [[ADD:%.*]] = add i32 [[SUB]], [[SUB1]], !dbg [[DBG429]]
// CHECK-DEBUG-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTSTEP]], align 4, !dbg [[DBG429]]
// CHECK-DEBUG-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], [[TMP9]], !dbg [[DBG429]]
// CHECK-DEBUG-NEXT:    br label [[COND_END:%.*]], !dbg [[DBG429]]
// CHECK-DEBUG:       cond.false:
// CHECK-DEBUG-NEXT:    br label [[COND_END]], !dbg [[DBG429]]
// CHECK-DEBUG:       cond.end:
// CHECK-DEBUG-NEXT:    [[COND:%.*]] = phi i32 [ [[DIV]], [[COND_TRUE]] ], [ 0, [[COND_FALSE]] ], !dbg [[DBG429]]
// CHECK-DEBUG-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DISTANCE_ADDR]], align 8, !dbg [[DBG429]]
// CHECK-DEBUG-NEXT:    store i32 [[COND]], ptr [[TMP10]], align 4, !dbg [[DBG429]]
// CHECK-DEBUG-NEXT:    ret void, !dbg [[DBG431:![0-9]+]]
//
//
// CHECK-DEBUG-LABEL: define {{[^@]+}}@__captured_stmt.20
// CHECK-DEBUG-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[LOOPVAR:%.*]], i32 noundef [[LOGICAL:%.*]], ptr noalias noundef [[__CONTEXT:%.*]]) #[[ATTR4]] !dbg [[DBG433:![0-9]+]] {
// CHECK-DEBUG-NEXT:  entry:
// CHECK-DEBUG-NEXT:    [[LOOPVAR_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    [[LOGICAL_ADDR:%.*]] = alloca i32, align 4
// CHECK-DEBUG-NEXT:    [[__CONTEXT_ADDR:%.*]] = alloca ptr, align 8
// CHECK-DEBUG-NEXT:    store ptr [[LOOPVAR]], ptr [[LOOPVAR_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[LOOPVAR_ADDR]], metadata [[META434:![0-9]+]], metadata !DIExpression()), !dbg [[DBG435:![0-9]+]]
// CHECK-DEBUG-NEXT:    store i32 [[LOGICAL]], ptr [[LOGICAL_ADDR]], align 4
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[LOGICAL_ADDR]], metadata [[META436:![0-9]+]], metadata !DIExpression()), !dbg [[DBG435]]
// CHECK-DEBUG-NEXT:    store ptr [[__CONTEXT]], ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    call void @llvm.dbg.declare(metadata ptr [[__CONTEXT_ADDR]], metadata [[META437:![0-9]+]], metadata !DIExpression()), !dbg [[DBG435]]
// CHECK-DEBUG-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[__CONTEXT_ADDR]], align 8
// CHECK-DEBUG-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_ANON_18:%.*]], ptr [[TMP0]], i32 0, i32 0, !dbg [[DBG438:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4, !dbg [[DBG438]]
// CHECK-DEBUG-NEXT:    [[TMP3:%.*]] = load i32, ptr [[LOGICAL_ADDR]], align 4, !dbg [[DBG440:![0-9]+]]
// CHECK-DEBUG-NEXT:    [[MUL:%.*]] = mul i32 1, [[TMP3]], !dbg [[DBG440]]
// CHECK-DEBUG-NEXT:    [[ADD:%.*]] = add i32 [[TMP2]], [[MUL]], !dbg [[DBG440]]
// CHECK-DEBUG-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[LOOPVAR_ADDR]], align 8, !dbg [[DBG440]]
// CHECK-DEBUG-NEXT:    store i32 [[ADD]], ptr [[TMP4]], align 4, !dbg [[DBG435]]
// CHECK-DEBUG-NEXT:    ret void, !dbg [[DBG438]]
//