clang/test/OpenMP/task_in_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 -no-enable-noundef-analysis -verify -Wno-vla -triple x86_64-apple-darwin10 -fopenmp -x c++ -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK1
   3 // RUN: %clang_cc1 -no-enable-noundef-analysis -fopenmp -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s
   4 // RUN: %clang_cc1 -no-enable-noundef-analysis -fopenmp -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK1
   5
   6 // RUN: %clang_cc1 -no-enable-noundef-analysis -verify -Wno-vla -triple x86_64-apple-darwin10 -fopenmp-simd -x c++ -emit-llvm %s -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
   7 // RUN: %clang_cc1 -no-enable-noundef-analysis -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s
   8 // RUN: %clang_cc1 -no-enable-noundef-analysis -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
   9 // expected-no-diagnostics
  10 #ifndef HEADER
  11 #define HEADER
  12
  13 typedef void **omp_allocator_handle_t;
  14 extern const omp_allocator_handle_t omp_null_allocator;
  15 extern const omp_allocator_handle_t omp_default_mem_alloc;
  16 extern const omp_allocator_handle_t omp_large_cap_mem_alloc;
  17 extern const omp_allocator_handle_t omp_const_mem_alloc;
  18 extern const omp_allocator_handle_t omp_high_bw_mem_alloc;
  19 extern const omp_allocator_handle_t omp_low_lat_mem_alloc;
  20 extern const omp_allocator_handle_t omp_cgroup_mem_alloc;
  21 extern const omp_allocator_handle_t omp_pteam_mem_alloc;
  22 extern const omp_allocator_handle_t omp_thread_mem_alloc;
  23
  24
  25 struct S {
  26   int a;
  27   S() : a(0) {}
  28   S(const S&) {}
  29   S& operator=(const S&) {return *this;}
  30   ~S() {}
  31   friend S operator+(const S&a, const S&b) {return a;}
  32 };
  33
  34
  35 int main(int argc, char **argv) {
  36   int a;
  37   float b;
  38   S c[5];
  39   short d[argc];
  40 #pragma omp taskgroup task_reduction(+: a, b, argc)
  41   {
  42 #pragma omp taskgroup task_reduction(-:c, d)
  43 #pragma omp parallel
  44 #pragma omp task in_reduction(+:a) in_reduction(-:d) allocate(omp_high_bw_mem_alloc: d)
  45     a += d[a];
  46   }
  47 #pragma omp task in_reduction(+:a)
  48   ++a;
  49   return 0;
  50 }
  51
  52
  53
  54
  55 #endif
  56 // CHECK1-LABEL: define {{[^@]+}}@main
  57 // CHECK1-SAME: (i32 [[ARGC:%.*]], ptr [[ARGV:%.*]]) #[[ATTR0:[0-9]+]] {
  58 // CHECK1-NEXT:  entry:
  59 // CHECK1-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
  60 // CHECK1-NEXT:    [[ARGC_ADDR:%.*]] = alloca i32, align 4
  61 // CHECK1-NEXT:    [[ARGV_ADDR:%.*]] = alloca ptr, align 8
  62 // CHECK1-NEXT:    [[A:%.*]] = alloca i32, align 4
  63 // CHECK1-NEXT:    [[B:%.*]] = alloca float, align 4
  64 // CHECK1-NEXT:    [[C:%.*]] = alloca [5 x %struct.S], align 16
  65 // CHECK1-NEXT:    [[SAVED_STACK:%.*]] = alloca ptr, align 8
  66 // CHECK1-NEXT:    [[__VLA_EXPR0:%.*]] = alloca i64, align 8
  67 // CHECK1-NEXT:    [[DOTRD_INPUT_:%.*]] = alloca [3 x %struct.kmp_taskred_input_t], align 8
  68 // CHECK1-NEXT:    [[DOTTASK_RED_:%.*]] = alloca ptr, align 8
  69 // CHECK1-NEXT:    [[DOTRD_INPUT_3:%.*]] = alloca [2 x %struct.kmp_taskred_input_t.0], align 8
  70 // CHECK1-NEXT:    [[DOTTASK_RED_6:%.*]] = alloca ptr, align 8
  71 // CHECK1-NEXT:    [[AGG_CAPTURED:%.*]] = alloca [[STRUCT_ANON_1:%.*]], align 8
  72 // CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
  73 // CHECK1-NEXT:    store i32 0, ptr [[RETVAL]], align 4
  74 // CHECK1-NEXT:    store i32 [[ARGC]], ptr [[ARGC_ADDR]], align 4
  75 // CHECK1-NEXT:    store ptr [[ARGV]], ptr [[ARGV_ADDR]], align 8
  76 // CHECK1-NEXT:    [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [5 x %struct.S], ptr [[C]], i32 0, i32 0
  77 // CHECK1-NEXT:    [[ARRAYCTOR_END:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[ARRAY_BEGIN]], i64 5
  78 // CHECK1-NEXT:    br label [[ARRAYCTOR_LOOP:%.*]]
  79 // CHECK1:       arrayctor.loop:
  80 // CHECK1-NEXT:    [[ARRAYCTOR_CUR:%.*]] = phi ptr [ [[ARRAY_BEGIN]], [[ENTRY:%.*]] ], [ [[ARRAYCTOR_NEXT:%.*]], [[ARRAYCTOR_LOOP]] ]
  81 // CHECK1-NEXT:    call void @_ZN1SC1Ev(ptr nonnull align 4 dereferenceable(4) [[ARRAYCTOR_CUR]])
  82 // CHECK1-NEXT:    [[ARRAYCTOR_NEXT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYCTOR_CUR]], i64 1
  83 // CHECK1-NEXT:    [[ARRAYCTOR_DONE:%.*]] = icmp eq ptr [[ARRAYCTOR_NEXT]], [[ARRAYCTOR_END]]
  84 // CHECK1-NEXT:    br i1 [[ARRAYCTOR_DONE]], label [[ARRAYCTOR_CONT:%.*]], label [[ARRAYCTOR_LOOP]]
  85 // CHECK1:       arrayctor.cont:
  86 // CHECK1-NEXT:    [[TMP1:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
  87 // CHECK1-NEXT:    [[TMP2:%.*]] = zext i32 [[TMP1]] to i64
  88 // CHECK1-NEXT:    [[TMP3:%.*]] = call ptr @llvm.stacksave.p0()
  89 // CHECK1-NEXT:    store ptr [[TMP3]], ptr [[SAVED_STACK]], align 8
  90 // CHECK1-NEXT:    [[VLA:%.*]] = alloca i16, i64 [[TMP2]], align 16
  91 // CHECK1-NEXT:    store i64 [[TMP2]], ptr [[__VLA_EXPR0]], align 8
  92 // CHECK1-NEXT:    call void @__kmpc_taskgroup(ptr @[[GLOB1]], i32 [[TMP0]])
  93 // CHECK1-NEXT:    [[DOTRD_INPUT_GEP_:%.*]] = getelementptr inbounds [3 x %struct.kmp_taskred_input_t], ptr [[DOTRD_INPUT_]], i64 0, i64 0
  94 // CHECK1-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T:%.*]], ptr [[DOTRD_INPUT_GEP_]], i32 0, i32 0
  95 // CHECK1-NEXT:    store ptr [[A]], ptr [[TMP4]], align 8
  96 // CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_]], i32 0, i32 1
  97 // CHECK1-NEXT:    store ptr [[A]], ptr [[TMP5]], align 8
  98 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_]], i32 0, i32 2
  99 // CHECK1-NEXT:    store i64 4, ptr [[TMP6]], align 8
 100 // CHECK1-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_]], i32 0, i32 3
 101 // CHECK1-NEXT:    store ptr @.red_init., ptr [[TMP7]], align 8
 102 // CHECK1-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_]], i32 0, i32 4
 103 // CHECK1-NEXT:    store ptr null, ptr [[TMP8]], align 8
 104 // CHECK1-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_]], i32 0, i32 5
 105 // CHECK1-NEXT:    store ptr @.red_comb., ptr [[TMP9]], align 8
 106 // CHECK1-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_]], i32 0, i32 6
 107 // CHECK1-NEXT:    call void @llvm.memset.p0.i64(ptr align 8 [[TMP10]], i8 0, i64 4, i1 false)
 108 // CHECK1-NEXT:    [[DOTRD_INPUT_GEP_1:%.*]] = getelementptr inbounds [3 x %struct.kmp_taskred_input_t], ptr [[DOTRD_INPUT_]], i64 0, i64 1
 109 // CHECK1-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_1]], i32 0, i32 0
 110 // CHECK1-NEXT:    store ptr [[B]], ptr [[TMP11]], align 8
 111 // CHECK1-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_1]], i32 0, i32 1
 112 // CHECK1-NEXT:    store ptr [[B]], ptr [[TMP12]], align 8
 113 // CHECK1-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_1]], i32 0, i32 2
 114 // CHECK1-NEXT:    store i64 4, ptr [[TMP13]], align 8
 115 // CHECK1-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_1]], i32 0, i32 3
 116 // CHECK1-NEXT:    store ptr @.red_init..1, ptr [[TMP14]], align 8
 117 // CHECK1-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_1]], i32 0, i32 4
 118 // CHECK1-NEXT:    store ptr null, ptr [[TMP15]], align 8
 119 // CHECK1-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_1]], i32 0, i32 5
 120 // CHECK1-NEXT:    store ptr @.red_comb..2, ptr [[TMP16]], align 8
 121 // CHECK1-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_1]], i32 0, i32 6
 122 // CHECK1-NEXT:    call void @llvm.memset.p0.i64(ptr align 8 [[TMP17]], i8 0, i64 4, i1 false)
 123 // CHECK1-NEXT:    [[DOTRD_INPUT_GEP_2:%.*]] = getelementptr inbounds [3 x %struct.kmp_taskred_input_t], ptr [[DOTRD_INPUT_]], i64 0, i64 2
 124 // CHECK1-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_2]], i32 0, i32 0
 125 // CHECK1-NEXT:    store ptr [[ARGC_ADDR]], ptr [[TMP18]], align 8
 126 // CHECK1-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_2]], i32 0, i32 1
 127 // CHECK1-NEXT:    store ptr [[ARGC_ADDR]], ptr [[TMP19]], align 8
 128 // CHECK1-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_2]], i32 0, i32 2
 129 // CHECK1-NEXT:    store i64 4, ptr [[TMP20]], align 8
 130 // CHECK1-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_2]], i32 0, i32 3
 131 // CHECK1-NEXT:    store ptr @.red_init..3, ptr [[TMP21]], align 8
 132 // CHECK1-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_2]], i32 0, i32 4
 133 // CHECK1-NEXT:    store ptr null, ptr [[TMP22]], align 8
 134 // CHECK1-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_2]], i32 0, i32 5
 135 // CHECK1-NEXT:    store ptr @.red_comb..4, ptr [[TMP23]], align 8
 136 // CHECK1-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T]], ptr [[DOTRD_INPUT_GEP_2]], i32 0, i32 6
 137 // CHECK1-NEXT:    call void @llvm.memset.p0.i64(ptr align 8 [[TMP24]], i8 0, i64 4, i1 false)
 138 // CHECK1-NEXT:    [[TMP25:%.*]] = call ptr @__kmpc_taskred_init(i32 [[TMP0]], i32 3, ptr [[DOTRD_INPUT_]])
 139 // CHECK1-NEXT:    store ptr [[TMP25]], ptr [[DOTTASK_RED_]], align 8
 140 // CHECK1-NEXT:    call void @__kmpc_taskgroup(ptr @[[GLOB1]], i32 [[TMP0]])
 141 // CHECK1-NEXT:    [[DOTRD_INPUT_GEP_4:%.*]] = getelementptr inbounds [2 x %struct.kmp_taskred_input_t.0], ptr [[DOTRD_INPUT_3]], i64 0, i64 0
 142 // CHECK1-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0:%.*]], ptr [[DOTRD_INPUT_GEP_4]], i32 0, i32 0
 143 // CHECK1-NEXT:    store ptr [[C]], ptr [[TMP26]], align 8
 144 // CHECK1-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_4]], i32 0, i32 1
 145 // CHECK1-NEXT:    store ptr [[C]], ptr [[TMP27]], align 8
 146 // CHECK1-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_4]], i32 0, i32 2
 147 // CHECK1-NEXT:    store i64 20, ptr [[TMP28]], align 8
 148 // CHECK1-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_4]], i32 0, i32 3
 149 // CHECK1-NEXT:    store ptr @.red_init..5, ptr [[TMP29]], align 8
 150 // CHECK1-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_4]], i32 0, i32 4
 151 // CHECK1-NEXT:    store ptr @.red_fini., ptr [[TMP30]], align 8
 152 // CHECK1-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_4]], i32 0, i32 5
 153 // CHECK1-NEXT:    store ptr @.red_comb..6, ptr [[TMP31]], align 8
 154 // CHECK1-NEXT:    [[TMP32:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_4]], i32 0, i32 6
 155 // CHECK1-NEXT:    call void @llvm.memset.p0.i64(ptr align 8 [[TMP32]], i8 0, i64 4, i1 false)
 156 // CHECK1-NEXT:    [[DOTRD_INPUT_GEP_5:%.*]] = getelementptr inbounds [2 x %struct.kmp_taskred_input_t.0], ptr [[DOTRD_INPUT_3]], i64 0, i64 1
 157 // CHECK1-NEXT:    [[TMP33:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_5]], i32 0, i32 0
 158 // CHECK1-NEXT:    store ptr [[VLA]], ptr [[TMP33]], align 8
 159 // CHECK1-NEXT:    [[TMP34:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_5]], i32 0, i32 1
 160 // CHECK1-NEXT:    store ptr [[VLA]], ptr [[TMP34]], align 8
 161 // CHECK1-NEXT:    [[TMP35:%.*]] = mul nuw i64 [[TMP2]], 2
 162 // CHECK1-NEXT:    [[TMP36:%.*]] = udiv exact i64 [[TMP35]], ptrtoint (ptr getelementptr (i16, ptr null, i32 1) to i64)
 163 // CHECK1-NEXT:    [[TMP37:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_5]], i32 0, i32 2
 164 // CHECK1-NEXT:    store i64 [[TMP35]], ptr [[TMP37]], align 8
 165 // CHECK1-NEXT:    [[TMP38:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_5]], i32 0, i32 3
 166 // CHECK1-NEXT:    store ptr @.red_init..7, ptr [[TMP38]], align 8
 167 // CHECK1-NEXT:    [[TMP39:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_5]], i32 0, i32 4
 168 // CHECK1-NEXT:    store ptr null, ptr [[TMP39]], align 8
 169 // CHECK1-NEXT:    [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_5]], i32 0, i32 5
 170 // CHECK1-NEXT:    store ptr @.red_comb..8, ptr [[TMP40]], align 8
 171 // CHECK1-NEXT:    [[TMP41:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASKRED_INPUT_T_0]], ptr [[DOTRD_INPUT_GEP_5]], i32 0, i32 6
 172 // CHECK1-NEXT:    store i32 1, ptr [[TMP41]], align 8
 173 // CHECK1-NEXT:    [[TMP42:%.*]] = call ptr @__kmpc_taskred_init(i32 [[TMP0]], i32 2, ptr [[DOTRD_INPUT_3]])
 174 // CHECK1-NEXT:    store ptr [[TMP42]], ptr [[DOTTASK_RED_6]], align 8
 175 // CHECK1-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 5, ptr @main.omp_outlined, ptr [[A]], i64 [[TMP2]], ptr [[VLA]], ptr [[DOTTASK_RED_]], ptr [[DOTTASK_RED_6]])
 176 // CHECK1-NEXT:    call void @__kmpc_end_taskgroup(ptr @[[GLOB1]], i32 [[TMP0]])
 177 // CHECK1-NEXT:    call void @__kmpc_end_taskgroup(ptr @[[GLOB1]], i32 [[TMP0]])
 178 // CHECK1-NEXT:    [[TMP43:%.*]] = getelementptr inbounds [[STRUCT_ANON_1]], ptr [[AGG_CAPTURED]], i32 0, i32 0
 179 // CHECK1-NEXT:    store ptr [[A]], ptr [[TMP43]], align 8
 180 // CHECK1-NEXT:    [[TMP44:%.*]] = call ptr @__kmpc_omp_task_alloc(ptr @[[GLOB1]], i32 [[TMP0]], i32 1, i64 40, i64 8, ptr @.omp_task_entry..10)
 181 // CHECK1-NEXT:    [[TMP45:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES_2:%.*]], ptr [[TMP44]], i32 0, i32 0
 182 // CHECK1-NEXT:    [[TMP46:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T:%.*]], ptr [[TMP45]], i32 0, i32 0
 183 // CHECK1-NEXT:    [[TMP47:%.*]] = load ptr, ptr [[TMP46]], align 8
 184 // CHECK1-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[TMP47]], ptr align 8 [[AGG_CAPTURED]], i64 8, i1 false)
 185 // CHECK1-NEXT:    [[TMP48:%.*]] = call i32 @__kmpc_omp_task(ptr @[[GLOB1]], i32 [[TMP0]], ptr [[TMP44]])
 186 // CHECK1-NEXT:    store i32 0, ptr [[RETVAL]], align 4
 187 // CHECK1-NEXT:    [[TMP49:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
 188 // CHECK1-NEXT:    call void @llvm.stackrestore.p0(ptr [[TMP49]])
 189 // CHECK1-NEXT:    [[ARRAY_BEGIN7:%.*]] = getelementptr inbounds [5 x %struct.S], ptr [[C]], i32 0, i32 0
 190 // CHECK1-NEXT:    [[TMP50:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAY_BEGIN7]], i64 5
 191 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 192 // CHECK1:       arraydestroy.body:
 193 // CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP50]], [[ARRAYCTOR_CONT]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
 194 // CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 195 // CHECK1-NEXT:    call void @_ZN1SD1Ev(ptr nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3:[0-9]+]]
 196 // CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN7]]
 197 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE8:%.*]], label [[ARRAYDESTROY_BODY]]
 198 // CHECK1:       arraydestroy.done8:
 199 // CHECK1-NEXT:    [[TMP51:%.*]] = load i32, ptr [[RETVAL]], align 4
 200 // CHECK1-NEXT:    ret i32 [[TMP51]]
 201 //
 202 //
 203 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SC1Ev
 204 // CHECK1-SAME: (ptr nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1:[0-9]+]] align 2 {
 205 // CHECK1-NEXT:  entry:
 206 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 207 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 208 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 209 // CHECK1-NEXT:    call void @_ZN1SC2Ev(ptr nonnull align 4 dereferenceable(4) [[THIS1]])
 210 // CHECK1-NEXT:    ret void
 211 //
 212 //
 213 // CHECK1-LABEL: define {{[^@]+}}@.red_init.
 214 // CHECK1-SAME: (ptr noalias [[TMP0:%.*]], ptr noalias [[TMP1:%.*]]) #[[ATTR5:[0-9]+]] {
 215 // CHECK1-NEXT:  entry:
 216 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 217 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 218 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 219 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 220 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 221 // CHECK1-NEXT:    store i32 0, ptr [[TMP2]], align 4
 222 // CHECK1-NEXT:    ret void
 223 //
 224 //
 225 // CHECK1-LABEL: define {{[^@]+}}@.red_comb.
 226 // CHECK1-SAME: (ptr [[TMP0:%.*]], ptr [[TMP1:%.*]]) #[[ATTR5]] {
 227 // CHECK1-NEXT:  entry:
 228 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 229 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 230 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 231 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 232 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 233 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 234 // CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP2]], align 4
 235 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP3]], align 4
 236 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP4]], [[TMP5]]
 237 // CHECK1-NEXT:    store i32 [[ADD]], ptr [[TMP2]], align 4
 238 // CHECK1-NEXT:    ret void
 239 //
 240 //
 241 // CHECK1-LABEL: define {{[^@]+}}@.red_init..1
 242 // CHECK1-SAME: (ptr noalias [[TMP0:%.*]], ptr noalias [[TMP1:%.*]]) #[[ATTR5]] {
 243 // CHECK1-NEXT:  entry:
 244 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 245 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 246 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 247 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 248 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 249 // CHECK1-NEXT:    store float 0.000000e+00, ptr [[TMP2]], align 4
 250 // CHECK1-NEXT:    ret void
 251 //
 252 //
 253 // CHECK1-LABEL: define {{[^@]+}}@.red_comb..2
 254 // CHECK1-SAME: (ptr [[TMP0:%.*]], ptr [[TMP1:%.*]]) #[[ATTR5]] {
 255 // CHECK1-NEXT:  entry:
 256 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 257 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 258 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 259 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 260 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 261 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 262 // CHECK1-NEXT:    [[TMP4:%.*]] = load float, ptr [[TMP2]], align 4
 263 // CHECK1-NEXT:    [[TMP5:%.*]] = load float, ptr [[TMP3]], align 4
 264 // CHECK1-NEXT:    [[ADD:%.*]] = fadd float [[TMP4]], [[TMP5]]
 265 // CHECK1-NEXT:    store float [[ADD]], ptr [[TMP2]], align 4
 266 // CHECK1-NEXT:    ret void
 267 //
 268 //
 269 // CHECK1-LABEL: define {{[^@]+}}@.red_init..3
 270 // CHECK1-SAME: (ptr noalias [[TMP0:%.*]], ptr noalias [[TMP1:%.*]]) #[[ATTR5]] {
 271 // CHECK1-NEXT:  entry:
 272 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 273 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 274 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 275 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 276 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 277 // CHECK1-NEXT:    store i32 0, ptr [[TMP2]], align 4
 278 // CHECK1-NEXT:    ret void
 279 //
 280 //
 281 // CHECK1-LABEL: define {{[^@]+}}@.red_comb..4
 282 // CHECK1-SAME: (ptr [[TMP0:%.*]], ptr [[TMP1:%.*]]) #[[ATTR5]] {
 283 // CHECK1-NEXT:  entry:
 284 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 285 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 286 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 287 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 288 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 289 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 290 // CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP2]], align 4
 291 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP3]], align 4
 292 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP4]], [[TMP5]]
 293 // CHECK1-NEXT:    store i32 [[ADD]], ptr [[TMP2]], align 4
 294 // CHECK1-NEXT:    ret void
 295 //
 296 //
 297 // CHECK1-LABEL: define {{[^@]+}}@.red_init..5
 298 // CHECK1-SAME: (ptr noalias [[TMP0:%.*]], ptr noalias [[TMP1:%.*]]) #[[ATTR5]] {
 299 // CHECK1-NEXT:  entry:
 300 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 301 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 302 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 303 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 304 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 305 // CHECK1-NEXT:    [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [5 x %struct.S], ptr [[TMP2]], i32 0, i32 0
 306 // CHECK1-NEXT:    [[TMP3:%.*]] = getelementptr [[STRUCT_S:%.*]], ptr [[ARRAY_BEGIN]], i64 5
 307 // CHECK1-NEXT:    [[OMP_ARRAYINIT_ISEMPTY:%.*]] = icmp eq ptr [[ARRAY_BEGIN]], [[TMP3]]
 308 // CHECK1-NEXT:    br i1 [[OMP_ARRAYINIT_ISEMPTY]], label [[OMP_ARRAYINIT_DONE:%.*]], label [[OMP_ARRAYINIT_BODY:%.*]]
 309 // CHECK1:       omp.arrayinit.body:
 310 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi ptr [ [[ARRAY_BEGIN]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYINIT_BODY]] ]
 311 // CHECK1-NEXT:    call void @_ZN1SC1Ev(ptr nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_DESTELEMENTPAST]])
 312 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr [[STRUCT_S]], ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
 313 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq ptr [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP3]]
 314 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYINIT_DONE]], label [[OMP_ARRAYINIT_BODY]]
 315 // CHECK1:       omp.arrayinit.done:
 316 // CHECK1-NEXT:    ret void
 317 //
 318 //
 319 // CHECK1-LABEL: define {{[^@]+}}@.red_fini.
 320 // CHECK1-SAME: (ptr [[TMP0:%.*]]) #[[ATTR5]] {
 321 // CHECK1-NEXT:  entry:
 322 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 323 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 324 // CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 325 // CHECK1-NEXT:    [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [5 x %struct.S], ptr [[TMP1]], i32 0, i32 0
 326 // CHECK1-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[ARRAY_BEGIN]], i64 5
 327 // CHECK1-NEXT:    br label [[ARRAYDESTROY_BODY:%.*]]
 328 // CHECK1:       arraydestroy.body:
 329 // CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi ptr [ [[TMP2]], [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
 330 // CHECK1-NEXT:    [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], ptr [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
 331 // CHECK1-NEXT:    call void @_ZN1SD1Ev(ptr nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
 332 // CHECK1-NEXT:    [[ARRAYDESTROY_DONE:%.*]] = icmp eq ptr [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
 333 // CHECK1-NEXT:    br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
 334 // CHECK1:       arraydestroy.done1:
 335 // CHECK1-NEXT:    ret void
 336 //
 337 //
 338 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SD1Ev
 339 // CHECK1-SAME: (ptr nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
 340 // CHECK1-NEXT:  entry:
 341 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 342 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 343 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 344 // CHECK1-NEXT:    call void @_ZN1SD2Ev(ptr nonnull align 4 dereferenceable(4) [[THIS1]]) #[[ATTR3]]
 345 // CHECK1-NEXT:    ret void
 346 //
 347 //
 348 // CHECK1-LABEL: define {{[^@]+}}@.red_comb..6
 349 // CHECK1-SAME: (ptr [[TMP0:%.*]], ptr [[TMP1:%.*]]) #[[ATTR5]] {
 350 // CHECK1-NEXT:  entry:
 351 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 352 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 353 // CHECK1-NEXT:    [[REF_TMP:%.*]] = alloca [[STRUCT_S:%.*]], align 4
 354 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 355 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 356 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 357 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 358 // CHECK1-NEXT:    [[TMP4:%.*]] = getelementptr [[STRUCT_S]], ptr [[TMP2]], i64 5
 359 // CHECK1-NEXT:    [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq ptr [[TMP2]], [[TMP4]]
 360 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE2:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
 361 // CHECK1:       omp.arraycpy.body:
 362 // CHECK1-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi ptr [ [[TMP3]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
 363 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi ptr [ [[TMP2]], [[ENTRY]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
 364 // CHECK1-NEXT:    call void @_ZplRK1SS1_(ptr sret([[STRUCT_S]]) align 4 [[REF_TMP]], ptr nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_DESTELEMENTPAST]], ptr nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_SRCELEMENTPAST]])
 365 // CHECK1-NEXT:    [[CALL:%.*]] = call nonnull align 4 dereferenceable(4) ptr @_ZN1SaSERKS_(ptr nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_DESTELEMENTPAST]], ptr nonnull align 4 dereferenceable(4) [[REF_TMP]])
 366 // CHECK1-NEXT:    call void @_ZN1SD1Ev(ptr nonnull align 4 dereferenceable(4) [[REF_TMP]]) #[[ATTR3]]
 367 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr [[STRUCT_S]], ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
 368 // CHECK1-NEXT:    [[OMP_ARRAYCPY_SRC_ELEMENT]] = getelementptr [[STRUCT_S]], ptr [[OMP_ARRAYCPY_SRCELEMENTPAST]], i32 1
 369 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq ptr [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP4]]
 370 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYCPY_DONE2]], label [[OMP_ARRAYCPY_BODY]]
 371 // CHECK1:       omp.arraycpy.done2:
 372 // CHECK1-NEXT:    ret void
 373 //
 374 //
 375 // CHECK1-LABEL: define {{[^@]+}}@_ZplRK1SS1_
 376 // CHECK1-SAME: (ptr noalias sret([[STRUCT_S:%.*]]) align 4 [[AGG_RESULT:%.*]], ptr nonnull align 4 dereferenceable(4) [[A:%.*]], ptr nonnull align 4 dereferenceable(4) [[B:%.*]]) #[[ATTR7:[0-9]+]] {
 377 // CHECK1-NEXT:  entry:
 378 // CHECK1-NEXT:    [[RESULT_PTR:%.*]] = alloca ptr, align 8
 379 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 380 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 381 // CHECK1-NEXT:    store ptr [[AGG_RESULT]], ptr [[RESULT_PTR]], align 8
 382 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 383 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 384 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 385 // CHECK1-NEXT:    call void @_ZN1SC1ERKS_(ptr nonnull align 4 dereferenceable(4) [[AGG_RESULT]], ptr nonnull align 4 dereferenceable(4) [[TMP0]])
 386 // CHECK1-NEXT:    ret void
 387 //
 388 //
 389 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SaSERKS_
 390 // CHECK1-SAME: (ptr nonnull align 4 dereferenceable(4) [[THIS:%.*]], ptr nonnull align 4 dereferenceable(4) [[TMP0:%.*]]) #[[ATTR7]] align 2 {
 391 // CHECK1-NEXT:  entry:
 392 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 393 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 394 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 395 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 396 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 397 // CHECK1-NEXT:    ret ptr [[THIS1]]
 398 //
 399 //
 400 // CHECK1-LABEL: define {{[^@]+}}@.red_init..7
 401 // CHECK1-SAME: (ptr noalias [[TMP0:%.*]], ptr noalias [[TMP1:%.*]]) #[[ATTR5]] {
 402 // CHECK1-NEXT:  entry:
 403 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 404 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 405 // CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 406 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 407 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 408 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 409 // CHECK1-NEXT:    [[TMP4:%.*]] = call ptr @__kmpc_threadprivate_cached(ptr @[[GLOB1]], i32 [[TMP2]], ptr @{{reduction_size[.].+[.]}})
 410 // CHECK1-NEXT:    [[TMP5:%.*]] = load i64, ptr [[TMP4]], align 8
 411 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr i16, ptr [[TMP3]], i64 [[TMP5]]
 412 // CHECK1-NEXT:    [[OMP_ARRAYINIT_ISEMPTY:%.*]] = icmp eq ptr [[TMP3]], [[TMP6]]
 413 // CHECK1-NEXT:    br i1 [[OMP_ARRAYINIT_ISEMPTY]], label [[OMP_ARRAYINIT_DONE:%.*]], label [[OMP_ARRAYINIT_BODY:%.*]]
 414 // CHECK1:       omp.arrayinit.body:
 415 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi ptr [ [[TMP3]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYINIT_BODY]] ]
 416 // CHECK1-NEXT:    store i16 0, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 2
 417 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr i16, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
 418 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq ptr [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP6]]
 419 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYINIT_DONE]], label [[OMP_ARRAYINIT_BODY]]
 420 // CHECK1:       omp.arrayinit.done:
 421 // CHECK1-NEXT:    ret void
 422 //
 423 //
 424 // CHECK1-LABEL: define {{[^@]+}}@.red_comb..8
 425 // CHECK1-SAME: (ptr [[TMP0:%.*]], ptr [[TMP1:%.*]]) #[[ATTR5]] {
 426 // CHECK1-NEXT:  entry:
 427 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 428 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 429 // CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 430 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 431 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 432 // CHECK1-NEXT:    [[TMP3:%.*]] = call ptr @__kmpc_threadprivate_cached(ptr @[[GLOB1]], i32 [[TMP2]], ptr @{{reduction_size[.].+[.]}})
 433 // CHECK1-NEXT:    [[TMP4:%.*]] = load i64, ptr [[TMP3]], align 8
 434 // CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 435 // CHECK1-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 436 // CHECK1-NEXT:    [[TMP7:%.*]] = getelementptr i16, ptr [[TMP5]], i64 [[TMP4]]
 437 // CHECK1-NEXT:    [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq ptr [[TMP5]], [[TMP7]]
 438 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE4:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
 439 // CHECK1:       omp.arraycpy.body:
 440 // CHECK1-NEXT:    [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi ptr [ [[TMP6]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
 441 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi ptr [ [[TMP5]], [[ENTRY]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
 442 // CHECK1-NEXT:    [[TMP8:%.*]] = load i16, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 2
 443 // CHECK1-NEXT:    [[CONV:%.*]] = sext i16 [[TMP8]] to i32
 444 // CHECK1-NEXT:    [[TMP9:%.*]] = load i16, ptr [[OMP_ARRAYCPY_SRCELEMENTPAST]], align 2
 445 // CHECK1-NEXT:    [[CONV2:%.*]] = sext i16 [[TMP9]] to i32
 446 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[CONV]], [[CONV2]]
 447 // CHECK1-NEXT:    [[CONV3:%.*]] = trunc i32 [[ADD]] to i16
 448 // CHECK1-NEXT:    store i16 [[CONV3]], ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], align 2
 449 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr i16, ptr [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
 450 // CHECK1-NEXT:    [[OMP_ARRAYCPY_SRC_ELEMENT]] = getelementptr i16, ptr [[OMP_ARRAYCPY_SRCELEMENTPAST]], i32 1
 451 // CHECK1-NEXT:    [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq ptr [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP7]]
 452 // CHECK1-NEXT:    br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYCPY_DONE4]], label [[OMP_ARRAYCPY_BODY]]
 453 // CHECK1:       omp.arraycpy.done4:
 454 // CHECK1-NEXT:    ret void
 455 //
 456 //
 457 // CHECK1-LABEL: define {{[^@]+}}@main.omp_outlined
 458 // CHECK1-SAME: (ptr noalias [[DOTGLOBAL_TID_:%.*]], ptr noalias [[DOTBOUND_TID_:%.*]], ptr nonnull align 4 dereferenceable(4) [[A:%.*]], i64 [[VLA:%.*]], ptr nonnull align 2 dereferenceable(2) [[D:%.*]], ptr nonnull align 8 dereferenceable(8) [[DOTTASK_RED_:%.*]], ptr nonnull align 8 dereferenceable(8) [[DOTTASK_RED_1:%.*]]) #[[ATTR8:[0-9]+]] {
 459 // CHECK1-NEXT:  entry:
 460 // CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 461 // CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 462 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 463 // CHECK1-NEXT:    [[VLA_ADDR:%.*]] = alloca i64, align 8
 464 // CHECK1-NEXT:    [[D_ADDR:%.*]] = alloca ptr, align 8
 465 // CHECK1-NEXT:    [[DOTTASK_RED__ADDR:%.*]] = alloca ptr, align 8
 466 // CHECK1-NEXT:    [[DOTTASK_RED__ADDR2:%.*]] = alloca ptr, align 8
 467 // CHECK1-NEXT:    [[AGG_CAPTURED:%.*]] = alloca [[STRUCT_ANON:%.*]], align 8
 468 // CHECK1-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 469 // CHECK1-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 470 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 471 // CHECK1-NEXT:    store i64 [[VLA]], ptr [[VLA_ADDR]], align 8
 472 // CHECK1-NEXT:    store ptr [[D]], ptr [[D_ADDR]], align 8
 473 // CHECK1-NEXT:    store ptr [[DOTTASK_RED_]], ptr [[DOTTASK_RED__ADDR]], align 8
 474 // CHECK1-NEXT:    store ptr [[DOTTASK_RED_1]], ptr [[DOTTASK_RED__ADDR2]], align 8
 475 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 476 // CHECK1-NEXT:    [[TMP1:%.*]] = load i64, ptr [[VLA_ADDR]], align 8
 477 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[D_ADDR]], align 8
 478 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTTASK_RED__ADDR]], align 8
 479 // CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[DOTTASK_RED__ADDR2]], align 8
 480 // CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 0
 481 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[TMP5]], align 8
 482 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 1
 483 // CHECK1-NEXT:    store i64 [[TMP1]], ptr [[TMP6]], align 8
 484 // CHECK1-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 2
 485 // CHECK1-NEXT:    store ptr [[TMP2]], ptr [[TMP7]], align 8
 486 // CHECK1-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 3
 487 // CHECK1-NEXT:    store ptr [[TMP3]], ptr [[TMP8]], align 8
 488 // CHECK1-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[AGG_CAPTURED]], i32 0, i32 4
 489 // CHECK1-NEXT:    store ptr [[TMP4]], ptr [[TMP9]], align 8
 490 // CHECK1-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 491 // CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 492 // CHECK1-NEXT:    [[TMP12:%.*]] = call ptr @__kmpc_omp_task_alloc(ptr @[[GLOB1]], i32 [[TMP11]], i32 1, i64 56, i64 40, ptr @.omp_task_entry.)
 493 // CHECK1-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], ptr [[TMP12]], i32 0, i32 0
 494 // CHECK1-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T:%.*]], ptr [[TMP13]], i32 0, i32 0
 495 // CHECK1-NEXT:    [[TMP15:%.*]] = load ptr, ptr [[TMP14]], align 8
 496 // CHECK1-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[TMP15]], ptr align 8 [[AGG_CAPTURED]], i64 40, i1 false)
 497 // CHECK1-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES]], ptr [[TMP12]], i32 0, i32 1
 498 // CHECK1-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T:%.*]], ptr [[TMP16]], i32 0, i32 0
 499 // CHECK1-NEXT:    [[TMP18:%.*]] = load ptr, ptr [[TMP3]], align 8
 500 // CHECK1-NEXT:    store ptr [[TMP18]], ptr [[TMP17]], align 8
 501 // CHECK1-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP16]], i32 0, i32 1
 502 // CHECK1-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[TMP4]], align 8
 503 // CHECK1-NEXT:    store ptr [[TMP20]], ptr [[TMP19]], align 8
 504 // CHECK1-NEXT:    [[TMP21:%.*]] = call i32 @__kmpc_omp_task(ptr @[[GLOB1]], i32 [[TMP11]], ptr [[TMP12]])
 505 // CHECK1-NEXT:    ret void
 506 //
 507 //
 508 // CHECK1-LABEL: define {{[^@]+}}@.omp_task_privates_map.
 509 // CHECK1-SAME: (ptr noalias [[TMP0:%.*]], ptr noalias [[TMP1:%.*]], ptr noalias [[TMP2:%.*]]) #[[ATTR9:[0-9]+]] {
 510 // CHECK1-NEXT:  entry:
 511 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 512 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 513 // CHECK1-NEXT:    [[DOTADDR2:%.*]] = alloca ptr, align 8
 514 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 515 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 516 // CHECK1-NEXT:    store ptr [[TMP2]], ptr [[DOTADDR2]], align 8
 517 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 518 // CHECK1-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T:%.*]], ptr [[TMP3]], i32 0, i32 0
 519 // CHECK1-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 520 // CHECK1-NEXT:    store ptr [[TMP4]], ptr [[TMP5]], align 8
 521 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[STRUCT__KMP_PRIVATES_T]], ptr [[TMP3]], i32 0, i32 1
 522 // CHECK1-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[DOTADDR2]], align 8
 523 // CHECK1-NEXT:    store ptr [[TMP6]], ptr [[TMP7]], align 8
 524 // CHECK1-NEXT:    ret void
 525 //
 526 //
 527 // CHECK1-LABEL: define {{[^@]+}}@.omp_task_entry.
 528 // CHECK1-SAME: (i32 [[TMP0:%.*]], ptr noalias [[TMP1:%.*]]) #[[ATTR5]] {
 529 // CHECK1-NEXT:  entry:
 530 // CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR_I:%.*]] = alloca i32, align 4
 531 // CHECK1-NEXT:    [[DOTPART_ID__ADDR_I:%.*]] = alloca ptr, align 8
 532 // CHECK1-NEXT:    [[DOTPRIVATES__ADDR_I:%.*]] = alloca ptr, align 8
 533 // CHECK1-NEXT:    [[DOTCOPY_FN__ADDR_I:%.*]] = alloca ptr, align 8
 534 // CHECK1-NEXT:    [[DOTTASK_T__ADDR_I:%.*]] = alloca ptr, align 8
 535 // CHECK1-NEXT:    [[__CONTEXT_ADDR_I:%.*]] = alloca ptr, align 8
 536 // CHECK1-NEXT:    [[DOTFIRSTPRIV_PTR_ADDR_I:%.*]] = alloca ptr, align 8
 537 // CHECK1-NEXT:    [[DOTFIRSTPRIV_PTR_ADDR1_I:%.*]] = alloca ptr, align 8
 538 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i32, align 4
 539 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 540 // CHECK1-NEXT:    store i32 [[TMP0]], ptr [[DOTADDR]], align 4
 541 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 542 // CHECK1-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTADDR]], align 4
 543 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 544 // CHECK1-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], ptr [[TMP3]], i32 0, i32 0
 545 // CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T:%.*]], ptr [[TMP4]], i32 0, i32 2
 546 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T]], ptr [[TMP4]], i32 0, i32 0
 547 // CHECK1-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[TMP6]], align 8
 548 // CHECK1-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES]], ptr [[TMP3]], i32 0, i32 1
 549 // CHECK1-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META3:![0-9]+]])
 550 // CHECK1-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META6:![0-9]+]])
 551 // CHECK1-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META8:![0-9]+]])
 552 // CHECK1-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META10:![0-9]+]])
 553 // CHECK1-NEXT:    store i32 [[TMP2]], ptr [[DOTGLOBAL_TID__ADDR_I]], align 4, !noalias !12
 554 // CHECK1-NEXT:    store ptr [[TMP5]], ptr [[DOTPART_ID__ADDR_I]], align 8, !noalias !12
 555 // CHECK1-NEXT:    store ptr [[TMP8]], ptr [[DOTPRIVATES__ADDR_I]], align 8, !noalias !12
 556 // CHECK1-NEXT:    store ptr @.omp_task_privates_map., ptr [[DOTCOPY_FN__ADDR_I]], align 8, !noalias !12
 557 // CHECK1-NEXT:    store ptr [[TMP3]], ptr [[DOTTASK_T__ADDR_I]], align 8, !noalias !12
 558 // CHECK1-NEXT:    store ptr [[TMP7]], ptr [[__CONTEXT_ADDR_I]], align 8, !noalias !12
 559 // CHECK1-NEXT:    [[TMP9:%.*]] = load ptr, ptr [[__CONTEXT_ADDR_I]], align 8, !noalias !12
 560 // CHECK1-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [[STRUCT_ANON:%.*]], ptr [[TMP9]], i32 0, i32 1
 561 // CHECK1-NEXT:    [[TMP11:%.*]] = load i64, ptr [[TMP10]], align 8
 562 // CHECK1-NEXT:    [[TMP12:%.*]] = load ptr, ptr [[DOTCOPY_FN__ADDR_I]], align 8, !noalias !12
 563 // CHECK1-NEXT:    [[TMP13:%.*]] = load ptr, ptr [[DOTPRIVATES__ADDR_I]], align 8, !noalias !12
 564 // CHECK1-NEXT:    call void [[TMP12]](ptr [[TMP13]], ptr [[DOTFIRSTPRIV_PTR_ADDR_I]], ptr [[DOTFIRSTPRIV_PTR_ADDR1_I]]) #[[ATTR3]]
 565 // CHECK1-NEXT:    [[TMP14:%.*]] = load ptr, ptr [[DOTFIRSTPRIV_PTR_ADDR_I]], align 8, !noalias !12
 566 // CHECK1-NEXT:    [[TMP15:%.*]] = load ptr, ptr [[DOTFIRSTPRIV_PTR_ADDR1_I]], align 8, !noalias !12
 567 // CHECK1-NEXT:    [[TMP16:%.*]] = load ptr, ptr [[TMP9]], align 8
 568 // CHECK1-NEXT:    [[TMP17:%.*]] = load ptr, ptr [[TMP14]], align 8
 569 // CHECK1-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTGLOBAL_TID__ADDR_I]], align 4, !noalias !12
 570 // CHECK1-NEXT:    [[TMP19:%.*]] = call ptr @__kmpc_task_reduction_get_th_data(i32 [[TMP18]], ptr [[TMP17]], ptr [[TMP16]])
 571 // CHECK1-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [[STRUCT_ANON]], ptr [[TMP9]], i32 0, i32 2
 572 // CHECK1-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[TMP20]], align 8
 573 // CHECK1-NEXT:    [[TMP22:%.*]] = mul nuw i64 [[TMP11]], 2
 574 // CHECK1-NEXT:    [[TMP23:%.*]] = udiv exact i64 [[TMP22]], ptrtoint (ptr getelementptr (i16, ptr null, i32 1) to i64)
 575 // CHECK1-NEXT:    [[TMP24:%.*]] = call ptr @__kmpc_threadprivate_cached(ptr @[[GLOB1]], i32 [[TMP18]], ptr @{{reduction_size[.].+[.]}})
 576 // CHECK1-NEXT:    store i64 [[TMP23]], ptr [[TMP24]], align 8
 577 // CHECK1-NEXT:    [[TMP25:%.*]] = load ptr, ptr [[TMP15]], align 8
 578 // CHECK1-NEXT:    [[TMP26:%.*]] = call ptr @__kmpc_task_reduction_get_th_data(i32 [[TMP18]], ptr [[TMP25]], ptr [[TMP21]])
 579 // CHECK1-NEXT:    [[TMP27:%.*]] = load i32, ptr [[TMP19]], align 4
 580 // CHECK1-NEXT:    [[IDXPROM_I:%.*]] = sext i32 [[TMP27]] to i64
 581 // CHECK1-NEXT:    [[ARRAYIDX_I:%.*]] = getelementptr inbounds i16, ptr [[TMP26]], i64 [[IDXPROM_I]]
 582 // CHECK1-NEXT:    [[TMP28:%.*]] = load i16, ptr [[ARRAYIDX_I]], align 2
 583 // CHECK1-NEXT:    [[CONV_I:%.*]] = sext i16 [[TMP28]] to i32
 584 // CHECK1-NEXT:    [[TMP29:%.*]] = load i32, ptr [[TMP19]], align 4
 585 // CHECK1-NEXT:    [[ADD_I:%.*]] = add nsw i32 [[TMP29]], [[CONV_I]]
 586 // CHECK1-NEXT:    store i32 [[ADD_I]], ptr [[TMP19]], align 4
 587 // CHECK1-NEXT:    ret i32 0
 588 //
 589 //
 590 // CHECK1-LABEL: define {{[^@]+}}@.omp_task_entry..10
 591 // CHECK1-SAME: (i32 [[TMP0:%.*]], ptr noalias [[TMP1:%.*]]) #[[ATTR5]] {
 592 // CHECK1-NEXT:  entry:
 593 // CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR_I:%.*]] = alloca i32, align 4
 594 // CHECK1-NEXT:    [[DOTPART_ID__ADDR_I:%.*]] = alloca ptr, align 8
 595 // CHECK1-NEXT:    [[DOTPRIVATES__ADDR_I:%.*]] = alloca ptr, align 8
 596 // CHECK1-NEXT:    [[DOTCOPY_FN__ADDR_I:%.*]] = alloca ptr, align 8
 597 // CHECK1-NEXT:    [[DOTTASK_T__ADDR_I:%.*]] = alloca ptr, align 8
 598 // CHECK1-NEXT:    [[__CONTEXT_ADDR_I:%.*]] = alloca ptr, align 8
 599 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca i32, align 4
 600 // CHECK1-NEXT:    [[DOTADDR1:%.*]] = alloca ptr, align 8
 601 // CHECK1-NEXT:    store i32 [[TMP0]], ptr [[DOTADDR]], align 4
 602 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
 603 // CHECK1-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTADDR]], align 4
 604 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
 605 // CHECK1-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T_WITH_PRIVATES_2:%.*]], ptr [[TMP3]], i32 0, i32 0
 606 // CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T:%.*]], ptr [[TMP4]], i32 0, i32 2
 607 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_T]], ptr [[TMP4]], i32 0, i32 0
 608 // CHECK1-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[TMP6]], align 8
 609 // CHECK1-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META15:![0-9]+]])
 610 // CHECK1-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META18:![0-9]+]])
 611 // CHECK1-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META20:![0-9]+]])
 612 // CHECK1-NEXT:    call void @llvm.experimental.noalias.scope.decl(metadata [[META22:![0-9]+]])
 613 // CHECK1-NEXT:    store i32 [[TMP2]], ptr [[DOTGLOBAL_TID__ADDR_I]], align 4, !noalias !24
 614 // CHECK1-NEXT:    store ptr [[TMP5]], ptr [[DOTPART_ID__ADDR_I]], align 8, !noalias !24
 615 // CHECK1-NEXT:    store ptr null, ptr [[DOTPRIVATES__ADDR_I]], align 8, !noalias !24
 616 // CHECK1-NEXT:    store ptr null, ptr [[DOTCOPY_FN__ADDR_I]], align 8, !noalias !24
 617 // CHECK1-NEXT:    store ptr [[TMP3]], ptr [[DOTTASK_T__ADDR_I]], align 8, !noalias !24
 618 // CHECK1-NEXT:    store ptr [[TMP7]], ptr [[__CONTEXT_ADDR_I]], align 8, !noalias !24
 619 // CHECK1-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[__CONTEXT_ADDR_I]], align 8, !noalias !24
 620 // CHECK1-NEXT:    [[TMP9:%.*]] = load ptr, ptr [[TMP8]], align 8
 621 // CHECK1-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTGLOBAL_TID__ADDR_I]], align 4, !noalias !24
 622 // CHECK1-NEXT:    [[TMP11:%.*]] = call ptr @__kmpc_task_reduction_get_th_data(i32 [[TMP10]], ptr null, ptr [[TMP9]])
 623 // CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[TMP11]], align 4
 624 // CHECK1-NEXT:    [[INC_I:%.*]] = add nsw i32 [[TMP12]], 1
 625 // CHECK1-NEXT:    store i32 [[INC_I]], ptr [[TMP11]], align 4
 626 // CHECK1-NEXT:    ret i32 0
 627 //
 628 //
 629 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SC2Ev
 630 // CHECK1-SAME: (ptr nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
 631 // CHECK1-NEXT:  entry:
 632 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 633 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 634 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 635 // CHECK1-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 636 // CHECK1-NEXT:    store i32 0, ptr [[A]], align 4
 637 // CHECK1-NEXT:    ret void
 638 //
 639 //
 640 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SD2Ev
 641 // CHECK1-SAME: (ptr nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
 642 // CHECK1-NEXT:  entry:
 643 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 644 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 645 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 646 // CHECK1-NEXT:    ret void
 647 //
 648 //
 649 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SC1ERKS_
 650 // CHECK1-SAME: (ptr nonnull align 4 dereferenceable(4) [[THIS:%.*]], ptr nonnull align 4 dereferenceable(4) [[TMP0:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
 651 // CHECK1-NEXT:  entry:
 652 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 653 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 654 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 655 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 656 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 657 // CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[DOTADDR]], align 8
 658 // CHECK1-NEXT:    call void @_ZN1SC2ERKS_(ptr nonnull align 4 dereferenceable(4) [[THIS1]], ptr nonnull align 4 dereferenceable(4) [[TMP1]])
 659 // CHECK1-NEXT:    ret void
 660 //
 661 //
 662 // CHECK1-LABEL: define {{[^@]+}}@_ZN1SC2ERKS_
 663 // CHECK1-SAME: (ptr nonnull align 4 dereferenceable(4) [[THIS:%.*]], ptr nonnull align 4 dereferenceable(4) [[TMP0:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
 664 // CHECK1-NEXT:  entry:
 665 // CHECK1-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 666 // CHECK1-NEXT:    [[DOTADDR:%.*]] = alloca ptr, align 8
 667 // CHECK1-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 668 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[DOTADDR]], align 8
 669 // CHECK1-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 670 // CHECK1-NEXT:    ret void
 671 //