clang/test/OpenMP/target_teams_generic_loop_reduction_codegen.cpp

   1 // 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 _
   2 // RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK1
   3 // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
   4 // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK1
   5 // RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK3
   6 // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
   7 // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK3
   8
   9 // RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK5
  10 // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
  11 // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++  -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK5
  12
  13 // RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
  14 // RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
  15 // RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
  16 // RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
  17 // RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
  18 // RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
  19
  20 // RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
  21 // RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
  22 // RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++  -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
  23
  24 // expected-no-diagnostics
  25 #ifndef HEADER
  26 #define HEADER
  27
  28 template <typename T>
  29 T tmain() {
  30   T t_var = T();
  31   T vec[] = {1, 2};
  32 #pragma omp target teams distribute parallel for reduction(+: t_var)
  33   for (int i = 0; i < 2; ++i) {
  34     t_var += (T) i;
  35   }
  36   return T();
  37 }
  38
  39 int main() {
  40   static int sivar;
  41 #ifdef LAMBDA
  42
  43   [&]() {
  44 #pragma omp target teams distribute parallel for reduction(+: sivar)
  45   for (int i = 0; i < 2; ++i) {
  46
  47     // Skip global and bound tid vars
  48
  49
  50
  51     // Skip global and bound tid vars, and prev lb and ub vars
  52     // skip loop vars
  53
  54
  55     sivar += i;
  56
  57     [&]() {
  58
  59       sivar += 4;
  60
  61     }();
  62   }
  63   }();
  64   return 0;
  65 #else
  66 #pragma omp target teams distribute parallel for reduction(+: sivar)
  67   for (int i = 0; i < 2; ++i) {
  68     sivar += i;
  69   }
  70   return tmain<int>();
  71 #endif
  72 }
  73
  74
  75
  76
  77 // Skip global and bound tid vars
  78
  79
  80 // Skip global and bound tid vars, and prev lb and ub
  81 // skip loop vars
  82
  83
  84
  85
  86 // Skip global and bound tid vars
  87
  88
  89 // Skip global and bound tid vars, and prev lb and ub vars
  90 // skip loop vars
  91
  92 #endif
  93 // CHECK1-LABEL: define {{[^@]+}}@main
  94 // CHECK1-SAME: () #[[ATTR0:[0-9]+]] {
  95 // CHECK1-NEXT:  entry:
  96 // CHECK1-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
  97 // CHECK1-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 8
  98 // CHECK1-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 8
  99 // CHECK1-NEXT:    [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 8
 100 // CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 101 // CHECK1-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
 102 // CHECK1-NEXT:    store i32 0, ptr [[RETVAL]], align 4
 103 // CHECK1-NEXT:    [[TMP0:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
 104 // CHECK1-NEXT:    store ptr @_ZZ4mainE5sivar, ptr [[TMP0]], align 8
 105 // CHECK1-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
 106 // CHECK1-NEXT:    store ptr @_ZZ4mainE5sivar, ptr [[TMP1]], align 8
 107 // CHECK1-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
 108 // CHECK1-NEXT:    store ptr null, ptr [[TMP2]], align 8
 109 // CHECK1-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
 110 // CHECK1-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
 111 // CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
 112 // CHECK1-NEXT:    store i32 2, ptr [[TMP5]], align 4
 113 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
 114 // CHECK1-NEXT:    store i32 1, ptr [[TMP6]], align 4
 115 // CHECK1-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
 116 // CHECK1-NEXT:    store ptr [[TMP3]], ptr [[TMP7]], align 8
 117 // CHECK1-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
 118 // CHECK1-NEXT:    store ptr [[TMP4]], ptr [[TMP8]], align 8
 119 // CHECK1-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
 120 // CHECK1-NEXT:    store ptr @.offload_sizes, ptr [[TMP9]], align 8
 121 // CHECK1-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
 122 // CHECK1-NEXT:    store ptr @.offload_maptypes, ptr [[TMP10]], align 8
 123 // CHECK1-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
 124 // CHECK1-NEXT:    store ptr null, ptr [[TMP11]], align 8
 125 // CHECK1-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
 126 // CHECK1-NEXT:    store ptr null, ptr [[TMP12]], align 8
 127 // CHECK1-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
 128 // CHECK1-NEXT:    store i64 2, ptr [[TMP13]], align 8
 129 // CHECK1-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
 130 // CHECK1-NEXT:    store i64 0, ptr [[TMP14]], align 8
 131 // CHECK1-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
 132 // CHECK1-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
 133 // CHECK1-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
 134 // CHECK1-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
 135 // CHECK1-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
 136 // CHECK1-NEXT:    store i32 0, ptr [[TMP17]], align 4
 137 // CHECK1-NEXT:    [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.region_id, ptr [[KERNEL_ARGS]])
 138 // CHECK1-NEXT:    [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
 139 // CHECK1-NEXT:    br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
 140 // CHECK1:       omp_offload.failed:
 141 // CHECK1-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66(ptr @_ZZ4mainE5sivar) #[[ATTR2:[0-9]+]]
 142 // CHECK1-NEXT:    br label [[OMP_OFFLOAD_CONT]]
 143 // CHECK1:       omp_offload.cont:
 144 // CHECK1-NEXT:    [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_v()
 145 // CHECK1-NEXT:    ret i32 [[CALL]]
 146 //
 147 //
 148 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66
 149 // CHECK1-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[SIVAR:%.*]]) #[[ATTR1:[0-9]+]] {
 150 // CHECK1-NEXT:  entry:
 151 // CHECK1-NEXT:    [[SIVAR_ADDR:%.*]] = alloca ptr, align 8
 152 // CHECK1-NEXT:    store ptr [[SIVAR]], ptr [[SIVAR_ADDR]], align 8
 153 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[SIVAR_ADDR]], align 8
 154 // CHECK1-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB4]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.omp_outlined, ptr [[TMP0]])
 155 // CHECK1-NEXT:    ret void
 156 //
 157 //
 158 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.omp_outlined
 159 // CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[SIVAR:%.*]]) #[[ATTR1]] {
 160 // CHECK1-NEXT:  entry:
 161 // CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 162 // CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 163 // CHECK1-NEXT:    [[SIVAR_ADDR:%.*]] = alloca ptr, align 8
 164 // CHECK1-NEXT:    [[SIVAR1:%.*]] = alloca i32, align 4
 165 // CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 166 // CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 167 // CHECK1-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 168 // CHECK1-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 169 // CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 170 // CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 171 // CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
 172 // CHECK1-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x ptr], align 8
 173 // CHECK1-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 174 // CHECK1-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 175 // CHECK1-NEXT:    store ptr [[SIVAR]], ptr [[SIVAR_ADDR]], align 8
 176 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[SIVAR_ADDR]], align 8
 177 // CHECK1-NEXT:    store i32 0, ptr [[SIVAR1]], align 4
 178 // CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 179 // CHECK1-NEXT:    store i32 1, ptr [[DOTOMP_COMB_UB]], align 4
 180 // CHECK1-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 181 // CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 182 // CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 183 // CHECK1-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
 184 // CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1:[0-9]+]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 185 // CHECK1-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 186 // CHECK1-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 1
 187 // CHECK1-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 188 // CHECK1:       cond.true:
 189 // CHECK1-NEXT:    br label [[COND_END:%.*]]
 190 // CHECK1:       cond.false:
 191 // CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 192 // CHECK1-NEXT:    br label [[COND_END]]
 193 // CHECK1:       cond.end:
 194 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
 195 // CHECK1-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 196 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 197 // CHECK1-NEXT:    store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4
 198 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 199 // CHECK1:       omp.inner.for.cond:
 200 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 201 // CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 202 // CHECK1-NEXT:    [[CMP2:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]]
 203 // CHECK1-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 204 // CHECK1:       omp.inner.for.body:
 205 // CHECK1-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 206 // CHECK1-NEXT:    [[TMP9:%.*]] = zext i32 [[TMP8]] to i64
 207 // CHECK1-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 208 // CHECK1-NEXT:    [[TMP11:%.*]] = zext i32 [[TMP10]] to i64
 209 // CHECK1-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB4]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.omp_outlined.omp_outlined, i64 [[TMP9]], i64 [[TMP11]], ptr [[SIVAR1]])
 210 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 211 // CHECK1:       omp.inner.for.inc:
 212 // CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 213 // CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 214 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP12]], [[TMP13]]
 215 // CHECK1-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 216 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
 217 // CHECK1:       omp.inner.for.end:
 218 // CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 219 // CHECK1:       omp.loop.exit:
 220 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]])
 221 // CHECK1-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0
 222 // CHECK1-NEXT:    store ptr [[SIVAR1]], ptr [[TMP14]], align 8
 223 // CHECK1-NEXT:    [[TMP15:%.*]] = call i32 @__kmpc_reduce_nowait(ptr @[[GLOB3:[0-9]+]], i32 [[TMP2]], i32 1, i64 8, ptr [[DOTOMP_REDUCTION_RED_LIST]], ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.omp_outlined.omp.reduction.reduction_func, ptr @.gomp_critical_user_.reduction.var)
 224 // CHECK1-NEXT:    switch i32 [[TMP15]], label [[DOTOMP_REDUCTION_DEFAULT:%.*]] [
 225 // CHECK1-NEXT:    i32 1, label [[DOTOMP_REDUCTION_CASE1:%.*]]
 226 // CHECK1-NEXT:    i32 2, label [[DOTOMP_REDUCTION_CASE2:%.*]]
 227 // CHECK1-NEXT:    ]
 228 // CHECK1:       .omp.reduction.case1:
 229 // CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[TMP0]], align 4
 230 // CHECK1-NEXT:    [[TMP17:%.*]] = load i32, ptr [[SIVAR1]], align 4
 231 // CHECK1-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP16]], [[TMP17]]
 232 // CHECK1-NEXT:    store i32 [[ADD3]], ptr [[TMP0]], align 4
 233 // CHECK1-NEXT:    call void @__kmpc_end_reduce_nowait(ptr @[[GLOB3]], i32 [[TMP2]], ptr @.gomp_critical_user_.reduction.var)
 234 // CHECK1-NEXT:    br label [[DOTOMP_REDUCTION_DEFAULT]]
 235 // CHECK1:       .omp.reduction.case2:
 236 // CHECK1-NEXT:    [[TMP18:%.*]] = load i32, ptr [[SIVAR1]], align 4
 237 // CHECK1-NEXT:    [[TMP19:%.*]] = atomicrmw add ptr [[TMP0]], i32 [[TMP18]] monotonic, align 4
 238 // CHECK1-NEXT:    br label [[DOTOMP_REDUCTION_DEFAULT]]
 239 // CHECK1:       .omp.reduction.default:
 240 // CHECK1-NEXT:    ret void
 241 //
 242 //
 243 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.omp_outlined.omp_outlined
 244 // CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[SIVAR:%.*]]) #[[ATTR1]] {
 245 // CHECK1-NEXT:  entry:
 246 // CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 247 // CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 248 // CHECK1-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
 249 // CHECK1-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
 250 // CHECK1-NEXT:    [[SIVAR_ADDR:%.*]] = alloca ptr, align 8
 251 // CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 252 // CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 253 // CHECK1-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 254 // CHECK1-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 255 // CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 256 // CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 257 // CHECK1-NEXT:    [[SIVAR2:%.*]] = alloca i32, align 4
 258 // CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
 259 // CHECK1-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x ptr], align 8
 260 // CHECK1-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 261 // CHECK1-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 262 // CHECK1-NEXT:    store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 263 // CHECK1-NEXT:    store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 264 // CHECK1-NEXT:    store ptr [[SIVAR]], ptr [[SIVAR_ADDR]], align 8
 265 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[SIVAR_ADDR]], align 8
 266 // CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 267 // CHECK1-NEXT:    store i32 1, ptr [[DOTOMP_UB]], align 4
 268 // CHECK1-NEXT:    [[TMP1:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 269 // CHECK1-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP1]] to i32
 270 // CHECK1-NEXT:    [[TMP2:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 271 // CHECK1-NEXT:    [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32
 272 // CHECK1-NEXT:    store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4
 273 // CHECK1-NEXT:    store i32 [[CONV1]], ptr [[DOTOMP_UB]], align 4
 274 // CHECK1-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 275 // CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 276 // CHECK1-NEXT:    store i32 0, ptr [[SIVAR2]], align 4
 277 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 278 // CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 4
 279 // CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2:[0-9]+]], i32 [[TMP4]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 280 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 281 // CHECK1-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP5]], 1
 282 // CHECK1-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 283 // CHECK1:       cond.true:
 284 // CHECK1-NEXT:    br label [[COND_END:%.*]]
 285 // CHECK1:       cond.false:
 286 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 287 // CHECK1-NEXT:    br label [[COND_END]]
 288 // CHECK1:       cond.end:
 289 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP6]], [[COND_FALSE]] ]
 290 // CHECK1-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 291 // CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 292 // CHECK1-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_IV]], align 4
 293 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 294 // CHECK1:       omp.inner.for.cond:
 295 // CHECK1-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 296 // CHECK1-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 297 // CHECK1-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]]
 298 // CHECK1-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 299 // CHECK1:       omp.inner.for.body:
 300 // CHECK1-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 301 // CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1
 302 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 303 // CHECK1-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
 304 // CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[I]], align 4
 305 // CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[SIVAR2]], align 4
 306 // CHECK1-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP12]], [[TMP11]]
 307 // CHECK1-NEXT:    store i32 [[ADD4]], ptr [[SIVAR2]], align 4
 308 // CHECK1-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 309 // CHECK1:       omp.body.continue:
 310 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 311 // CHECK1:       omp.inner.for.inc:
 312 // CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 313 // CHECK1-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP13]], 1
 314 // CHECK1-NEXT:    store i32 [[ADD5]], ptr [[DOTOMP_IV]], align 4
 315 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
 316 // CHECK1:       omp.inner.for.end:
 317 // CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 318 // CHECK1:       omp.loop.exit:
 319 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP4]])
 320 // CHECK1-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0
 321 // CHECK1-NEXT:    store ptr [[SIVAR2]], ptr [[TMP14]], align 8
 322 // CHECK1-NEXT:    [[TMP15:%.*]] = call i32 @__kmpc_reduce_nowait(ptr @[[GLOB3]], i32 [[TMP4]], i32 1, i64 8, ptr [[DOTOMP_REDUCTION_RED_LIST]], ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.omp_outlined.omp_outlined.omp.reduction.reduction_func, ptr @.gomp_critical_user_.reduction.var)
 323 // CHECK1-NEXT:    switch i32 [[TMP15]], label [[DOTOMP_REDUCTION_DEFAULT:%.*]] [
 324 // CHECK1-NEXT:    i32 1, label [[DOTOMP_REDUCTION_CASE1:%.*]]
 325 // CHECK1-NEXT:    i32 2, label [[DOTOMP_REDUCTION_CASE2:%.*]]
 326 // CHECK1-NEXT:    ]
 327 // CHECK1:       .omp.reduction.case1:
 328 // CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[TMP0]], align 4
 329 // CHECK1-NEXT:    [[TMP17:%.*]] = load i32, ptr [[SIVAR2]], align 4
 330 // CHECK1-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP16]], [[TMP17]]
 331 // CHECK1-NEXT:    store i32 [[ADD6]], ptr [[TMP0]], align 4
 332 // CHECK1-NEXT:    call void @__kmpc_end_reduce_nowait(ptr @[[GLOB3]], i32 [[TMP4]], ptr @.gomp_critical_user_.reduction.var)
 333 // CHECK1-NEXT:    br label [[DOTOMP_REDUCTION_DEFAULT]]
 334 // CHECK1:       .omp.reduction.case2:
 335 // CHECK1-NEXT:    [[TMP18:%.*]] = load i32, ptr [[SIVAR2]], align 4
 336 // CHECK1-NEXT:    [[TMP19:%.*]] = atomicrmw add ptr [[TMP0]], i32 [[TMP18]] monotonic, align 4
 337 // CHECK1-NEXT:    br label [[DOTOMP_REDUCTION_DEFAULT]]
 338 // CHECK1:       .omp.reduction.default:
 339 // CHECK1-NEXT:    ret void
 340 //
 341 //
 342 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.omp_outlined.omp_outlined.omp.reduction.reduction_func
 343 // CHECK1-SAME: (ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]]) #[[ATTR3:[0-9]+]] {
 344 // CHECK1-NEXT:  entry:
 345 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 346 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 347 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 348 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 349 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 350 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 351 // CHECK1-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP3]], i64 0, i64 0
 352 // CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[TMP4]], align 8
 353 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP2]], i64 0, i64 0
 354 // CHECK1-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[TMP6]], align 8
 355 // CHECK1-NEXT:    [[TMP8:%.*]] = load i32, ptr [[TMP7]], align 4
 356 // CHECK1-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP5]], align 4
 357 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP8]], [[TMP9]]
 358 // CHECK1-NEXT:    store i32 [[ADD]], ptr [[TMP7]], align 4
 359 // CHECK1-NEXT:    ret void
 360 //
 361 //
 362 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.omp_outlined.omp.reduction.reduction_func
 363 // CHECK1-SAME: (ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]]) #[[ATTR3]] {
 364 // CHECK1-NEXT:  entry:
 365 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 366 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 367 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 368 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 369 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 370 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 371 // CHECK1-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP3]], i64 0, i64 0
 372 // CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[TMP4]], align 8
 373 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP2]], i64 0, i64 0
 374 // CHECK1-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[TMP6]], align 8
 375 // CHECK1-NEXT:    [[TMP8:%.*]] = load i32, ptr [[TMP7]], align 4
 376 // CHECK1-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP5]], align 4
 377 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP8]], [[TMP9]]
 378 // CHECK1-NEXT:    store i32 [[ADD]], ptr [[TMP7]], align 4
 379 // CHECK1-NEXT:    ret void
 380 //
 381 //
 382 // CHECK1-LABEL: define {{[^@]+}}@_Z5tmainIiET_v
 383 // CHECK1-SAME: () #[[ATTR5:[0-9]+]] comdat {
 384 // CHECK1-NEXT:  entry:
 385 // CHECK1-NEXT:    [[T_VAR:%.*]] = alloca i32, align 4
 386 // CHECK1-NEXT:    [[VEC:%.*]] = alloca [2 x i32], align 4
 387 // CHECK1-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 8
 388 // CHECK1-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 8
 389 // CHECK1-NEXT:    [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 8
 390 // CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 391 // CHECK1-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
 392 // CHECK1-NEXT:    store i32 0, ptr [[T_VAR]], align 4
 393 // CHECK1-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 4 [[VEC]], ptr align 4 @__const._Z5tmainIiET_v.vec, i64 8, i1 false)
 394 // CHECK1-NEXT:    [[TMP0:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
 395 // CHECK1-NEXT:    store ptr [[T_VAR]], ptr [[TMP0]], align 8
 396 // CHECK1-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
 397 // CHECK1-NEXT:    store ptr [[T_VAR]], ptr [[TMP1]], align 8
 398 // CHECK1-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
 399 // CHECK1-NEXT:    store ptr null, ptr [[TMP2]], align 8
 400 // CHECK1-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
 401 // CHECK1-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
 402 // CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
 403 // CHECK1-NEXT:    store i32 2, ptr [[TMP5]], align 4
 404 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
 405 // CHECK1-NEXT:    store i32 1, ptr [[TMP6]], align 4
 406 // CHECK1-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
 407 // CHECK1-NEXT:    store ptr [[TMP3]], ptr [[TMP7]], align 8
 408 // CHECK1-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
 409 // CHECK1-NEXT:    store ptr [[TMP4]], ptr [[TMP8]], align 8
 410 // CHECK1-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
 411 // CHECK1-NEXT:    store ptr @.offload_sizes.1, ptr [[TMP9]], align 8
 412 // CHECK1-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
 413 // CHECK1-NEXT:    store ptr @.offload_maptypes.2, ptr [[TMP10]], align 8
 414 // CHECK1-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
 415 // CHECK1-NEXT:    store ptr null, ptr [[TMP11]], align 8
 416 // CHECK1-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
 417 // CHECK1-NEXT:    store ptr null, ptr [[TMP12]], align 8
 418 // CHECK1-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
 419 // CHECK1-NEXT:    store i64 2, ptr [[TMP13]], align 8
 420 // CHECK1-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
 421 // CHECK1-NEXT:    store i64 0, ptr [[TMP14]], align 8
 422 // CHECK1-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
 423 // CHECK1-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
 424 // CHECK1-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
 425 // CHECK1-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
 426 // CHECK1-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
 427 // CHECK1-NEXT:    store i32 0, ptr [[TMP17]], align 4
 428 // CHECK1-NEXT:    [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
 429 // CHECK1-NEXT:    [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
 430 // CHECK1-NEXT:    br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
 431 // CHECK1:       omp_offload.failed:
 432 // CHECK1-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32(ptr [[T_VAR]]) #[[ATTR2]]
 433 // CHECK1-NEXT:    br label [[OMP_OFFLOAD_CONT]]
 434 // CHECK1:       omp_offload.cont:
 435 // CHECK1-NEXT:    ret i32 0
 436 //
 437 //
 438 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32
 439 // CHECK1-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[T_VAR:%.*]]) #[[ATTR1]] {
 440 // CHECK1-NEXT:  entry:
 441 // CHECK1-NEXT:    [[T_VAR_ADDR:%.*]] = alloca ptr, align 8
 442 // CHECK1-NEXT:    store ptr [[T_VAR]], ptr [[T_VAR_ADDR]], align 8
 443 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[T_VAR_ADDR]], align 8
 444 // CHECK1-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB4]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.omp_outlined, ptr [[TMP0]])
 445 // CHECK1-NEXT:    ret void
 446 //
 447 //
 448 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.omp_outlined
 449 // CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[T_VAR:%.*]]) #[[ATTR1]] {
 450 // CHECK1-NEXT:  entry:
 451 // CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 452 // CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 453 // CHECK1-NEXT:    [[T_VAR_ADDR:%.*]] = alloca ptr, align 8
 454 // CHECK1-NEXT:    [[T_VAR1:%.*]] = alloca i32, align 4
 455 // CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 456 // CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 457 // CHECK1-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 458 // CHECK1-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 459 // CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 460 // CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 461 // CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
 462 // CHECK1-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x ptr], align 8
 463 // CHECK1-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 464 // CHECK1-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 465 // CHECK1-NEXT:    store ptr [[T_VAR]], ptr [[T_VAR_ADDR]], align 8
 466 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[T_VAR_ADDR]], align 8
 467 // CHECK1-NEXT:    store i32 0, ptr [[T_VAR1]], align 4
 468 // CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 469 // CHECK1-NEXT:    store i32 1, ptr [[DOTOMP_COMB_UB]], align 4
 470 // CHECK1-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 471 // CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 472 // CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 473 // CHECK1-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
 474 // CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 475 // CHECK1-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 476 // CHECK1-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 1
 477 // CHECK1-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 478 // CHECK1:       cond.true:
 479 // CHECK1-NEXT:    br label [[COND_END:%.*]]
 480 // CHECK1:       cond.false:
 481 // CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 482 // CHECK1-NEXT:    br label [[COND_END]]
 483 // CHECK1:       cond.end:
 484 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
 485 // CHECK1-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 486 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 487 // CHECK1-NEXT:    store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4
 488 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 489 // CHECK1:       omp.inner.for.cond:
 490 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 491 // CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 492 // CHECK1-NEXT:    [[CMP2:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]]
 493 // CHECK1-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 494 // CHECK1:       omp.inner.for.body:
 495 // CHECK1-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 496 // CHECK1-NEXT:    [[TMP9:%.*]] = zext i32 [[TMP8]] to i64
 497 // CHECK1-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 498 // CHECK1-NEXT:    [[TMP11:%.*]] = zext i32 [[TMP10]] to i64
 499 // CHECK1-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB4]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.omp_outlined.omp_outlined, i64 [[TMP9]], i64 [[TMP11]], ptr [[T_VAR1]])
 500 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 501 // CHECK1:       omp.inner.for.inc:
 502 // CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 503 // CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 504 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP12]], [[TMP13]]
 505 // CHECK1-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 506 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
 507 // CHECK1:       omp.inner.for.end:
 508 // CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 509 // CHECK1:       omp.loop.exit:
 510 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]])
 511 // CHECK1-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0
 512 // CHECK1-NEXT:    store ptr [[T_VAR1]], ptr [[TMP14]], align 8
 513 // CHECK1-NEXT:    [[TMP15:%.*]] = call i32 @__kmpc_reduce_nowait(ptr @[[GLOB3]], i32 [[TMP2]], i32 1, i64 8, ptr [[DOTOMP_REDUCTION_RED_LIST]], ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.omp_outlined.omp.reduction.reduction_func, ptr @.gomp_critical_user_.reduction.var)
 514 // CHECK1-NEXT:    switch i32 [[TMP15]], label [[DOTOMP_REDUCTION_DEFAULT:%.*]] [
 515 // CHECK1-NEXT:    i32 1, label [[DOTOMP_REDUCTION_CASE1:%.*]]
 516 // CHECK1-NEXT:    i32 2, label [[DOTOMP_REDUCTION_CASE2:%.*]]
 517 // CHECK1-NEXT:    ]
 518 // CHECK1:       .omp.reduction.case1:
 519 // CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[TMP0]], align 4
 520 // CHECK1-NEXT:    [[TMP17:%.*]] = load i32, ptr [[T_VAR1]], align 4
 521 // CHECK1-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP16]], [[TMP17]]
 522 // CHECK1-NEXT:    store i32 [[ADD3]], ptr [[TMP0]], align 4
 523 // CHECK1-NEXT:    call void @__kmpc_end_reduce_nowait(ptr @[[GLOB3]], i32 [[TMP2]], ptr @.gomp_critical_user_.reduction.var)
 524 // CHECK1-NEXT:    br label [[DOTOMP_REDUCTION_DEFAULT]]
 525 // CHECK1:       .omp.reduction.case2:
 526 // CHECK1-NEXT:    [[TMP18:%.*]] = load i32, ptr [[T_VAR1]], align 4
 527 // CHECK1-NEXT:    [[TMP19:%.*]] = atomicrmw add ptr [[TMP0]], i32 [[TMP18]] monotonic, align 4
 528 // CHECK1-NEXT:    br label [[DOTOMP_REDUCTION_DEFAULT]]
 529 // CHECK1:       .omp.reduction.default:
 530 // CHECK1-NEXT:    ret void
 531 //
 532 //
 533 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.omp_outlined.omp_outlined
 534 // CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[T_VAR:%.*]]) #[[ATTR1]] {
 535 // CHECK1-NEXT:  entry:
 536 // CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 537 // CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 538 // CHECK1-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
 539 // CHECK1-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
 540 // CHECK1-NEXT:    [[T_VAR_ADDR:%.*]] = alloca ptr, align 8
 541 // CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 542 // CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 543 // CHECK1-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 544 // CHECK1-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 545 // CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 546 // CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 547 // CHECK1-NEXT:    [[T_VAR2:%.*]] = alloca i32, align 4
 548 // CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
 549 // CHECK1-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x ptr], align 8
 550 // CHECK1-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 551 // CHECK1-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 552 // CHECK1-NEXT:    store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 553 // CHECK1-NEXT:    store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 554 // CHECK1-NEXT:    store ptr [[T_VAR]], ptr [[T_VAR_ADDR]], align 8
 555 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[T_VAR_ADDR]], align 8
 556 // CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 557 // CHECK1-NEXT:    store i32 1, ptr [[DOTOMP_UB]], align 4
 558 // CHECK1-NEXT:    [[TMP1:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 559 // CHECK1-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP1]] to i32
 560 // CHECK1-NEXT:    [[TMP2:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 561 // CHECK1-NEXT:    [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32
 562 // CHECK1-NEXT:    store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4
 563 // CHECK1-NEXT:    store i32 [[CONV1]], ptr [[DOTOMP_UB]], align 4
 564 // CHECK1-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 565 // CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 566 // CHECK1-NEXT:    store i32 0, ptr [[T_VAR2]], align 4
 567 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 568 // CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 4
 569 // CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP4]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 570 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 571 // CHECK1-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP5]], 1
 572 // CHECK1-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 573 // CHECK1:       cond.true:
 574 // CHECK1-NEXT:    br label [[COND_END:%.*]]
 575 // CHECK1:       cond.false:
 576 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 577 // CHECK1-NEXT:    br label [[COND_END]]
 578 // CHECK1:       cond.end:
 579 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP6]], [[COND_FALSE]] ]
 580 // CHECK1-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 581 // CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 582 // CHECK1-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_IV]], align 4
 583 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 584 // CHECK1:       omp.inner.for.cond:
 585 // CHECK1-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 586 // CHECK1-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 587 // CHECK1-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]]
 588 // CHECK1-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 589 // CHECK1:       omp.inner.for.body:
 590 // CHECK1-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 591 // CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1
 592 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 593 // CHECK1-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
 594 // CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[I]], align 4
 595 // CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[T_VAR2]], align 4
 596 // CHECK1-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP12]], [[TMP11]]
 597 // CHECK1-NEXT:    store i32 [[ADD4]], ptr [[T_VAR2]], align 4
 598 // CHECK1-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 599 // CHECK1:       omp.body.continue:
 600 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 601 // CHECK1:       omp.inner.for.inc:
 602 // CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 603 // CHECK1-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP13]], 1
 604 // CHECK1-NEXT:    store i32 [[ADD5]], ptr [[DOTOMP_IV]], align 4
 605 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
 606 // CHECK1:       omp.inner.for.end:
 607 // CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 608 // CHECK1:       omp.loop.exit:
 609 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP4]])
 610 // CHECK1-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0
 611 // CHECK1-NEXT:    store ptr [[T_VAR2]], ptr [[TMP14]], align 8
 612 // CHECK1-NEXT:    [[TMP15:%.*]] = call i32 @__kmpc_reduce_nowait(ptr @[[GLOB3]], i32 [[TMP4]], i32 1, i64 8, ptr [[DOTOMP_REDUCTION_RED_LIST]], ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.omp_outlined.omp_outlined.omp.reduction.reduction_func, ptr @.gomp_critical_user_.reduction.var)
 613 // CHECK1-NEXT:    switch i32 [[TMP15]], label [[DOTOMP_REDUCTION_DEFAULT:%.*]] [
 614 // CHECK1-NEXT:    i32 1, label [[DOTOMP_REDUCTION_CASE1:%.*]]
 615 // CHECK1-NEXT:    i32 2, label [[DOTOMP_REDUCTION_CASE2:%.*]]
 616 // CHECK1-NEXT:    ]
 617 // CHECK1:       .omp.reduction.case1:
 618 // CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[TMP0]], align 4
 619 // CHECK1-NEXT:    [[TMP17:%.*]] = load i32, ptr [[T_VAR2]], align 4
 620 // CHECK1-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP16]], [[TMP17]]
 621 // CHECK1-NEXT:    store i32 [[ADD6]], ptr [[TMP0]], align 4
 622 // CHECK1-NEXT:    call void @__kmpc_end_reduce_nowait(ptr @[[GLOB3]], i32 [[TMP4]], ptr @.gomp_critical_user_.reduction.var)
 623 // CHECK1-NEXT:    br label [[DOTOMP_REDUCTION_DEFAULT]]
 624 // CHECK1:       .omp.reduction.case2:
 625 // CHECK1-NEXT:    [[TMP18:%.*]] = load i32, ptr [[T_VAR2]], align 4
 626 // CHECK1-NEXT:    [[TMP19:%.*]] = atomicrmw add ptr [[TMP0]], i32 [[TMP18]] monotonic, align 4
 627 // CHECK1-NEXT:    br label [[DOTOMP_REDUCTION_DEFAULT]]
 628 // CHECK1:       .omp.reduction.default:
 629 // CHECK1-NEXT:    ret void
 630 //
 631 //
 632 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.omp_outlined.omp_outlined.omp.reduction.reduction_func
 633 // CHECK1-SAME: (ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]]) #[[ATTR3]] {
 634 // CHECK1-NEXT:  entry:
 635 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 636 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 637 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 638 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 639 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 640 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 641 // CHECK1-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP3]], i64 0, i64 0
 642 // CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[TMP4]], align 8
 643 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP2]], i64 0, i64 0
 644 // CHECK1-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[TMP6]], align 8
 645 // CHECK1-NEXT:    [[TMP8:%.*]] = load i32, ptr [[TMP7]], align 4
 646 // CHECK1-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP5]], align 4
 647 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP8]], [[TMP9]]
 648 // CHECK1-NEXT:    store i32 [[ADD]], ptr [[TMP7]], align 4
 649 // CHECK1-NEXT:    ret void
 650 //
 651 //
 652 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.omp_outlined.omp.reduction.reduction_func
 653 // CHECK1-SAME: (ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]]) #[[ATTR3]] {
 654 // CHECK1-NEXT:  entry:
 655 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 656 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 657 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 658 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 659 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 660 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 661 // CHECK1-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP3]], i64 0, i64 0
 662 // CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[TMP4]], align 8
 663 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP2]], i64 0, i64 0
 664 // CHECK1-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[TMP6]], align 8
 665 // CHECK1-NEXT:    [[TMP8:%.*]] = load i32, ptr [[TMP7]], align 4
 666 // CHECK1-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP5]], align 4
 667 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP8]], [[TMP9]]
 668 // CHECK1-NEXT:    store i32 [[ADD]], ptr [[TMP7]], align 4
 669 // CHECK1-NEXT:    ret void
 670 //
 671 //
 672 // CHECK1-LABEL: define {{[^@]+}}@.omp_offloading.requires_reg
 673 // CHECK1-SAME: () #[[ATTR7:[0-9]+]] {
 674 // CHECK1-NEXT:  entry:
 675 // CHECK1-NEXT:    call void @__tgt_register_requires(i64 1)
 676 // CHECK1-NEXT:    ret void
 677 //
 678 //
 679 // CHECK3-LABEL: define {{[^@]+}}@main
 680 // CHECK3-SAME: () #[[ATTR0:[0-9]+]] {
 681 // CHECK3-NEXT:  entry:
 682 // CHECK3-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
 683 // CHECK3-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 4
 684 // CHECK3-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 4
 685 // CHECK3-NEXT:    [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 4
 686 // CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 687 // CHECK3-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
 688 // CHECK3-NEXT:    store i32 0, ptr [[RETVAL]], align 4
 689 // CHECK3-NEXT:    [[TMP0:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
 690 // CHECK3-NEXT:    store ptr @_ZZ4mainE5sivar, ptr [[TMP0]], align 4
 691 // CHECK3-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
 692 // CHECK3-NEXT:    store ptr @_ZZ4mainE5sivar, ptr [[TMP1]], align 4
 693 // CHECK3-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
 694 // CHECK3-NEXT:    store ptr null, ptr [[TMP2]], align 4
 695 // CHECK3-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
 696 // CHECK3-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
 697 // CHECK3-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
 698 // CHECK3-NEXT:    store i32 2, ptr [[TMP5]], align 4
 699 // CHECK3-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
 700 // CHECK3-NEXT:    store i32 1, ptr [[TMP6]], align 4
 701 // CHECK3-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
 702 // CHECK3-NEXT:    store ptr [[TMP3]], ptr [[TMP7]], align 4
 703 // CHECK3-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
 704 // CHECK3-NEXT:    store ptr [[TMP4]], ptr [[TMP8]], align 4
 705 // CHECK3-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
 706 // CHECK3-NEXT:    store ptr @.offload_sizes, ptr [[TMP9]], align 4
 707 // CHECK3-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
 708 // CHECK3-NEXT:    store ptr @.offload_maptypes, ptr [[TMP10]], align 4
 709 // CHECK3-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
 710 // CHECK3-NEXT:    store ptr null, ptr [[TMP11]], align 4
 711 // CHECK3-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
 712 // CHECK3-NEXT:    store ptr null, ptr [[TMP12]], align 4
 713 // CHECK3-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
 714 // CHECK3-NEXT:    store i64 2, ptr [[TMP13]], align 8
 715 // CHECK3-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
 716 // CHECK3-NEXT:    store i64 0, ptr [[TMP14]], align 8
 717 // CHECK3-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
 718 // CHECK3-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
 719 // CHECK3-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
 720 // CHECK3-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
 721 // CHECK3-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
 722 // CHECK3-NEXT:    store i32 0, ptr [[TMP17]], align 4
 723 // CHECK3-NEXT:    [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.region_id, ptr [[KERNEL_ARGS]])
 724 // CHECK3-NEXT:    [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
 725 // CHECK3-NEXT:    br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
 726 // CHECK3:       omp_offload.failed:
 727 // CHECK3-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66(ptr @_ZZ4mainE5sivar) #[[ATTR2:[0-9]+]]
 728 // CHECK3-NEXT:    br label [[OMP_OFFLOAD_CONT]]
 729 // CHECK3:       omp_offload.cont:
 730 // CHECK3-NEXT:    [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
 731 // CHECK3-NEXT:    ret i32 [[CALL]]
 732 //
 733 //
 734 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66
 735 // CHECK3-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[SIVAR:%.*]]) #[[ATTR1:[0-9]+]] {
 736 // CHECK3-NEXT:  entry:
 737 // CHECK3-NEXT:    [[SIVAR_ADDR:%.*]] = alloca ptr, align 4
 738 // CHECK3-NEXT:    store ptr [[SIVAR]], ptr [[SIVAR_ADDR]], align 4
 739 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[SIVAR_ADDR]], align 4
 740 // CHECK3-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB4]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.omp_outlined, ptr [[TMP0]])
 741 // CHECK3-NEXT:    ret void
 742 //
 743 //
 744 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.omp_outlined
 745 // CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[SIVAR:%.*]]) #[[ATTR1]] {
 746 // CHECK3-NEXT:  entry:
 747 // CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 748 // CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 749 // CHECK3-NEXT:    [[SIVAR_ADDR:%.*]] = alloca ptr, align 4
 750 // CHECK3-NEXT:    [[SIVAR1:%.*]] = alloca i32, align 4
 751 // CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 752 // CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 753 // CHECK3-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 754 // CHECK3-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 755 // CHECK3-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 756 // CHECK3-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 757 // CHECK3-NEXT:    [[I:%.*]] = alloca i32, align 4
 758 // CHECK3-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x ptr], align 4
 759 // CHECK3-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 760 // CHECK3-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 761 // CHECK3-NEXT:    store ptr [[SIVAR]], ptr [[SIVAR_ADDR]], align 4
 762 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[SIVAR_ADDR]], align 4
 763 // CHECK3-NEXT:    store i32 0, ptr [[SIVAR1]], align 4
 764 // CHECK3-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 765 // CHECK3-NEXT:    store i32 1, ptr [[DOTOMP_COMB_UB]], align 4
 766 // CHECK3-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 767 // CHECK3-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 768 // CHECK3-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 769 // CHECK3-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
 770 // CHECK3-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1:[0-9]+]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 771 // CHECK3-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 772 // CHECK3-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 1
 773 // CHECK3-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 774 // CHECK3:       cond.true:
 775 // CHECK3-NEXT:    br label [[COND_END:%.*]]
 776 // CHECK3:       cond.false:
 777 // CHECK3-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 778 // CHECK3-NEXT:    br label [[COND_END]]
 779 // CHECK3:       cond.end:
 780 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
 781 // CHECK3-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 782 // CHECK3-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 783 // CHECK3-NEXT:    store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4
 784 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 785 // CHECK3:       omp.inner.for.cond:
 786 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 787 // CHECK3-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 788 // CHECK3-NEXT:    [[CMP2:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]]
 789 // CHECK3-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 790 // CHECK3:       omp.inner.for.body:
 791 // CHECK3-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 792 // CHECK3-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 793 // CHECK3-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB4]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.omp_outlined.omp_outlined, i32 [[TMP8]], i32 [[TMP9]], ptr [[SIVAR1]])
 794 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 795 // CHECK3:       omp.inner.for.inc:
 796 // CHECK3-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 797 // CHECK3-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 798 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP10]], [[TMP11]]
 799 // CHECK3-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 800 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
 801 // CHECK3:       omp.inner.for.end:
 802 // CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 803 // CHECK3:       omp.loop.exit:
 804 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]])
 805 // CHECK3-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0
 806 // CHECK3-NEXT:    store ptr [[SIVAR1]], ptr [[TMP12]], align 4
 807 // CHECK3-NEXT:    [[TMP13:%.*]] = call i32 @__kmpc_reduce_nowait(ptr @[[GLOB3:[0-9]+]], i32 [[TMP2]], i32 1, i32 4, ptr [[DOTOMP_REDUCTION_RED_LIST]], ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.omp_outlined.omp.reduction.reduction_func, ptr @.gomp_critical_user_.reduction.var)
 808 // CHECK3-NEXT:    switch i32 [[TMP13]], label [[DOTOMP_REDUCTION_DEFAULT:%.*]] [
 809 // CHECK3-NEXT:    i32 1, label [[DOTOMP_REDUCTION_CASE1:%.*]]
 810 // CHECK3-NEXT:    i32 2, label [[DOTOMP_REDUCTION_CASE2:%.*]]
 811 // CHECK3-NEXT:    ]
 812 // CHECK3:       .omp.reduction.case1:
 813 // CHECK3-NEXT:    [[TMP14:%.*]] = load i32, ptr [[TMP0]], align 4
 814 // CHECK3-NEXT:    [[TMP15:%.*]] = load i32, ptr [[SIVAR1]], align 4
 815 // CHECK3-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP14]], [[TMP15]]
 816 // CHECK3-NEXT:    store i32 [[ADD3]], ptr [[TMP0]], align 4
 817 // CHECK3-NEXT:    call void @__kmpc_end_reduce_nowait(ptr @[[GLOB3]], i32 [[TMP2]], ptr @.gomp_critical_user_.reduction.var)
 818 // CHECK3-NEXT:    br label [[DOTOMP_REDUCTION_DEFAULT]]
 819 // CHECK3:       .omp.reduction.case2:
 820 // CHECK3-NEXT:    [[TMP16:%.*]] = load i32, ptr [[SIVAR1]], align 4
 821 // CHECK3-NEXT:    [[TMP17:%.*]] = atomicrmw add ptr [[TMP0]], i32 [[TMP16]] monotonic, align 4
 822 // CHECK3-NEXT:    br label [[DOTOMP_REDUCTION_DEFAULT]]
 823 // CHECK3:       .omp.reduction.default:
 824 // CHECK3-NEXT:    ret void
 825 //
 826 //
 827 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.omp_outlined.omp_outlined
 828 // CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[SIVAR:%.*]]) #[[ATTR1]] {
 829 // CHECK3-NEXT:  entry:
 830 // CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 831 // CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 832 // CHECK3-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
 833 // CHECK3-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
 834 // CHECK3-NEXT:    [[SIVAR_ADDR:%.*]] = alloca ptr, align 4
 835 // CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 836 // CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 837 // CHECK3-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 838 // CHECK3-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 839 // CHECK3-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 840 // CHECK3-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 841 // CHECK3-NEXT:    [[SIVAR1:%.*]] = alloca i32, align 4
 842 // CHECK3-NEXT:    [[I:%.*]] = alloca i32, align 4
 843 // CHECK3-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x ptr], align 4
 844 // CHECK3-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 845 // CHECK3-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 846 // CHECK3-NEXT:    store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 847 // CHECK3-NEXT:    store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 848 // CHECK3-NEXT:    store ptr [[SIVAR]], ptr [[SIVAR_ADDR]], align 4
 849 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[SIVAR_ADDR]], align 4
 850 // CHECK3-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 851 // CHECK3-NEXT:    store i32 1, ptr [[DOTOMP_UB]], align 4
 852 // CHECK3-NEXT:    [[TMP1:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 853 // CHECK3-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 854 // CHECK3-NEXT:    store i32 [[TMP1]], ptr [[DOTOMP_LB]], align 4
 855 // CHECK3-NEXT:    store i32 [[TMP2]], ptr [[DOTOMP_UB]], align 4
 856 // CHECK3-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 857 // CHECK3-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 858 // CHECK3-NEXT:    store i32 0, ptr [[SIVAR1]], align 4
 859 // CHECK3-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 860 // CHECK3-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 4
 861 // CHECK3-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2:[0-9]+]], i32 [[TMP4]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 862 // CHECK3-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 863 // CHECK3-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP5]], 1
 864 // CHECK3-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 865 // CHECK3:       cond.true:
 866 // CHECK3-NEXT:    br label [[COND_END:%.*]]
 867 // CHECK3:       cond.false:
 868 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 869 // CHECK3-NEXT:    br label [[COND_END]]
 870 // CHECK3:       cond.end:
 871 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP6]], [[COND_FALSE]] ]
 872 // CHECK3-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 873 // CHECK3-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 874 // CHECK3-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_IV]], align 4
 875 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 876 // CHECK3:       omp.inner.for.cond:
 877 // CHECK3-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 878 // CHECK3-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 879 // CHECK3-NEXT:    [[CMP2:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]]
 880 // CHECK3-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 881 // CHECK3:       omp.inner.for.body:
 882 // CHECK3-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 883 // CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1
 884 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 885 // CHECK3-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
 886 // CHECK3-NEXT:    [[TMP11:%.*]] = load i32, ptr [[I]], align 4
 887 // CHECK3-NEXT:    [[TMP12:%.*]] = load i32, ptr [[SIVAR1]], align 4
 888 // CHECK3-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP12]], [[TMP11]]
 889 // CHECK3-NEXT:    store i32 [[ADD3]], ptr [[SIVAR1]], align 4
 890 // CHECK3-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
 891 // CHECK3:       omp.body.continue:
 892 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 893 // CHECK3:       omp.inner.for.inc:
 894 // CHECK3-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 895 // CHECK3-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP13]], 1
 896 // CHECK3-NEXT:    store i32 [[ADD4]], ptr [[DOTOMP_IV]], align 4
 897 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
 898 // CHECK3:       omp.inner.for.end:
 899 // CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 900 // CHECK3:       omp.loop.exit:
 901 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP4]])
 902 // CHECK3-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0
 903 // CHECK3-NEXT:    store ptr [[SIVAR1]], ptr [[TMP14]], align 4
 904 // CHECK3-NEXT:    [[TMP15:%.*]] = call i32 @__kmpc_reduce_nowait(ptr @[[GLOB3]], i32 [[TMP4]], i32 1, i32 4, ptr [[DOTOMP_REDUCTION_RED_LIST]], ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.omp_outlined.omp_outlined.omp.reduction.reduction_func, ptr @.gomp_critical_user_.reduction.var)
 905 // CHECK3-NEXT:    switch i32 [[TMP15]], label [[DOTOMP_REDUCTION_DEFAULT:%.*]] [
 906 // CHECK3-NEXT:    i32 1, label [[DOTOMP_REDUCTION_CASE1:%.*]]
 907 // CHECK3-NEXT:    i32 2, label [[DOTOMP_REDUCTION_CASE2:%.*]]
 908 // CHECK3-NEXT:    ]
 909 // CHECK3:       .omp.reduction.case1:
 910 // CHECK3-NEXT:    [[TMP16:%.*]] = load i32, ptr [[TMP0]], align 4
 911 // CHECK3-NEXT:    [[TMP17:%.*]] = load i32, ptr [[SIVAR1]], align 4
 912 // CHECK3-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP16]], [[TMP17]]
 913 // CHECK3-NEXT:    store i32 [[ADD5]], ptr [[TMP0]], align 4
 914 // CHECK3-NEXT:    call void @__kmpc_end_reduce_nowait(ptr @[[GLOB3]], i32 [[TMP4]], ptr @.gomp_critical_user_.reduction.var)
 915 // CHECK3-NEXT:    br label [[DOTOMP_REDUCTION_DEFAULT]]
 916 // CHECK3:       .omp.reduction.case2:
 917 // CHECK3-NEXT:    [[TMP18:%.*]] = load i32, ptr [[SIVAR1]], align 4
 918 // CHECK3-NEXT:    [[TMP19:%.*]] = atomicrmw add ptr [[TMP0]], i32 [[TMP18]] monotonic, align 4
 919 // CHECK3-NEXT:    br label [[DOTOMP_REDUCTION_DEFAULT]]
 920 // CHECK3:       .omp.reduction.default:
 921 // CHECK3-NEXT:    ret void
 922 //
 923 //
 924 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.omp_outlined.omp_outlined.omp.reduction.reduction_func
 925 // CHECK3-SAME: (ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]]) #[[ATTR3:[0-9]+]] {
 926 // CHECK3-NEXT:  entry:
 927 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 4
 928 // CHECK3-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 4
 929 // CHECK3-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 4
 930 // CHECK3-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 4
 931 // CHECK3-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 4
 932 // CHECK3-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR1]], align 4
 933 // CHECK3-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP3]], i32 0, i32 0
 934 // CHECK3-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[TMP4]], align 4
 935 // CHECK3-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP2]], i32 0, i32 0
 936 // CHECK3-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[TMP6]], align 4
 937 // CHECK3-NEXT:    [[TMP8:%.*]] = load i32, ptr [[TMP7]], align 4
 938 // CHECK3-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP5]], align 4
 939 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP8]], [[TMP9]]
 940 // CHECK3-NEXT:    store i32 [[ADD]], ptr [[TMP7]], align 4
 941 // CHECK3-NEXT:    ret void
 942 //
 943 //
 944 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l66.omp_outlined.omp.reduction.reduction_func
 945 // CHECK3-SAME: (ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]]) #[[ATTR3]] {
 946 // CHECK3-NEXT:  entry:
 947 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 4
 948 // CHECK3-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 4
 949 // CHECK3-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 4
 950 // CHECK3-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 4
 951 // CHECK3-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 4
 952 // CHECK3-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR1]], align 4
 953 // CHECK3-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP3]], i32 0, i32 0
 954 // CHECK3-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[TMP4]], align 4
 955 // CHECK3-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP2]], i32 0, i32 0
 956 // CHECK3-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[TMP6]], align 4
 957 // CHECK3-NEXT:    [[TMP8:%.*]] = load i32, ptr [[TMP7]], align 4
 958 // CHECK3-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP5]], align 4
 959 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP8]], [[TMP9]]
 960 // CHECK3-NEXT:    store i32 [[ADD]], ptr [[TMP7]], align 4
 961 // CHECK3-NEXT:    ret void
 962 //
 963 //
 964 // CHECK3-LABEL: define {{[^@]+}}@_Z5tmainIiET_v
 965 // CHECK3-SAME: () #[[ATTR5:[0-9]+]] comdat {
 966 // CHECK3-NEXT:  entry:
 967 // CHECK3-NEXT:    [[T_VAR:%.*]] = alloca i32, align 4
 968 // CHECK3-NEXT:    [[VEC:%.*]] = alloca [2 x i32], align 4
 969 // CHECK3-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 4
 970 // CHECK3-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 4
 971 // CHECK3-NEXT:    [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 4
 972 // CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 973 // CHECK3-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
 974 // CHECK3-NEXT:    store i32 0, ptr [[T_VAR]], align 4
 975 // CHECK3-NEXT:    call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[VEC]], ptr align 4 @__const._Z5tmainIiET_v.vec, i32 8, i1 false)
 976 // CHECK3-NEXT:    [[TMP0:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
 977 // CHECK3-NEXT:    store ptr [[T_VAR]], ptr [[TMP0]], align 4
 978 // CHECK3-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
 979 // CHECK3-NEXT:    store ptr [[T_VAR]], ptr [[TMP1]], align 4
 980 // CHECK3-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
 981 // CHECK3-NEXT:    store ptr null, ptr [[TMP2]], align 4
 982 // CHECK3-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
 983 // CHECK3-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
 984 // CHECK3-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
 985 // CHECK3-NEXT:    store i32 2, ptr [[TMP5]], align 4
 986 // CHECK3-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
 987 // CHECK3-NEXT:    store i32 1, ptr [[TMP6]], align 4
 988 // CHECK3-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
 989 // CHECK3-NEXT:    store ptr [[TMP3]], ptr [[TMP7]], align 4
 990 // CHECK3-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
 991 // CHECK3-NEXT:    store ptr [[TMP4]], ptr [[TMP8]], align 4
 992 // CHECK3-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
 993 // CHECK3-NEXT:    store ptr @.offload_sizes.1, ptr [[TMP9]], align 4
 994 // CHECK3-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
 995 // CHECK3-NEXT:    store ptr @.offload_maptypes.2, ptr [[TMP10]], align 4
 996 // CHECK3-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
 997 // CHECK3-NEXT:    store ptr null, ptr [[TMP11]], align 4
 998 // CHECK3-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
 999 // CHECK3-NEXT:    store ptr null, ptr [[TMP12]], align 4
1000 // CHECK3-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
1001 // CHECK3-NEXT:    store i64 2, ptr [[TMP13]], align 8
1002 // CHECK3-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
1003 // CHECK3-NEXT:    store i64 0, ptr [[TMP14]], align 8
1004 // CHECK3-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
1005 // CHECK3-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4
1006 // CHECK3-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
1007 // CHECK3-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4
1008 // CHECK3-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
1009 // CHECK3-NEXT:    store i32 0, ptr [[TMP17]], align 4
1010 // CHECK3-NEXT:    [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB4]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.region_id, ptr [[KERNEL_ARGS]])
1011 // CHECK3-NEXT:    [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0
1012 // CHECK3-NEXT:    br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
1013 // CHECK3:       omp_offload.failed:
1014 // CHECK3-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32(ptr [[T_VAR]]) #[[ATTR2]]
1015 // CHECK3-NEXT:    br label [[OMP_OFFLOAD_CONT]]
1016 // CHECK3:       omp_offload.cont:
1017 // CHECK3-NEXT:    ret i32 0
1018 //
1019 //
1020 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32
1021 // CHECK3-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[T_VAR:%.*]]) #[[ATTR1]] {
1022 // CHECK3-NEXT:  entry:
1023 // CHECK3-NEXT:    [[T_VAR_ADDR:%.*]] = alloca ptr, align 4
1024 // CHECK3-NEXT:    store ptr [[T_VAR]], ptr [[T_VAR_ADDR]], align 4
1025 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[T_VAR_ADDR]], align 4
1026 // CHECK3-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB4]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.omp_outlined, ptr [[TMP0]])
1027 // CHECK3-NEXT:    ret void
1028 //
1029 //
1030 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.omp_outlined
1031 // CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[T_VAR:%.*]]) #[[ATTR1]] {
1032 // CHECK3-NEXT:  entry:
1033 // CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
1034 // CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
1035 // CHECK3-NEXT:    [[T_VAR_ADDR:%.*]] = alloca ptr, align 4
1036 // CHECK3-NEXT:    [[T_VAR1:%.*]] = alloca i32, align 4
1037 // CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
1038 // CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
1039 // CHECK3-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
1040 // CHECK3-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
1041 // CHECK3-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
1042 // CHECK3-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
1043 // CHECK3-NEXT:    [[I:%.*]] = alloca i32, align 4
1044 // CHECK3-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x ptr], align 4
1045 // CHECK3-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
1046 // CHECK3-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
1047 // CHECK3-NEXT:    store ptr [[T_VAR]], ptr [[T_VAR_ADDR]], align 4
1048 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[T_VAR_ADDR]], align 4
1049 // CHECK3-NEXT:    store i32 0, ptr [[T_VAR1]], align 4
1050 // CHECK3-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
1051 // CHECK3-NEXT:    store i32 1, ptr [[DOTOMP_COMB_UB]], align 4
1052 // CHECK3-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
1053 // CHECK3-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
1054 // CHECK3-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
1055 // CHECK3-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
1056 // CHECK3-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
1057 // CHECK3-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
1058 // CHECK3-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 1
1059 // CHECK3-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
1060 // CHECK3:       cond.true:
1061 // CHECK3-NEXT:    br label [[COND_END:%.*]]
1062 // CHECK3:       cond.false:
1063 // CHECK3-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
1064 // CHECK3-NEXT:    br label [[COND_END]]
1065 // CHECK3:       cond.end:
1066 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
1067 // CHECK3-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
1068 // CHECK3-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
1069 // CHECK3-NEXT:    store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4
1070 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
1071 // CHECK3:       omp.inner.for.cond:
1072 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
1073 // CHECK3-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
1074 // CHECK3-NEXT:    [[CMP2:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]]
1075 // CHECK3-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
1076 // CHECK3:       omp.inner.for.body:
1077 // CHECK3-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
1078 // CHECK3-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
1079 // CHECK3-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB4]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.omp_outlined.omp_outlined, i32 [[TMP8]], i32 [[TMP9]], ptr [[T_VAR1]])
1080 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
1081 // CHECK3:       omp.inner.for.inc:
1082 // CHECK3-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
1083 // CHECK3-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
1084 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP10]], [[TMP11]]
1085 // CHECK3-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
1086 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
1087 // CHECK3:       omp.inner.for.end:
1088 // CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
1089 // CHECK3:       omp.loop.exit:
1090 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]])
1091 // CHECK3-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0
1092 // CHECK3-NEXT:    store ptr [[T_VAR1]], ptr [[TMP12]], align 4
1093 // CHECK3-NEXT:    [[TMP13:%.*]] = call i32 @__kmpc_reduce_nowait(ptr @[[GLOB3]], i32 [[TMP2]], i32 1, i32 4, ptr [[DOTOMP_REDUCTION_RED_LIST]], ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.omp_outlined.omp.reduction.reduction_func, ptr @.gomp_critical_user_.reduction.var)
1094 // CHECK3-NEXT:    switch i32 [[TMP13]], label [[DOTOMP_REDUCTION_DEFAULT:%.*]] [
1095 // CHECK3-NEXT:    i32 1, label [[DOTOMP_REDUCTION_CASE1:%.*]]
1096 // CHECK3-NEXT:    i32 2, label [[DOTOMP_REDUCTION_CASE2:%.*]]
1097 // CHECK3-NEXT:    ]
1098 // CHECK3:       .omp.reduction.case1:
1099 // CHECK3-NEXT:    [[TMP14:%.*]] = load i32, ptr [[TMP0]], align 4
1100 // CHECK3-NEXT:    [[TMP15:%.*]] = load i32, ptr [[T_VAR1]], align 4
1101 // CHECK3-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP14]], [[TMP15]]
1102 // CHECK3-NEXT:    store i32 [[ADD3]], ptr [[TMP0]], align 4
1103 // CHECK3-NEXT:    call void @__kmpc_end_reduce_nowait(ptr @[[GLOB3]], i32 [[TMP2]], ptr @.gomp_critical_user_.reduction.var)
1104 // CHECK3-NEXT:    br label [[DOTOMP_REDUCTION_DEFAULT]]
1105 // CHECK3:       .omp.reduction.case2:
1106 // CHECK3-NEXT:    [[TMP16:%.*]] = load i32, ptr [[T_VAR1]], align 4
1107 // CHECK3-NEXT:    [[TMP17:%.*]] = atomicrmw add ptr [[TMP0]], i32 [[TMP16]] monotonic, align 4
1108 // CHECK3-NEXT:    br label [[DOTOMP_REDUCTION_DEFAULT]]
1109 // CHECK3:       .omp.reduction.default:
1110 // CHECK3-NEXT:    ret void
1111 //
1112 //
1113 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.omp_outlined.omp_outlined
1114 // CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[T_VAR:%.*]]) #[[ATTR1]] {
1115 // CHECK3-NEXT:  entry:
1116 // CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
1117 // CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
1118 // CHECK3-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
1119 // CHECK3-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
1120 // CHECK3-NEXT:    [[T_VAR_ADDR:%.*]] = alloca ptr, align 4
1121 // CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
1122 // CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
1123 // CHECK3-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
1124 // CHECK3-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
1125 // CHECK3-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
1126 // CHECK3-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
1127 // CHECK3-NEXT:    [[T_VAR1:%.*]] = alloca i32, align 4
1128 // CHECK3-NEXT:    [[I:%.*]] = alloca i32, align 4
1129 // CHECK3-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x ptr], align 4
1130 // CHECK3-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
1131 // CHECK3-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
1132 // CHECK3-NEXT:    store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4
1133 // CHECK3-NEXT:    store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4
1134 // CHECK3-NEXT:    store ptr [[T_VAR]], ptr [[T_VAR_ADDR]], align 4
1135 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[T_VAR_ADDR]], align 4
1136 // CHECK3-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
1137 // CHECK3-NEXT:    store i32 1, ptr [[DOTOMP_UB]], align 4
1138 // CHECK3-NEXT:    [[TMP1:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4
1139 // CHECK3-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
1140 // CHECK3-NEXT:    store i32 [[TMP1]], ptr [[DOTOMP_LB]], align 4
1141 // CHECK3-NEXT:    store i32 [[TMP2]], ptr [[DOTOMP_UB]], align 4
1142 // CHECK3-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
1143 // CHECK3-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
1144 // CHECK3-NEXT:    store i32 0, ptr [[T_VAR1]], align 4
1145 // CHECK3-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
1146 // CHECK3-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 4
1147 // CHECK3-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP4]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
1148 // CHECK3-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
1149 // CHECK3-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP5]], 1
1150 // CHECK3-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
1151 // CHECK3:       cond.true:
1152 // CHECK3-NEXT:    br label [[COND_END:%.*]]
1153 // CHECK3:       cond.false:
1154 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
1155 // CHECK3-NEXT:    br label [[COND_END]]
1156 // CHECK3:       cond.end:
1157 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP6]], [[COND_FALSE]] ]
1158 // CHECK3-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
1159 // CHECK3-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
1160 // CHECK3-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_IV]], align 4
1161 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
1162 // CHECK3:       omp.inner.for.cond:
1163 // CHECK3-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
1164 // CHECK3-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
1165 // CHECK3-NEXT:    [[CMP2:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]]
1166 // CHECK3-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
1167 // CHECK3:       omp.inner.for.body:
1168 // CHECK3-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
1169 // CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1
1170 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
1171 // CHECK3-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
1172 // CHECK3-NEXT:    [[TMP11:%.*]] = load i32, ptr [[I]], align 4
1173 // CHECK3-NEXT:    [[TMP12:%.*]] = load i32, ptr [[T_VAR1]], align 4
1174 // CHECK3-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP12]], [[TMP11]]
1175 // CHECK3-NEXT:    store i32 [[ADD3]], ptr [[T_VAR1]], align 4
1176 // CHECK3-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
1177 // CHECK3:       omp.body.continue:
1178 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
1179 // CHECK3:       omp.inner.for.inc:
1180 // CHECK3-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
1181 // CHECK3-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP13]], 1
1182 // CHECK3-NEXT:    store i32 [[ADD4]], ptr [[DOTOMP_IV]], align 4
1183 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
1184 // CHECK3:       omp.inner.for.end:
1185 // CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
1186 // CHECK3:       omp.loop.exit:
1187 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP4]])
1188 // CHECK3-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOMP_REDUCTION_RED_LIST]], i32 0, i32 0
1189 // CHECK3-NEXT:    store ptr [[T_VAR1]], ptr [[TMP14]], align 4
1190 // CHECK3-NEXT:    [[TMP15:%.*]] = call i32 @__kmpc_reduce_nowait(ptr @[[GLOB3]], i32 [[TMP4]], i32 1, i32 4, ptr [[DOTOMP_REDUCTION_RED_LIST]], ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.omp_outlined.omp_outlined.omp.reduction.reduction_func, ptr @.gomp_critical_user_.reduction.var)
1191 // CHECK3-NEXT:    switch i32 [[TMP15]], label [[DOTOMP_REDUCTION_DEFAULT:%.*]] [
1192 // CHECK3-NEXT:    i32 1, label [[DOTOMP_REDUCTION_CASE1:%.*]]
1193 // CHECK3-NEXT:    i32 2, label [[DOTOMP_REDUCTION_CASE2:%.*]]
1194 // CHECK3-NEXT:    ]
1195 // CHECK3:       .omp.reduction.case1:
1196 // CHECK3-NEXT:    [[TMP16:%.*]] = load i32, ptr [[TMP0]], align 4
1197 // CHECK3-NEXT:    [[TMP17:%.*]] = load i32, ptr [[T_VAR1]], align 4
1198 // CHECK3-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP16]], [[TMP17]]
1199 // CHECK3-NEXT:    store i32 [[ADD5]], ptr [[TMP0]], align 4
1200 // CHECK3-NEXT:    call void @__kmpc_end_reduce_nowait(ptr @[[GLOB3]], i32 [[TMP4]], ptr @.gomp_critical_user_.reduction.var)
1201 // CHECK3-NEXT:    br label [[DOTOMP_REDUCTION_DEFAULT]]
1202 // CHECK3:       .omp.reduction.case2:
1203 // CHECK3-NEXT:    [[TMP18:%.*]] = load i32, ptr [[T_VAR1]], align 4
1204 // CHECK3-NEXT:    [[TMP19:%.*]] = atomicrmw add ptr [[TMP0]], i32 [[TMP18]] monotonic, align 4
1205 // CHECK3-NEXT:    br label [[DOTOMP_REDUCTION_DEFAULT]]
1206 // CHECK3:       .omp.reduction.default:
1207 // CHECK3-NEXT:    ret void
1208 //
1209 //
1210 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.omp_outlined.omp_outlined.omp.reduction.reduction_func
1211 // CHECK3-SAME: (ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]]) #[[ATTR3]] {
1212 // CHECK3-NEXT:  entry:
1213 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 4
1214 // CHECK3-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 4
1215 // CHECK3-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 4
1216 // CHECK3-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 4
1217 // CHECK3-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 4
1218 // CHECK3-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR1]], align 4
1219 // CHECK3-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP3]], i32 0, i32 0
1220 // CHECK3-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[TMP4]], align 4
1221 // CHECK3-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP2]], i32 0, i32 0
1222 // CHECK3-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[TMP6]], align 4
1223 // CHECK3-NEXT:    [[TMP8:%.*]] = load i32, ptr [[TMP7]], align 4
1224 // CHECK3-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP5]], align 4
1225 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP8]], [[TMP9]]
1226 // CHECK3-NEXT:    store i32 [[ADD]], ptr [[TMP7]], align 4
1227 // CHECK3-NEXT:    ret void
1228 //
1229 //
1230 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l32.omp_outlined.omp.reduction.reduction_func
1231 // CHECK3-SAME: (ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]]) #[[ATTR3]] {
1232 // CHECK3-NEXT:  entry:
1233 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 4
1234 // CHECK3-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 4
1235 // CHECK3-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 4
1236 // CHECK3-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 4
1237 // CHECK3-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 4
1238 // CHECK3-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR1]], align 4
1239 // CHECK3-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP3]], i32 0, i32 0
1240 // CHECK3-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[TMP4]], align 4
1241 // CHECK3-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP2]], i32 0, i32 0
1242 // CHECK3-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[TMP6]], align 4
1243 // CHECK3-NEXT:    [[TMP8:%.*]] = load i32, ptr [[TMP7]], align 4
1244 // CHECK3-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP5]], align 4
1245 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP8]], [[TMP9]]
1246 // CHECK3-NEXT:    store i32 [[ADD]], ptr [[TMP7]], align 4
1247 // CHECK3-NEXT:    ret void
1248 //
1249 //
1250 // CHECK3-LABEL: define {{[^@]+}}@.omp_offloading.requires_reg
1251 // CHECK3-SAME: () #[[ATTR7:[0-9]+]] {
1252 // CHECK3-NEXT:  entry:
1253 // CHECK3-NEXT:    call void @__tgt_register_requires(i64 1)
1254 // CHECK3-NEXT:    ret void
1255 //
1256 //
1257 // CHECK5-LABEL: define {{[^@]+}}@main
1258 // CHECK5-SAME: () #[[ATTR0:[0-9]+]] {
1259 // CHECK5-NEXT:  entry:
1260 // CHECK5-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
1261 // CHECK5-NEXT:    [[REF_TMP:%.*]] = alloca [[CLASS_ANON:%.*]], align 1
1262 // CHECK5-NEXT:    store i32 0, ptr [[RETVAL]], align 4
1263 // CHECK5-NEXT:    call void @"_ZZ4mainENK3$_0clEv"(ptr noundef nonnull align 1 dereferenceable(1) [[REF_TMP]])
1264 // CHECK5-NEXT:    ret i32 0
1265 //
1266 //
1267 // CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l44
1268 // CHECK5-SAME: (ptr noundef nonnull align 4 dereferenceable(4) [[SIVAR:%.*]]) #[[ATTR2:[0-9]+]] {
1269 // CHECK5-NEXT:  entry:
1270 // CHECK5-NEXT:    [[SIVAR_ADDR:%.*]] = alloca ptr, align 8
1271 // CHECK5-NEXT:    store ptr [[SIVAR]], ptr [[SIVAR_ADDR]], align 8
1272 // CHECK5-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[SIVAR_ADDR]], align 8
1273 // CHECK5-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB4:[0-9]+]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l44.omp_outlined, ptr [[TMP0]])
1274 // CHECK5-NEXT:    ret void
1275 //
1276 //
1277 // CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l44.omp_outlined
1278 // CHECK5-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[SIVAR:%.*]]) #[[ATTR2]] {
1279 // CHECK5-NEXT:  entry:
1280 // CHECK5-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
1281 // CHECK5-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
1282 // CHECK5-NEXT:    [[SIVAR_ADDR:%.*]] = alloca ptr, align 8
1283 // CHECK5-NEXT:    [[SIVAR1:%.*]] = alloca i32, align 4
1284 // CHECK5-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
1285 // CHECK5-NEXT:    [[TMP:%.*]] = alloca i32, align 4
1286 // CHECK5-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
1287 // CHECK5-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
1288 // CHECK5-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
1289 // CHECK5-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
1290 // CHECK5-NEXT:    [[I:%.*]] = alloca i32, align 4
1291 // CHECK5-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x ptr], align 8
1292 // CHECK5-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
1293 // CHECK5-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
1294 // CHECK5-NEXT:    store ptr [[SIVAR]], ptr [[SIVAR_ADDR]], align 8
1295 // CHECK5-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[SIVAR_ADDR]], align 8
1296 // CHECK5-NEXT:    store i32 0, ptr [[SIVAR1]], align 4
1297 // CHECK5-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
1298 // CHECK5-NEXT:    store i32 1, ptr [[DOTOMP_COMB_UB]], align 4
1299 // CHECK5-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
1300 // CHECK5-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
1301 // CHECK5-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
1302 // CHECK5-NEXT:    [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
1303 // CHECK5-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1:[0-9]+]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
1304 // CHECK5-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
1305 // CHECK5-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 1
1306 // CHECK5-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
1307 // CHECK5:       cond.true:
1308 // CHECK5-NEXT:    br label [[COND_END:%.*]]
1309 // CHECK5:       cond.false:
1310 // CHECK5-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
1311 // CHECK5-NEXT:    br label [[COND_END]]
1312 // CHECK5:       cond.end:
1313 // CHECK5-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ]
1314 // CHECK5-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
1315 // CHECK5-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
1316 // CHECK5-NEXT:    store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4
1317 // CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
1318 // CHECK5:       omp.inner.for.cond:
1319 // CHECK5-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
1320 // CHECK5-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
1321 // CHECK5-NEXT:    [[CMP2:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]]
1322 // CHECK5-NEXT:    br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
1323 // CHECK5:       omp.inner.for.body:
1324 // CHECK5-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
1325 // CHECK5-NEXT:    [[TMP9:%.*]] = zext i32 [[TMP8]] to i64
1326 // CHECK5-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
1327 // CHECK5-NEXT:    [[TMP11:%.*]] = zext i32 [[TMP10]] to i64
1328 // CHECK5-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB4]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l44.omp_outlined.omp_outlined, i64 [[TMP9]], i64 [[TMP11]], ptr [[SIVAR1]])
1329 // CHECK5-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
1330 // CHECK5:       omp.inner.for.inc:
1331 // CHECK5-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
1332 // CHECK5-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
1333 // CHECK5-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP12]], [[TMP13]]
1334 // CHECK5-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
1335 // CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND]]
1336 // CHECK5:       omp.inner.for.end:
1337 // CHECK5-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
1338 // CHECK5:       omp.loop.exit:
1339 // CHECK5-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]])
1340 // CHECK5-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0
1341 // CHECK5-NEXT:    store ptr [[SIVAR1]], ptr [[TMP14]], align 8
1342 // CHECK5-NEXT:    [[TMP15:%.*]] = call i32 @__kmpc_reduce_nowait(ptr @[[GLOB3:[0-9]+]], i32 [[TMP2]], i32 1, i64 8, ptr [[DOTOMP_REDUCTION_RED_LIST]], ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l44.omp_outlined.omp.reduction.reduction_func, ptr @.gomp_critical_user_.reduction.var)
1343 // CHECK5-NEXT:    switch i32 [[TMP15]], label [[DOTOMP_REDUCTION_DEFAULT:%.*]] [
1344 // CHECK5-NEXT:    i32 1, label [[DOTOMP_REDUCTION_CASE1:%.*]]
1345 // CHECK5-NEXT:    i32 2, label [[DOTOMP_REDUCTION_CASE2:%.*]]
1346 // CHECK5-NEXT:    ]
1347 // CHECK5:       .omp.reduction.case1:
1348 // CHECK5-NEXT:    [[TMP16:%.*]] = load i32, ptr [[TMP0]], align 4
1349 // CHECK5-NEXT:    [[TMP17:%.*]] = load i32, ptr [[SIVAR1]], align 4
1350 // CHECK5-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP16]], [[TMP17]]
1351 // CHECK5-NEXT:    store i32 [[ADD3]], ptr [[TMP0]], align 4
1352 // CHECK5-NEXT:    call void @__kmpc_end_reduce_nowait(ptr @[[GLOB3]], i32 [[TMP2]], ptr @.gomp_critical_user_.reduction.var)
1353 // CHECK5-NEXT:    br label [[DOTOMP_REDUCTION_DEFAULT]]
1354 // CHECK5:       .omp.reduction.case2:
1355 // CHECK5-NEXT:    [[TMP18:%.*]] = load i32, ptr [[SIVAR1]], align 4
1356 // CHECK5-NEXT:    [[TMP19:%.*]] = atomicrmw add ptr [[TMP0]], i32 [[TMP18]] monotonic, align 4
1357 // CHECK5-NEXT:    br label [[DOTOMP_REDUCTION_DEFAULT]]
1358 // CHECK5:       .omp.reduction.default:
1359 // CHECK5-NEXT:    ret void
1360 //
1361 //
1362 // CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l44.omp_outlined.omp_outlined
1363 // CHECK5-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[SIVAR:%.*]]) #[[ATTR2]] {
1364 // CHECK5-NEXT:  entry:
1365 // CHECK5-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
1366 // CHECK5-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
1367 // CHECK5-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
1368 // CHECK5-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
1369 // CHECK5-NEXT:    [[SIVAR_ADDR:%.*]] = alloca ptr, align 8
1370 // CHECK5-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
1371 // CHECK5-NEXT:    [[TMP:%.*]] = alloca i32, align 4
1372 // CHECK5-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
1373 // CHECK5-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
1374 // CHECK5-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
1375 // CHECK5-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
1376 // CHECK5-NEXT:    [[SIVAR2:%.*]] = alloca i32, align 4
1377 // CHECK5-NEXT:    [[I:%.*]] = alloca i32, align 4
1378 // CHECK5-NEXT:    [[REF_TMP:%.*]] = alloca [[CLASS_ANON_0:%.*]], align 8
1379 // CHECK5-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x ptr], align 8
1380 // CHECK5-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
1381 // CHECK5-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
1382 // CHECK5-NEXT:    store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8
1383 // CHECK5-NEXT:    store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8
1384 // CHECK5-NEXT:    store ptr [[SIVAR]], ptr [[SIVAR_ADDR]], align 8
1385 // CHECK5-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[SIVAR_ADDR]], align 8
1386 // CHECK5-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
1387 // CHECK5-NEXT:    store i32 1, ptr [[DOTOMP_UB]], align 4
1388 // CHECK5-NEXT:    [[TMP1:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8
1389 // CHECK5-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP1]] to i32
1390 // CHECK5-NEXT:    [[TMP2:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
1391 // CHECK5-NEXT:    [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32
1392 // CHECK5-NEXT:    store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4
1393 // CHECK5-NEXT:    store i32 [[CONV1]], ptr [[DOTOMP_UB]], align 4
1394 // CHECK5-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
1395 // CHECK5-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
1396 // CHECK5-NEXT:    store i32 0, ptr [[SIVAR2]], align 4
1397 // CHECK5-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
1398 // CHECK5-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 4
1399 // CHECK5-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2:[0-9]+]], i32 [[TMP4]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
1400 // CHECK5-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
1401 // CHECK5-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP5]], 1
1402 // CHECK5-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
1403 // CHECK5:       cond.true:
1404 // CHECK5-NEXT:    br label [[COND_END:%.*]]
1405 // CHECK5:       cond.false:
1406 // CHECK5-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
1407 // CHECK5-NEXT:    br label [[COND_END]]
1408 // CHECK5:       cond.end:
1409 // CHECK5-NEXT:    [[COND:%.*]] = phi i32 [ 1, [[COND_TRUE]] ], [ [[TMP6]], [[COND_FALSE]] ]
1410 // CHECK5-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
1411 // CHECK5-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
1412 // CHECK5-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_IV]], align 4
1413 // CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
1414 // CHECK5:       omp.inner.for.cond:
1415 // CHECK5-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
1416 // CHECK5-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
1417 // CHECK5-NEXT:    [[CMP3:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]]
1418 // CHECK5-NEXT:    br i1 [[CMP3]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
1419 // CHECK5:       omp.inner.for.body:
1420 // CHECK5-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
1421 // CHECK5-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1
1422 // CHECK5-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
1423 // CHECK5-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
1424 // CHECK5-NEXT:    [[TMP11:%.*]] = load i32, ptr [[I]], align 4
1425 // CHECK5-NEXT:    [[TMP12:%.*]] = load i32, ptr [[SIVAR2]], align 4
1426 // CHECK5-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP12]], [[TMP11]]
1427 // CHECK5-NEXT:    store i32 [[ADD4]], ptr [[SIVAR2]], align 4
1428 // CHECK5-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], ptr [[REF_TMP]], i32 0, i32 0
1429 // CHECK5-NEXT:    store ptr [[SIVAR2]], ptr [[TMP13]], align 8
1430 // CHECK5-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE_clEv"(ptr noundef nonnull align 8 dereferenceable(8) [[REF_TMP]])
1431 // CHECK5-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
1432 // CHECK5:       omp.body.continue:
1433 // CHECK5-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
1434 // CHECK5:       omp.inner.for.inc:
1435 // CHECK5-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
1436 // CHECK5-NEXT:    [[ADD5:%.*]] = add nsw i32 [[TMP14]], 1
1437 // CHECK5-NEXT:    store i32 [[ADD5]], ptr [[DOTOMP_IV]], align 4
1438 // CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND]]
1439 // CHECK5:       omp.inner.for.end:
1440 // CHECK5-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
1441 // CHECK5:       omp.loop.exit:
1442 // CHECK5-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP4]])
1443 // CHECK5-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOMP_REDUCTION_RED_LIST]], i64 0, i64 0
1444 // CHECK5-NEXT:    store ptr [[SIVAR2]], ptr [[TMP15]], align 8
1445 // CHECK5-NEXT:    [[TMP16:%.*]] = call i32 @__kmpc_reduce_nowait(ptr @[[GLOB3]], i32 [[TMP4]], i32 1, i64 8, ptr [[DOTOMP_REDUCTION_RED_LIST]], ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l44.omp_outlined.omp_outlined.omp.reduction.reduction_func, ptr @.gomp_critical_user_.reduction.var)
1446 // CHECK5-NEXT:    switch i32 [[TMP16]], label [[DOTOMP_REDUCTION_DEFAULT:%.*]] [
1447 // CHECK5-NEXT:    i32 1, label [[DOTOMP_REDUCTION_CASE1:%.*]]
1448 // CHECK5-NEXT:    i32 2, label [[DOTOMP_REDUCTION_CASE2:%.*]]
1449 // CHECK5-NEXT:    ]
1450 // CHECK5:       .omp.reduction.case1:
1451 // CHECK5-NEXT:    [[TMP17:%.*]] = load i32, ptr [[TMP0]], align 4
1452 // CHECK5-NEXT:    [[TMP18:%.*]] = load i32, ptr [[SIVAR2]], align 4
1453 // CHECK5-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP17]], [[TMP18]]
1454 // CHECK5-NEXT:    store i32 [[ADD6]], ptr [[TMP0]], align 4
1455 // CHECK5-NEXT:    call void @__kmpc_end_reduce_nowait(ptr @[[GLOB3]], i32 [[TMP4]], ptr @.gomp_critical_user_.reduction.var)
1456 // CHECK5-NEXT:    br label [[DOTOMP_REDUCTION_DEFAULT]]
1457 // CHECK5:       .omp.reduction.case2:
1458 // CHECK5-NEXT:    [[TMP19:%.*]] = load i32, ptr [[SIVAR2]], align 4
1459 // CHECK5-NEXT:    [[TMP20:%.*]] = atomicrmw add ptr [[TMP0]], i32 [[TMP19]] monotonic, align 4
1460 // CHECK5-NEXT:    br label [[DOTOMP_REDUCTION_DEFAULT]]
1461 // CHECK5:       .omp.reduction.default:
1462 // CHECK5-NEXT:    ret void
1463 //
1464 //
1465 // CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l44.omp_outlined.omp_outlined.omp.reduction.reduction_func
1466 // CHECK5-SAME: (ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]]) #[[ATTR4:[0-9]+]] {
1467 // CHECK5-NEXT:  entry:
1468 // CHECK5-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
1469 // CHECK5-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
1470 // CHECK5-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
1471 // CHECK5-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
1472 // CHECK5-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
1473 // CHECK5-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
1474 // CHECK5-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP3]], i64 0, i64 0
1475 // CHECK5-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[TMP4]], align 8
1476 // CHECK5-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP2]], i64 0, i64 0
1477 // CHECK5-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[TMP6]], align 8
1478 // CHECK5-NEXT:    [[TMP8:%.*]] = load i32, ptr [[TMP7]], align 4
1479 // CHECK5-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP5]], align 4
1480 // CHECK5-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP8]], [[TMP9]]
1481 // CHECK5-NEXT:    store i32 [[ADD]], ptr [[TMP7]], align 4
1482 // CHECK5-NEXT:    ret void
1483 //
1484 //
1485 // CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l44.omp_outlined.omp.reduction.reduction_func
1486 // CHECK5-SAME: (ptr noundef [[TMP0:%.*]], ptr noundef [[TMP1:%.*]]) #[[ATTR4]] {
1487 // CHECK5-NEXT:  entry:
1488 // CHECK5-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
1489 // CHECK5-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
1490 // CHECK5-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
1491 // CHECK5-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
1492 // CHECK5-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
1493 // CHECK5-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
1494 // CHECK5-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP3]], i64 0, i64 0
1495 // CHECK5-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[TMP4]], align 8
1496 // CHECK5-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP2]], i64 0, i64 0
1497 // CHECK5-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[TMP6]], align 8
1498 // CHECK5-NEXT:    [[TMP8:%.*]] = load i32, ptr [[TMP7]], align 4
1499 // CHECK5-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP5]], align 4
1500 // CHECK5-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP8]], [[TMP9]]
1501 // CHECK5-NEXT:    store i32 [[ADD]], ptr [[TMP7]], align 4
1502 // CHECK5-NEXT:    ret void
1503 //
1504 //
1505 // CHECK5-LABEL: define {{[^@]+}}@.omp_offloading.requires_reg
1506 // CHECK5-SAME: () #[[ATTR6:[0-9]+]] {
1507 // CHECK5-NEXT:  entry:
1508 // CHECK5-NEXT:    call void @__tgt_register_requires(i64 1)
1509 // CHECK5-NEXT:    ret void
1510 //