// 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 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fopenmp-targets=amdgcn-amd-amdhsa -fopenmp-target-ignore-env-vars -fopenmp-assume-no-nested-parallelism -fopenmp-assume-no-thread-state -emit-llvm-bc %s -o %t-ppc-host.bc
// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple amdgcn-amd-amdhsa -fopenmp-targets=amdgcn-amd-amdhsa -fopenmp-target-ignore-env-vars -fopenmp-assume-no-nested-parallelism -fopenmp-assume-no-thread-state -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s
// expected-no-diagnostics

#include <stdint.h>

int main()
{
  int N = 100;

  double a[N], b[N];
  int bint[N];
  unsigned cint[N];

  int8_t int8_sum = 0;
  int16_t int16_sum = 0;
  int32_t int32_sum = 0;
  uint32_t uint32_sum = 0;
  int64_t int64_sum = 0;
  uint64_t uint64_sum = 0;

  for (int i=0; i<N; i++)
    a[i]=i;
  for (int i=0; i<N; i++) {
    bint[i] = i+1;
    cint[i] = i+2;
  }

  double sum1, sum2, sum3, sum4;
  sum1 = sum2 = sum3 = sum4 = 0;

#pragma omp target teams distribute parallel for map(tofrom:sum1) reduction(+:sum1)
  for (int j = 0; j< N; j=j+1)
    sum1 += a[j];

#pragma omp target teams distribute parallel for map(tofrom:sum2) reduction(+:sum2)
  for (int j = 0; j< N; j=j+2)
    sum2 += a[j];

#pragma omp target teams distribute parallel for map(tofrom:sum1) reduction(+:sum1)
  for (int j = 0; j< N; j=j+1)
    for (int i = 0; i < N; ++i)
      sum1 += a[i];

#pragma omp target teams distribute parallel for map(tofrom:sum1) reduction(+:sum1)
  for (int j = 0; j< N; j=j+1) {
    sum1 += a[j];
    b[j] = a[j];
  }

#pragma omp target teams distribute parallel for map(tofrom:sum1) reduction(+:sum1) collapse(2)
  for (int j = 0; j< N; j=j+2)
    for (int i = j; i < N; i=i+3)
      sum1 += a[i];

#pragma omp target teams distribute parallel for map(tofrom:sum1) reduction(+:sum1) schedule(static,1)
  for (int j = 0; j< N; j=j+1)
    sum1 += a[j];

#pragma omp target teams distribute parallel for map(tofrom:sum1) reduction(+:sum1) schedule(dynamic,1)
  for (int j = 0; j< N; j=j+1)
    sum1 += a[j];

#pragma omp target teams distribute parallel for reduction(+:sum3) num_teams(100)
  for (int j = 0; j< N; j=j+1)
    sum3 += a[j];

#pragma omp target teams distribute parallel for map(tofrom:sum2) reduction(+:sum2) thread_limit(512)
  for (int j = 0; j< N; j=j+1)
    sum2 += a[j];

#pragma omp target teams distribute parallel for reduction(+:int32_sum)
  for (int j = 0; j< N; j=j+1)
    int32_sum += bint[j] + cint[j];

#pragma omp target teams distribute parallel for map(tofrom:uint32_sum) reduction(+:uint32_sum)
  for (int j = 0; j< N; j=j+1)
    uint32_sum += bint[j] + cint[j];

#pragma omp target teams distribute parallel for map(tofrom:int64_sum) reduction(+:int64_sum)
  for (int j = 0; j< N; j=j+1)
    int64_sum += bint[j] + cint[j];

#pragma omp target teams distribute parallel for map(tofrom:uint64_sum) reduction(+:uint64_sum)
  for (int j = 0; j< N; j=j+1)
    uint64_sum += bint[j] + cint[j];

#pragma omp target teams distribute parallel for map(tofrom:sum1) reduction(+:sum1)
  for (int j = 0; j< N; j=j+1) {
    sum1 += a[j];
#pragma omp simd
    for (int p = 0; p < N; p++)
      a[p]=b[p];
  }

}
// CHECK-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l33
// CHECK-SAME: (i64 noundef [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[SUM1:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]], ptr noalias noundef [[DYN_PTR:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[SUM1_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[VLA_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[J:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[N_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[N_ADDR]] to ptr
// CHECK-NEXT:    [[SUM1_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[SUM1_ADDR]] to ptr
// CHECK-NEXT:    [[VLA_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VLA_ADDR]] to ptr
// CHECK-NEXT:    [[A_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[A_ADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR1_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR1]] to ptr
// CHECK-NEXT:    [[DYN_PTR_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DYN_PTR_ADDR]] to ptr
// CHECK-NEXT:    [[J_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[J]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR__ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_2_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_2]] to ptr
// CHECK-NEXT:    [[DOTOMP_LB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_LB]] to ptr
// CHECK-NEXT:    [[DOTOMP_UB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_UB]] to ptr
// CHECK-NEXT:    [[DOTOMP_IV_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_IV]] to ptr
// CHECK-NEXT:    store i64 [[N]], ptr [[N_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[SUM1]], ptr [[SUM1_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store i64 [[VLA]], ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[A]], ptr [[A_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[SUM1_ADDR_ASCAST]], align 8, !nonnull [[META18:![0-9]+]], !align [[META19:![0-9]+]]
// CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[A_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META19]]
// CHECK-NEXT:    call void @__kmpc_specialized_kernel_init()
// CHECK-NEXT:    [[TMP5:%.*]] = alloca double, align 8, addrspace(5)
// CHECK-NEXT:    store double 0.000000e+00, ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[N_ADDR_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP6]], ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], 0
// CHECK-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP10:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[GPU_BLOCK_ID:%.*]] = call i32 @llvm.amdgcn.workgroup.id.x()
// CHECK-NEXT:    [[TMP11:%.*]] = mul i32 [[GPU_BLOCK_ID]], [[NVPTX_NUM_THREADS]]
// CHECK-NEXT:    [[TMP12:%.*]] = add i32 [[TMP11]], [[TMP10]]
// CHECK-NEXT:    [[TMP13:%.*]] = mul i32 [[TMP12]], 1
// CHECK-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP15:%.*]] = add i32 [[TMP13]], [[TMP14]]
// CHECK-NEXT:    [[TMP16:%.*]] = call i32 @__kmpc_get_hardware_num_blocks()
// CHECK-NEXT:    [[TMP17:%.*]] = zext i32 [[TMP15]] to i64
// CHECK-NEXT:    store i32 [[TMP15]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND:%.*]]
// CHECK:       for.cond:
// CHECK-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[CMP:%.*]] = icmp sle i32 [[TMP18]], [[TMP19]]
// CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// CHECK:       for.body:
// CHECK-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP20]], 1
// CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NEXT:    store i32 [[ADD]], ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP21:%.*]] = load i32, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP21]] to i64
// CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP4]], i64 [[IDXPROM]]
// CHECK-NEXT:    [[TMP22:%.*]] = load double, ptr [[ARRAYIDX]], align 8
// CHECK-NEXT:    [[TMP23:%.*]] = load double, ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    [[TMP24:%.*]] = fadd double [[TMP23]], [[TMP22]]
// CHECK-NEXT:    store double [[TMP24]], ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    br label [[FOR_INC:%.*]]
// CHECK:       for.inc:
// CHECK-NEXT:    [[NVPTX_NUM_THREADS4:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[TMP25:%.*]] = mul i32 [[NVPTX_NUM_THREADS4]], [[TMP16]]
// CHECK-NEXT:    [[TMP26:%.*]] = mul i32 [[TMP25]], 1
// CHECK-NEXT:    [[TMP27:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP28:%.*]] = add i32 [[TMP26]], [[TMP27]]
// CHECK-NEXT:    store i32 [[TMP28]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP20:![0-9]+]]
// CHECK:       for.end:
// CHECK-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP30:%.*]] = load ptr, ptr [[DOTADDR1_ASCAST]], align 8
// CHECK-NEXT:    [[TMP31:%.*]] = load double, ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    call void @__kmpc_xteamr_d_16x64(double [[TMP31]], ptr [[TMP2]], ptr [[TMP29]], ptr [[TMP30]], ptr @__kmpc_rfun_sum_d, ptr @__kmpc_rfun_sum_lds_d, double 0.000000e+00, i64 [[TMP17]], i32 [[TMP16]], i32 1)
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l37
// CHECK-SAME: (i64 noundef [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[SUM2:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]], ptr noalias noundef [[DYN_PTR:%.*]]) #[[ATTR0]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[SUM2_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[VLA_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[J:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[N_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[N_ADDR]] to ptr
// CHECK-NEXT:    [[SUM2_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[SUM2_ADDR]] to ptr
// CHECK-NEXT:    [[VLA_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VLA_ADDR]] to ptr
// CHECK-NEXT:    [[A_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[A_ADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR1_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR1]] to ptr
// CHECK-NEXT:    [[DYN_PTR_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DYN_PTR_ADDR]] to ptr
// CHECK-NEXT:    [[J_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[J]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR__ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_2_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_2]] to ptr
// CHECK-NEXT:    [[DOTOMP_LB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_LB]] to ptr
// CHECK-NEXT:    [[DOTOMP_UB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_UB]] to ptr
// CHECK-NEXT:    [[DOTOMP_IV_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_IV]] to ptr
// CHECK-NEXT:    store i64 [[N]], ptr [[N_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[SUM2]], ptr [[SUM2_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store i64 [[VLA]], ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[A]], ptr [[A_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[SUM2_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META19]]
// CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[A_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META19]]
// CHECK-NEXT:    call void @__kmpc_specialized_kernel_init()
// CHECK-NEXT:    [[TMP5:%.*]] = alloca double, align 8, addrspace(5)
// CHECK-NEXT:    store double 0.000000e+00, ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[N_ADDR_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP6]], ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[SUB:%.*]] = sub i32 [[TMP7]], -1
// CHECK-NEXT:    [[DIV:%.*]] = udiv i32 [[SUB]], 2
// CHECK-NEXT:    [[SUB3:%.*]] = sub i32 [[DIV]], 1
// CHECK-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP10:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[GPU_BLOCK_ID:%.*]] = call i32 @llvm.amdgcn.workgroup.id.x()
// CHECK-NEXT:    [[TMP11:%.*]] = mul i32 [[GPU_BLOCK_ID]], [[NVPTX_NUM_THREADS]]
// CHECK-NEXT:    [[TMP12:%.*]] = add i32 [[TMP11]], [[TMP10]]
// CHECK-NEXT:    [[TMP13:%.*]] = mul i32 [[TMP12]], 1
// CHECK-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP15:%.*]] = add i32 [[TMP13]], [[TMP14]]
// CHECK-NEXT:    [[TMP16:%.*]] = call i32 @__kmpc_get_hardware_num_blocks()
// CHECK-NEXT:    [[TMP17:%.*]] = zext i32 [[TMP15]] to i64
// CHECK-NEXT:    store i32 [[TMP15]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND:%.*]]
// CHECK:       for.cond:
// CHECK-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP19]], 1
// CHECK-NEXT:    [[CMP:%.*]] = icmp ult i32 [[TMP18]], [[ADD]]
// CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// CHECK:       for.body:
// CHECK-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[MUL:%.*]] = mul i32 [[TMP20]], 2
// CHECK-NEXT:    [[ADD4:%.*]] = add i32 0, [[MUL]]
// CHECK-NEXT:    store i32 [[ADD4]], ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP21:%.*]] = load i32, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP21]] to i64
// CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP4]], i64 [[IDXPROM]]
// CHECK-NEXT:    [[TMP22:%.*]] = load double, ptr [[ARRAYIDX]], align 8
// CHECK-NEXT:    [[TMP23:%.*]] = load double, ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    [[TMP24:%.*]] = fadd double [[TMP23]], [[TMP22]]
// CHECK-NEXT:    store double [[TMP24]], ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    br label [[FOR_INC:%.*]]
// CHECK:       for.inc:
// CHECK-NEXT:    [[NVPTX_NUM_THREADS5:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[TMP25:%.*]] = mul i32 [[NVPTX_NUM_THREADS5]], [[TMP16]]
// CHECK-NEXT:    [[TMP26:%.*]] = mul i32 [[TMP25]], 1
// CHECK-NEXT:    [[TMP27:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP28:%.*]] = add i32 [[TMP26]], [[TMP27]]
// CHECK-NEXT:    store i32 [[TMP28]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP22:![0-9]+]]
// CHECK:       for.end:
// CHECK-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP30:%.*]] = load ptr, ptr [[DOTADDR1_ASCAST]], align 8
// CHECK-NEXT:    [[TMP31:%.*]] = load double, ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    call void @__kmpc_xteamr_d_16x64(double [[TMP31]], ptr [[TMP2]], ptr [[TMP29]], ptr [[TMP30]], ptr @__kmpc_rfun_sum_d, ptr @__kmpc_rfun_sum_lds_d, double 0.000000e+00, i64 [[TMP17]], i32 [[TMP16]], i32 1)
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l41
// CHECK-SAME: (i64 noundef [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[SUM1:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]], ptr noalias noundef [[DYN_PTR:%.*]]) #[[ATTR0]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[SUM1_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[VLA_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[J:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[I:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[N_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[N_ADDR]] to ptr
// CHECK-NEXT:    [[SUM1_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[SUM1_ADDR]] to ptr
// CHECK-NEXT:    [[VLA_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VLA_ADDR]] to ptr
// CHECK-NEXT:    [[A_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[A_ADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR1_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR1]] to ptr
// CHECK-NEXT:    [[DYN_PTR_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DYN_PTR_ADDR]] to ptr
// CHECK-NEXT:    [[J_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[J]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR__ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_2_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_2]] to ptr
// CHECK-NEXT:    [[DOTOMP_LB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_LB]] to ptr
// CHECK-NEXT:    [[DOTOMP_UB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_UB]] to ptr
// CHECK-NEXT:    [[DOTOMP_IV_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_IV]] to ptr
// CHECK-NEXT:    [[I_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[I]] to ptr
// CHECK-NEXT:    store i64 [[N]], ptr [[N_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[SUM1]], ptr [[SUM1_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store i64 [[VLA]], ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[A]], ptr [[A_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[SUM1_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META19]]
// CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[A_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META19]]
// CHECK-NEXT:    call void @__kmpc_specialized_kernel_init()
// CHECK-NEXT:    [[TMP5:%.*]] = alloca double, align 8, addrspace(5)
// CHECK-NEXT:    store double 0.000000e+00, ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[N_ADDR_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP6]], ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], 0
// CHECK-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP10:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[GPU_BLOCK_ID:%.*]] = call i32 @llvm.amdgcn.workgroup.id.x()
// CHECK-NEXT:    [[TMP11:%.*]] = mul i32 [[GPU_BLOCK_ID]], [[NVPTX_NUM_THREADS]]
// CHECK-NEXT:    [[TMP12:%.*]] = add i32 [[TMP11]], [[TMP10]]
// CHECK-NEXT:    [[TMP13:%.*]] = mul i32 [[TMP12]], 1
// CHECK-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP15:%.*]] = add i32 [[TMP13]], [[TMP14]]
// CHECK-NEXT:    [[TMP16:%.*]] = call i32 @__kmpc_get_hardware_num_blocks()
// CHECK-NEXT:    [[TMP17:%.*]] = zext i32 [[TMP15]] to i64
// CHECK-NEXT:    store i32 [[TMP15]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND:%.*]]
// CHECK:       for.cond:
// CHECK-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[CMP:%.*]] = icmp sle i32 [[TMP18]], [[TMP19]]
// CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END9:%.*]]
// CHECK:       for.body:
// CHECK-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP20]], 1
// CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NEXT:    store i32 [[ADD]], ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[I_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND4:%.*]]
// CHECK:       for.cond4:
// CHECK-NEXT:    [[TMP21:%.*]] = load i32, ptr [[I_ASCAST]], align 4
// CHECK-NEXT:    [[TMP22:%.*]] = load i32, ptr [[N_ADDR_ASCAST]], align 4
// CHECK-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP21]], [[TMP22]]
// CHECK-NEXT:    br i1 [[CMP5]], label [[FOR_BODY6:%.*]], label [[FOR_END:%.*]]
// CHECK:       for.body6:
// CHECK-NEXT:    [[TMP23:%.*]] = load i32, ptr [[I_ASCAST]], align 4
// CHECK-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP23]] to i64
// CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP4]], i64 [[IDXPROM]]
// CHECK-NEXT:    [[TMP24:%.*]] = load double, ptr [[ARRAYIDX]], align 8
// CHECK-NEXT:    [[TMP25:%.*]] = load double, ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    [[TMP26:%.*]] = fadd double [[TMP25]], [[TMP24]]
// CHECK-NEXT:    store double [[TMP26]], ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    br label [[FOR_INC:%.*]]
// CHECK:       for.inc:
// CHECK-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I_ASCAST]], align 4
// CHECK-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP27]], 1
// CHECK-NEXT:    store i32 [[INC]], ptr [[I_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND4]], !llvm.loop [[LOOP23:![0-9]+]]
// CHECK:       for.end:
// CHECK-NEXT:    br label [[FOR_INC7:%.*]]
// CHECK:       for.inc7:
// CHECK-NEXT:    [[NVPTX_NUM_THREADS8:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[TMP28:%.*]] = mul i32 [[NVPTX_NUM_THREADS8]], [[TMP16]]
// CHECK-NEXT:    [[TMP29:%.*]] = mul i32 [[TMP28]], 1
// CHECK-NEXT:    [[TMP30:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP31:%.*]] = add i32 [[TMP29]], [[TMP30]]
// CHECK-NEXT:    store i32 [[TMP31]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP24:![0-9]+]]
// CHECK:       for.end9:
// CHECK-NEXT:    [[TMP32:%.*]] = load ptr, ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP33:%.*]] = load ptr, ptr [[DOTADDR1_ASCAST]], align 8
// CHECK-NEXT:    [[TMP34:%.*]] = load double, ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    call void @__kmpc_xteamr_d_16x64(double [[TMP34]], ptr [[TMP2]], ptr [[TMP32]], ptr [[TMP33]], ptr @__kmpc_rfun_sum_d, ptr @__kmpc_rfun_sum_lds_d, double 0.000000e+00, i64 [[TMP17]], i32 [[TMP16]], i32 1)
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l46
// CHECK-SAME: (i64 noundef [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[SUM1:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], i64 noundef [[VLA1:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]], ptr noalias noundef [[DYN_PTR:%.*]]) #[[ATTR0]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[SUM1_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[VLA_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[VLA_ADDR2:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR3:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[J:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_4:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[N_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[N_ADDR]] to ptr
// CHECK-NEXT:    [[SUM1_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[SUM1_ADDR]] to ptr
// CHECK-NEXT:    [[VLA_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VLA_ADDR]] to ptr
// CHECK-NEXT:    [[A_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[A_ADDR]] to ptr
// CHECK-NEXT:    [[VLA_ADDR2_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VLA_ADDR2]] to ptr
// CHECK-NEXT:    [[B_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[B_ADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR3_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR3]] to ptr
// CHECK-NEXT:    [[DYN_PTR_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DYN_PTR_ADDR]] to ptr
// CHECK-NEXT:    [[J_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[J]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR__ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_4_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_4]] to ptr
// CHECK-NEXT:    [[DOTOMP_LB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_LB]] to ptr
// CHECK-NEXT:    [[DOTOMP_UB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_UB]] to ptr
// CHECK-NEXT:    [[DOTOMP_IV_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_IV]] to ptr
// CHECK-NEXT:    store i64 [[N]], ptr [[N_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[SUM1]], ptr [[SUM1_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store i64 [[VLA]], ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[A]], ptr [[A_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store i64 [[VLA1]], ptr [[VLA_ADDR2_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[B]], ptr [[B_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR3_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[SUM1_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META19]]
// CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[A_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META19]]
// CHECK-NEXT:    [[TMP5:%.*]] = load i64, ptr [[VLA_ADDR2_ASCAST]], align 8
// CHECK-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[B_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META19]]
// CHECK-NEXT:    call void @__kmpc_specialized_kernel_init()
// CHECK-NEXT:    [[TMP7:%.*]] = alloca double, align 8, addrspace(5)
// CHECK-NEXT:    store double 0.000000e+00, ptr addrspace(5) [[TMP7]], align 8
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[N_ADDR_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP8]], ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP9]], 0
// CHECK-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK-NEXT:    [[SUB5:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK-NEXT:    store i32 [[SUB5]], ptr [[DOTCAPTURE_EXPR_4_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_4_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP10]], ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP11]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP12:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[GPU_BLOCK_ID:%.*]] = call i32 @llvm.amdgcn.workgroup.id.x()
// CHECK-NEXT:    [[TMP13:%.*]] = mul i32 [[GPU_BLOCK_ID]], [[NVPTX_NUM_THREADS]]
// CHECK-NEXT:    [[TMP14:%.*]] = add i32 [[TMP13]], [[TMP12]]
// CHECK-NEXT:    [[TMP15:%.*]] = mul i32 [[TMP14]], 1
// CHECK-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP17:%.*]] = add i32 [[TMP15]], [[TMP16]]
// CHECK-NEXT:    [[TMP18:%.*]] = call i32 @__kmpc_get_hardware_num_blocks()
// CHECK-NEXT:    [[TMP19:%.*]] = zext i32 [[TMP17]] to i64
// CHECK-NEXT:    store i32 [[TMP17]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND:%.*]]
// CHECK:       for.cond:
// CHECK-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[CMP:%.*]] = icmp sle i32 [[TMP20]], [[TMP21]]
// CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// CHECK:       for.body:
// CHECK-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP22]], 1
// CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NEXT:    store i32 [[ADD]], ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP23:%.*]] = load i32, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP23]] to i64
// CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP4]], i64 [[IDXPROM]]
// CHECK-NEXT:    [[TMP24:%.*]] = load double, ptr [[ARRAYIDX]], align 8
// CHECK-NEXT:    [[TMP25:%.*]] = load double, ptr addrspace(5) [[TMP7]], align 8
// CHECK-NEXT:    [[TMP26:%.*]] = fadd double [[TMP25]], [[TMP24]]
// CHECK-NEXT:    store double [[TMP26]], ptr addrspace(5) [[TMP7]], align 8
// CHECK-NEXT:    [[TMP27:%.*]] = load i32, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[IDXPROM6:%.*]] = sext i32 [[TMP27]] to i64
// CHECK-NEXT:    [[ARRAYIDX7:%.*]] = getelementptr inbounds double, ptr [[TMP4]], i64 [[IDXPROM6]]
// CHECK-NEXT:    [[TMP28:%.*]] = load double, ptr [[ARRAYIDX7]], align 8
// CHECK-NEXT:    [[TMP29:%.*]] = load i32, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[IDXPROM8:%.*]] = sext i32 [[TMP29]] to i64
// CHECK-NEXT:    [[ARRAYIDX9:%.*]] = getelementptr inbounds double, ptr [[TMP6]], i64 [[IDXPROM8]]
// CHECK-NEXT:    store double [[TMP28]], ptr [[ARRAYIDX9]], align 8
// CHECK-NEXT:    br label [[FOR_INC:%.*]]
// CHECK:       for.inc:
// CHECK-NEXT:    [[NVPTX_NUM_THREADS10:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[TMP30:%.*]] = mul i32 [[NVPTX_NUM_THREADS10]], [[TMP18]]
// CHECK-NEXT:    [[TMP31:%.*]] = mul i32 [[TMP30]], 1
// CHECK-NEXT:    [[TMP32:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP33:%.*]] = add i32 [[TMP31]], [[TMP32]]
// CHECK-NEXT:    store i32 [[TMP33]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP25:![0-9]+]]
// CHECK:       for.end:
// CHECK-NEXT:    [[TMP34:%.*]] = load ptr, ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP35:%.*]] = load ptr, ptr [[DOTADDR3_ASCAST]], align 8
// CHECK-NEXT:    [[TMP36:%.*]] = load double, ptr addrspace(5) [[TMP7]], align 8
// CHECK-NEXT:    call void @__kmpc_xteamr_d_16x64(double [[TMP36]], ptr [[TMP2]], ptr [[TMP34]], ptr [[TMP35]], ptr @__kmpc_rfun_sum_d, ptr @__kmpc_rfun_sum_lds_d, double 0.000000e+00, i64 [[TMP19]], i32 [[TMP18]], i32 1)
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l52
// CHECK-SAME: (i64 noundef [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[SUM1:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]], ptr noalias noundef [[DYN_PTR:%.*]]) #[[ATTR0]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[SUM1_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[VLA_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[J:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[I:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTLB_MIN:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTLB_MAX:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTMIN_LESS_MAX:%.*]] = alloca i8, align 1, addrspace(5)
// CHECK-NEXT:    [[DOTUPPER:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTLOWER:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_LB:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_UB:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[N_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[N_ADDR]] to ptr
// CHECK-NEXT:    [[SUM1_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[SUM1_ADDR]] to ptr
// CHECK-NEXT:    [[VLA_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VLA_ADDR]] to ptr
// CHECK-NEXT:    [[A_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[A_ADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR1_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR1]] to ptr
// CHECK-NEXT:    [[DYN_PTR_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DYN_PTR_ADDR]] to ptr
// CHECK-NEXT:    [[J_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[J]] to ptr
// CHECK-NEXT:    [[I_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[I]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR__ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_]] to ptr
// CHECK-NEXT:    [[DOTLB_MIN_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTLB_MIN]] to ptr
// CHECK-NEXT:    [[DOTLB_MAX_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTLB_MAX]] to ptr
// CHECK-NEXT:    [[DOTMIN_LESS_MAX_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTMIN_LESS_MAX]] to ptr
// CHECK-NEXT:    [[DOTUPPER_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTUPPER]] to ptr
// CHECK-NEXT:    [[DOTLOWER_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTLOWER]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_2_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_2]] to ptr
// CHECK-NEXT:    [[DOTOMP_LB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_LB]] to ptr
// CHECK-NEXT:    [[DOTOMP_UB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_UB]] to ptr
// CHECK-NEXT:    [[DOTOMP_IV_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_IV]] to ptr
// CHECK-NEXT:    store i64 [[N]], ptr [[N_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[SUM1]], ptr [[SUM1_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store i64 [[VLA]], ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[A]], ptr [[A_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[SUM1_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META19]]
// CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[A_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META19]]
// CHECK-NEXT:    call void @__kmpc_specialized_kernel_init()
// CHECK-NEXT:    [[TMP5:%.*]] = alloca double, align 8, addrspace(5)
// CHECK-NEXT:    store double 0.000000e+00, ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP6]], ptr [[I_ASCAST]], align 4
// CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[N_ADDR_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP7]], ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP8]], ptr [[DOTLB_MIN_ASCAST]], align 4
// CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP9]], 1
// CHECK-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 2
// CHECK-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 2
// CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NEXT:    store i32 [[ADD]], ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP10:%.*]] = load i32, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP10]], ptr [[DOTLB_MAX_ASCAST]], align 4
// CHECK-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTLB_MIN_ASCAST]], align 4
// CHECK-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTLB_MAX_ASCAST]], align 4
// CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP11]], [[TMP12]]
// CHECK-NEXT:    [[STOREDV:%.*]] = zext i1 [[CMP]] to i8
// CHECK-NEXT:    store i8 [[STOREDV]], ptr [[DOTMIN_LESS_MAX_ASCAST]], align 1
// CHECK-NEXT:    [[TMP13:%.*]] = load i32, ptr [[N_ADDR_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP13]], ptr [[DOTUPPER_ASCAST]], align 4
// CHECK-NEXT:    [[TMP14:%.*]] = load i8, ptr [[DOTMIN_LESS_MAX_ASCAST]], align 1
// CHECK-NEXT:    [[LOADEDV:%.*]] = icmp ne i8 [[TMP14]], 0
// CHECK-NEXT:    br i1 [[LOADEDV]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK:       cond.true:
// CHECK-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTLB_MIN_ASCAST]], align 4
// CHECK-NEXT:    br label [[COND_END:%.*]]
// CHECK:       cond.false:
// CHECK-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTLB_MAX_ASCAST]], align 4
// CHECK-NEXT:    br label [[COND_END]]
// CHECK:       cond.end:
// CHECK-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP15]], [[COND_TRUE]] ], [ [[TMP16]], [[COND_FALSE]] ]
// CHECK-NEXT:    store i32 [[COND]], ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[COND]], ptr [[DOTLOWER_ASCAST]], align 4
// CHECK-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[SUB3:%.*]] = sub i32 [[TMP17]], -1
// CHECK-NEXT:    [[DIV4:%.*]] = udiv i32 [[SUB3]], 2
// CHECK-NEXT:    [[CONV:%.*]] = zext i32 [[DIV4]] to i64
// CHECK-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTUPPER_ASCAST]], align 4
// CHECK-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTLOWER_ASCAST]], align 4
// CHECK-NEXT:    [[SUB5:%.*]] = sub i32 [[TMP18]], [[TMP19]]
// CHECK-NEXT:    [[SUB6:%.*]] = sub i32 [[SUB5]], 1
// CHECK-NEXT:    [[ADD7:%.*]] = add i32 [[SUB6]], 3
// CHECK-NEXT:    [[DIV8:%.*]] = udiv i32 [[ADD7]], 3
// CHECK-NEXT:    [[CONV9:%.*]] = zext i32 [[DIV8]] to i64
// CHECK-NEXT:    [[MUL10:%.*]] = mul nsw i64 [[CONV]], [[CONV9]]
// CHECK-NEXT:    [[SUB11:%.*]] = sub nsw i64 [[MUL10]], 1
// CHECK-NEXT:    store i64 [[SUB11]], ptr [[DOTCAPTURE_EXPR_2_ASCAST]], align 8
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP20:%.*]] = load i32, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP20]], ptr [[I_ASCAST]], align 4
// CHECK-NEXT:    store i64 0, ptr [[DOTOMP_LB_ASCAST]], align 8
// CHECK-NEXT:    [[TMP21:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR_2_ASCAST]], align 8
// CHECK-NEXT:    store i64 [[TMP21]], ptr [[DOTOMP_UB_ASCAST]], align 8
// CHECK-NEXT:    [[TMP22:%.*]] = load i64, ptr [[DOTOMP_LB_ASCAST]], align 8
// CHECK-NEXT:    store i64 [[TMP22]], ptr [[DOTOMP_IV_ASCAST]], align 8
// CHECK-NEXT:    [[TMP23:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[GPU_BLOCK_ID:%.*]] = call i32 @llvm.amdgcn.workgroup.id.x()
// CHECK-NEXT:    [[TMP24:%.*]] = mul i32 [[GPU_BLOCK_ID]], [[NVPTX_NUM_THREADS]]
// CHECK-NEXT:    [[TMP25:%.*]] = add i32 [[TMP24]], [[TMP23]]
// CHECK-NEXT:    [[TMP26:%.*]] = zext i32 [[TMP25]] to i64
// CHECK-NEXT:    [[TMP27:%.*]] = mul i64 [[TMP26]], 1
// CHECK-NEXT:    [[TMP28:%.*]] = load i64, ptr [[DOTOMP_IV_ASCAST]], align 8
// CHECK-NEXT:    [[TMP29:%.*]] = add i64 [[TMP27]], [[TMP28]]
// CHECK-NEXT:    [[TMP30:%.*]] = call i32 @__kmpc_get_hardware_num_blocks()
// CHECK-NEXT:    store i64 [[TMP29]], ptr [[DOTOMP_IV_ASCAST]], align 8
// CHECK-NEXT:    br label [[FOR_COND:%.*]]
// CHECK:       for.cond:
// CHECK-NEXT:    [[TMP31:%.*]] = load i64, ptr [[DOTOMP_IV_ASCAST]], align 8
// CHECK-NEXT:    [[TMP32:%.*]] = load i64, ptr [[DOTOMP_UB_ASCAST]], align 8
// CHECK-NEXT:    [[CMP12:%.*]] = icmp sle i64 [[TMP31]], [[TMP32]]
// CHECK-NEXT:    br i1 [[CMP12]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// CHECK:       for.body:
// CHECK-NEXT:    [[TMP33:%.*]] = load i64, ptr [[DOTOMP_IV_ASCAST]], align 8
// CHECK-NEXT:    [[TMP34:%.*]] = load i32, ptr [[DOTUPPER_ASCAST]], align 4
// CHECK-NEXT:    [[TMP35:%.*]] = load i32, ptr [[DOTLOWER_ASCAST]], align 4
// CHECK-NEXT:    [[SUB13:%.*]] = sub i32 [[TMP34]], [[TMP35]]
// CHECK-NEXT:    [[SUB14:%.*]] = sub i32 [[SUB13]], 1
// CHECK-NEXT:    [[ADD15:%.*]] = add i32 [[SUB14]], 3
// CHECK-NEXT:    [[DIV16:%.*]] = udiv i32 [[ADD15]], 3
// CHECK-NEXT:    [[MUL17:%.*]] = mul i32 1, [[DIV16]]
// CHECK-NEXT:    [[CONV18:%.*]] = zext i32 [[MUL17]] to i64
// CHECK-NEXT:    [[DIV19:%.*]] = sdiv i64 [[TMP33]], [[CONV18]]
// CHECK-NEXT:    [[MUL20:%.*]] = mul nsw i64 [[DIV19]], 2
// CHECK-NEXT:    [[ADD21:%.*]] = add nsw i64 0, [[MUL20]]
// CHECK-NEXT:    [[CONV22:%.*]] = trunc i64 [[ADD21]] to i32
// CHECK-NEXT:    store i32 [[CONV22]], ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP36:%.*]] = load i32, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[CONV23:%.*]] = sext i32 [[TMP36]] to i64
// CHECK-NEXT:    [[TMP37:%.*]] = load i64, ptr [[DOTOMP_IV_ASCAST]], align 8
// CHECK-NEXT:    [[TMP38:%.*]] = load i64, ptr [[DOTOMP_IV_ASCAST]], align 8
// CHECK-NEXT:    [[TMP39:%.*]] = load i32, ptr [[DOTUPPER_ASCAST]], align 4
// CHECK-NEXT:    [[TMP40:%.*]] = load i32, ptr [[DOTLOWER_ASCAST]], align 4
// CHECK-NEXT:    [[SUB24:%.*]] = sub i32 [[TMP39]], [[TMP40]]
// CHECK-NEXT:    [[SUB25:%.*]] = sub i32 [[SUB24]], 1
// CHECK-NEXT:    [[ADD26:%.*]] = add i32 [[SUB25]], 3
// CHECK-NEXT:    [[DIV27:%.*]] = udiv i32 [[ADD26]], 3
// CHECK-NEXT:    [[MUL28:%.*]] = mul i32 1, [[DIV27]]
// CHECK-NEXT:    [[CONV29:%.*]] = zext i32 [[MUL28]] to i64
// CHECK-NEXT:    [[DIV30:%.*]] = sdiv i64 [[TMP38]], [[CONV29]]
// CHECK-NEXT:    [[TMP41:%.*]] = load i32, ptr [[DOTUPPER_ASCAST]], align 4
// CHECK-NEXT:    [[TMP42:%.*]] = load i32, ptr [[DOTLOWER_ASCAST]], align 4
// CHECK-NEXT:    [[SUB31:%.*]] = sub i32 [[TMP41]], [[TMP42]]
// CHECK-NEXT:    [[SUB32:%.*]] = sub i32 [[SUB31]], 1
// CHECK-NEXT:    [[ADD33:%.*]] = add i32 [[SUB32]], 3
// CHECK-NEXT:    [[DIV34:%.*]] = udiv i32 [[ADD33]], 3
// CHECK-NEXT:    [[MUL35:%.*]] = mul i32 1, [[DIV34]]
// CHECK-NEXT:    [[CONV36:%.*]] = zext i32 [[MUL35]] to i64
// CHECK-NEXT:    [[MUL37:%.*]] = mul nsw i64 [[DIV30]], [[CONV36]]
// CHECK-NEXT:    [[SUB38:%.*]] = sub nsw i64 [[TMP37]], [[MUL37]]
// CHECK-NEXT:    [[MUL39:%.*]] = mul nsw i64 [[SUB38]], 3
// CHECK-NEXT:    [[ADD40:%.*]] = add nsw i64 [[CONV23]], [[MUL39]]
// CHECK-NEXT:    [[CONV41:%.*]] = trunc i64 [[ADD40]] to i32
// CHECK-NEXT:    store i32 [[CONV41]], ptr [[I_ASCAST]], align 4
// CHECK-NEXT:    [[TMP43:%.*]] = load i32, ptr [[I_ASCAST]], align 4
// CHECK-NEXT:    [[TMP44:%.*]] = load i32, ptr [[N_ADDR_ASCAST]], align 4
// CHECK-NEXT:    [[CMP42:%.*]] = icmp slt i32 [[TMP43]], [[TMP44]]
// CHECK-NEXT:    br i1 [[CMP42]], label [[OMP_BODY_NEXT:%.*]], label [[FOR_INC:%.*]]
// CHECK:       omp.body.next:
// CHECK-NEXT:    [[TMP45:%.*]] = load i32, ptr [[I_ASCAST]], align 4
// CHECK-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP45]] to i64
// CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP4]], i64 [[IDXPROM]]
// CHECK-NEXT:    [[TMP46:%.*]] = load double, ptr [[ARRAYIDX]], align 8
// CHECK-NEXT:    [[TMP47:%.*]] = load double, ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    [[TMP48:%.*]] = fadd double [[TMP47]], [[TMP46]]
// CHECK-NEXT:    store double [[TMP48]], ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    br label [[FOR_INC]]
// CHECK:       for.inc:
// CHECK-NEXT:    [[NVPTX_NUM_THREADS43:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[TMP49:%.*]] = mul i32 [[NVPTX_NUM_THREADS43]], [[TMP30]]
// CHECK-NEXT:    [[TMP50:%.*]] = zext i32 [[TMP49]] to i64
// CHECK-NEXT:    [[TMP51:%.*]] = mul i64 [[TMP50]], 1
// CHECK-NEXT:    [[TMP52:%.*]] = load i64, ptr [[DOTOMP_IV_ASCAST]], align 8
// CHECK-NEXT:    [[TMP53:%.*]] = add i64 [[TMP51]], [[TMP52]]
// CHECK-NEXT:    store i64 [[TMP53]], ptr [[DOTOMP_IV_ASCAST]], align 8
// CHECK-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP26:![0-9]+]]
// CHECK:       for.end:
// CHECK-NEXT:    [[TMP54:%.*]] = load ptr, ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP55:%.*]] = load ptr, ptr [[DOTADDR1_ASCAST]], align 8
// CHECK-NEXT:    [[TMP56:%.*]] = load double, ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    call void @__kmpc_xteamr_d_16x64(double [[TMP56]], ptr [[TMP2]], ptr [[TMP54]], ptr [[TMP55]], ptr @__kmpc_rfun_sum_d, ptr @__kmpc_rfun_sum_lds_d, double 0.000000e+00, i64 [[TMP29]], i32 [[TMP30]], i32 1)
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l57
// CHECK-SAME: (i64 noundef [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[SUM1:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]], ptr noalias noundef [[DYN_PTR:%.*]]) #[[ATTR0]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[SUM1_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[VLA_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[J:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[N_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[N_ADDR]] to ptr
// CHECK-NEXT:    [[SUM1_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[SUM1_ADDR]] to ptr
// CHECK-NEXT:    [[VLA_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VLA_ADDR]] to ptr
// CHECK-NEXT:    [[A_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[A_ADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR1_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR1]] to ptr
// CHECK-NEXT:    [[DYN_PTR_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DYN_PTR_ADDR]] to ptr
// CHECK-NEXT:    [[J_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[J]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR__ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_2_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_2]] to ptr
// CHECK-NEXT:    [[DOTOMP_LB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_LB]] to ptr
// CHECK-NEXT:    [[DOTOMP_UB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_UB]] to ptr
// CHECK-NEXT:    [[DOTOMP_IV_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_IV]] to ptr
// CHECK-NEXT:    store i64 [[N]], ptr [[N_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[SUM1]], ptr [[SUM1_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store i64 [[VLA]], ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[A]], ptr [[A_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[SUM1_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META19]]
// CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[A_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META19]]
// CHECK-NEXT:    call void @__kmpc_specialized_kernel_init()
// CHECK-NEXT:    [[TMP5:%.*]] = alloca double, align 8, addrspace(5)
// CHECK-NEXT:    store double 0.000000e+00, ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[N_ADDR_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP6]], ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], 0
// CHECK-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP10:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[GPU_BLOCK_ID:%.*]] = call i32 @llvm.amdgcn.workgroup.id.x()
// CHECK-NEXT:    [[TMP11:%.*]] = mul i32 [[GPU_BLOCK_ID]], [[NVPTX_NUM_THREADS]]
// CHECK-NEXT:    [[TMP12:%.*]] = add i32 [[TMP11]], [[TMP10]]
// CHECK-NEXT:    [[TMP13:%.*]] = mul i32 [[TMP12]], 1
// CHECK-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP15:%.*]] = add i32 [[TMP13]], [[TMP14]]
// CHECK-NEXT:    [[TMP16:%.*]] = call i32 @__kmpc_get_hardware_num_blocks()
// CHECK-NEXT:    [[TMP17:%.*]] = zext i32 [[TMP15]] to i64
// CHECK-NEXT:    store i32 [[TMP15]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND:%.*]]
// CHECK:       for.cond:
// CHECK-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[CMP:%.*]] = icmp sle i32 [[TMP18]], [[TMP19]]
// CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// CHECK:       for.body:
// CHECK-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP20]], 1
// CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NEXT:    store i32 [[ADD]], ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP21:%.*]] = load i32, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP21]] to i64
// CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP4]], i64 [[IDXPROM]]
// CHECK-NEXT:    [[TMP22:%.*]] = load double, ptr [[ARRAYIDX]], align 8
// CHECK-NEXT:    [[TMP23:%.*]] = load double, ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    [[TMP24:%.*]] = fadd double [[TMP23]], [[TMP22]]
// CHECK-NEXT:    store double [[TMP24]], ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    br label [[FOR_INC:%.*]]
// CHECK:       for.inc:
// CHECK-NEXT:    [[NVPTX_NUM_THREADS4:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[TMP25:%.*]] = mul i32 [[NVPTX_NUM_THREADS4]], [[TMP16]]
// CHECK-NEXT:    [[TMP26:%.*]] = mul i32 [[TMP25]], 1
// CHECK-NEXT:    [[TMP27:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP28:%.*]] = add i32 [[TMP26]], [[TMP27]]
// CHECK-NEXT:    store i32 [[TMP28]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP27:![0-9]+]]
// CHECK:       for.end:
// CHECK-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP30:%.*]] = load ptr, ptr [[DOTADDR1_ASCAST]], align 8
// CHECK-NEXT:    [[TMP31:%.*]] = load double, ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    call void @__kmpc_xteamr_d_16x64(double [[TMP31]], ptr [[TMP2]], ptr [[TMP29]], ptr [[TMP30]], ptr @__kmpc_rfun_sum_d, ptr @__kmpc_rfun_sum_lds_d, double 0.000000e+00, i64 [[TMP17]], i32 [[TMP16]], i32 1)
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l61
// CHECK-SAME: (i64 noundef [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[SUM1:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]], ptr noalias noundef [[DYN_PTR:%.*]]) #[[ATTR0]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[SUM1_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[VLA_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[J:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[N_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[N_ADDR]] to ptr
// CHECK-NEXT:    [[SUM1_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[SUM1_ADDR]] to ptr
// CHECK-NEXT:    [[VLA_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VLA_ADDR]] to ptr
// CHECK-NEXT:    [[A_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[A_ADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR1_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR1]] to ptr
// CHECK-NEXT:    [[DYN_PTR_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DYN_PTR_ADDR]] to ptr
// CHECK-NEXT:    [[J_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[J]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR__ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_2_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_2]] to ptr
// CHECK-NEXT:    [[DOTOMP_LB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_LB]] to ptr
// CHECK-NEXT:    [[DOTOMP_UB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_UB]] to ptr
// CHECK-NEXT:    [[DOTOMP_IV_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_IV]] to ptr
// CHECK-NEXT:    store i64 [[N]], ptr [[N_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[SUM1]], ptr [[SUM1_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store i64 [[VLA]], ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[A]], ptr [[A_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[SUM1_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META19]]
// CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[A_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META19]]
// CHECK-NEXT:    call void @__kmpc_specialized_kernel_init()
// CHECK-NEXT:    [[TMP5:%.*]] = alloca double, align 8, addrspace(5)
// CHECK-NEXT:    store double 0.000000e+00, ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[N_ADDR_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP6]], ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], 0
// CHECK-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP10:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[GPU_BLOCK_ID:%.*]] = call i32 @llvm.amdgcn.workgroup.id.x()
// CHECK-NEXT:    [[TMP11:%.*]] = mul i32 [[GPU_BLOCK_ID]], [[NVPTX_NUM_THREADS]]
// CHECK-NEXT:    [[TMP12:%.*]] = add i32 [[TMP11]], [[TMP10]]
// CHECK-NEXT:    [[TMP13:%.*]] = mul i32 [[TMP12]], 1
// CHECK-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP15:%.*]] = add i32 [[TMP13]], [[TMP14]]
// CHECK-NEXT:    [[TMP16:%.*]] = call i32 @__kmpc_get_hardware_num_blocks()
// CHECK-NEXT:    [[TMP17:%.*]] = zext i32 [[TMP15]] to i64
// CHECK-NEXT:    store i32 [[TMP15]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND:%.*]]
// CHECK:       for.cond:
// CHECK-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[CMP:%.*]] = icmp sle i32 [[TMP18]], [[TMP19]]
// CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// CHECK:       for.body:
// CHECK-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP20]], 1
// CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NEXT:    store i32 [[ADD]], ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP21:%.*]] = load i32, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP21]] to i64
// CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP4]], i64 [[IDXPROM]]
// CHECK-NEXT:    [[TMP22:%.*]] = load double, ptr [[ARRAYIDX]], align 8
// CHECK-NEXT:    [[TMP23:%.*]] = load double, ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    [[TMP24:%.*]] = fadd double [[TMP23]], [[TMP22]]
// CHECK-NEXT:    store double [[TMP24]], ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    br label [[FOR_INC:%.*]]
// CHECK:       for.inc:
// CHECK-NEXT:    [[NVPTX_NUM_THREADS4:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[TMP25:%.*]] = mul i32 [[NVPTX_NUM_THREADS4]], [[TMP16]]
// CHECK-NEXT:    [[TMP26:%.*]] = mul i32 [[TMP25]], 1
// CHECK-NEXT:    [[TMP27:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP28:%.*]] = add i32 [[TMP26]], [[TMP27]]
// CHECK-NEXT:    store i32 [[TMP28]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP28:![0-9]+]]
// CHECK:       for.end:
// CHECK-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP30:%.*]] = load ptr, ptr [[DOTADDR1_ASCAST]], align 8
// CHECK-NEXT:    [[TMP31:%.*]] = load double, ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    call void @__kmpc_xteamr_d_16x64(double [[TMP31]], ptr [[TMP2]], ptr [[TMP29]], ptr [[TMP30]], ptr @__kmpc_rfun_sum_d, ptr @__kmpc_rfun_sum_lds_d, double 0.000000e+00, i64 [[TMP17]], i32 [[TMP16]], i32 1)
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l65
// CHECK-SAME: (i64 noundef [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[SUM3:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]], ptr noalias noundef [[DYN_PTR:%.*]]) #[[ATTR0]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[SUM3_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[VLA_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[J:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[N_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[N_ADDR]] to ptr
// CHECK-NEXT:    [[SUM3_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[SUM3_ADDR]] to ptr
// CHECK-NEXT:    [[VLA_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VLA_ADDR]] to ptr
// CHECK-NEXT:    [[A_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[A_ADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR1_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR1]] to ptr
// CHECK-NEXT:    [[DYN_PTR_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DYN_PTR_ADDR]] to ptr
// CHECK-NEXT:    [[J_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[J]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR__ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_2_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_2]] to ptr
// CHECK-NEXT:    [[DOTOMP_LB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_LB]] to ptr
// CHECK-NEXT:    [[DOTOMP_UB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_UB]] to ptr
// CHECK-NEXT:    [[DOTOMP_IV_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_IV]] to ptr
// CHECK-NEXT:    store i64 [[N]], ptr [[N_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[SUM3]], ptr [[SUM3_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store i64 [[VLA]], ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[A]], ptr [[A_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[SUM3_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META19]]
// CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[A_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META19]]
// CHECK-NEXT:    call void @__kmpc_specialized_kernel_init()
// CHECK-NEXT:    [[TMP5:%.*]] = alloca double, align 8, addrspace(5)
// CHECK-NEXT:    store double 0.000000e+00, ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[N_ADDR_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP6]], ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], 0
// CHECK-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP10:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[GPU_BLOCK_ID:%.*]] = call i32 @llvm.amdgcn.workgroup.id.x()
// CHECK-NEXT:    [[TMP11:%.*]] = mul i32 [[GPU_BLOCK_ID]], [[NVPTX_NUM_THREADS]]
// CHECK-NEXT:    [[TMP12:%.*]] = add i32 [[TMP11]], [[TMP10]]
// CHECK-NEXT:    [[TMP13:%.*]] = mul i32 [[TMP12]], 1
// CHECK-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP15:%.*]] = add i32 [[TMP13]], [[TMP14]]
// CHECK-NEXT:    [[TMP16:%.*]] = call i32 @__kmpc_get_hardware_num_blocks()
// CHECK-NEXT:    [[TMP17:%.*]] = zext i32 [[TMP15]] to i64
// CHECK-NEXT:    store i32 [[TMP15]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND:%.*]]
// CHECK:       for.cond:
// CHECK-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[CMP:%.*]] = icmp sle i32 [[TMP18]], [[TMP19]]
// CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// CHECK:       for.body:
// CHECK-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP20]], 1
// CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NEXT:    store i32 [[ADD]], ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP21:%.*]] = load i32, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP21]] to i64
// CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP4]], i64 [[IDXPROM]]
// CHECK-NEXT:    [[TMP22:%.*]] = load double, ptr [[ARRAYIDX]], align 8
// CHECK-NEXT:    [[TMP23:%.*]] = load double, ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    [[TMP24:%.*]] = fadd double [[TMP23]], [[TMP22]]
// CHECK-NEXT:    store double [[TMP24]], ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    br label [[FOR_INC:%.*]]
// CHECK:       for.inc:
// CHECK-NEXT:    [[NVPTX_NUM_THREADS4:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[TMP25:%.*]] = mul i32 [[NVPTX_NUM_THREADS4]], [[TMP16]]
// CHECK-NEXT:    [[TMP26:%.*]] = mul i32 [[TMP25]], 1
// CHECK-NEXT:    [[TMP27:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP28:%.*]] = add i32 [[TMP26]], [[TMP27]]
// CHECK-NEXT:    store i32 [[TMP28]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP29:![0-9]+]]
// CHECK:       for.end:
// CHECK-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP30:%.*]] = load ptr, ptr [[DOTADDR1_ASCAST]], align 8
// CHECK-NEXT:    [[TMP31:%.*]] = load double, ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    call void @__kmpc_xteamr_d_16x64(double [[TMP31]], ptr [[TMP2]], ptr [[TMP29]], ptr [[TMP30]], ptr @__kmpc_rfun_sum_d, ptr @__kmpc_rfun_sum_lds_d, double 0.000000e+00, i64 [[TMP17]], i32 [[TMP16]], i32 1)
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l69
// CHECK-SAME: (i64 noundef [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[SUM2:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]], ptr noalias noundef [[DYN_PTR:%.*]]) #[[ATTR0]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[SUM2_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[VLA_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[J:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[N_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[N_ADDR]] to ptr
// CHECK-NEXT:    [[SUM2_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[SUM2_ADDR]] to ptr
// CHECK-NEXT:    [[VLA_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VLA_ADDR]] to ptr
// CHECK-NEXT:    [[A_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[A_ADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR1_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR1]] to ptr
// CHECK-NEXT:    [[DYN_PTR_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DYN_PTR_ADDR]] to ptr
// CHECK-NEXT:    [[J_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[J]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR__ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_2_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_2]] to ptr
// CHECK-NEXT:    [[DOTOMP_LB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_LB]] to ptr
// CHECK-NEXT:    [[DOTOMP_UB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_UB]] to ptr
// CHECK-NEXT:    [[DOTOMP_IV_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_IV]] to ptr
// CHECK-NEXT:    store i64 [[N]], ptr [[N_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[SUM2]], ptr [[SUM2_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store i64 [[VLA]], ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[A]], ptr [[A_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[SUM2_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META19]]
// CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[A_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META19]]
// CHECK-NEXT:    call void @__kmpc_specialized_kernel_init()
// CHECK-NEXT:    [[TMP5:%.*]] = alloca double, align 8, addrspace(5)
// CHECK-NEXT:    store double 0.000000e+00, ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[N_ADDR_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP6]], ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], 0
// CHECK-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP10:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[GPU_BLOCK_ID:%.*]] = call i32 @llvm.amdgcn.workgroup.id.x()
// CHECK-NEXT:    [[TMP11:%.*]] = mul i32 [[GPU_BLOCK_ID]], [[NVPTX_NUM_THREADS]]
// CHECK-NEXT:    [[TMP12:%.*]] = add i32 [[TMP11]], [[TMP10]]
// CHECK-NEXT:    [[TMP13:%.*]] = mul i32 [[TMP12]], 1
// CHECK-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP15:%.*]] = add i32 [[TMP13]], [[TMP14]]
// CHECK-NEXT:    [[TMP16:%.*]] = call i32 @__kmpc_get_hardware_num_blocks()
// CHECK-NEXT:    [[TMP17:%.*]] = zext i32 [[TMP15]] to i64
// CHECK-NEXT:    store i32 [[TMP15]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND:%.*]]
// CHECK:       for.cond:
// CHECK-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[CMP:%.*]] = icmp sle i32 [[TMP18]], [[TMP19]]
// CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// CHECK:       for.body:
// CHECK-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP20]], 1
// CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NEXT:    store i32 [[ADD]], ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP21:%.*]] = load i32, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP21]] to i64
// CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP4]], i64 [[IDXPROM]]
// CHECK-NEXT:    [[TMP22:%.*]] = load double, ptr [[ARRAYIDX]], align 8
// CHECK-NEXT:    [[TMP23:%.*]] = load double, ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    [[TMP24:%.*]] = fadd double [[TMP23]], [[TMP22]]
// CHECK-NEXT:    store double [[TMP24]], ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    br label [[FOR_INC:%.*]]
// CHECK:       for.inc:
// CHECK-NEXT:    [[NVPTX_NUM_THREADS4:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[TMP25:%.*]] = mul i32 [[NVPTX_NUM_THREADS4]], [[TMP16]]
// CHECK-NEXT:    [[TMP26:%.*]] = mul i32 [[TMP25]], 1
// CHECK-NEXT:    [[TMP27:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP28:%.*]] = add i32 [[TMP26]], [[TMP27]]
// CHECK-NEXT:    store i32 [[TMP28]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP30:![0-9]+]]
// CHECK:       for.end:
// CHECK-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP30:%.*]] = load ptr, ptr [[DOTADDR1_ASCAST]], align 8
// CHECK-NEXT:    [[TMP31:%.*]] = load double, ptr addrspace(5) [[TMP5]], align 8
// CHECK-NEXT:    call void @__kmpc_xteamr_d_16x64(double [[TMP31]], ptr [[TMP2]], ptr [[TMP29]], ptr [[TMP30]], ptr @__kmpc_rfun_sum_d, ptr @__kmpc_rfun_sum_lds_d, double 0.000000e+00, i64 [[TMP17]], i32 [[TMP16]], i32 1)
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l73
// CHECK-SAME: (i64 noundef [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[INT32_SUM:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[BINT:%.*]], i64 noundef [[VLA1:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[CINT:%.*]], ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]], ptr noalias noundef [[DYN_PTR:%.*]]) #[[ATTR0]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[INT32_SUM_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[VLA_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[BINT_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[VLA_ADDR2:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[CINT_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR3:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[J:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_4:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[N_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[N_ADDR]] to ptr
// CHECK-NEXT:    [[INT32_SUM_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[INT32_SUM_ADDR]] to ptr
// CHECK-NEXT:    [[VLA_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VLA_ADDR]] to ptr
// CHECK-NEXT:    [[BINT_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[BINT_ADDR]] to ptr
// CHECK-NEXT:    [[VLA_ADDR2_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VLA_ADDR2]] to ptr
// CHECK-NEXT:    [[CINT_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[CINT_ADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR3_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR3]] to ptr
// CHECK-NEXT:    [[DYN_PTR_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DYN_PTR_ADDR]] to ptr
// CHECK-NEXT:    [[J_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[J]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR__ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_4_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_4]] to ptr
// CHECK-NEXT:    [[DOTOMP_LB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_LB]] to ptr
// CHECK-NEXT:    [[DOTOMP_UB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_UB]] to ptr
// CHECK-NEXT:    [[DOTOMP_IV_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_IV]] to ptr
// CHECK-NEXT:    store i64 [[N]], ptr [[N_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[INT32_SUM]], ptr [[INT32_SUM_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store i64 [[VLA]], ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[BINT]], ptr [[BINT_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store i64 [[VLA1]], ptr [[VLA_ADDR2_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[CINT]], ptr [[CINT_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR3_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[INT32_SUM_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META31:![0-9]+]]
// CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[BINT_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META31]]
// CHECK-NEXT:    [[TMP5:%.*]] = load i64, ptr [[VLA_ADDR2_ASCAST]], align 8
// CHECK-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[CINT_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META31]]
// CHECK-NEXT:    call void @__kmpc_specialized_kernel_init()
// CHECK-NEXT:    [[TMP7:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    store i32 0, ptr addrspace(5) [[TMP7]], align 4
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[N_ADDR_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP8]], ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP9]], 0
// CHECK-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK-NEXT:    [[SUB5:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK-NEXT:    store i32 [[SUB5]], ptr [[DOTCAPTURE_EXPR_4_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_4_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP10]], ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP11]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP12:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[GPU_BLOCK_ID:%.*]] = call i32 @llvm.amdgcn.workgroup.id.x()
// CHECK-NEXT:    [[TMP13:%.*]] = mul i32 [[GPU_BLOCK_ID]], [[NVPTX_NUM_THREADS]]
// CHECK-NEXT:    [[TMP14:%.*]] = add i32 [[TMP13]], [[TMP12]]
// CHECK-NEXT:    [[TMP15:%.*]] = mul i32 [[TMP14]], 1
// CHECK-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP17:%.*]] = add i32 [[TMP15]], [[TMP16]]
// CHECK-NEXT:    [[TMP18:%.*]] = call i32 @__kmpc_get_hardware_num_blocks()
// CHECK-NEXT:    [[TMP19:%.*]] = zext i32 [[TMP17]] to i64
// CHECK-NEXT:    store i32 [[TMP17]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND:%.*]]
// CHECK:       for.cond:
// CHECK-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[CMP:%.*]] = icmp sle i32 [[TMP20]], [[TMP21]]
// CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// CHECK:       for.body:
// CHECK-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP22]], 1
// CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NEXT:    store i32 [[ADD]], ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP23:%.*]] = load i32, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP23]] to i64
// CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP4]], i64 [[IDXPROM]]
// CHECK-NEXT:    [[TMP24:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
// CHECK-NEXT:    [[TMP25:%.*]] = load i32, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[IDXPROM6:%.*]] = sext i32 [[TMP25]] to i64
// CHECK-NEXT:    [[ARRAYIDX7:%.*]] = getelementptr inbounds i32, ptr [[TMP6]], i64 [[IDXPROM6]]
// CHECK-NEXT:    [[TMP26:%.*]] = load i32, ptr [[ARRAYIDX7]], align 4
// CHECK-NEXT:    [[ADD8:%.*]] = add i32 [[TMP24]], [[TMP26]]
// CHECK-NEXT:    [[TMP27:%.*]] = load i32, ptr addrspace(5) [[TMP7]], align 4
// CHECK-NEXT:    [[TMP28:%.*]] = add i32 [[TMP27]], [[ADD8]]
// CHECK-NEXT:    store i32 [[TMP28]], ptr addrspace(5) [[TMP7]], align 4
// CHECK-NEXT:    br label [[FOR_INC:%.*]]
// CHECK:       for.inc:
// CHECK-NEXT:    [[NVPTX_NUM_THREADS9:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[TMP29:%.*]] = mul i32 [[NVPTX_NUM_THREADS9]], [[TMP18]]
// CHECK-NEXT:    [[TMP30:%.*]] = mul i32 [[TMP29]], 1
// CHECK-NEXT:    [[TMP31:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP32:%.*]] = add i32 [[TMP30]], [[TMP31]]
// CHECK-NEXT:    store i32 [[TMP32]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP32:![0-9]+]]
// CHECK:       for.end:
// CHECK-NEXT:    [[TMP33:%.*]] = load ptr, ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP34:%.*]] = load ptr, ptr [[DOTADDR3_ASCAST]], align 8
// CHECK-NEXT:    [[TMP35:%.*]] = load i32, ptr addrspace(5) [[TMP7]], align 4
// CHECK-NEXT:    call void @__kmpc_xteamr_i_16x64(i32 [[TMP35]], ptr [[TMP2]], ptr [[TMP33]], ptr [[TMP34]], ptr @__kmpc_rfun_sum_i, ptr @__kmpc_rfun_sum_lds_i, i32 0, i64 [[TMP19]], i32 [[TMP18]], i32 1)
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l77
// CHECK-SAME: (i64 noundef [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[UINT32_SUM:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[BINT:%.*]], i64 noundef [[VLA1:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[CINT:%.*]], ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]], ptr noalias noundef [[DYN_PTR:%.*]]) #[[ATTR0]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[UINT32_SUM_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[VLA_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[BINT_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[VLA_ADDR2:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[CINT_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR3:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[J:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_4:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[N_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[N_ADDR]] to ptr
// CHECK-NEXT:    [[UINT32_SUM_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[UINT32_SUM_ADDR]] to ptr
// CHECK-NEXT:    [[VLA_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VLA_ADDR]] to ptr
// CHECK-NEXT:    [[BINT_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[BINT_ADDR]] to ptr
// CHECK-NEXT:    [[VLA_ADDR2_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VLA_ADDR2]] to ptr
// CHECK-NEXT:    [[CINT_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[CINT_ADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR3_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR3]] to ptr
// CHECK-NEXT:    [[DYN_PTR_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DYN_PTR_ADDR]] to ptr
// CHECK-NEXT:    [[J_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[J]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR__ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_4_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_4]] to ptr
// CHECK-NEXT:    [[DOTOMP_LB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_LB]] to ptr
// CHECK-NEXT:    [[DOTOMP_UB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_UB]] to ptr
// CHECK-NEXT:    [[DOTOMP_IV_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_IV]] to ptr
// CHECK-NEXT:    store i64 [[N]], ptr [[N_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[UINT32_SUM]], ptr [[UINT32_SUM_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store i64 [[VLA]], ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[BINT]], ptr [[BINT_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store i64 [[VLA1]], ptr [[VLA_ADDR2_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[CINT]], ptr [[CINT_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR3_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[UINT32_SUM_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META31]]
// CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[BINT_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META31]]
// CHECK-NEXT:    [[TMP5:%.*]] = load i64, ptr [[VLA_ADDR2_ASCAST]], align 8
// CHECK-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[CINT_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META31]]
// CHECK-NEXT:    call void @__kmpc_specialized_kernel_init()
// CHECK-NEXT:    [[TMP7:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    store i32 0, ptr addrspace(5) [[TMP7]], align 4
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[N_ADDR_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP8]], ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP9]], 0
// CHECK-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK-NEXT:    [[SUB5:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK-NEXT:    store i32 [[SUB5]], ptr [[DOTCAPTURE_EXPR_4_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_4_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP10]], ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP11]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP12:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[GPU_BLOCK_ID:%.*]] = call i32 @llvm.amdgcn.workgroup.id.x()
// CHECK-NEXT:    [[TMP13:%.*]] = mul i32 [[GPU_BLOCK_ID]], [[NVPTX_NUM_THREADS]]
// CHECK-NEXT:    [[TMP14:%.*]] = add i32 [[TMP13]], [[TMP12]]
// CHECK-NEXT:    [[TMP15:%.*]] = mul i32 [[TMP14]], 1
// CHECK-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP17:%.*]] = add i32 [[TMP15]], [[TMP16]]
// CHECK-NEXT:    [[TMP18:%.*]] = call i32 @__kmpc_get_hardware_num_blocks()
// CHECK-NEXT:    [[TMP19:%.*]] = zext i32 [[TMP17]] to i64
// CHECK-NEXT:    store i32 [[TMP17]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND:%.*]]
// CHECK:       for.cond:
// CHECK-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[CMP:%.*]] = icmp sle i32 [[TMP20]], [[TMP21]]
// CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// CHECK:       for.body:
// CHECK-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP22]], 1
// CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NEXT:    store i32 [[ADD]], ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP23:%.*]] = load i32, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP23]] to i64
// CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP4]], i64 [[IDXPROM]]
// CHECK-NEXT:    [[TMP24:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
// CHECK-NEXT:    [[TMP25:%.*]] = load i32, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[IDXPROM6:%.*]] = sext i32 [[TMP25]] to i64
// CHECK-NEXT:    [[ARRAYIDX7:%.*]] = getelementptr inbounds i32, ptr [[TMP6]], i64 [[IDXPROM6]]
// CHECK-NEXT:    [[TMP26:%.*]] = load i32, ptr [[ARRAYIDX7]], align 4
// CHECK-NEXT:    [[ADD8:%.*]] = add i32 [[TMP24]], [[TMP26]]
// CHECK-NEXT:    [[TMP27:%.*]] = load i32, ptr addrspace(5) [[TMP7]], align 4
// CHECK-NEXT:    [[TMP28:%.*]] = add i32 [[TMP27]], [[ADD8]]
// CHECK-NEXT:    store i32 [[TMP28]], ptr addrspace(5) [[TMP7]], align 4
// CHECK-NEXT:    br label [[FOR_INC:%.*]]
// CHECK:       for.inc:
// CHECK-NEXT:    [[NVPTX_NUM_THREADS9:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[TMP29:%.*]] = mul i32 [[NVPTX_NUM_THREADS9]], [[TMP18]]
// CHECK-NEXT:    [[TMP30:%.*]] = mul i32 [[TMP29]], 1
// CHECK-NEXT:    [[TMP31:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP32:%.*]] = add i32 [[TMP30]], [[TMP31]]
// CHECK-NEXT:    store i32 [[TMP32]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP33:![0-9]+]]
// CHECK:       for.end:
// CHECK-NEXT:    [[TMP33:%.*]] = load ptr, ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP34:%.*]] = load ptr, ptr [[DOTADDR3_ASCAST]], align 8
// CHECK-NEXT:    [[TMP35:%.*]] = load i32, ptr addrspace(5) [[TMP7]], align 4
// CHECK-NEXT:    call void @__kmpc_xteamr_i_16x64(i32 [[TMP35]], ptr [[TMP2]], ptr [[TMP33]], ptr [[TMP34]], ptr @__kmpc_rfun_sum_i, ptr @__kmpc_rfun_sum_lds_i, i32 0, i64 [[TMP19]], i32 [[TMP18]], i32 1)
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l81
// CHECK-SAME: (i64 noundef [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[INT64_SUM:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[BINT:%.*]], i64 noundef [[VLA1:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[CINT:%.*]], ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]], ptr noalias noundef [[DYN_PTR:%.*]]) #[[ATTR0]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[INT64_SUM_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[VLA_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[BINT_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[VLA_ADDR2:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[CINT_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR3:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[J:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_4:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[N_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[N_ADDR]] to ptr
// CHECK-NEXT:    [[INT64_SUM_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[INT64_SUM_ADDR]] to ptr
// CHECK-NEXT:    [[VLA_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VLA_ADDR]] to ptr
// CHECK-NEXT:    [[BINT_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[BINT_ADDR]] to ptr
// CHECK-NEXT:    [[VLA_ADDR2_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VLA_ADDR2]] to ptr
// CHECK-NEXT:    [[CINT_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[CINT_ADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR3_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR3]] to ptr
// CHECK-NEXT:    [[DYN_PTR_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DYN_PTR_ADDR]] to ptr
// CHECK-NEXT:    [[J_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[J]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR__ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_4_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_4]] to ptr
// CHECK-NEXT:    [[DOTOMP_LB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_LB]] to ptr
// CHECK-NEXT:    [[DOTOMP_UB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_UB]] to ptr
// CHECK-NEXT:    [[DOTOMP_IV_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_IV]] to ptr
// CHECK-NEXT:    store i64 [[N]], ptr [[N_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[INT64_SUM]], ptr [[INT64_SUM_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store i64 [[VLA]], ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[BINT]], ptr [[BINT_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store i64 [[VLA1]], ptr [[VLA_ADDR2_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[CINT]], ptr [[CINT_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR3_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[INT64_SUM_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META19]]
// CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[BINT_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META31]]
// CHECK-NEXT:    [[TMP5:%.*]] = load i64, ptr [[VLA_ADDR2_ASCAST]], align 8
// CHECK-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[CINT_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META31]]
// CHECK-NEXT:    call void @__kmpc_specialized_kernel_init()
// CHECK-NEXT:    [[TMP7:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    store i64 0, ptr addrspace(5) [[TMP7]], align 8
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[N_ADDR_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP8]], ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP9]], 0
// CHECK-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK-NEXT:    [[SUB5:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK-NEXT:    store i32 [[SUB5]], ptr [[DOTCAPTURE_EXPR_4_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_4_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP10]], ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP11]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP12:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[GPU_BLOCK_ID:%.*]] = call i32 @llvm.amdgcn.workgroup.id.x()
// CHECK-NEXT:    [[TMP13:%.*]] = mul i32 [[GPU_BLOCK_ID]], [[NVPTX_NUM_THREADS]]
// CHECK-NEXT:    [[TMP14:%.*]] = add i32 [[TMP13]], [[TMP12]]
// CHECK-NEXT:    [[TMP15:%.*]] = mul i32 [[TMP14]], 1
// CHECK-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP17:%.*]] = add i32 [[TMP15]], [[TMP16]]
// CHECK-NEXT:    [[TMP18:%.*]] = call i32 @__kmpc_get_hardware_num_blocks()
// CHECK-NEXT:    [[TMP19:%.*]] = zext i32 [[TMP17]] to i64
// CHECK-NEXT:    store i32 [[TMP17]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND:%.*]]
// CHECK:       for.cond:
// CHECK-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[CMP:%.*]] = icmp sle i32 [[TMP20]], [[TMP21]]
// CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// CHECK:       for.body:
// CHECK-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP22]], 1
// CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NEXT:    store i32 [[ADD]], ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP23:%.*]] = load i32, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP23]] to i64
// CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP4]], i64 [[IDXPROM]]
// CHECK-NEXT:    [[TMP24:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
// CHECK-NEXT:    [[TMP25:%.*]] = load i32, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[IDXPROM6:%.*]] = sext i32 [[TMP25]] to i64
// CHECK-NEXT:    [[ARRAYIDX7:%.*]] = getelementptr inbounds i32, ptr [[TMP6]], i64 [[IDXPROM6]]
// CHECK-NEXT:    [[TMP26:%.*]] = load i32, ptr [[ARRAYIDX7]], align 4
// CHECK-NEXT:    [[ADD8:%.*]] = add i32 [[TMP24]], [[TMP26]]
// CHECK-NEXT:    [[TMP27:%.*]] = load i64, ptr addrspace(5) [[TMP7]], align 8
// CHECK-NEXT:    [[TMP28:%.*]] = zext i32 [[ADD8]] to i64
// CHECK-NEXT:    [[TMP29:%.*]] = add i64 [[TMP27]], [[TMP28]]
// CHECK-NEXT:    store i64 [[TMP29]], ptr addrspace(5) [[TMP7]], align 8
// CHECK-NEXT:    br label [[FOR_INC:%.*]]
// CHECK:       for.inc:
// CHECK-NEXT:    [[NVPTX_NUM_THREADS9:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[TMP30:%.*]] = mul i32 [[NVPTX_NUM_THREADS9]], [[TMP18]]
// CHECK-NEXT:    [[TMP31:%.*]] = mul i32 [[TMP30]], 1
// CHECK-NEXT:    [[TMP32:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP33:%.*]] = add i32 [[TMP31]], [[TMP32]]
// CHECK-NEXT:    store i32 [[TMP33]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP34:![0-9]+]]
// CHECK:       for.end:
// CHECK-NEXT:    [[TMP34:%.*]] = load ptr, ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP35:%.*]] = load ptr, ptr [[DOTADDR3_ASCAST]], align 8
// CHECK-NEXT:    [[TMP36:%.*]] = load i64, ptr addrspace(5) [[TMP7]], align 8
// CHECK-NEXT:    call void @__kmpc_xteamr_l_16x64(i64 [[TMP36]], ptr [[TMP2]], ptr [[TMP34]], ptr [[TMP35]], ptr @__kmpc_rfun_sum_l, ptr @__kmpc_rfun_sum_lds_l, i64 0, i64 [[TMP19]], i32 [[TMP18]], i32 1)
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l85
// CHECK-SAME: (i64 noundef [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[UINT64_SUM:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[BINT:%.*]], i64 noundef [[VLA1:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[CINT:%.*]], ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]], ptr noalias noundef [[DYN_PTR:%.*]]) #[[ATTR0]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[UINT64_SUM_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[VLA_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[BINT_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[VLA_ADDR2:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[CINT_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR3:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[J:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_4:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[N_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[N_ADDR]] to ptr
// CHECK-NEXT:    [[UINT64_SUM_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[UINT64_SUM_ADDR]] to ptr
// CHECK-NEXT:    [[VLA_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VLA_ADDR]] to ptr
// CHECK-NEXT:    [[BINT_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[BINT_ADDR]] to ptr
// CHECK-NEXT:    [[VLA_ADDR2_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VLA_ADDR2]] to ptr
// CHECK-NEXT:    [[CINT_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[CINT_ADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR3_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR3]] to ptr
// CHECK-NEXT:    [[DYN_PTR_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DYN_PTR_ADDR]] to ptr
// CHECK-NEXT:    [[J_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[J]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR__ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_4_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_4]] to ptr
// CHECK-NEXT:    [[DOTOMP_LB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_LB]] to ptr
// CHECK-NEXT:    [[DOTOMP_UB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_UB]] to ptr
// CHECK-NEXT:    [[DOTOMP_IV_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_IV]] to ptr
// CHECK-NEXT:    store i64 [[N]], ptr [[N_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[UINT64_SUM]], ptr [[UINT64_SUM_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store i64 [[VLA]], ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[BINT]], ptr [[BINT_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store i64 [[VLA1]], ptr [[VLA_ADDR2_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[CINT]], ptr [[CINT_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR3_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[UINT64_SUM_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META19]]
// CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[BINT_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META31]]
// CHECK-NEXT:    [[TMP5:%.*]] = load i64, ptr [[VLA_ADDR2_ASCAST]], align 8
// CHECK-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[CINT_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META31]]
// CHECK-NEXT:    call void @__kmpc_specialized_kernel_init()
// CHECK-NEXT:    [[TMP7:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    store i64 0, ptr addrspace(5) [[TMP7]], align 8
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[N_ADDR_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP8]], ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP9]], 0
// CHECK-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK-NEXT:    [[SUB5:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK-NEXT:    store i32 [[SUB5]], ptr [[DOTCAPTURE_EXPR_4_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_4_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP10]], ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP11]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP12:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[GPU_BLOCK_ID:%.*]] = call i32 @llvm.amdgcn.workgroup.id.x()
// CHECK-NEXT:    [[TMP13:%.*]] = mul i32 [[GPU_BLOCK_ID]], [[NVPTX_NUM_THREADS]]
// CHECK-NEXT:    [[TMP14:%.*]] = add i32 [[TMP13]], [[TMP12]]
// CHECK-NEXT:    [[TMP15:%.*]] = mul i32 [[TMP14]], 1
// CHECK-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP17:%.*]] = add i32 [[TMP15]], [[TMP16]]
// CHECK-NEXT:    [[TMP18:%.*]] = call i32 @__kmpc_get_hardware_num_blocks()
// CHECK-NEXT:    [[TMP19:%.*]] = zext i32 [[TMP17]] to i64
// CHECK-NEXT:    store i32 [[TMP17]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND:%.*]]
// CHECK:       for.cond:
// CHECK-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[CMP:%.*]] = icmp sle i32 [[TMP20]], [[TMP21]]
// CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// CHECK:       for.body:
// CHECK-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP22]], 1
// CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NEXT:    store i32 [[ADD]], ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP23:%.*]] = load i32, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP23]] to i64
// CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP4]], i64 [[IDXPROM]]
// CHECK-NEXT:    [[TMP24:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
// CHECK-NEXT:    [[TMP25:%.*]] = load i32, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[IDXPROM6:%.*]] = sext i32 [[TMP25]] to i64
// CHECK-NEXT:    [[ARRAYIDX7:%.*]] = getelementptr inbounds i32, ptr [[TMP6]], i64 [[IDXPROM6]]
// CHECK-NEXT:    [[TMP26:%.*]] = load i32, ptr [[ARRAYIDX7]], align 4
// CHECK-NEXT:    [[ADD8:%.*]] = add i32 [[TMP24]], [[TMP26]]
// CHECK-NEXT:    [[TMP27:%.*]] = load i64, ptr addrspace(5) [[TMP7]], align 8
// CHECK-NEXT:    [[TMP28:%.*]] = zext i32 [[ADD8]] to i64
// CHECK-NEXT:    [[TMP29:%.*]] = add i64 [[TMP27]], [[TMP28]]
// CHECK-NEXT:    store i64 [[TMP29]], ptr addrspace(5) [[TMP7]], align 8
// CHECK-NEXT:    br label [[FOR_INC:%.*]]
// CHECK:       for.inc:
// CHECK-NEXT:    [[NVPTX_NUM_THREADS9:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[TMP30:%.*]] = mul i32 [[NVPTX_NUM_THREADS9]], [[TMP18]]
// CHECK-NEXT:    [[TMP31:%.*]] = mul i32 [[TMP30]], 1
// CHECK-NEXT:    [[TMP32:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP33:%.*]] = add i32 [[TMP31]], [[TMP32]]
// CHECK-NEXT:    store i32 [[TMP33]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP35:![0-9]+]]
// CHECK:       for.end:
// CHECK-NEXT:    [[TMP34:%.*]] = load ptr, ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP35:%.*]] = load ptr, ptr [[DOTADDR3_ASCAST]], align 8
// CHECK-NEXT:    [[TMP36:%.*]] = load i64, ptr addrspace(5) [[TMP7]], align 8
// CHECK-NEXT:    call void @__kmpc_xteamr_l_16x64(i64 [[TMP36]], ptr [[TMP2]], ptr [[TMP34]], ptr [[TMP35]], ptr @__kmpc_rfun_sum_l, ptr @__kmpc_rfun_sum_lds_l, i64 0, i64 [[TMP19]], i32 [[TMP18]], i32 1)
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l89
// CHECK-SAME: (i64 noundef [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[SUM1:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], i64 noundef [[VLA1:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]], ptr noalias noundef [[DYN_PTR:%.*]]) #[[ATTR0]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[SUM1_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[VLA_ADDR:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[VLA_ADDR2:%.*]] = alloca i64, align 8, addrspace(5)
// CHECK-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DOTADDR3:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
// CHECK-NEXT:    [[J:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_4:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[TMP:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_6:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_7:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[P:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[DOTOMP_IV12:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[P13:%.*]] = alloca i32, align 4, addrspace(5)
// CHECK-NEXT:    [[N_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[N_ADDR]] to ptr
// CHECK-NEXT:    [[SUM1_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[SUM1_ADDR]] to ptr
// CHECK-NEXT:    [[VLA_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VLA_ADDR]] to ptr
// CHECK-NEXT:    [[A_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[A_ADDR]] to ptr
// CHECK-NEXT:    [[VLA_ADDR2_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VLA_ADDR2]] to ptr
// CHECK-NEXT:    [[B_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[B_ADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR]] to ptr
// CHECK-NEXT:    [[DOTADDR3_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTADDR3]] to ptr
// CHECK-NEXT:    [[DYN_PTR_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DYN_PTR_ADDR]] to ptr
// CHECK-NEXT:    [[J_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[J]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR__ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_4_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_4]] to ptr
// CHECK-NEXT:    [[DOTOMP_LB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_LB]] to ptr
// CHECK-NEXT:    [[DOTOMP_UB_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_UB]] to ptr
// CHECK-NEXT:    [[DOTOMP_IV_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_IV]] to ptr
// CHECK-NEXT:    [[TMP_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[TMP]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_6_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_6]] to ptr
// CHECK-NEXT:    [[DOTCAPTURE_EXPR_7_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCAPTURE_EXPR_7]] to ptr
// CHECK-NEXT:    [[P_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[P]] to ptr
// CHECK-NEXT:    [[DOTOMP_IV12_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTOMP_IV12]] to ptr
// CHECK-NEXT:    [[P13_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[P13]] to ptr
// CHECK-NEXT:    store i64 [[N]], ptr [[N_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[SUM1]], ptr [[SUM1_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store i64 [[VLA]], ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[A]], ptr [[A_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store i64 [[VLA1]], ptr [[VLA_ADDR2_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[B]], ptr [[B_ADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR3_ASCAST]], align 8
// CHECK-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[SUM1_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META19]]
// CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr [[VLA_ADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[A_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META19]]
// CHECK-NEXT:    [[TMP5:%.*]] = load i64, ptr [[VLA_ADDR2_ASCAST]], align 8
// CHECK-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[B_ADDR_ASCAST]], align 8, !nonnull [[META18]], !align [[META19]]
// CHECK-NEXT:    call void @__kmpc_specialized_kernel_init()
// CHECK-NEXT:    [[TMP7:%.*]] = alloca double, align 8, addrspace(5)
// CHECK-NEXT:    store double 0.000000e+00, ptr addrspace(5) [[TMP7]], align 8
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[N_ADDR_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP8]], ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ASCAST]], align 4
// CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP9]], 0
// CHECK-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
// CHECK-NEXT:    [[SUB5:%.*]] = sub nsw i32 [[DIV]], 1
// CHECK-NEXT:    store i32 [[SUB5]], ptr [[DOTCAPTURE_EXPR_4_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_4_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP10]], ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_LB_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP11]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP12:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
// CHECK-NEXT:    [[NVPTX_NUM_THREADS:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[GPU_BLOCK_ID:%.*]] = call i32 @llvm.amdgcn.workgroup.id.x()
// CHECK-NEXT:    [[TMP13:%.*]] = mul i32 [[GPU_BLOCK_ID]], [[NVPTX_NUM_THREADS]]
// CHECK-NEXT:    [[TMP14:%.*]] = add i32 [[TMP13]], [[TMP12]]
// CHECK-NEXT:    [[TMP15:%.*]] = mul i32 [[TMP14]], 1
// CHECK-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP17:%.*]] = add i32 [[TMP15]], [[TMP16]]
// CHECK-NEXT:    [[TMP18:%.*]] = call i32 @__kmpc_get_hardware_num_blocks()
// CHECK-NEXT:    [[TMP19:%.*]] = zext i32 [[TMP17]] to i64
// CHECK-NEXT:    store i32 [[TMP17]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND:%.*]]
// CHECK:       for.cond:
// CHECK-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_UB_ASCAST]], align 4
// CHECK-NEXT:    [[CMP:%.*]] = icmp sle i32 [[TMP20]], [[TMP21]]
// CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// CHECK:       for.body:
// CHECK-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP22]], 1
// CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NEXT:    store i32 [[ADD]], ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[TMP23:%.*]] = load i32, ptr [[J_ASCAST]], align 4
// CHECK-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP23]] to i64
// CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP4]], i64 [[IDXPROM]]
// CHECK-NEXT:    [[TMP24:%.*]] = load double, ptr [[ARRAYIDX]], align 8
// CHECK-NEXT:    [[TMP25:%.*]] = load double, ptr addrspace(5) [[TMP7]], align 8
// CHECK-NEXT:    [[TMP26:%.*]] = fadd double [[TMP25]], [[TMP24]]
// CHECK-NEXT:    store double [[TMP26]], ptr addrspace(5) [[TMP7]], align 8
// CHECK-NEXT:    [[TMP27:%.*]] = load i32, ptr [[N_ADDR_ASCAST]], align 4
// CHECK-NEXT:    store i32 [[TMP27]], ptr [[DOTCAPTURE_EXPR_6_ASCAST]], align 4
// CHECK-NEXT:    [[TMP28:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_6_ASCAST]], align 4
// CHECK-NEXT:    [[SUB8:%.*]] = sub nsw i32 [[TMP28]], 0
// CHECK-NEXT:    [[DIV9:%.*]] = sdiv i32 [[SUB8]], 1
// CHECK-NEXT:    [[SUB10:%.*]] = sub nsw i32 [[DIV9]], 1
// CHECK-NEXT:    store i32 [[SUB10]], ptr [[DOTCAPTURE_EXPR_7_ASCAST]], align 4
// CHECK-NEXT:    store i32 0, ptr [[P_ASCAST]], align 4
// CHECK-NEXT:    [[TMP29:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_6_ASCAST]], align 4
// CHECK-NEXT:    [[CMP11:%.*]] = icmp slt i32 0, [[TMP29]]
// CHECK-NEXT:    br i1 [[CMP11]], label [[SIMD_IF_THEN:%.*]], label [[SIMD_IF_END:%.*]]
// CHECK:       simd.if.then:
// CHECK-NEXT:    store i32 0, ptr [[DOTOMP_IV12_ASCAST]], align 4
// CHECK-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK:       omp.inner.for.cond:
// CHECK-NEXT:    [[TMP30:%.*]] = load i32, ptr [[DOTOMP_IV12_ASCAST]], align 4, !llvm.access.group [[ACC_GRP36:![0-9]+]]
// CHECK-NEXT:    [[TMP31:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_7_ASCAST]], align 4, !llvm.access.group [[ACC_GRP36]]
// CHECK-NEXT:    [[ADD14:%.*]] = add nsw i32 [[TMP31]], 1
// CHECK-NEXT:    [[CMP15:%.*]] = icmp slt i32 [[TMP30]], [[ADD14]]
// CHECK-NEXT:    br i1 [[CMP15]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK:       omp.inner.for.body:
// CHECK-NEXT:    [[TMP32:%.*]] = load i32, ptr [[DOTOMP_IV12_ASCAST]], align 4, !llvm.access.group [[ACC_GRP36]]
// CHECK-NEXT:    [[MUL16:%.*]] = mul nsw i32 [[TMP32]], 1
// CHECK-NEXT:    [[ADD17:%.*]] = add nsw i32 0, [[MUL16]]
// CHECK-NEXT:    store i32 [[ADD17]], ptr [[P13_ASCAST]], align 4, !llvm.access.group [[ACC_GRP36]]
// CHECK-NEXT:    [[TMP33:%.*]] = load i32, ptr [[P13_ASCAST]], align 4, !llvm.access.group [[ACC_GRP36]]
// CHECK-NEXT:    [[IDXPROM18:%.*]] = sext i32 [[TMP33]] to i64
// CHECK-NEXT:    [[ARRAYIDX19:%.*]] = getelementptr inbounds double, ptr [[TMP6]], i64 [[IDXPROM18]]
// CHECK-NEXT:    [[TMP34:%.*]] = load double, ptr [[ARRAYIDX19]], align 8, !llvm.access.group [[ACC_GRP36]]
// CHECK-NEXT:    [[TMP35:%.*]] = load i32, ptr [[P13_ASCAST]], align 4, !llvm.access.group [[ACC_GRP36]]
// CHECK-NEXT:    [[IDXPROM20:%.*]] = sext i32 [[TMP35]] to i64
// CHECK-NEXT:    [[ARRAYIDX21:%.*]] = getelementptr inbounds double, ptr [[TMP4]], i64 [[IDXPROM20]]
// CHECK-NEXT:    store double [[TMP34]], ptr [[ARRAYIDX21]], align 8, !llvm.access.group [[ACC_GRP36]]
// CHECK-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK:       omp.body.continue:
// CHECK-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK:       omp.inner.for.inc:
// CHECK-NEXT:    [[TMP36:%.*]] = load i32, ptr [[DOTOMP_IV12_ASCAST]], align 4, !llvm.access.group [[ACC_GRP36]]
// CHECK-NEXT:    [[ADD22:%.*]] = add nsw i32 [[TMP36]], 1
// CHECK-NEXT:    store i32 [[ADD22]], ptr [[DOTOMP_IV12_ASCAST]], align 4, !llvm.access.group [[ACC_GRP36]]
// CHECK-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP37:![0-9]+]]
// CHECK:       omp.inner.for.end:
// CHECK-NEXT:    [[TMP37:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_6_ASCAST]], align 4
// CHECK-NEXT:    [[SUB23:%.*]] = sub nsw i32 [[TMP37]], 0
// CHECK-NEXT:    [[DIV24:%.*]] = sdiv i32 [[SUB23]], 1
// CHECK-NEXT:    [[MUL25:%.*]] = mul nsw i32 [[DIV24]], 1
// CHECK-NEXT:    [[ADD26:%.*]] = add nsw i32 0, [[MUL25]]
// CHECK-NEXT:    store i32 [[ADD26]], ptr [[P13_ASCAST]], align 4
// CHECK-NEXT:    br label [[SIMD_IF_END]]
// CHECK:       simd.if.end:
// CHECK-NEXT:    br label [[FOR_INC:%.*]]
// CHECK:       for.inc:
// CHECK-NEXT:    [[NVPTX_NUM_THREADS27:%.*]] = call i32 @__kmpc_get_hardware_num_threads_in_block()
// CHECK-NEXT:    [[TMP38:%.*]] = mul i32 [[NVPTX_NUM_THREADS27]], [[TMP18]]
// CHECK-NEXT:    [[TMP39:%.*]] = mul i32 [[TMP38]], 1
// CHECK-NEXT:    [[TMP40:%.*]] = load i32, ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    [[TMP41:%.*]] = add i32 [[TMP39]], [[TMP40]]
// CHECK-NEXT:    store i32 [[TMP41]], ptr [[DOTOMP_IV_ASCAST]], align 4
// CHECK-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP40:![0-9]+]]
// CHECK:       for.end:
// CHECK-NEXT:    [[TMP42:%.*]] = load ptr, ptr [[DOTADDR_ASCAST]], align 8
// CHECK-NEXT:    [[TMP43:%.*]] = load ptr, ptr [[DOTADDR3_ASCAST]], align 8
// CHECK-NEXT:    [[TMP44:%.*]] = load double, ptr addrspace(5) [[TMP7]], align 8
// CHECK-NEXT:    call void @__kmpc_xteamr_d_16x64(double [[TMP44]], ptr [[TMP2]], ptr [[TMP42]], ptr [[TMP43]], ptr @__kmpc_rfun_sum_d, ptr @__kmpc_rfun_sum_lds_d, double 0.000000e+00, i64 [[TMP19]], i32 [[TMP18]], i32 1)
// CHECK-NEXT:    ret void
//