clang/test/OpenMP/bug60602.cpp

   1 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --include-generated-funcs --replace-value-regex "__omp_offloading_[0-9a-z]+_[0-9a-z]+" "reduction_size[.].+[.]" "pl_cond[.].+[.|,]" --prefix-filecheck-ir-name _ --version 2
   2 // Test host codegen.
   3 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s
   4 // expected-no-diagnostics
   5
   6 int kernel_within_loop(int *a, int *b, int N, int num_iters) {
   7   int i;
   8   for (i = 0; i < num_iters; ++i) {
   9 #pragma omp target parallel for map(a[0:N]) map(b[0:N])
  10     for (int j = 0; j< N; j++)
  11       a[j] = b[j];
  12
  13 #pragma omp target teams distribute parallel for map(a[0:N]) map(b[0:N])
  14     for (int j = 0; j< N; j+=3)
  15       a[j] = b[j] * 2;
  16   }
  17   return a[N-1];
  18 }
  19 // CHECK-LABEL: define dso_local noundef signext i32 @_Z18kernel_within_loopPiS_ii
  20 // CHECK-SAME: (ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], i32 noundef signext [[N:%.*]], i32 noundef signext [[NUM_ITERS:%.*]]) #[[ATTR0:[0-9]+]] {
  21 // CHECK-NEXT:  entry:
  22 // CHECK-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
  23 // CHECK-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
  24 // CHECK-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
  25 // CHECK-NEXT:    [[NUM_ITERS_ADDR:%.*]] = alloca i32, align 4
  26 // CHECK-NEXT:    [[I:%.*]] = alloca i32, align 4
  27 // CHECK-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
  28 // CHECK-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [3 x ptr], align 8
  29 // CHECK-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [3 x ptr], align 8
  30 // CHECK-NEXT:    [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [3 x ptr], align 8
  31 // CHECK-NEXT:    [[DOTOFFLOAD_SIZES:%.*]] = alloca [3 x i64], align 8
  32 // CHECK-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
  33 // CHECK-NEXT:    [[N_CASTED3:%.*]] = alloca i64, align 8
  34 // CHECK-NEXT:    [[DOTOFFLOAD_BASEPTRS8:%.*]] = alloca [3 x ptr], align 8
  35 // CHECK-NEXT:    [[DOTOFFLOAD_PTRS9:%.*]] = alloca [3 x ptr], align 8
  36 // CHECK-NEXT:    [[DOTOFFLOAD_MAPPERS10:%.*]] = alloca [3 x ptr], align 8
  37 // CHECK-NEXT:    [[DOTOFFLOAD_SIZES11:%.*]] = alloca [3 x i64], align 8
  38 // CHECK-NEXT:    [[TMP:%.*]] = alloca i32, align 4
  39 // CHECK-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
  40 // CHECK-NEXT:    [[DOTCAPTURE_EXPR_12:%.*]] = alloca i32, align 4
  41 // CHECK-NEXT:    [[KERNEL_ARGS14:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
  42 // CHECK-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
  43 // CHECK-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
  44 // CHECK-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
  45 // CHECK-NEXT:    store i32 [[NUM_ITERS]], ptr [[NUM_ITERS_ADDR]], align 4
  46 // CHECK-NEXT:    store i32 0, ptr [[I]], align 4
  47 // CHECK-NEXT:    br label [[FOR_COND:%.*]]
  48 // CHECK:       for.cond:
  49 // CHECK-NEXT:    [[TMP0:%.*]] = load i32, ptr [[I]], align 4
  50 // CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[NUM_ITERS_ADDR]], align 4
  51 // CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP0]], [[TMP1]]
  52 // CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
  53 // CHECK:       for.body:
  54 // CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4
  55 // CHECK-NEXT:    store i32 [[TMP2]], ptr [[N_CASTED]], align 4
  56 // CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr [[N_CASTED]], align 8
  57 // CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[A_ADDR]], align 8
  58 // CHECK-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[B_ADDR]], align 8
  59 // CHECK-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[A_ADDR]], align 8
  60 // CHECK-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[A_ADDR]], align 8
  61 // CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds nuw i32, ptr [[TMP7]], i64 0
  62 // CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[N_ADDR]], align 4
  63 // CHECK-NEXT:    [[CONV:%.*]] = sext i32 [[TMP8]] to i64
  64 // CHECK-NEXT:    [[TMP9:%.*]] = mul nuw i64 [[CONV]], 4
  65 // CHECK-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[B_ADDR]], align 8
  66 // CHECK-NEXT:    [[TMP11:%.*]] = load ptr, ptr [[B_ADDR]], align 8
  67 // CHECK-NEXT:    [[ARRAYIDX1:%.*]] = getelementptr inbounds nuw i32, ptr [[TMP11]], i64 0
  68 // CHECK-NEXT:    [[TMP12:%.*]] = load i32, ptr [[N_ADDR]], align 4
  69 // CHECK-NEXT:    [[CONV2:%.*]] = sext i32 [[TMP12]] to i64
  70 // CHECK-NEXT:    [[TMP13:%.*]] = mul nuw i64 [[CONV2]], 4
  71 // CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES]], ptr align 8 @.offload_sizes, i64 24, i1 false)
  72 // CHECK-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
  73 // CHECK-NEXT:    store i64 [[TMP3]], ptr [[TMP14]], align 8
  74 // CHECK-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
  75 // CHECK-NEXT:    store i64 [[TMP3]], ptr [[TMP15]], align 8
  76 // CHECK-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
  77 // CHECK-NEXT:    store ptr null, ptr [[TMP16]], align 8
  78 // CHECK-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
  79 // CHECK-NEXT:    store ptr [[TMP6]], ptr [[TMP17]], align 8
  80 // CHECK-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
  81 // CHECK-NEXT:    store ptr [[ARRAYIDX]], ptr [[TMP18]], align 8
  82 // CHECK-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 1
  83 // CHECK-NEXT:    store i64 [[TMP9]], ptr [[TMP19]], align 8
  84 // CHECK-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1
  85 // CHECK-NEXT:    store ptr null, ptr [[TMP20]], align 8
  86 // CHECK-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
  87 // CHECK-NEXT:    store ptr [[TMP10]], ptr [[TMP21]], align 8
  88 // CHECK-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
  89 // CHECK-NEXT:    store ptr [[ARRAYIDX1]], ptr [[TMP22]], align 8
  90 // CHECK-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 2
  91 // CHECK-NEXT:    store i64 [[TMP13]], ptr [[TMP23]], align 8
  92 // CHECK-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2
  93 // CHECK-NEXT:    store ptr null, ptr [[TMP24]], align 8
  94 // CHECK-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
  95 // CHECK-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
  96 // CHECK-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
  97 // CHECK-NEXT:    [[TMP28:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
  98 // CHECK-NEXT:    store i32 3, ptr [[TMP28]], align 4
  99 // CHECK-NEXT:    [[TMP29:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
 100 // CHECK-NEXT:    store i32 3, ptr [[TMP29]], align 4
 101 // CHECK-NEXT:    [[TMP30:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
 102 // CHECK-NEXT:    store ptr [[TMP25]], ptr [[TMP30]], align 8
 103 // CHECK-NEXT:    [[TMP31:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
 104 // CHECK-NEXT:    store ptr [[TMP26]], ptr [[TMP31]], align 8
 105 // CHECK-NEXT:    [[TMP32:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
 106 // CHECK-NEXT:    store ptr [[TMP27]], ptr [[TMP32]], align 8
 107 // CHECK-NEXT:    [[TMP33:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
 108 // CHECK-NEXT:    store ptr @.offload_maptypes, ptr [[TMP33]], align 8
 109 // CHECK-NEXT:    [[TMP34:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
 110 // CHECK-NEXT:    store ptr null, ptr [[TMP34]], align 8
 111 // CHECK-NEXT:    [[TMP35:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
 112 // CHECK-NEXT:    store ptr null, ptr [[TMP35]], align 8
 113 // CHECK-NEXT:    [[TMP36:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
 114 // CHECK-NEXT:    store i64 0, ptr [[TMP36]], align 8
 115 // CHECK-NEXT:    [[TMP37:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
 116 // CHECK-NEXT:    store i64 0, ptr [[TMP37]], align 8
 117 // CHECK-NEXT:    [[TMP38:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
 118 // CHECK-NEXT:    store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP38]], align 4
 119 // CHECK-NEXT:    [[TMP39:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
 120 // CHECK-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP39]], align 4
 121 // CHECK-NEXT:    [[TMP40:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
 122 // CHECK-NEXT:    store i32 0, ptr [[TMP40]], align 4
 123 // CHECK-NEXT:    [[TMP41:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18kernel_within_loopPiS_ii_l9.region_id, ptr [[KERNEL_ARGS]])
 124 // CHECK-NEXT:    [[TMP42:%.*]] = icmp ne i32 [[TMP41]], 0
 125 // CHECK-NEXT:    br i1 [[TMP42]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
 126 // CHECK:       omp_offload.failed:
 127 // CHECK-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18kernel_within_loopPiS_ii_l9(i64 [[TMP3]], ptr [[TMP4]], ptr [[TMP5]]) #[[ATTR2:[0-9]+]]
 128 // CHECK-NEXT:    br label [[OMP_OFFLOAD_CONT]]
 129 // CHECK:       omp_offload.cont:
 130 // CHECK-NEXT:    [[TMP43:%.*]] = load i32, ptr [[N_ADDR]], align 4
 131 // CHECK-NEXT:    store i32 [[TMP43]], ptr [[N_CASTED3]], align 4
 132 // CHECK-NEXT:    [[TMP44:%.*]] = load i64, ptr [[N_CASTED3]], align 8
 133 // CHECK-NEXT:    [[TMP45:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 134 // CHECK-NEXT:    [[TMP46:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 135 // CHECK-NEXT:    [[TMP47:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 136 // CHECK-NEXT:    [[TMP48:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 137 // CHECK-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds nuw i32, ptr [[TMP48]], i64 0
 138 // CHECK-NEXT:    [[TMP49:%.*]] = load i32, ptr [[N_ADDR]], align 4
 139 // CHECK-NEXT:    [[CONV5:%.*]] = sext i32 [[TMP49]] to i64
 140 // CHECK-NEXT:    [[TMP50:%.*]] = mul nuw i64 [[CONV5]], 4
 141 // CHECK-NEXT:    [[TMP51:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 142 // CHECK-NEXT:    [[TMP52:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 143 // CHECK-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds nuw i32, ptr [[TMP52]], i64 0
 144 // CHECK-NEXT:    [[TMP53:%.*]] = load i32, ptr [[N_ADDR]], align 4
 145 // CHECK-NEXT:    [[CONV7:%.*]] = sext i32 [[TMP53]] to i64
 146 // CHECK-NEXT:    [[TMP54:%.*]] = mul nuw i64 [[CONV7]], 4
 147 // CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES11]], ptr align 8 @.offload_sizes.1, i64 24, i1 false)
 148 // CHECK-NEXT:    [[TMP55:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
 149 // CHECK-NEXT:    store i64 [[TMP44]], ptr [[TMP55]], align 8
 150 // CHECK-NEXT:    [[TMP56:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
 151 // CHECK-NEXT:    store i64 [[TMP44]], ptr [[TMP56]], align 8
 152 // CHECK-NEXT:    [[TMP57:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0
 153 // CHECK-NEXT:    store ptr null, ptr [[TMP57]], align 8
 154 // CHECK-NEXT:    [[TMP58:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1
 155 // CHECK-NEXT:    store ptr [[TMP47]], ptr [[TMP58]], align 8
 156 // CHECK-NEXT:    [[TMP59:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1
 157 // CHECK-NEXT:    store ptr [[ARRAYIDX4]], ptr [[TMP59]], align 8
 158 // CHECK-NEXT:    [[TMP60:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES11]], i32 0, i32 1
 159 // CHECK-NEXT:    store i64 [[TMP50]], ptr [[TMP60]], align 8
 160 // CHECK-NEXT:    [[TMP61:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1
 161 // CHECK-NEXT:    store ptr null, ptr [[TMP61]], align 8
 162 // CHECK-NEXT:    [[TMP62:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 2
 163 // CHECK-NEXT:    store ptr [[TMP51]], ptr [[TMP62]], align 8
 164 // CHECK-NEXT:    [[TMP63:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 2
 165 // CHECK-NEXT:    store ptr [[ARRAYIDX6]], ptr [[TMP63]], align 8
 166 // CHECK-NEXT:    [[TMP64:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES11]], i32 0, i32 2
 167 // CHECK-NEXT:    store i64 [[TMP54]], ptr [[TMP64]], align 8
 168 // CHECK-NEXT:    [[TMP65:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 2
 169 // CHECK-NEXT:    store ptr null, ptr [[TMP65]], align 8
 170 // CHECK-NEXT:    [[TMP66:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0
 171 // CHECK-NEXT:    [[TMP67:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0
 172 // CHECK-NEXT:    [[TMP68:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES11]], i32 0, i32 0
 173 // CHECK-NEXT:    [[TMP69:%.*]] = load i32, ptr [[N_ADDR]], align 4
 174 // CHECK-NEXT:    store i32 [[TMP69]], ptr [[DOTCAPTURE_EXPR_]], align 4
 175 // CHECK-NEXT:    [[TMP70:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 176 // CHECK-NEXT:    [[SUB:%.*]] = sub i32 [[TMP70]], -2
 177 // CHECK-NEXT:    [[DIV:%.*]] = udiv i32 [[SUB]], 3
 178 // CHECK-NEXT:    [[SUB13:%.*]] = sub i32 [[DIV]], 1
 179 // CHECK-NEXT:    store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_12]], align 4
 180 // CHECK-NEXT:    [[TMP71:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_12]], align 4
 181 // CHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP71]], 1
 182 // CHECK-NEXT:    [[TMP72:%.*]] = zext i32 [[ADD]] to i64
 183 // CHECK-NEXT:    [[TMP73:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
 184 // CHECK-NEXT:    store i32 3, ptr [[TMP73]], align 4
 185 // CHECK-NEXT:    [[TMP74:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
 186 // CHECK-NEXT:    store i32 3, ptr [[TMP74]], align 4
 187 // CHECK-NEXT:    [[TMP75:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
 188 // CHECK-NEXT:    store ptr [[TMP66]], ptr [[TMP75]], align 8
 189 // CHECK-NEXT:    [[TMP76:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
 190 // CHECK-NEXT:    store ptr [[TMP67]], ptr [[TMP76]], align 8
 191 // CHECK-NEXT:    [[TMP77:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
 192 // CHECK-NEXT:    store ptr [[TMP68]], ptr [[TMP77]], align 8
 193 // CHECK-NEXT:    [[TMP78:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
 194 // CHECK-NEXT:    store ptr @.offload_maptypes.2, ptr [[TMP78]], align 8
 195 // CHECK-NEXT:    [[TMP79:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
 196 // CHECK-NEXT:    store ptr null, ptr [[TMP79]], align 8
 197 // CHECK-NEXT:    [[TMP80:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
 198 // CHECK-NEXT:    store ptr null, ptr [[TMP80]], align 8
 199 // CHECK-NEXT:    [[TMP81:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
 200 // CHECK-NEXT:    store i64 [[TMP72]], ptr [[TMP81]], align 8
 201 // CHECK-NEXT:    [[TMP82:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 9
 202 // CHECK-NEXT:    store i64 0, ptr [[TMP82]], align 8
 203 // CHECK-NEXT:    [[TMP83:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 10
 204 // CHECK-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP83]], align 4
 205 // CHECK-NEXT:    [[TMP84:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 11
 206 // CHECK-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP84]], align 4
 207 // CHECK-NEXT:    [[TMP85:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 12
 208 // CHECK-NEXT:    store i32 0, ptr [[TMP85]], align 4
 209 // CHECK-NEXT:    [[TMP86:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18kernel_within_loopPiS_ii_l13.region_id, ptr [[KERNEL_ARGS14]])
 210 // CHECK-NEXT:    [[TMP87:%.*]] = icmp ne i32 [[TMP86]], 0
 211 // CHECK-NEXT:    br i1 [[TMP87]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
 212 // CHECK:       omp_offload.failed15:
 213 // CHECK-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18kernel_within_loopPiS_ii_l13(i64 [[TMP44]], ptr [[TMP45]], ptr [[TMP46]]) #[[ATTR2]]
 214 // CHECK-NEXT:    br label [[OMP_OFFLOAD_CONT16]]
 215 // CHECK:       omp_offload.cont16:
 216 // CHECK-NEXT:    br label [[FOR_INC:%.*]]
 217 // CHECK:       for.inc:
 218 // CHECK-NEXT:    [[TMP88:%.*]] = load i32, ptr [[I]], align 4
 219 // CHECK-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP88]], 1
 220 // CHECK-NEXT:    store i32 [[INC]], ptr [[I]], align 4
 221 // CHECK-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
 222 // CHECK:       for.end:
 223 // CHECK-NEXT:    [[TMP89:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 224 // CHECK-NEXT:    [[TMP90:%.*]] = load i32, ptr [[N_ADDR]], align 4
 225 // CHECK-NEXT:    [[SUB17:%.*]] = sub nsw i32 [[TMP90]], 1
 226 // CHECK-NEXT:    [[IDXPROM:%.*]] = sext i32 [[SUB17]] to i64
 227 // CHECK-NEXT:    [[ARRAYIDX18:%.*]] = getelementptr inbounds i32, ptr [[TMP89]], i64 [[IDXPROM]]
 228 // CHECK-NEXT:    [[TMP91:%.*]] = load i32, ptr [[ARRAYIDX18]], align 4
 229 // CHECK-NEXT:    ret i32 [[TMP91]]
 230 //
 231 //
 232 // CHECK-LABEL: define internal void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18kernel_within_loopPiS_ii_l9
 233 // CHECK-SAME: (i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]]) #[[ATTR1:[0-9]+]] {
 234 // CHECK-NEXT:  entry:
 235 // CHECK-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 236 // CHECK-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 237 // CHECK-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 238 // CHECK-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
 239 // CHECK-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 240 // CHECK-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 241 // CHECK-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 242 // CHECK-NEXT:    [[TMP0:%.*]] = load i32, ptr [[N_ADDR]], align 4
 243 // CHECK-NEXT:    store i32 [[TMP0]], ptr [[N_CASTED]], align 4
 244 // CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr [[N_CASTED]], align 8
 245 // CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 246 // CHECK-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 247 // CHECK-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18kernel_within_loopPiS_ii_l9.omp_outlined, i64 [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 248 // CHECK-NEXT:    ret void
 249 //
 250 //
 251 // CHECK-LABEL: define internal void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18kernel_within_loopPiS_ii_l9.omp_outlined
 252 // CHECK-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]]) #[[ATTR1]] {
 253 // CHECK-NEXT:  entry:
 254 // CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 255 // CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 256 // CHECK-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 257 // CHECK-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 258 // CHECK-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 259 // CHECK-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 260 // CHECK-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 261 // CHECK-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 262 // CHECK-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 263 // CHECK-NEXT:    [[J:%.*]] = alloca i32, align 4
 264 // CHECK-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 265 // CHECK-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 266 // CHECK-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 267 // CHECK-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 268 // CHECK-NEXT:    [[J3:%.*]] = alloca i32, align 4
 269 // CHECK-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 270 // CHECK-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 271 // CHECK-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 272 // CHECK-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 273 // CHECK-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 274 // CHECK-NEXT:    [[TMP0:%.*]] = load i32, ptr [[N_ADDR]], align 4
 275 // CHECK-NEXT:    store i32 [[TMP0]], ptr [[DOTCAPTURE_EXPR_]], align 4
 276 // CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 277 // CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP1]], 0
 278 // CHECK-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 279 // CHECK-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 280 // CHECK-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 281 // CHECK-NEXT:    store i32 0, ptr [[J]], align 4
 282 // CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 283 // CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP2]]
 284 // CHECK-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 285 // CHECK:       omp.precond.then:
 286 // CHECK-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 287 // CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 288 // CHECK-NEXT:    store i32 [[TMP3]], ptr [[DOTOMP_UB]], align 4
 289 // CHECK-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 290 // CHECK-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 291 // CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 292 // CHECK-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP4]], align 4
 293 // CHECK-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1:[0-9]+]], i32 [[TMP5]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 294 // CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 295 // CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 296 // CHECK-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
 297 // CHECK-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 298 // CHECK:       cond.true:
 299 // CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 300 // CHECK-NEXT:    br label [[COND_END:%.*]]
 301 // CHECK:       cond.false:
 302 // CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 303 // CHECK-NEXT:    br label [[COND_END]]
 304 // CHECK:       cond.end:
 305 // CHECK-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP8]], [[COND_TRUE]] ], [ [[TMP9]], [[COND_FALSE]] ]
 306 // CHECK-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 307 // CHECK-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 308 // CHECK-NEXT:    store i32 [[TMP10]], ptr [[DOTOMP_IV]], align 4
 309 // CHECK-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 310 // CHECK:       omp.inner.for.cond:
 311 // CHECK-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 312 // CHECK-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 313 // CHECK-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP11]], [[TMP12]]
 314 // CHECK-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 315 // CHECK:       omp.inner.for.body:
 316 // CHECK-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 317 // CHECK-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP13]], 1
 318 // CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 319 // CHECK-NEXT:    store i32 [[ADD]], ptr [[J3]], align 4
 320 // CHECK-NEXT:    [[TMP14:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 321 // CHECK-NEXT:    [[TMP15:%.*]] = load i32, ptr [[J3]], align 4
 322 // CHECK-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP15]] to i64
 323 // CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP14]], i64 [[IDXPROM]]
 324 // CHECK-NEXT:    [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 325 // CHECK-NEXT:    [[TMP17:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 326 // CHECK-NEXT:    [[TMP18:%.*]] = load i32, ptr [[J3]], align 4
 327 // CHECK-NEXT:    [[IDXPROM6:%.*]] = sext i32 [[TMP18]] to i64
 328 // CHECK-NEXT:    [[ARRAYIDX7:%.*]] = getelementptr inbounds i32, ptr [[TMP17]], i64 [[IDXPROM6]]
 329 // CHECK-NEXT:    store i32 [[TMP16]], ptr [[ARRAYIDX7]], align 4
 330 // CHECK-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 331 // CHECK:       omp.body.continue:
 332 // CHECK-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 333 // CHECK:       omp.inner.for.inc:
 334 // CHECK-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 335 // CHECK-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP19]], 1
 336 // CHECK-NEXT:    store i32 [[ADD8]], ptr [[DOTOMP_IV]], align 4
 337 // CHECK-NEXT:    br label [[OMP_INNER_FOR_COND]]
 338 // CHECK:       omp.inner.for.end:
 339 // CHECK-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 340 // CHECK:       omp.loop.exit:
 341 // CHECK-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 342 // CHECK-NEXT:    [[TMP21:%.*]] = load i32, ptr [[TMP20]], align 4
 343 // CHECK-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP21]])
 344 // CHECK-NEXT:    br label [[OMP_PRECOND_END]]
 345 // CHECK:       omp.precond.end:
 346 // CHECK-NEXT:    ret void
 347 //
 348 //
 349 // CHECK-LABEL: define internal void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18kernel_within_loopPiS_ii_l13
 350 // CHECK-SAME: (i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]]) #[[ATTR1]] {
 351 // CHECK-NEXT:  entry:
 352 // CHECK-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 353 // CHECK-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 354 // CHECK-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 355 // CHECK-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
 356 // CHECK-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 357 // CHECK-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 358 // CHECK-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 359 // CHECK-NEXT:    [[TMP0:%.*]] = load i32, ptr [[N_ADDR]], align 4
 360 // CHECK-NEXT:    store i32 [[TMP0]], ptr [[N_CASTED]], align 4
 361 // CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr [[N_CASTED]], align 8
 362 // CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 363 // CHECK-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 364 // CHECK-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB2]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18kernel_within_loopPiS_ii_l13.omp_outlined, i64 [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 365 // CHECK-NEXT:    ret void
 366 //
 367 //
 368 // CHECK-LABEL: define internal void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18kernel_within_loopPiS_ii_l13.omp_outlined
 369 // CHECK-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]]) #[[ATTR1]] {
 370 // CHECK-NEXT:  entry:
 371 // CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 372 // CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 373 // CHECK-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 374 // CHECK-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 375 // CHECK-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 376 // CHECK-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 377 // CHECK-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 378 // CHECK-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 379 // CHECK-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 380 // CHECK-NEXT:    [[J:%.*]] = alloca i32, align 4
 381 // CHECK-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 382 // CHECK-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 383 // CHECK-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 384 // CHECK-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 385 // CHECK-NEXT:    [[J3:%.*]] = alloca i32, align 4
 386 // CHECK-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
 387 // CHECK-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 388 // CHECK-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 389 // CHECK-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 390 // CHECK-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 391 // CHECK-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 392 // CHECK-NEXT:    [[TMP0:%.*]] = load i32, ptr [[N_ADDR]], align 4
 393 // CHECK-NEXT:    store i32 [[TMP0]], ptr [[DOTCAPTURE_EXPR_]], align 4
 394 // CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 395 // CHECK-NEXT:    [[SUB:%.*]] = sub i32 [[TMP1]], -2
 396 // CHECK-NEXT:    [[DIV:%.*]] = udiv i32 [[SUB]], 3
 397 // CHECK-NEXT:    [[SUB2:%.*]] = sub i32 [[DIV]], 1
 398 // CHECK-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 399 // CHECK-NEXT:    store i32 0, ptr [[J]], align 4
 400 // CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 401 // CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP2]]
 402 // CHECK-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 403 // CHECK:       omp.precond.then:
 404 // CHECK-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 405 // CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 406 // CHECK-NEXT:    store i32 [[TMP3]], ptr [[DOTOMP_COMB_UB]], align 4
 407 // CHECK-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 408 // CHECK-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 409 // CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 410 // CHECK-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP4]], align 4
 411 // CHECK-NEXT:    call void @__kmpc_for_static_init_4u(ptr @[[GLOB3:[0-9]+]], i32 [[TMP5]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 412 // CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 413 // CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 414 // CHECK-NEXT:    [[CMP4:%.*]] = icmp ugt i32 [[TMP6]], [[TMP7]]
 415 // CHECK-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 416 // CHECK:       cond.true:
 417 // CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 418 // CHECK-NEXT:    br label [[COND_END:%.*]]
 419 // CHECK:       cond.false:
 420 // CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 421 // CHECK-NEXT:    br label [[COND_END]]
 422 // CHECK:       cond.end:
 423 // CHECK-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP8]], [[COND_TRUE]] ], [ [[TMP9]], [[COND_FALSE]] ]
 424 // CHECK-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 425 // CHECK-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 426 // CHECK-NEXT:    store i32 [[TMP10]], ptr [[DOTOMP_IV]], align 4
 427 // CHECK-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 428 // CHECK:       omp.inner.for.cond:
 429 // CHECK-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 430 // CHECK-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 431 // CHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP12]], 1
 432 // CHECK-NEXT:    [[CMP5:%.*]] = icmp ult i32 [[TMP11]], [[ADD]]
 433 // CHECK-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 434 // CHECK:       omp.inner.for.body:
 435 // CHECK-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 436 // CHECK-NEXT:    [[TMP14:%.*]] = zext i32 [[TMP13]] to i64
 437 // CHECK-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 438 // CHECK-NEXT:    [[TMP16:%.*]] = zext i32 [[TMP15]] to i64
 439 // CHECK-NEXT:    [[TMP17:%.*]] = load i32, ptr [[N_ADDR]], align 4
 440 // CHECK-NEXT:    store i32 [[TMP17]], ptr [[N_CASTED]], align 4
 441 // CHECK-NEXT:    [[TMP18:%.*]] = load i64, ptr [[N_CASTED]], align 8
 442 // CHECK-NEXT:    [[TMP19:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 443 // CHECK-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 444 // CHECK-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18kernel_within_loopPiS_ii_l13.omp_outlined.omp_outlined, i64 [[TMP14]], i64 [[TMP16]], i64 [[TMP18]], ptr [[TMP19]], ptr [[TMP20]])
 445 // CHECK-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 446 // CHECK:       omp.inner.for.inc:
 447 // CHECK-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 448 // CHECK-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 449 // CHECK-NEXT:    [[ADD6:%.*]] = add i32 [[TMP21]], [[TMP22]]
 450 // CHECK-NEXT:    store i32 [[ADD6]], ptr [[DOTOMP_IV]], align 4
 451 // CHECK-NEXT:    br label [[OMP_INNER_FOR_COND]]
 452 // CHECK:       omp.inner.for.end:
 453 // CHECK-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 454 // CHECK:       omp.loop.exit:
 455 // CHECK-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 456 // CHECK-NEXT:    [[TMP24:%.*]] = load i32, ptr [[TMP23]], align 4
 457 // CHECK-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB3]], i32 [[TMP24]])
 458 // CHECK-NEXT:    br label [[OMP_PRECOND_END]]
 459 // CHECK:       omp.precond.end:
 460 // CHECK-NEXT:    ret void
 461 //
 462 //
 463 // CHECK-LABEL: define internal void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z18kernel_within_loopPiS_ii_l13.omp_outlined.omp_outlined
 464 // CHECK-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]]) #[[ATTR1]] {
 465 // CHECK-NEXT:  entry:
 466 // CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 467 // CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 468 // CHECK-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
 469 // CHECK-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
 470 // CHECK-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 471 // CHECK-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 472 // CHECK-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 473 // CHECK-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 474 // CHECK-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 475 // CHECK-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 476 // CHECK-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 477 // CHECK-NEXT:    [[J:%.*]] = alloca i32, align 4
 478 // CHECK-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 479 // CHECK-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 480 // CHECK-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 481 // CHECK-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 482 // CHECK-NEXT:    [[J4:%.*]] = alloca i32, align 4
 483 // CHECK-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 484 // CHECK-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 485 // CHECK-NEXT:    store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 486 // CHECK-NEXT:    store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 487 // CHECK-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 488 // CHECK-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 489 // CHECK-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 490 // CHECK-NEXT:    [[TMP0:%.*]] = load i32, ptr [[N_ADDR]], align 4
 491 // CHECK-NEXT:    store i32 [[TMP0]], ptr [[DOTCAPTURE_EXPR_]], align 4
 492 // CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 493 // CHECK-NEXT:    [[SUB:%.*]] = sub i32 [[TMP1]], -2
 494 // CHECK-NEXT:    [[DIV:%.*]] = udiv i32 [[SUB]], 3
 495 // CHECK-NEXT:    [[SUB2:%.*]] = sub i32 [[DIV]], 1
 496 // CHECK-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 497 // CHECK-NEXT:    store i32 0, ptr [[J]], align 4
 498 // CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 499 // CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP2]]
 500 // CHECK-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 501 // CHECK:       omp.precond.then:
 502 // CHECK-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 503 // CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 504 // CHECK-NEXT:    store i32 [[TMP3]], ptr [[DOTOMP_UB]], align 4
 505 // CHECK-NEXT:    [[TMP4:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 506 // CHECK-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP4]] to i32
 507 // CHECK-NEXT:    [[TMP5:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 508 // CHECK-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP5]] to i32
 509 // CHECK-NEXT:    store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4
 510 // CHECK-NEXT:    store i32 [[CONV3]], ptr [[DOTOMP_UB]], align 4
 511 // CHECK-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 512 // CHECK-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 513 // CHECK-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 514 // CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[TMP6]], align 4
 515 // CHECK-NEXT:    call void @__kmpc_for_static_init_4u(ptr @[[GLOB1]], i32 [[TMP7]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 516 // CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 517 // CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 518 // CHECK-NEXT:    [[CMP5:%.*]] = icmp ugt i32 [[TMP8]], [[TMP9]]
 519 // CHECK-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 520 // CHECK:       cond.true:
 521 // CHECK-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 522 // CHECK-NEXT:    br label [[COND_END:%.*]]
 523 // CHECK:       cond.false:
 524 // CHECK-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 525 // CHECK-NEXT:    br label [[COND_END]]
 526 // CHECK:       cond.end:
 527 // CHECK-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP10]], [[COND_TRUE]] ], [ [[TMP11]], [[COND_FALSE]] ]
 528 // CHECK-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 529 // CHECK-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 530 // CHECK-NEXT:    store i32 [[TMP12]], ptr [[DOTOMP_IV]], align 4
 531 // CHECK-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 532 // CHECK:       omp.inner.for.cond:
 533 // CHECK-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 534 // CHECK-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 535 // CHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP14]], 1
 536 // CHECK-NEXT:    [[CMP6:%.*]] = icmp ult i32 [[TMP13]], [[ADD]]
 537 // CHECK-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 538 // CHECK:       omp.inner.for.body:
 539 // CHECK-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 540 // CHECK-NEXT:    [[MUL:%.*]] = mul i32 [[TMP15]], 3
 541 // CHECK-NEXT:    [[ADD7:%.*]] = add i32 0, [[MUL]]
 542 // CHECK-NEXT:    store i32 [[ADD7]], ptr [[J4]], align 4
 543 // CHECK-NEXT:    [[TMP16:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 544 // CHECK-NEXT:    [[TMP17:%.*]] = load i32, ptr [[J4]], align 4
 545 // CHECK-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP17]] to i64
 546 // CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP16]], i64 [[IDXPROM]]
 547 // CHECK-NEXT:    [[TMP18:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 548 // CHECK-NEXT:    [[MUL8:%.*]] = mul nsw i32 [[TMP18]], 2
 549 // CHECK-NEXT:    [[TMP19:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 550 // CHECK-NEXT:    [[TMP20:%.*]] = load i32, ptr [[J4]], align 4
 551 // CHECK-NEXT:    [[IDXPROM9:%.*]] = sext i32 [[TMP20]] to i64
 552 // CHECK-NEXT:    [[ARRAYIDX10:%.*]] = getelementptr inbounds i32, ptr [[TMP19]], i64 [[IDXPROM9]]
 553 // CHECK-NEXT:    store i32 [[MUL8]], ptr [[ARRAYIDX10]], align 4
 554 // CHECK-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 555 // CHECK:       omp.body.continue:
 556 // CHECK-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 557 // CHECK:       omp.inner.for.inc:
 558 // CHECK-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 559 // CHECK-NEXT:    [[ADD11:%.*]] = add i32 [[TMP21]], 1
 560 // CHECK-NEXT:    store i32 [[ADD11]], ptr [[DOTOMP_IV]], align 4
 561 // CHECK-NEXT:    br label [[OMP_INNER_FOR_COND]]
 562 // CHECK:       omp.inner.for.end:
 563 // CHECK-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 564 // CHECK:       omp.loop.exit:
 565 // CHECK-NEXT:    [[TMP22:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 566 // CHECK-NEXT:    [[TMP23:%.*]] = load i32, ptr [[TMP22]], align 4
 567 // CHECK-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP23]])
 568 // CHECK-NEXT:    br label [[OMP_PRECOND_END]]
 569 // CHECK:       omp.precond.end:
 570 // CHECK-NEXT:    ret void
 571 //