clang/test/OpenMP/target_task_affinity_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 // Test host codegen.
   3 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK1
   4 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
   5 // RUN: %clang_cc1 -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
   6 // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK3
   7 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
   8 // RUN: %clang_cc1 -DCK1 -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
   9 // RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK5
  10 // RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
  11 // RUN: %clang_cc1 -DCK1 -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 --check-prefix=CHECK5
  12 // RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK7
  13 // RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
  14 // RUN: %clang_cc1 -DCK1 -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 --check-prefix=CHECK7
  15 // Test target codegen
  16 // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
  17 // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix=CHECK9
  18 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
  19 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK9
  20 // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
  21 // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix=CHECK11
  22 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
  23 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK11
  24 // RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
  25 // RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix=CHECK5
  26 // RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
  27 // RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK5
  28 // RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
  29 // RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix=CHECK7
  30 // RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
  31 // RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK7
  32 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
  33 // expected-no-diagnostics
  34 #ifndef HEADER
  35 #define HEADER
  36
  37 int main() {
  38   int *A;
  39   int *B;
  40   #pragma omp target data map(tofrom: A[0:1024])
  41
  42 #pragma omp target data use_device_ptr(A)
  43 {
  44   #pragma omp target defaultmap(none) is_device_ptr(A) map(tofrom: B[0:1024])
  45   {
  46     #pragma omp task shared(B) affinity(A[0:1024])
  47       {
  48         for (int i = 0; i < 1024; i++) {
  49           B[i] = 2 * A[i];
  50         }
  51       }
  52     #pragma omp taskwait
  53   }
  54 }
  55   return 0;
  56 }
  57
  58 #endif
  59 // CHECK1-LABEL: define {{[^@]+}}@main
  60 // CHECK1-SAME: () #[[ATTR0:[0-9]+]] {
  61 // CHECK1-NEXT:  entry:
  62 // CHECK1-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
  63 // CHECK1-NEXT:    [[A:%.*]] = alloca ptr, align 8
  64 // CHECK1-NEXT:    [[B:%.*]] = alloca ptr, align 8
  65 // CHECK1-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 8
  66 // CHECK1-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 8
  67 // CHECK1-NEXT:    [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 8
  68 // CHECK1-NEXT:    [[DOTOFFLOAD_BASEPTRS1:%.*]] = alloca [1 x ptr], align 8
  69 // CHECK1-NEXT:    [[DOTOFFLOAD_PTRS2:%.*]] = alloca [1 x ptr], align 8
  70 // CHECK1-NEXT:    [[DOTOFFLOAD_MAPPERS3:%.*]] = alloca [1 x ptr], align 8
  71 // CHECK1-NEXT:    [[TMP0:%.*]] = alloca ptr, align 8
  72 // CHECK1-NEXT:    [[DOTOFFLOAD_BASEPTRS5:%.*]] = alloca [2 x ptr], align 8
  73 // CHECK1-NEXT:    [[DOTOFFLOAD_PTRS6:%.*]] = alloca [2 x ptr], align 8
  74 // CHECK1-NEXT:    [[DOTOFFLOAD_MAPPERS7:%.*]] = alloca [2 x ptr], align 8
  75 // CHECK1-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
  76 // CHECK1-NEXT:    store i32 0, ptr [[RETVAL]], align 4
  77 // CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A]], align 8
  78 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A]], align 8
  79 // CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds nuw i32, ptr [[TMP2]], i64 0
  80 // CHECK1-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
  81 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[TMP3]], align 8
  82 // CHECK1-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
  83 // CHECK1-NEXT:    store ptr [[ARRAYIDX]], ptr [[TMP4]], align 8
  84 // CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
  85 // CHECK1-NEXT:    store ptr null, ptr [[TMP5]], align 8
  86 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
  87 // CHECK1-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
  88 // CHECK1-NEXT:    call void @__tgt_target_data_begin_mapper(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 1, ptr [[TMP6]], ptr [[TMP7]], ptr @.offload_sizes, ptr @.offload_maptypes, ptr null, ptr null)
  89 // CHECK1-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[A]], align 8
  90 // CHECK1-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
  91 // CHECK1-NEXT:    store ptr [[TMP8]], ptr [[TMP9]], align 8
  92 // CHECK1-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
  93 // CHECK1-NEXT:    store ptr [[TMP8]], ptr [[TMP10]], align 8
  94 // CHECK1-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
  95 // CHECK1-NEXT:    store ptr null, ptr [[TMP11]], align 8
  96 // CHECK1-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
  97 // CHECK1-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
  98 // CHECK1-NEXT:    call void @__tgt_target_data_begin_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP12]], ptr [[TMP13]], ptr @.offload_sizes.1, ptr @.offload_maptypes.2, ptr null, ptr null)
  99 // CHECK1-NEXT:    [[TMP14:%.*]] = load ptr, ptr [[TMP9]], align 8
 100 // CHECK1-NEXT:    store ptr [[TMP14]], ptr [[TMP0]], align 8
 101 // CHECK1-NEXT:    [[TMP15:%.*]] = load ptr, ptr [[B]], align 8
 102 // CHECK1-NEXT:    [[TMP16:%.*]] = load ptr, ptr [[TMP0]], align 8
 103 // CHECK1-NEXT:    [[TMP17:%.*]] = load ptr, ptr [[B]], align 8
 104 // CHECK1-NEXT:    [[TMP18:%.*]] = load ptr, ptr [[B]], align 8
 105 // CHECK1-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds nuw i32, ptr [[TMP18]], i64 0
 106 // CHECK1-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
 107 // CHECK1-NEXT:    store ptr [[TMP17]], ptr [[TMP19]], align 8
 108 // CHECK1-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
 109 // CHECK1-NEXT:    store ptr [[ARRAYIDX4]], ptr [[TMP20]], align 8
 110 // CHECK1-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 0
 111 // CHECK1-NEXT:    store ptr null, ptr [[TMP21]], align 8
 112 // CHECK1-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
 113 // CHECK1-NEXT:    store ptr [[TMP16]], ptr [[TMP22]], align 8
 114 // CHECK1-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
 115 // CHECK1-NEXT:    store ptr [[TMP16]], ptr [[TMP23]], align 8
 116 // CHECK1-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i64 0, i64 1
 117 // CHECK1-NEXT:    store ptr null, ptr [[TMP24]], align 8
 118 // CHECK1-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
 119 // CHECK1-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
 120 // CHECK1-NEXT:    [[TMP27:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
 121 // CHECK1-NEXT:    store i32 3, ptr [[TMP27]], align 4
 122 // CHECK1-NEXT:    [[TMP28:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
 123 // CHECK1-NEXT:    store i32 2, ptr [[TMP28]], align 4
 124 // CHECK1-NEXT:    [[TMP29:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
 125 // CHECK1-NEXT:    store ptr [[TMP25]], ptr [[TMP29]], align 8
 126 // CHECK1-NEXT:    [[TMP30:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
 127 // CHECK1-NEXT:    store ptr [[TMP26]], ptr [[TMP30]], align 8
 128 // CHECK1-NEXT:    [[TMP31:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
 129 // CHECK1-NEXT:    store ptr @.offload_sizes.3, ptr [[TMP31]], align 8
 130 // CHECK1-NEXT:    [[TMP32:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
 131 // CHECK1-NEXT:    store ptr @.offload_maptypes.4, ptr [[TMP32]], align 8
 132 // CHECK1-NEXT:    [[TMP33:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
 133 // CHECK1-NEXT:    store ptr null, ptr [[TMP33]], align 8
 134 // CHECK1-NEXT:    [[TMP34:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
 135 // CHECK1-NEXT:    store ptr null, ptr [[TMP34]], align 8
 136 // CHECK1-NEXT:    [[TMP35:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
 137 // CHECK1-NEXT:    store i64 0, ptr [[TMP35]], align 8
 138 // CHECK1-NEXT:    [[TMP36:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
 139 // CHECK1-NEXT:    store i64 0, ptr [[TMP36]], align 8
 140 // CHECK1-NEXT:    [[TMP37:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
 141 // CHECK1-NEXT:    store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP37]], align 4
 142 // CHECK1-NEXT:    [[TMP38:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
 143 // CHECK1-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP38]], align 4
 144 // CHECK1-NEXT:    [[TMP39:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
 145 // CHECK1-NEXT:    store i32 0, ptr [[TMP39]], align 4
 146 // CHECK1-NEXT:    [[TMP40:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l44.region_id, ptr [[KERNEL_ARGS]])
 147 // CHECK1-NEXT:    [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
 148 // CHECK1-NEXT:    br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
 149 // CHECK1:       omp_offload.failed:
 150 // CHECK1-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l44(ptr [[TMP15]], ptr [[TMP16]]) #[[ATTR1:[0-9]+]]
 151 // CHECK1-NEXT:    br label [[OMP_OFFLOAD_CONT]]
 152 // CHECK1:       omp_offload.cont:
 153 // CHECK1-NEXT:    [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
 154 // CHECK1-NEXT:    [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
 155 // CHECK1-NEXT:    call void @__tgt_target_data_end_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP42]], ptr [[TMP43]], ptr @.offload_sizes.1, ptr @.offload_maptypes.2, ptr null, ptr null)
 156 // CHECK1-NEXT:    [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
 157 // CHECK1-NEXT:    [[TMP45:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
 158 // CHECK1-NEXT:    call void @__tgt_target_data_end_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP44]], ptr [[TMP45]], ptr @.offload_sizes, ptr @.offload_maptypes, ptr null, ptr null)
 159 // CHECK1-NEXT:    ret i32 0
 160 //
 161 //
 162 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l44
 163 // CHECK1-SAME: (ptr noundef [[B:%.*]], ptr noundef [[A:%.*]]) #[[ATTR2:[0-9]+]] {
 164 // CHECK1-NEXT:  entry:
 165 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 166 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 167 // CHECK1-NEXT:    [[AGG_CAPTURED:%.*]] = alloca [[STRUCT_ANON:%.*]], align 8
 168 // CHECK1-NEXT:    [[DOTAFFS_ARR_ADDR:%.*]] = alloca [1 x %struct.kmp_task_affinity_info_t], align 8
 169 // CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 170 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 171 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 172 // CHECK1-NEXT:    [[TMP1:%.*]] = getelementptr inbounds nuw [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 0
 173 // CHECK1-NEXT:    store ptr [[B_ADDR]], ptr [[TMP1]], align 8
 174 // CHECK1-NEXT:    [[TMP2:%.*]] = call ptr @__kmpc_omp_task_alloc(ptr @[[GLOB1]], i32 [[TMP0]], i32 1, i64 48, i64 8, ptr @.omp_task_entry.)
 175 // CHECK1-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x %struct.kmp_task_affinity_info_t], ptr [[DOTAFFS_ARR_ADDR]], i64 0, i64 0
 176 // CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 177 // CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds nuw i32, ptr [[TMP4]], i64 0
 178 // CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 179 // CHECK1-NEXT:    [[ARRAYIDX1:%.*]] = getelementptr inbounds nuw i32, ptr [[TMP5]], i64 1023
 180 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr i32, ptr [[ARRAYIDX1]], i32 1
 181 // CHECK1-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[ARRAYIDX]] to i64
 182 // CHECK1-NEXT:    [[TMP8:%.*]] = ptrtoint ptr [[TMP6]] to i64
 183 // CHECK1-NEXT:    [[TMP9:%.*]] = sub nuw i64 [[TMP8]], [[TMP7]]
 184 // CHECK1-NEXT:    [[TMP10:%.*]] = getelementptr [[STRUCT_KMP_TASK_AFFINITY_INFO_T:%.*]], ptr [[TMP3]], i64 0
 185 // CHECK1-NEXT:    [[TMP11:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_AFFINITY_INFO_T]], ptr [[TMP10]], i32 0, i32 0
 186 // CHECK1-NEXT:    [[TMP12:%.*]] = ptrtoint ptr [[ARRAYIDX]] to i64
 187 // CHECK1-NEXT:    store i64 [[TMP12]], ptr [[TMP11]], align 8
 188 // CHECK1-NEXT:    [[TMP13:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_AFFINITY_INFO_T]], ptr [[TMP10]], i32 0, i32 1
 189 // CHECK1-NEXT:    store i64 [[TMP9]], ptr [[TMP13]], align 8
 190 // CHECK1-NEXT:    [[TMP14:%.*]] = call i32 @__kmpc_omp_reg_task_with_affinity(ptr @[[GLOB1]], i32 [[TMP0]], ptr [[TMP2]], i32 1, ptr [[TMP3]])
 191 // CHECK1-NEXT:    [[TMP15:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], ptr [[TMP2]], i32 0, i32 0
 192 // CHECK1-NEXT:    [[TMP16:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T:%.*]], ptr [[TMP15]], i32 0, i32 0
 193 // CHECK1-NEXT:    [[TMP17:%.*]] = load ptr, ptr [[TMP16]], align 8
 194 // CHECK1-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[TMP17]], ptr align 8 [[AGG_CAPTURED]], i64 8, i1 false)
 195 // CHECK1-NEXT:    [[TMP18:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T_WITH_PRIVATES]], ptr [[TMP2]], i32 0, i32 1
 196 // CHECK1-NEXT:    [[TMP19:%.*]] = getelementptr inbounds nuw [[STRUCT__KMP_PRIVATES_T:%.*]], ptr [[TMP18]], i32 0, i32 0
 197 // CHECK1-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 198 // CHECK1-NEXT:    store ptr [[TMP20]], ptr [[TMP19]], align 8
 199 // CHECK1-NEXT:    [[TMP21:%.*]] = call i32 @__kmpc_omp_task(ptr @[[GLOB1]], i32 [[TMP0]], ptr [[TMP2]])
 200 // CHECK1-NEXT:    [[TMP22:%.*]] = call i32 @__kmpc_omp_taskwait(ptr @[[GLOB1]], i32 [[TMP0]])
 201 // CHECK1-NEXT:    ret void
 202 //
 203 //
 204 // CHECK1-LABEL: define {{[^@]+}}@.omp_task_privates_map.
 205 // CHECK1-SAME: (ptr noalias noundef [[TMP0:%.*]], ptr noalias noundef [[TMP1:%.*]]) #[[ATTR3:[0-9]+]] {
 206 // CHECK1-NEXT:  entry:
 207 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 208 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 209 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 210 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 211 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 212 // CHECK1-NEXT:    [[TMP3:%.*]] = getelementptr inbounds nuw [[STRUCT__KMP_PRIVATES_T:%.*]], ptr [[TMP2]], i32 0, i32 0
 213 // CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 214 // CHECK1-NEXT:    store ptr [[TMP3]], ptr [[TMP4]], align 8
 215 // CHECK1-NEXT:    ret void
 216 //
 217 //
 218 // CHECK1-LABEL: define {{[^@]+}}@.omp_task_entry.
 219 // CHECK1-SAME: (i32 noundef signext [[TMP0:%.*]], ptr noalias noundef [[TMP1:%.*]]) #[[ATTR4:[0-9]+]] {
 220 // CHECK1-NEXT:  entry:
 221 // CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR_I:%.*]] = alloca i32, align 4
 222 // CHECK1-NEXT:    [[DOTPART_ID__ADDR_I:%.*]] = alloca ptr, align 8
 223 // CHECK1-NEXT:    [[DOTPRIVATES__ADDR_I:%.*]] = alloca ptr, align 8
 224 // CHECK1-NEXT:    [[DOTCOPY_FN__ADDR_I:%.*]] = alloca ptr, align 8
 225 // CHECK1-NEXT:    [[DOTTASK_T__ADDR_I:%.*]] = alloca ptr, align 8
 226 // CHECK1-NEXT:    [[__CONTEXT_ADDR_I:%.*]] = alloca ptr, align 8
 227 // CHECK1-NEXT:    [[DOTFIRSTPRIV_PTR_ADDR_I:%.*]] = alloca ptr, align 8
 228 // CHECK1-NEXT:    [[I_I:%.*]] = alloca i32, align 4
 229 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i32, align 4
 230 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 231 // CHECK1-NEXT:    store i32 [[TMP0]], ptr [[DOTADDR]], align 4
 232 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 233 // CHECK1-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTADDR]], align 4
 234 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 235 // CHECK1-NEXT:    [[TMP4:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], ptr [[TMP3]], i32 0, i32 0
 236 // CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T:%.*]], ptr [[TMP4]], i32 0, i32 2
 237 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T]], ptr [[TMP4]], i32 0, i32 0
 238 // CHECK1-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[TMP6]], align 8
 239 // CHECK1-NEXT:    [[TMP8:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T_WITH_PRIVATES]], ptr [[TMP3]], i32 0, i32 1
 240 // CHECK1-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META4:![0-9]+]])
 241 // CHECK1-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META7:![0-9]+]])
 242 // CHECK1-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META9:![0-9]+]])
 243 // CHECK1-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META11:![0-9]+]])
 244 // CHECK1-NEXT:    store i32 [[TMP2]], ptr [[DOTGLOBAL_TID__ADDR_I]], align 4, !noalias [[META13:![0-9]+]]
 245 // CHECK1-NEXT:    store ptr [[TMP5]], ptr [[DOTPART_ID__ADDR_I]], align 8, !noalias [[META13]]
 246 // CHECK1-NEXT:    store ptr [[TMP8]], ptr [[DOTPRIVATES__ADDR_I]], align 8, !noalias [[META13]]
 247 // CHECK1-NEXT:    store ptr @.omp_task_privates_map., ptr [[DOTCOPY_FN__ADDR_I]], align 8, !noalias [[META13]]
 248 // CHECK1-NEXT:    store ptr [[TMP3]], ptr [[DOTTASK_T__ADDR_I]], align 8, !noalias [[META13]]
 249 // CHECK1-NEXT:    store ptr [[TMP7]], ptr [[__CONTEXT_ADDR_I]], align 8, !noalias [[META13]]
 250 // CHECK1-NEXT:    [[TMP9:%.*]] = load ptr, ptr [[__CONTEXT_ADDR_I]], align 8, !noalias [[META13]]
 251 // CHECK1-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTCOPY_FN__ADDR_I]], align 8, !noalias [[META13]]
 252 // CHECK1-NEXT:    [[TMP11:%.*]] = load ptr, ptr [[DOTPRIVATES__ADDR_I]], align 8, !noalias [[META13]]
 253 // CHECK1-NEXT:    call void [[TMP10]](ptr [[TMP11]], ptr [[DOTFIRSTPRIV_PTR_ADDR_I]]) #[[ATTR1]]
 254 // CHECK1-NEXT:    [[TMP12:%.*]] = load ptr, ptr [[DOTFIRSTPRIV_PTR_ADDR_I]], align 8, !noalias [[META13]]
 255 // CHECK1-NEXT:    store i32 0, ptr [[I_I]], align 4, !noalias [[META13]]
 256 // CHECK1-NEXT:    br label [[FOR_COND_I:%.*]]
 257 // CHECK1:       for.cond.i:
 258 // CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[I_I]], align 4, !noalias [[META13]]
 259 // CHECK1-NEXT:    [[CMP_I:%.*]] = icmp slt i32 [[TMP13]], 1024
 260 // CHECK1-NEXT:    br i1 [[CMP_I]], label [[FOR_BODY_I:%.*]], label [[DOTOMP_OUTLINED__EXIT:%.*]]
 261 // CHECK1:       for.body.i:
 262 // CHECK1-NEXT:    [[TMP14:%.*]] = load ptr, ptr [[TMP12]], align 8
 263 // CHECK1-NEXT:    [[TMP15:%.*]] = load i32, ptr [[I_I]], align 4, !noalias [[META13]]
 264 // CHECK1-NEXT:    [[IDXPROM_I:%.*]] = sext i32 [[TMP15]] to i64
 265 // CHECK1-NEXT:    [[ARRAYIDX_I:%.*]] = getelementptr inbounds i32, ptr [[TMP14]], i64 [[IDXPROM_I]]
 266 // CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX_I]], align 4
 267 // CHECK1-NEXT:    [[MUL_I:%.*]] = mul nsw i32 2, [[TMP16]]
 268 // CHECK1-NEXT:    [[TMP17:%.*]] = load ptr, ptr [[TMP9]], align 8
 269 // CHECK1-NEXT:    [[TMP18:%.*]] = load ptr, ptr [[TMP17]], align 8
 270 // CHECK1-NEXT:    [[TMP19:%.*]] = load i32, ptr [[I_I]], align 4, !noalias [[META13]]
 271 // CHECK1-NEXT:    [[IDXPROM1_I:%.*]] = sext i32 [[TMP19]] to i64
 272 // CHECK1-NEXT:    [[ARRAYIDX2_I:%.*]] = getelementptr inbounds i32, ptr [[TMP18]], i64 [[IDXPROM1_I]]
 273 // CHECK1-NEXT:    store i32 [[MUL_I]], ptr [[ARRAYIDX2_I]], align 4
 274 // CHECK1-NEXT:    [[TMP20:%.*]] = load i32, ptr [[I_I]], align 4, !noalias [[META13]]
 275 // CHECK1-NEXT:    [[INC_I:%.*]] = add nsw i32 [[TMP20]], 1
 276 // CHECK1-NEXT:    store i32 [[INC_I]], ptr [[I_I]], align 4, !noalias [[META13]]
 277 // CHECK1-NEXT:    br label [[FOR_COND_I]], !llvm.loop [[LOOP14:![0-9]+]]
 278 // CHECK1:       .omp_outlined..exit:
 279 // CHECK1-NEXT:    ret i32 0
 280 //
 281 //
 282 // CHECK3-LABEL: define {{[^@]+}}@main
 283 // CHECK3-SAME: () #[[ATTR0:[0-9]+]] {
 284 // CHECK3-NEXT:  entry:
 285 // CHECK3-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
 286 // CHECK3-NEXT:    [[A:%.*]] = alloca ptr, align 4
 287 // CHECK3-NEXT:    [[B:%.*]] = alloca ptr, align 4
 288 // CHECK3-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 4
 289 // CHECK3-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 4
 290 // CHECK3-NEXT:    [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 4
 291 // CHECK3-NEXT:    [[DOTOFFLOAD_BASEPTRS1:%.*]] = alloca [1 x ptr], align 4
 292 // CHECK3-NEXT:    [[DOTOFFLOAD_PTRS2:%.*]] = alloca [1 x ptr], align 4
 293 // CHECK3-NEXT:    [[DOTOFFLOAD_MAPPERS3:%.*]] = alloca [1 x ptr], align 4
 294 // CHECK3-NEXT:    [[TMP0:%.*]] = alloca ptr, align 4
 295 // CHECK3-NEXT:    [[DOTOFFLOAD_BASEPTRS5:%.*]] = alloca [2 x ptr], align 4
 296 // CHECK3-NEXT:    [[DOTOFFLOAD_PTRS6:%.*]] = alloca [2 x ptr], align 4
 297 // CHECK3-NEXT:    [[DOTOFFLOAD_MAPPERS7:%.*]] = alloca [2 x ptr], align 4
 298 // CHECK3-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
 299 // CHECK3-NEXT:    store i32 0, ptr [[RETVAL]], align 4
 300 // CHECK3-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A]], align 4
 301 // CHECK3-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A]], align 4
 302 // CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds nuw i32, ptr [[TMP2]], i32 0
 303 // CHECK3-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
 304 // CHECK3-NEXT:    store ptr [[TMP1]], ptr [[TMP3]], align 4
 305 // CHECK3-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
 306 // CHECK3-NEXT:    store ptr [[ARRAYIDX]], ptr [[TMP4]], align 4
 307 // CHECK3-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
 308 // CHECK3-NEXT:    store ptr null, ptr [[TMP5]], align 4
 309 // CHECK3-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
 310 // CHECK3-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
 311 // CHECK3-NEXT:    call void @__tgt_target_data_begin_mapper(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 1, ptr [[TMP6]], ptr [[TMP7]], ptr @.offload_sizes, ptr @.offload_maptypes, ptr null, ptr null)
 312 // CHECK3-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[A]], align 4
 313 // CHECK3-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
 314 // CHECK3-NEXT:    store ptr [[TMP8]], ptr [[TMP9]], align 4
 315 // CHECK3-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
 316 // CHECK3-NEXT:    store ptr [[TMP8]], ptr [[TMP10]], align 4
 317 // CHECK3-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0
 318 // CHECK3-NEXT:    store ptr null, ptr [[TMP11]], align 4
 319 // CHECK3-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
 320 // CHECK3-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
 321 // CHECK3-NEXT:    call void @__tgt_target_data_begin_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP12]], ptr [[TMP13]], ptr @.offload_sizes.1, ptr @.offload_maptypes.2, ptr null, ptr null)
 322 // CHECK3-NEXT:    [[TMP14:%.*]] = load ptr, ptr [[TMP9]], align 4
 323 // CHECK3-NEXT:    store ptr [[TMP14]], ptr [[TMP0]], align 4
 324 // CHECK3-NEXT:    [[TMP15:%.*]] = load ptr, ptr [[B]], align 4
 325 // CHECK3-NEXT:    [[TMP16:%.*]] = load ptr, ptr [[TMP0]], align 4
 326 // CHECK3-NEXT:    [[TMP17:%.*]] = load ptr, ptr [[B]], align 4
 327 // CHECK3-NEXT:    [[TMP18:%.*]] = load ptr, ptr [[B]], align 4
 328 // CHECK3-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds nuw i32, ptr [[TMP18]], i32 0
 329 // CHECK3-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
 330 // CHECK3-NEXT:    store ptr [[TMP17]], ptr [[TMP19]], align 4
 331 // CHECK3-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
 332 // CHECK3-NEXT:    store ptr [[ARRAYIDX4]], ptr [[TMP20]], align 4
 333 // CHECK3-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 0
 334 // CHECK3-NEXT:    store ptr null, ptr [[TMP21]], align 4
 335 // CHECK3-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 1
 336 // CHECK3-NEXT:    store ptr [[TMP16]], ptr [[TMP22]], align 4
 337 // CHECK3-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 1
 338 // CHECK3-NEXT:    store ptr [[TMP16]], ptr [[TMP23]], align 4
 339 // CHECK3-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS7]], i32 0, i32 1
 340 // CHECK3-NEXT:    store ptr null, ptr [[TMP24]], align 4
 341 // CHECK3-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS5]], i32 0, i32 0
 342 // CHECK3-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS6]], i32 0, i32 0
 343 // CHECK3-NEXT:    [[TMP27:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
 344 // CHECK3-NEXT:    store i32 3, ptr [[TMP27]], align 4
 345 // CHECK3-NEXT:    [[TMP28:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
 346 // CHECK3-NEXT:    store i32 2, ptr [[TMP28]], align 4
 347 // CHECK3-NEXT:    [[TMP29:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
 348 // CHECK3-NEXT:    store ptr [[TMP25]], ptr [[TMP29]], align 4
 349 // CHECK3-NEXT:    [[TMP30:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
 350 // CHECK3-NEXT:    store ptr [[TMP26]], ptr [[TMP30]], align 4
 351 // CHECK3-NEXT:    [[TMP31:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
 352 // CHECK3-NEXT:    store ptr @.offload_sizes.3, ptr [[TMP31]], align 4
 353 // CHECK3-NEXT:    [[TMP32:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
 354 // CHECK3-NEXT:    store ptr @.offload_maptypes.4, ptr [[TMP32]], align 4
 355 // CHECK3-NEXT:    [[TMP33:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
 356 // CHECK3-NEXT:    store ptr null, ptr [[TMP33]], align 4
 357 // CHECK3-NEXT:    [[TMP34:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
 358 // CHECK3-NEXT:    store ptr null, ptr [[TMP34]], align 4
 359 // CHECK3-NEXT:    [[TMP35:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
 360 // CHECK3-NEXT:    store i64 0, ptr [[TMP35]], align 8
 361 // CHECK3-NEXT:    [[TMP36:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
 362 // CHECK3-NEXT:    store i64 0, ptr [[TMP36]], align 8
 363 // CHECK3-NEXT:    [[TMP37:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
 364 // CHECK3-NEXT:    store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP37]], align 4
 365 // CHECK3-NEXT:    [[TMP38:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
 366 // CHECK3-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP38]], align 4
 367 // CHECK3-NEXT:    [[TMP39:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
 368 // CHECK3-NEXT:    store i32 0, ptr [[TMP39]], align 4
 369 // CHECK3-NEXT:    [[TMP40:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l44.region_id, ptr [[KERNEL_ARGS]])
 370 // CHECK3-NEXT:    [[TMP41:%.*]] = icmp ne i32 [[TMP40]], 0
 371 // CHECK3-NEXT:    br i1 [[TMP41]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
 372 // CHECK3:       omp_offload.failed:
 373 // CHECK3-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l44(ptr [[TMP15]], ptr [[TMP16]]) #[[ATTR1:[0-9]+]]
 374 // CHECK3-NEXT:    br label [[OMP_OFFLOAD_CONT]]
 375 // CHECK3:       omp_offload.cont:
 376 // CHECK3-NEXT:    [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
 377 // CHECK3-NEXT:    [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
 378 // CHECK3-NEXT:    call void @__tgt_target_data_end_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP42]], ptr [[TMP43]], ptr @.offload_sizes.1, ptr @.offload_maptypes.2, ptr null, ptr null)
 379 // CHECK3-NEXT:    [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
 380 // CHECK3-NEXT:    [[TMP45:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
 381 // CHECK3-NEXT:    call void @__tgt_target_data_end_mapper(ptr @[[GLOB1]], i64 -1, i32 1, ptr [[TMP44]], ptr [[TMP45]], ptr @.offload_sizes, ptr @.offload_maptypes, ptr null, ptr null)
 382 // CHECK3-NEXT:    ret i32 0
 383 //
 384 //
 385 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l44
 386 // CHECK3-SAME: (ptr noundef [[B:%.*]], ptr noundef [[A:%.*]]) #[[ATTR2:[0-9]+]] {
 387 // CHECK3-NEXT:  entry:
 388 // CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 389 // CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 390 // CHECK3-NEXT:    [[AGG_CAPTURED:%.*]] = alloca [[STRUCT_ANON:%.*]], align 4
 391 // CHECK3-NEXT:    [[DOTAFFS_ARR_ADDR:%.*]] = alloca [1 x %struct.kmp_task_affinity_info_t], align 4
 392 // CHECK3-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 393 // CHECK3-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 394 // CHECK3-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 395 // CHECK3-NEXT:    [[TMP1:%.*]] = getelementptr inbounds nuw [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 0
 396 // CHECK3-NEXT:    store ptr [[B_ADDR]], ptr [[TMP1]], align 4
 397 // CHECK3-NEXT:    [[TMP2:%.*]] = call ptr @__kmpc_omp_task_alloc(ptr @[[GLOB1]], i32 [[TMP0]], i32 1, i32 24, i32 4, ptr @.omp_task_entry.)
 398 // CHECK3-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x %struct.kmp_task_affinity_info_t], ptr [[DOTAFFS_ARR_ADDR]], i32 0, i32 0
 399 // CHECK3-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 400 // CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds nuw i32, ptr [[TMP4]], i32 0
 401 // CHECK3-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 402 // CHECK3-NEXT:    [[ARRAYIDX1:%.*]] = getelementptr inbounds nuw i32, ptr [[TMP5]], i32 1023
 403 // CHECK3-NEXT:    [[TMP6:%.*]] = getelementptr i32, ptr [[ARRAYIDX1]], i32 1
 404 // CHECK3-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[ARRAYIDX]] to i32
 405 // CHECK3-NEXT:    [[TMP8:%.*]] = ptrtoint ptr [[TMP6]] to i32
 406 // CHECK3-NEXT:    [[TMP9:%.*]] = sub nuw i32 [[TMP8]], [[TMP7]]
 407 // CHECK3-NEXT:    [[TMP10:%.*]] = getelementptr [[STRUCT_KMP_TASK_AFFINITY_INFO_T:%.*]], ptr [[TMP3]], i32 0
 408 // CHECK3-NEXT:    [[TMP11:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_AFFINITY_INFO_T]], ptr [[TMP10]], i32 0, i32 0
 409 // CHECK3-NEXT:    [[TMP12:%.*]] = ptrtoint ptr [[ARRAYIDX]] to i32
 410 // CHECK3-NEXT:    store i32 [[TMP12]], ptr [[TMP11]], align 4
 411 // CHECK3-NEXT:    [[TMP13:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_AFFINITY_INFO_T]], ptr [[TMP10]], i32 0, i32 1
 412 // CHECK3-NEXT:    store i32 [[TMP9]], ptr [[TMP13]], align 4
 413 // CHECK3-NEXT:    [[TMP14:%.*]] = call i32 @__kmpc_omp_reg_task_with_affinity(ptr @[[GLOB1]], i32 [[TMP0]], ptr [[TMP2]], i32 1, ptr [[TMP3]])
 414 // CHECK3-NEXT:    [[TMP15:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], ptr [[TMP2]], i32 0, i32 0
 415 // CHECK3-NEXT:    [[TMP16:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T:%.*]], ptr [[TMP15]], i32 0, i32 0
 416 // CHECK3-NEXT:    [[TMP17:%.*]] = load ptr, ptr [[TMP16]], align 4
 417 // CHECK3-NEXT:    call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[TMP17]], ptr align 4 [[AGG_CAPTURED]], i32 4, i1 false)
 418 // CHECK3-NEXT:    [[TMP18:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T_WITH_PRIVATES]], ptr [[TMP2]], i32 0, i32 1
 419 // CHECK3-NEXT:    [[TMP19:%.*]] = getelementptr inbounds nuw [[STRUCT__KMP_PRIVATES_T:%.*]], ptr [[TMP18]], i32 0, i32 0
 420 // CHECK3-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 421 // CHECK3-NEXT:    store ptr [[TMP20]], ptr [[TMP19]], align 4
 422 // CHECK3-NEXT:    [[TMP21:%.*]] = call i32 @__kmpc_omp_task(ptr @[[GLOB1]], i32 [[TMP0]], ptr [[TMP2]])
 423 // CHECK3-NEXT:    [[TMP22:%.*]] = call i32 @__kmpc_omp_taskwait(ptr @[[GLOB1]], i32 [[TMP0]])
 424 // CHECK3-NEXT:    ret void
 425 //
 426 //
 427 // CHECK3-LABEL: define {{[^@]+}}@.omp_task_privates_map.
 428 // CHECK3-SAME: (ptr noalias noundef [[TMP0:%.*]], ptr noalias noundef [[TMP1:%.*]]) #[[ATTR3:[0-9]+]] {
 429 // CHECK3-NEXT:  entry:
 430 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 4
 431 // CHECK3-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 4
 432 // CHECK3-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 4
 433 // CHECK3-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 4
 434 // CHECK3-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 4
 435 // CHECK3-NEXT:    [[TMP3:%.*]] = getelementptr inbounds nuw [[STRUCT__KMP_PRIVATES_T:%.*]], ptr [[TMP2]], i32 0, i32 0
 436 // CHECK3-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR1]], align 4
 437 // CHECK3-NEXT:    store ptr [[TMP3]], ptr [[TMP4]], align 4
 438 // CHECK3-NEXT:    ret void
 439 //
 440 //
 441 // CHECK3-LABEL: define {{[^@]+}}@.omp_task_entry.
 442 // CHECK3-SAME: (i32 noundef [[TMP0:%.*]], ptr noalias noundef [[TMP1:%.*]]) #[[ATTR4:[0-9]+]] {
 443 // CHECK3-NEXT:  entry:
 444 // CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR_I:%.*]] = alloca i32, align 4
 445 // CHECK3-NEXT:    [[DOTPART_ID__ADDR_I:%.*]] = alloca ptr, align 4
 446 // CHECK3-NEXT:    [[DOTPRIVATES__ADDR_I:%.*]] = alloca ptr, align 4
 447 // CHECK3-NEXT:    [[DOTCOPY_FN__ADDR_I:%.*]] = alloca ptr, align 4
 448 // CHECK3-NEXT:    [[DOTTASK_T__ADDR_I:%.*]] = alloca ptr, align 4
 449 // CHECK3-NEXT:    [[__CONTEXT_ADDR_I:%.*]] = alloca ptr, align 4
 450 // CHECK3-NEXT:    [[DOTFIRSTPRIV_PTR_ADDR_I:%.*]] = alloca ptr, align 4
 451 // CHECK3-NEXT:    [[I_I:%.*]] = alloca i32, align 4
 452 // CHECK3-NEXT:    [[DOTADDR:%.*]] = alloca i32, align 4
 453 // CHECK3-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 4
 454 // CHECK3-NEXT:    store i32 [[TMP0]], ptr [[DOTADDR]], align 4
 455 // CHECK3-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 4
 456 // CHECK3-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTADDR]], align 4
 457 // CHECK3-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR1]], align 4
 458 // CHECK3-NEXT:    [[TMP4:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], ptr [[TMP3]], i32 0, i32 0
 459 // CHECK3-NEXT:    [[TMP5:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T:%.*]], ptr [[TMP4]], i32 0, i32 2
 460 // CHECK3-NEXT:    [[TMP6:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T]], ptr [[TMP4]], i32 0, i32 0
 461 // CHECK3-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[TMP6]], align 4
 462 // CHECK3-NEXT:    [[TMP8:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T_WITH_PRIVATES]], ptr [[TMP3]], i32 0, i32 1
 463 // CHECK3-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META5:![0-9]+]])
 464 // CHECK3-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META8:![0-9]+]])
 465 // CHECK3-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META10:![0-9]+]])
 466 // CHECK3-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META12:![0-9]+]])
 467 // CHECK3-NEXT:    store i32 [[TMP2]], ptr [[DOTGLOBAL_TID__ADDR_I]], align 4, !noalias [[META14:![0-9]+]]
 468 // CHECK3-NEXT:    store ptr [[TMP5]], ptr [[DOTPART_ID__ADDR_I]], align 4, !noalias [[META14]]
 469 // CHECK3-NEXT:    store ptr [[TMP8]], ptr [[DOTPRIVATES__ADDR_I]], align 4, !noalias [[META14]]
 470 // CHECK3-NEXT:    store ptr @.omp_task_privates_map., ptr [[DOTCOPY_FN__ADDR_I]], align 4, !noalias [[META14]]
 471 // CHECK3-NEXT:    store ptr [[TMP3]], ptr [[DOTTASK_T__ADDR_I]], align 4, !noalias [[META14]]
 472 // CHECK3-NEXT:    store ptr [[TMP7]], ptr [[__CONTEXT_ADDR_I]], align 4, !noalias [[META14]]
 473 // CHECK3-NEXT:    [[TMP9:%.*]] = load ptr, ptr [[__CONTEXT_ADDR_I]], align 4, !noalias [[META14]]
 474 // CHECK3-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTCOPY_FN__ADDR_I]], align 4, !noalias [[META14]]
 475 // CHECK3-NEXT:    [[TMP11:%.*]] = load ptr, ptr [[DOTPRIVATES__ADDR_I]], align 4, !noalias [[META14]]
 476 // CHECK3-NEXT:    call void [[TMP10]](ptr [[TMP11]], ptr [[DOTFIRSTPRIV_PTR_ADDR_I]]) #[[ATTR1]]
 477 // CHECK3-NEXT:    [[TMP12:%.*]] = load ptr, ptr [[DOTFIRSTPRIV_PTR_ADDR_I]], align 4, !noalias [[META14]]
 478 // CHECK3-NEXT:    store i32 0, ptr [[I_I]], align 4, !noalias [[META14]]
 479 // CHECK3-NEXT:    br label [[FOR_COND_I:%.*]]
 480 // CHECK3:       for.cond.i:
 481 // CHECK3-NEXT:    [[TMP13:%.*]] = load i32, ptr [[I_I]], align 4, !noalias [[META14]]
 482 // CHECK3-NEXT:    [[CMP_I:%.*]] = icmp slt i32 [[TMP13]], 1024
 483 // CHECK3-NEXT:    br i1 [[CMP_I]], label [[FOR_BODY_I:%.*]], label [[DOTOMP_OUTLINED__EXIT:%.*]]
 484 // CHECK3:       for.body.i:
 485 // CHECK3-NEXT:    [[TMP14:%.*]] = load ptr, ptr [[TMP12]], align 4
 486 // CHECK3-NEXT:    [[TMP15:%.*]] = load i32, ptr [[I_I]], align 4, !noalias [[META14]]
 487 // CHECK3-NEXT:    [[ARRAYIDX_I:%.*]] = getelementptr inbounds i32, ptr [[TMP14]], i32 [[TMP15]]
 488 // CHECK3-NEXT:    [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX_I]], align 4
 489 // CHECK3-NEXT:    [[MUL_I:%.*]] = mul nsw i32 2, [[TMP16]]
 490 // CHECK3-NEXT:    [[TMP17:%.*]] = load ptr, ptr [[TMP9]], align 4
 491 // CHECK3-NEXT:    [[TMP18:%.*]] = load ptr, ptr [[TMP17]], align 4
 492 // CHECK3-NEXT:    [[TMP19:%.*]] = load i32, ptr [[I_I]], align 4, !noalias [[META14]]
 493 // CHECK3-NEXT:    [[ARRAYIDX1_I:%.*]] = getelementptr inbounds i32, ptr [[TMP18]], i32 [[TMP19]]
 494 // CHECK3-NEXT:    store i32 [[MUL_I]], ptr [[ARRAYIDX1_I]], align 4
 495 // CHECK3-NEXT:    [[TMP20:%.*]] = load i32, ptr [[I_I]], align 4, !noalias [[META14]]
 496 // CHECK3-NEXT:    [[INC_I:%.*]] = add nsw i32 [[TMP20]], 1
 497 // CHECK3-NEXT:    store i32 [[INC_I]], ptr [[I_I]], align 4, !noalias [[META14]]
 498 // CHECK3-NEXT:    br label [[FOR_COND_I]], !llvm.loop [[LOOP15:![0-9]+]]
 499 // CHECK3:       .omp_outlined..exit:
 500 // CHECK3-NEXT:    ret i32 0
 501 //
 502 //
 503 // CHECK5-LABEL: define {{[^@]+}}@main
 504 // CHECK5-SAME: () #[[ATTR0:[0-9]+]] {
 505 // CHECK5-NEXT:  entry:
 506 // CHECK5-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
 507 // CHECK5-NEXT:    [[A:%.*]] = alloca ptr, align 8
 508 // CHECK5-NEXT:    [[B:%.*]] = alloca ptr, align 8
 509 // CHECK5-NEXT:    [[I:%.*]] = alloca i32, align 4
 510 // CHECK5-NEXT:    store i32 0, ptr [[RETVAL]], align 4
 511 // CHECK5-NEXT:    store i32 0, ptr [[I]], align 4
 512 // CHECK5-NEXT:    br label [[FOR_COND:%.*]]
 513 // CHECK5:       for.cond:
 514 // CHECK5-NEXT:    [[TMP0:%.*]] = load i32, ptr [[I]], align 4
 515 // CHECK5-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP0]], 1024
 516 // CHECK5-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
 517 // CHECK5:       for.body:
 518 // CHECK5-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A]], align 8
 519 // CHECK5-NEXT:    [[TMP2:%.*]] = load i32, ptr [[I]], align 4
 520 // CHECK5-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP2]] to i64
 521 // CHECK5-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i64 [[IDXPROM]]
 522 // CHECK5-NEXT:    [[TMP3:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 523 // CHECK5-NEXT:    [[MUL:%.*]] = mul nsw i32 2, [[TMP3]]
 524 // CHECK5-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[B]], align 8
 525 // CHECK5-NEXT:    [[TMP5:%.*]] = load i32, ptr [[I]], align 4
 526 // CHECK5-NEXT:    [[IDXPROM1:%.*]] = sext i32 [[TMP5]] to i64
 527 // CHECK5-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, ptr [[TMP4]], i64 [[IDXPROM1]]
 528 // CHECK5-NEXT:    store i32 [[MUL]], ptr [[ARRAYIDX2]], align 4
 529 // CHECK5-NEXT:    br label [[FOR_INC:%.*]]
 530 // CHECK5:       for.inc:
 531 // CHECK5-NEXT:    [[TMP6:%.*]] = load i32, ptr [[I]], align 4
 532 // CHECK5-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP6]], 1
 533 // CHECK5-NEXT:    store i32 [[INC]], ptr [[I]], align 4
 534 // CHECK5-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP2:![0-9]+]]
 535 // CHECK5:       for.end:
 536 // CHECK5-NEXT:    ret i32 0
 537 //
 538 //
 539 // CHECK7-LABEL: define {{[^@]+}}@main
 540 // CHECK7-SAME: () #[[ATTR0:[0-9]+]] {
 541 // CHECK7-NEXT:  entry:
 542 // CHECK7-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
 543 // CHECK7-NEXT:    [[A:%.*]] = alloca ptr, align 4
 544 // CHECK7-NEXT:    [[B:%.*]] = alloca ptr, align 4
 545 // CHECK7-NEXT:    [[I:%.*]] = alloca i32, align 4
 546 // CHECK7-NEXT:    store i32 0, ptr [[RETVAL]], align 4
 547 // CHECK7-NEXT:    store i32 0, ptr [[I]], align 4
 548 // CHECK7-NEXT:    br label [[FOR_COND:%.*]]
 549 // CHECK7:       for.cond:
 550 // CHECK7-NEXT:    [[TMP0:%.*]] = load i32, ptr [[I]], align 4
 551 // CHECK7-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP0]], 1024
 552 // CHECK7-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
 553 // CHECK7:       for.body:
 554 // CHECK7-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A]], align 4
 555 // CHECK7-NEXT:    [[TMP2:%.*]] = load i32, ptr [[I]], align 4
 556 // CHECK7-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 [[TMP2]]
 557 // CHECK7-NEXT:    [[TMP3:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 558 // CHECK7-NEXT:    [[MUL:%.*]] = mul nsw i32 2, [[TMP3]]
 559 // CHECK7-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[B]], align 4
 560 // CHECK7-NEXT:    [[TMP5:%.*]] = load i32, ptr [[I]], align 4
 561 // CHECK7-NEXT:    [[ARRAYIDX1:%.*]] = getelementptr inbounds i32, ptr [[TMP4]], i32 [[TMP5]]
 562 // CHECK7-NEXT:    store i32 [[MUL]], ptr [[ARRAYIDX1]], align 4
 563 // CHECK7-NEXT:    br label [[FOR_INC:%.*]]
 564 // CHECK7:       for.inc:
 565 // CHECK7-NEXT:    [[TMP6:%.*]] = load i32, ptr [[I]], align 4
 566 // CHECK7-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP6]], 1
 567 // CHECK7-NEXT:    store i32 [[INC]], ptr [[I]], align 4
 568 // CHECK7-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP3:![0-9]+]]
 569 // CHECK7:       for.end:
 570 // CHECK7-NEXT:    ret i32 0
 571 //
 572 //
 573 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l44
 574 // CHECK9-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], ptr noundef [[B:%.*]], ptr noundef [[A:%.*]]) #[[ATTR0:[0-9]+]] {
 575 // CHECK9-NEXT:  entry:
 576 // CHECK9-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8
 577 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 578 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 579 // CHECK9-NEXT:    [[AGG_CAPTURED:%.*]] = alloca [[STRUCT_ANON:%.*]], align 8
 580 // CHECK9-NEXT:    [[DOTAFFS_ARR_ADDR:%.*]] = alloca [1 x %struct.kmp_task_affinity_info_t], align 8
 581 // CHECK9-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 582 // CHECK9-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 8
 583 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 584 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 585 // CHECK9-NEXT:    [[TMP1:%.*]] = getelementptr inbounds nuw [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 0
 586 // CHECK9-NEXT:    store ptr [[B_ADDR]], ptr [[TMP1]], align 8
 587 // CHECK9-NEXT:    [[TMP2:%.*]] = call ptr @__kmpc_omp_task_alloc(ptr @[[GLOB1]], i32 [[TMP0]], i32 1, i64 48, i64 8, ptr @.omp_task_entry.)
 588 // CHECK9-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x %struct.kmp_task_affinity_info_t], ptr [[DOTAFFS_ARR_ADDR]], i64 0, i64 0
 589 // CHECK9-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 590 // CHECK9-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds nuw i32, ptr [[TMP4]], i64 0
 591 // CHECK9-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 592 // CHECK9-NEXT:    [[ARRAYIDX1:%.*]] = getelementptr inbounds nuw i32, ptr [[TMP5]], i64 1023
 593 // CHECK9-NEXT:    [[TMP6:%.*]] = getelementptr i32, ptr [[ARRAYIDX1]], i32 1
 594 // CHECK9-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[ARRAYIDX]] to i64
 595 // CHECK9-NEXT:    [[TMP8:%.*]] = ptrtoint ptr [[TMP6]] to i64
 596 // CHECK9-NEXT:    [[TMP9:%.*]] = sub nuw i64 [[TMP8]], [[TMP7]]
 597 // CHECK9-NEXT:    [[TMP10:%.*]] = getelementptr [[STRUCT_KMP_TASK_AFFINITY_INFO_T:%.*]], ptr [[TMP3]], i64 0
 598 // CHECK9-NEXT:    [[TMP11:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_AFFINITY_INFO_T]], ptr [[TMP10]], i32 0, i32 0
 599 // CHECK9-NEXT:    [[TMP12:%.*]] = ptrtoint ptr [[ARRAYIDX]] to i64
 600 // CHECK9-NEXT:    store i64 [[TMP12]], ptr [[TMP11]], align 8
 601 // CHECK9-NEXT:    [[TMP13:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_AFFINITY_INFO_T]], ptr [[TMP10]], i32 0, i32 1
 602 // CHECK9-NEXT:    store i64 [[TMP9]], ptr [[TMP13]], align 8
 603 // CHECK9-NEXT:    [[TMP14:%.*]] = call i32 @__kmpc_omp_reg_task_with_affinity(ptr @[[GLOB1]], i32 [[TMP0]], ptr [[TMP2]], i32 1, ptr [[TMP3]])
 604 // CHECK9-NEXT:    [[TMP15:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], ptr [[TMP2]], i32 0, i32 0
 605 // CHECK9-NEXT:    [[TMP16:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T:%.*]], ptr [[TMP15]], i32 0, i32 0
 606 // CHECK9-NEXT:    [[TMP17:%.*]] = load ptr, ptr [[TMP16]], align 8
 607 // CHECK9-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[TMP17]], ptr align 8 [[AGG_CAPTURED]], i64 8, i1 false)
 608 // CHECK9-NEXT:    [[TMP18:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T_WITH_PRIVATES]], ptr [[TMP2]], i32 0, i32 1
 609 // CHECK9-NEXT:    [[TMP19:%.*]] = getelementptr inbounds nuw [[STRUCT__KMP_PRIVATES_T:%.*]], ptr [[TMP18]], i32 0, i32 0
 610 // CHECK9-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 611 // CHECK9-NEXT:    store ptr [[TMP20]], ptr [[TMP19]], align 8
 612 // CHECK9-NEXT:    [[TMP21:%.*]] = call i32 @__kmpc_omp_task(ptr @[[GLOB1]], i32 [[TMP0]], ptr [[TMP2]])
 613 // CHECK9-NEXT:    [[TMP22:%.*]] = call i32 @__kmpc_omp_taskwait(ptr @[[GLOB1]], i32 [[TMP0]])
 614 // CHECK9-NEXT:    ret void
 615 //
 616 //
 617 // CHECK9-LABEL: define {{[^@]+}}@.omp_task_privates_map.
 618 // CHECK9-SAME: (ptr noalias noundef [[TMP0:%.*]], ptr noalias noundef [[TMP1:%.*]]) #[[ATTR1:[0-9]+]] {
 619 // CHECK9-NEXT:  entry:
 620 // CHECK9-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 621 // CHECK9-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 622 // CHECK9-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 623 // CHECK9-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 624 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 625 // CHECK9-NEXT:    [[TMP3:%.*]] = getelementptr inbounds nuw [[STRUCT__KMP_PRIVATES_T:%.*]], ptr [[TMP2]], i32 0, i32 0
 626 // CHECK9-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 627 // CHECK9-NEXT:    store ptr [[TMP3]], ptr [[TMP4]], align 8
 628 // CHECK9-NEXT:    ret void
 629 //
 630 //
 631 // CHECK9-LABEL: define {{[^@]+}}@.omp_task_entry.
 632 // CHECK9-SAME: (i32 noundef signext [[TMP0:%.*]], ptr noalias noundef [[TMP1:%.*]]) #[[ATTR2:[0-9]+]] {
 633 // CHECK9-NEXT:  entry:
 634 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR_I:%.*]] = alloca i32, align 4
 635 // CHECK9-NEXT:    [[DOTPART_ID__ADDR_I:%.*]] = alloca ptr, align 8
 636 // CHECK9-NEXT:    [[DOTPRIVATES__ADDR_I:%.*]] = alloca ptr, align 8
 637 // CHECK9-NEXT:    [[DOTCOPY_FN__ADDR_I:%.*]] = alloca ptr, align 8
 638 // CHECK9-NEXT:    [[DOTTASK_T__ADDR_I:%.*]] = alloca ptr, align 8
 639 // CHECK9-NEXT:    [[__CONTEXT_ADDR_I:%.*]] = alloca ptr, align 8
 640 // CHECK9-NEXT:    [[DOTFIRSTPRIV_PTR_ADDR_I:%.*]] = alloca ptr, align 8
 641 // CHECK9-NEXT:    [[I_I:%.*]] = alloca i32, align 4
 642 // CHECK9-NEXT:    [[DOTADDR:%.*]] = alloca i32, align 4
 643 // CHECK9-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 644 // CHECK9-NEXT:    store i32 [[TMP0]], ptr [[DOTADDR]], align 4
 645 // CHECK9-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 646 // CHECK9-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTADDR]], align 4
 647 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 648 // CHECK9-NEXT:    [[TMP4:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], ptr [[TMP3]], i32 0, i32 0
 649 // CHECK9-NEXT:    [[TMP5:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T:%.*]], ptr [[TMP4]], i32 0, i32 2
 650 // CHECK9-NEXT:    [[TMP6:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T]], ptr [[TMP4]], i32 0, i32 0
 651 // CHECK9-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[TMP6]], align 8
 652 // CHECK9-NEXT:    [[TMP8:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T_WITH_PRIVATES]], ptr [[TMP3]], i32 0, i32 1
 653 // CHECK9-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META5:![0-9]+]])
 654 // CHECK9-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META8:![0-9]+]])
 655 // CHECK9-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META10:![0-9]+]])
 656 // CHECK9-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META12:![0-9]+]])
 657 // CHECK9-NEXT:    store i32 [[TMP2]], ptr [[DOTGLOBAL_TID__ADDR_I]], align 4, !noalias [[META14:![0-9]+]]
 658 // CHECK9-NEXT:    store ptr [[TMP5]], ptr [[DOTPART_ID__ADDR_I]], align 8, !noalias [[META14]]
 659 // CHECK9-NEXT:    store ptr [[TMP8]], ptr [[DOTPRIVATES__ADDR_I]], align 8, !noalias [[META14]]
 660 // CHECK9-NEXT:    store ptr @.omp_task_privates_map., ptr [[DOTCOPY_FN__ADDR_I]], align 8, !noalias [[META14]]
 661 // CHECK9-NEXT:    store ptr [[TMP3]], ptr [[DOTTASK_T__ADDR_I]], align 8, !noalias [[META14]]
 662 // CHECK9-NEXT:    store ptr [[TMP7]], ptr [[__CONTEXT_ADDR_I]], align 8, !noalias [[META14]]
 663 // CHECK9-NEXT:    [[TMP9:%.*]] = load ptr, ptr [[__CONTEXT_ADDR_I]], align 8, !noalias [[META14]]
 664 // CHECK9-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTCOPY_FN__ADDR_I]], align 8, !noalias [[META14]]
 665 // CHECK9-NEXT:    [[TMP11:%.*]] = load ptr, ptr [[DOTPRIVATES__ADDR_I]], align 8, !noalias [[META14]]
 666 // CHECK9-NEXT:    call void [[TMP10]](ptr [[TMP11]], ptr [[DOTFIRSTPRIV_PTR_ADDR_I]]) #[[ATTR3:[0-9]+]]
 667 // CHECK9-NEXT:    [[TMP12:%.*]] = load ptr, ptr [[DOTFIRSTPRIV_PTR_ADDR_I]], align 8, !noalias [[META14]]
 668 // CHECK9-NEXT:    store i32 0, ptr [[I_I]], align 4, !noalias [[META14]]
 669 // CHECK9-NEXT:    br label [[FOR_COND_I:%.*]]
 670 // CHECK9:       for.cond.i:
 671 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[I_I]], align 4, !noalias [[META14]]
 672 // CHECK9-NEXT:    [[CMP_I:%.*]] = icmp slt i32 [[TMP13]], 1024
 673 // CHECK9-NEXT:    br i1 [[CMP_I]], label [[FOR_BODY_I:%.*]], label [[DOTOMP_OUTLINED__EXIT:%.*]]
 674 // CHECK9:       for.body.i:
 675 // CHECK9-NEXT:    [[TMP14:%.*]] = load ptr, ptr [[TMP12]], align 8
 676 // CHECK9-NEXT:    [[TMP15:%.*]] = load i32, ptr [[I_I]], align 4, !noalias [[META14]]
 677 // CHECK9-NEXT:    [[IDXPROM_I:%.*]] = sext i32 [[TMP15]] to i64
 678 // CHECK9-NEXT:    [[ARRAYIDX_I:%.*]] = getelementptr inbounds i32, ptr [[TMP14]], i64 [[IDXPROM_I]]
 679 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX_I]], align 4
 680 // CHECK9-NEXT:    [[MUL_I:%.*]] = mul nsw i32 2, [[TMP16]]
 681 // CHECK9-NEXT:    [[TMP17:%.*]] = load ptr, ptr [[TMP9]], align 8
 682 // CHECK9-NEXT:    [[TMP18:%.*]] = load ptr, ptr [[TMP17]], align 8
 683 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[I_I]], align 4, !noalias [[META14]]
 684 // CHECK9-NEXT:    [[IDXPROM1_I:%.*]] = sext i32 [[TMP19]] to i64
 685 // CHECK9-NEXT:    [[ARRAYIDX2_I:%.*]] = getelementptr inbounds i32, ptr [[TMP18]], i64 [[IDXPROM1_I]]
 686 // CHECK9-NEXT:    store i32 [[MUL_I]], ptr [[ARRAYIDX2_I]], align 4
 687 // CHECK9-NEXT:    [[TMP20:%.*]] = load i32, ptr [[I_I]], align 4, !noalias [[META14]]
 688 // CHECK9-NEXT:    [[INC_I:%.*]] = add nsw i32 [[TMP20]], 1
 689 // CHECK9-NEXT:    store i32 [[INC_I]], ptr [[I_I]], align 4, !noalias [[META14]]
 690 // CHECK9-NEXT:    br label [[FOR_COND_I]], !llvm.loop [[LOOP15:![0-9]+]]
 691 // CHECK9:       .omp_outlined..exit:
 692 // CHECK9-NEXT:    ret i32 0
 693 //
 694 //
 695 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l44
 696 // CHECK11-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], ptr noundef [[B:%.*]], ptr noundef [[A:%.*]]) #[[ATTR0:[0-9]+]] {
 697 // CHECK11-NEXT:  entry:
 698 // CHECK11-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 4
 699 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 700 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 701 // CHECK11-NEXT:    [[AGG_CAPTURED:%.*]] = alloca [[STRUCT_ANON:%.*]], align 4
 702 // CHECK11-NEXT:    [[DOTAFFS_ARR_ADDR:%.*]] = alloca [1 x %struct.kmp_task_affinity_info_t], align 4
 703 // CHECK11-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 704 // CHECK11-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 4
 705 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 706 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 707 // CHECK11-NEXT:    [[TMP1:%.*]] = getelementptr inbounds nuw [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 0
 708 // CHECK11-NEXT:    store ptr [[B_ADDR]], ptr [[TMP1]], align 4
 709 // CHECK11-NEXT:    [[TMP2:%.*]] = call ptr @__kmpc_omp_task_alloc(ptr @[[GLOB1]], i32 [[TMP0]], i32 1, i32 24, i32 4, ptr @.omp_task_entry.)
 710 // CHECK11-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x %struct.kmp_task_affinity_info_t], ptr [[DOTAFFS_ARR_ADDR]], i32 0, i32 0
 711 // CHECK11-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 712 // CHECK11-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds nuw i32, ptr [[TMP4]], i32 0
 713 // CHECK11-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 714 // CHECK11-NEXT:    [[ARRAYIDX1:%.*]] = getelementptr inbounds nuw i32, ptr [[TMP5]], i32 1023
 715 // CHECK11-NEXT:    [[TMP6:%.*]] = getelementptr i32, ptr [[ARRAYIDX1]], i32 1
 716 // CHECK11-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[ARRAYIDX]] to i32
 717 // CHECK11-NEXT:    [[TMP8:%.*]] = ptrtoint ptr [[TMP6]] to i32
 718 // CHECK11-NEXT:    [[TMP9:%.*]] = sub nuw i32 [[TMP8]], [[TMP7]]
 719 // CHECK11-NEXT:    [[TMP10:%.*]] = getelementptr [[STRUCT_KMP_TASK_AFFINITY_INFO_T:%.*]], ptr [[TMP3]], i32 0
 720 // CHECK11-NEXT:    [[TMP11:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_AFFINITY_INFO_T]], ptr [[TMP10]], i32 0, i32 0
 721 // CHECK11-NEXT:    [[TMP12:%.*]] = ptrtoint ptr [[ARRAYIDX]] to i32
 722 // CHECK11-NEXT:    store i32 [[TMP12]], ptr [[TMP11]], align 4
 723 // CHECK11-NEXT:    [[TMP13:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_AFFINITY_INFO_T]], ptr [[TMP10]], i32 0, i32 1
 724 // CHECK11-NEXT:    store i32 [[TMP9]], ptr [[TMP13]], align 4
 725 // CHECK11-NEXT:    [[TMP14:%.*]] = call i32 @__kmpc_omp_reg_task_with_affinity(ptr @[[GLOB1]], i32 [[TMP0]], ptr [[TMP2]], i32 1, ptr [[TMP3]])
 726 // CHECK11-NEXT:    [[TMP15:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], ptr [[TMP2]], i32 0, i32 0
 727 // CHECK11-NEXT:    [[TMP16:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T:%.*]], ptr [[TMP15]], i32 0, i32 0
 728 // CHECK11-NEXT:    [[TMP17:%.*]] = load ptr, ptr [[TMP16]], align 4
 729 // CHECK11-NEXT:    call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[TMP17]], ptr align 4 [[AGG_CAPTURED]], i32 4, i1 false)
 730 // CHECK11-NEXT:    [[TMP18:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T_WITH_PRIVATES]], ptr [[TMP2]], i32 0, i32 1
 731 // CHECK11-NEXT:    [[TMP19:%.*]] = getelementptr inbounds nuw [[STRUCT__KMP_PRIVATES_T:%.*]], ptr [[TMP18]], i32 0, i32 0
 732 // CHECK11-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 733 // CHECK11-NEXT:    store ptr [[TMP20]], ptr [[TMP19]], align 4
 734 // CHECK11-NEXT:    [[TMP21:%.*]] = call i32 @__kmpc_omp_task(ptr @[[GLOB1]], i32 [[TMP0]], ptr [[TMP2]])
 735 // CHECK11-NEXT:    [[TMP22:%.*]] = call i32 @__kmpc_omp_taskwait(ptr @[[GLOB1]], i32 [[TMP0]])
 736 // CHECK11-NEXT:    ret void
 737 //
 738 //
 739 // CHECK11-LABEL: define {{[^@]+}}@.omp_task_privates_map.
 740 // CHECK11-SAME: (ptr noalias noundef [[TMP0:%.*]], ptr noalias noundef [[TMP1:%.*]]) #[[ATTR1:[0-9]+]] {
 741 // CHECK11-NEXT:  entry:
 742 // CHECK11-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 4
 743 // CHECK11-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 4
 744 // CHECK11-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 4
 745 // CHECK11-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 4
 746 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 4
 747 // CHECK11-NEXT:    [[TMP3:%.*]] = getelementptr inbounds nuw [[STRUCT__KMP_PRIVATES_T:%.*]], ptr [[TMP2]], i32 0, i32 0
 748 // CHECK11-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTADDR1]], align 4
 749 // CHECK11-NEXT:    store ptr [[TMP3]], ptr [[TMP4]], align 4
 750 // CHECK11-NEXT:    ret void
 751 //
 752 //
 753 // CHECK11-LABEL: define {{[^@]+}}@.omp_task_entry.
 754 // CHECK11-SAME: (i32 noundef [[TMP0:%.*]], ptr noalias noundef [[TMP1:%.*]]) #[[ATTR2:[0-9]+]] {
 755 // CHECK11-NEXT:  entry:
 756 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR_I:%.*]] = alloca i32, align 4
 757 // CHECK11-NEXT:    [[DOTPART_ID__ADDR_I:%.*]] = alloca ptr, align 4
 758 // CHECK11-NEXT:    [[DOTPRIVATES__ADDR_I:%.*]] = alloca ptr, align 4
 759 // CHECK11-NEXT:    [[DOTCOPY_FN__ADDR_I:%.*]] = alloca ptr, align 4
 760 // CHECK11-NEXT:    [[DOTTASK_T__ADDR_I:%.*]] = alloca ptr, align 4
 761 // CHECK11-NEXT:    [[__CONTEXT_ADDR_I:%.*]] = alloca ptr, align 4
 762 // CHECK11-NEXT:    [[DOTFIRSTPRIV_PTR_ADDR_I:%.*]] = alloca ptr, align 4
 763 // CHECK11-NEXT:    [[I_I:%.*]] = alloca i32, align 4
 764 // CHECK11-NEXT:    [[DOTADDR:%.*]] = alloca i32, align 4
 765 // CHECK11-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 4
 766 // CHECK11-NEXT:    store i32 [[TMP0]], ptr [[DOTADDR]], align 4
 767 // CHECK11-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 4
 768 // CHECK11-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTADDR]], align 4
 769 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR1]], align 4
 770 // CHECK11-NEXT:    [[TMP4:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], ptr [[TMP3]], i32 0, i32 0
 771 // CHECK11-NEXT:    [[TMP5:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T:%.*]], ptr [[TMP4]], i32 0, i32 2
 772 // CHECK11-NEXT:    [[TMP6:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T]], ptr [[TMP4]], i32 0, i32 0
 773 // CHECK11-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[TMP6]], align 4
 774 // CHECK11-NEXT:    [[TMP8:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T_WITH_PRIVATES]], ptr [[TMP3]], i32 0, i32 1
 775 // CHECK11-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META6:![0-9]+]])
 776 // CHECK11-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META9:![0-9]+]])
 777 // CHECK11-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META11:![0-9]+]])
 778 // CHECK11-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META13:![0-9]+]])
 779 // CHECK11-NEXT:    store i32 [[TMP2]], ptr [[DOTGLOBAL_TID__ADDR_I]], align 4, !noalias [[META15:![0-9]+]]
 780 // CHECK11-NEXT:    store ptr [[TMP5]], ptr [[DOTPART_ID__ADDR_I]], align 4, !noalias [[META15]]
 781 // CHECK11-NEXT:    store ptr [[TMP8]], ptr [[DOTPRIVATES__ADDR_I]], align 4, !noalias [[META15]]
 782 // CHECK11-NEXT:    store ptr @.omp_task_privates_map., ptr [[DOTCOPY_FN__ADDR_I]], align 4, !noalias [[META15]]
 783 // CHECK11-NEXT:    store ptr [[TMP3]], ptr [[DOTTASK_T__ADDR_I]], align 4, !noalias [[META15]]
 784 // CHECK11-NEXT:    store ptr [[TMP7]], ptr [[__CONTEXT_ADDR_I]], align 4, !noalias [[META15]]
 785 // CHECK11-NEXT:    [[TMP9:%.*]] = load ptr, ptr [[__CONTEXT_ADDR_I]], align 4, !noalias [[META15]]
 786 // CHECK11-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTCOPY_FN__ADDR_I]], align 4, !noalias [[META15]]
 787 // CHECK11-NEXT:    [[TMP11:%.*]] = load ptr, ptr [[DOTPRIVATES__ADDR_I]], align 4, !noalias [[META15]]
 788 // CHECK11-NEXT:    call void [[TMP10]](ptr [[TMP11]], ptr [[DOTFIRSTPRIV_PTR_ADDR_I]]) #[[ATTR3:[0-9]+]]
 789 // CHECK11-NEXT:    [[TMP12:%.*]] = load ptr, ptr [[DOTFIRSTPRIV_PTR_ADDR_I]], align 4, !noalias [[META15]]
 790 // CHECK11-NEXT:    store i32 0, ptr [[I_I]], align 4, !noalias [[META15]]
 791 // CHECK11-NEXT:    br label [[FOR_COND_I:%.*]]
 792 // CHECK11:       for.cond.i:
 793 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[I_I]], align 4, !noalias [[META15]]
 794 // CHECK11-NEXT:    [[CMP_I:%.*]] = icmp slt i32 [[TMP13]], 1024
 795 // CHECK11-NEXT:    br i1 [[CMP_I]], label [[FOR_BODY_I:%.*]], label [[DOTOMP_OUTLINED__EXIT:%.*]]
 796 // CHECK11:       for.body.i:
 797 // CHECK11-NEXT:    [[TMP14:%.*]] = load ptr, ptr [[TMP12]], align 4
 798 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[I_I]], align 4, !noalias [[META15]]
 799 // CHECK11-NEXT:    [[ARRAYIDX_I:%.*]] = getelementptr inbounds i32, ptr [[TMP14]], i32 [[TMP15]]
 800 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX_I]], align 4
 801 // CHECK11-NEXT:    [[MUL_I:%.*]] = mul nsw i32 2, [[TMP16]]
 802 // CHECK11-NEXT:    [[TMP17:%.*]] = load ptr, ptr [[TMP9]], align 4
 803 // CHECK11-NEXT:    [[TMP18:%.*]] = load ptr, ptr [[TMP17]], align 4
 804 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[I_I]], align 4, !noalias [[META15]]
 805 // CHECK11-NEXT:    [[ARRAYIDX1_I:%.*]] = getelementptr inbounds i32, ptr [[TMP18]], i32 [[TMP19]]
 806 // CHECK11-NEXT:    store i32 [[MUL_I]], ptr [[ARRAYIDX1_I]], align 4
 807 // CHECK11-NEXT:    [[TMP20:%.*]] = load i32, ptr [[I_I]], align 4, !noalias [[META15]]
 808 // CHECK11-NEXT:    [[INC_I:%.*]] = add nsw i32 [[TMP20]], 1
 809 // CHECK11-NEXT:    store i32 [[INC_I]], ptr [[I_I]], align 4, !noalias [[META15]]
 810 // CHECK11-NEXT:    br label [[FOR_COND_I]], !llvm.loop [[LOOP16:![0-9]+]]
 811 // CHECK11:       .omp_outlined..exit:
 812 // CHECK11-NEXT:    ret i32 0
 813 //