clang/test/OpenMP/allocate_allocator_modifier_codegen.cpp

   1 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --include-generated-funcs --replace-value-regex "__omp_offloading_[0-9a-z]+_[0-9a-z]+" "reduction_size[.].+[.]" "pl_cond[.].+[.|,]" --prefix-filecheck-ir-name _ --version 5
   2 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=52 -triple x86_64-unknown-linux-gnu -emit-llvm -o - %s | FileCheck %s
   3 // RUN: %clang_cc1 -fopenmp -fopenmp-version=52 -triple x86_64-unknown-linux-gnu -x c++ -std=c++11 -emit-pch -o %t %s
   4 // RUN: %clang_cc1 -fopenmp -fopenmp-version=52 -triple x86_64-unknown-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s
   5 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=52 -triple x86_64-unknown-linux-gnu -emit-llvm -o - %s | FileCheck %s
   6 // RUN: %clang_cc1 -fopenmp -fopenmp-version=52 -fnoopenmp-use-tls -triple x86_64-unknown-linux-gnu -x c++ -std=c++11 -emit-pch -o %t %s
   7 // RUN: %clang_cc1 -fopenmp -fopenmp-version=52 -fnoopenmp-use-tls -triple x86_64-unknown-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix CHECK-TLS %s
   8
   9 // RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=52 -triple x86_64-unknown-linux-gnu -emit-llvm -o - %s | FileCheck --check-prefix SIMD-ONLY0 %s
  10 // RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=52 -triple x86_64-unknown-linux-gnu -x c++ -std=c++11 -emit-pch -o %t %s
  11 // RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=52 -triple x86_64-unknown-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
  12 // RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=52 -triple x86_64-unknown-linux-gnu -emit-llvm -o - %s | FileCheck --check-prefix SIMD-ONLY0 %s
  13 // RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=52 -fnoopenmp-use-tls -triple x86_64-unknown-linux-gnu -x c++ -std=c++11 -emit-pch -o %t %s
  14 // RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=52 -fnoopenmp-use-tls -triple x86_64-unknown-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
  15 // expected-no-diagnostics
  16
  17 #ifndef HEADER
  18 #define HEADER
  19
  20 enum omp_allocator_handle_t {
  21   omp_null_allocator = 0,
  22   omp_default_mem_alloc = 1,
  23   omp_large_cap_mem_alloc = 2,
  24   omp_const_mem_alloc = 3,
  25   omp_high_bw_mem_alloc = 4,
  26   omp_low_lat_mem_alloc = 5,
  27   omp_cgroup_mem_alloc = 6,
  28   omp_pteam_mem_alloc = 7,
  29   omp_thread_mem_alloc = 8,
  30   KMP_ALLOCATOR_MAX_HANDLE = __UINTPTR_MAX__
  31 };
  32
  33 template <class T>
  34 struct ST {
  35   static T m;
  36 };
  37
  38 template <class T, omp_allocator_handle_t TY> T foo() {
  39   T v;
  40  #pragma omp scope private(v) allocate(allocator(TY):v)
  41   v = ST<T>::m;
  42   return v;
  43 }
  44
  45 namespace ns {
  46 int a;
  47 }
  48
  49 int main() {
  50   static int a;
  51   static int temp;
  52   #pragma omp scope private(ns::a) allocate(allocator(omp_pteam_mem_alloc):ns::a)
  53   ns::a++;
  54
  55  #pragma omp scope private(a) allocate(allocator(omp_thread_mem_alloc):a)
  56   a = 2;
  57   double b = 3;
  58   #pragma omp scope private(temp) allocate(temp)
  59   temp += foo<int, omp_cgroup_mem_alloc>();
  60   return temp+ns::a;
  61 }
  62
  63 extern template int ST<int>::m;
  64
  65 int b;
  66
  67 void bar(int a, float &z) {
  68   #pragma omp scope private(a,z) allocate(allocator(omp_default_mem_alloc):a,z)
  69   a += b;
  70 }
  71 #endif
  72 // CHECK-LABEL: define dso_local noundef i32 @main(
  73 // CHECK-SAME: ) #[[ATTR0:[0-9]+]] {
  74 // CHECK-NEXT:  [[ENTRY:.*:]]
  75 // CHECK-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
  76 // CHECK-NEXT:    [[B:%.*]] = alloca double, align 8
  77 // CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
  78 // CHECK-NEXT:    store i32 0, ptr [[RETVAL]], align 4
  79 // CHECK-NEXT:    [[DOTA__VOID_ADDR:%.*]] = call ptr @__kmpc_alloc(i32 [[TMP0]], i64 4, ptr inttoptr (i64 7 to ptr))
  80 // CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[DOTA__VOID_ADDR]], align 4
  81 // CHECK-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP1]], 1
  82 // CHECK-NEXT:    store i32 [[INC]], ptr [[DOTA__VOID_ADDR]], align 4
  83 // CHECK-NEXT:    call void @__kmpc_free(i32 [[TMP0]], ptr [[DOTA__VOID_ADDR]], ptr inttoptr (i64 7 to ptr))
  84 // CHECK-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2:[0-9]+]], i32 [[TMP0]])
  85 // CHECK-NEXT:    [[DOTA__VOID_ADDR1:%.*]] = call ptr @__kmpc_alloc(i32 [[TMP0]], i64 4, ptr inttoptr (i64 8 to ptr))
  86 // CHECK-NEXT:    store i32 2, ptr [[DOTA__VOID_ADDR1]], align 4
  87 // CHECK-NEXT:    call void @__kmpc_free(i32 [[TMP0]], ptr [[DOTA__VOID_ADDR1]], ptr inttoptr (i64 8 to ptr))
  88 // CHECK-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP0]])
  89 // CHECK-NEXT:    store double 3.000000e+00, ptr [[B]], align 8
  90 // CHECK-NEXT:    [[DOTTEMP__VOID_ADDR:%.*]] = call ptr @__kmpc_alloc(i32 [[TMP0]], i64 4, ptr null)
  91 // CHECK-NEXT:    [[CALL:%.*]] = call noundef i32 @_Z3fooIiL22omp_allocator_handle_t6EET_v()
  92 // CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTTEMP__VOID_ADDR]], align 4
  93 // CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], [[CALL]]
  94 // CHECK-NEXT:    store i32 [[ADD]], ptr [[DOTTEMP__VOID_ADDR]], align 4
  95 // CHECK-NEXT:    call void @__kmpc_free(i32 [[TMP0]], ptr [[DOTTEMP__VOID_ADDR]], ptr null)
  96 // CHECK-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP0]])
  97 // CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr @_ZZ4mainE4temp, align 4
  98 // CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr @_ZN2ns1aE, align 4
  99 // CHECK-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP3]], [[TMP4]]
 100 // CHECK-NEXT:    ret i32 [[ADD2]]
 101 //
 102 //
 103 // CHECK-LABEL: define linkonce_odr noundef i32 @_Z3fooIiL22omp_allocator_handle_t6EET_v(
 104 // CHECK-SAME: ) #[[ATTR3:[0-9]+]] comdat {
 105 // CHECK-NEXT:  [[ENTRY:.*:]]
 106 // CHECK-NEXT:    [[V:%.*]] = alloca i32, align 4
 107 // CHECK-NEXT:    [[V1:%.*]] = alloca i32, align 4
 108 // CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 109 // CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr @_ZN2STIiE1mE, align 4
 110 // CHECK-NEXT:    store i32 [[TMP1]], ptr [[V1]], align 4
 111 // CHECK-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP0]])
 112 // CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr [[V]], align 4
 113 // CHECK-NEXT:    ret i32 [[TMP2]]
 114 //
 115 //
 116 // CHECK-LABEL: define dso_local void @_Z3bariRf(
 117 // CHECK-SAME: i32 noundef [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[Z:%.*]]) #[[ATTR3]] {
 118 // CHECK-NEXT:  [[ENTRY:.*:]]
 119 // CHECK-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
 120 // CHECK-NEXT:    [[Z_ADDR:%.*]] = alloca ptr, align 8
 121 // CHECK-NEXT:    [[TMP:%.*]] = alloca ptr, align 8
 122 // CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 123 // CHECK-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
 124 // CHECK-NEXT:    store ptr [[Z]], ptr [[Z_ADDR]], align 8
 125 // CHECK-NEXT:    [[DOTA__VOID_ADDR:%.*]] = call ptr @__kmpc_alloc(i32 [[TMP0]], i64 4, ptr inttoptr (i64 1 to ptr))
 126 // CHECK-NEXT:    [[DOTZ__VOID_ADDR:%.*]] = call ptr @__kmpc_alloc(i32 [[TMP0]], i64 4, ptr inttoptr (i64 1 to ptr))
 127 // CHECK-NEXT:    store ptr [[DOTZ__VOID_ADDR]], ptr [[TMP]], align 8
 128 // CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr @b, align 4
 129 // CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTA__VOID_ADDR]], align 4
 130 // CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], [[TMP1]]
 131 // CHECK-NEXT:    store i32 [[ADD]], ptr [[DOTA__VOID_ADDR]], align 4
 132 // CHECK-NEXT:    call void @__kmpc_free(i32 [[TMP0]], ptr [[DOTZ__VOID_ADDR]], ptr inttoptr (i64 1 to ptr))
 133 // CHECK-NEXT:    call void @__kmpc_free(i32 [[TMP0]], ptr [[DOTA__VOID_ADDR]], ptr inttoptr (i64 1 to ptr))
 134 // CHECK-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP0]])
 135 // CHECK-NEXT:    ret void
 136 //
 137 //
 138 // CHECK-TLS-LABEL: define dso_local noundef i32 @main(
 139 // CHECK-TLS-SAME: ) #[[ATTR0:[0-9]+]] {
 140 // CHECK-TLS-NEXT:  [[ENTRY:.*:]]
 141 // CHECK-TLS-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
 142 // CHECK-TLS-NEXT:    [[B:%.*]] = alloca double, align 8
 143 // CHECK-TLS-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 144 // CHECK-TLS-NEXT:    store i32 0, ptr [[RETVAL]], align 4
 145 // CHECK-TLS-NEXT:    [[DOTA__VOID_ADDR:%.*]] = call ptr @__kmpc_alloc(i32 [[TMP0]], i64 4, ptr inttoptr (i64 7 to ptr))
 146 // CHECK-TLS-NEXT:    [[TMP1:%.*]] = load i32, ptr [[DOTA__VOID_ADDR]], align 4
 147 // CHECK-TLS-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP1]], 1
 148 // CHECK-TLS-NEXT:    store i32 [[INC]], ptr [[DOTA__VOID_ADDR]], align 4
 149 // CHECK-TLS-NEXT:    call void @__kmpc_free(i32 [[TMP0]], ptr [[DOTA__VOID_ADDR]], ptr inttoptr (i64 7 to ptr))
 150 // CHECK-TLS-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2:[0-9]+]], i32 [[TMP0]])
 151 // CHECK-TLS-NEXT:    [[DOTA__VOID_ADDR1:%.*]] = call ptr @__kmpc_alloc(i32 [[TMP0]], i64 4, ptr inttoptr (i64 8 to ptr))
 152 // CHECK-TLS-NEXT:    store i32 2, ptr [[DOTA__VOID_ADDR1]], align 4
 153 // CHECK-TLS-NEXT:    call void @__kmpc_free(i32 [[TMP0]], ptr [[DOTA__VOID_ADDR1]], ptr inttoptr (i64 8 to ptr))
 154 // CHECK-TLS-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP0]])
 155 // CHECK-TLS-NEXT:    store double 3.000000e+00, ptr [[B]], align 8
 156 // CHECK-TLS-NEXT:    [[DOTTEMP__VOID_ADDR:%.*]] = call ptr @__kmpc_alloc(i32 [[TMP0]], i64 4, ptr null)
 157 // CHECK-TLS-NEXT:    [[CALL:%.*]] = call noundef i32 @_Z3fooIiL22omp_allocator_handle_t6EET_v()
 158 // CHECK-TLS-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTTEMP__VOID_ADDR]], align 4
 159 // CHECK-TLS-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], [[CALL]]
 160 // CHECK-TLS-NEXT:    store i32 [[ADD]], ptr [[DOTTEMP__VOID_ADDR]], align 4
 161 // CHECK-TLS-NEXT:    call void @__kmpc_free(i32 [[TMP0]], ptr [[DOTTEMP__VOID_ADDR]], ptr null)
 162 // CHECK-TLS-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP0]])
 163 // CHECK-TLS-NEXT:    [[TMP3:%.*]] = load i32, ptr @_ZZ4mainE4temp, align 4
 164 // CHECK-TLS-NEXT:    [[TMP4:%.*]] = load i32, ptr @_ZN2ns1aE, align 4
 165 // CHECK-TLS-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP3]], [[TMP4]]
 166 // CHECK-TLS-NEXT:    ret i32 [[ADD2]]
 167 //
 168 //
 169 // CHECK-TLS-LABEL: define linkonce_odr noundef i32 @_Z3fooIiL22omp_allocator_handle_t6EET_v(
 170 // CHECK-TLS-SAME: ) #[[ATTR3:[0-9]+]] comdat {
 171 // CHECK-TLS-NEXT:  [[ENTRY:.*:]]
 172 // CHECK-TLS-NEXT:    [[V:%.*]] = alloca i32, align 4
 173 // CHECK-TLS-NEXT:    [[V1:%.*]] = alloca i32, align 4
 174 // CHECK-TLS-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 175 // CHECK-TLS-NEXT:    [[TMP1:%.*]] = load i32, ptr @_ZN2STIiE1mE, align 4
 176 // CHECK-TLS-NEXT:    store i32 [[TMP1]], ptr [[V1]], align 4
 177 // CHECK-TLS-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP0]])
 178 // CHECK-TLS-NEXT:    [[TMP2:%.*]] = load i32, ptr [[V]], align 4
 179 // CHECK-TLS-NEXT:    ret i32 [[TMP2]]
 180 //
 181 //
 182 // CHECK-TLS-LABEL: define dso_local void @_Z3bariRf(
 183 // CHECK-TLS-SAME: i32 noundef [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[Z:%.*]]) #[[ATTR3]] {
 184 // CHECK-TLS-NEXT:  [[ENTRY:.*:]]
 185 // CHECK-TLS-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
 186 // CHECK-TLS-NEXT:    [[Z_ADDR:%.*]] = alloca ptr, align 8
 187 // CHECK-TLS-NEXT:    [[TMP:%.*]] = alloca ptr, align 8
 188 // CHECK-TLS-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 189 // CHECK-TLS-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
 190 // CHECK-TLS-NEXT:    store ptr [[Z]], ptr [[Z_ADDR]], align 8
 191 // CHECK-TLS-NEXT:    [[DOTA__VOID_ADDR:%.*]] = call ptr @__kmpc_alloc(i32 [[TMP0]], i64 4, ptr inttoptr (i64 1 to ptr))
 192 // CHECK-TLS-NEXT:    [[DOTZ__VOID_ADDR:%.*]] = call ptr @__kmpc_alloc(i32 [[TMP0]], i64 4, ptr inttoptr (i64 1 to ptr))
 193 // CHECK-TLS-NEXT:    store ptr [[DOTZ__VOID_ADDR]], ptr [[TMP]], align 8
 194 // CHECK-TLS-NEXT:    [[TMP1:%.*]] = load i32, ptr @b, align 4
 195 // CHECK-TLS-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTA__VOID_ADDR]], align 4
 196 // CHECK-TLS-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], [[TMP1]]
 197 // CHECK-TLS-NEXT:    store i32 [[ADD]], ptr [[DOTA__VOID_ADDR]], align 4
 198 // CHECK-TLS-NEXT:    call void @__kmpc_free(i32 [[TMP0]], ptr [[DOTZ__VOID_ADDR]], ptr inttoptr (i64 1 to ptr))
 199 // CHECK-TLS-NEXT:    call void @__kmpc_free(i32 [[TMP0]], ptr [[DOTA__VOID_ADDR]], ptr inttoptr (i64 1 to ptr))
 200 // CHECK-TLS-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB2]], i32 [[TMP0]])
 201 // CHECK-TLS-NEXT:    ret void
 202 //
 203 //
 204 // SIMD-ONLY0-LABEL: define dso_local noundef i32 @main(
 205 // SIMD-ONLY0-SAME: ) #[[ATTR0:[0-9]+]] {
 206 // SIMD-ONLY0-NEXT:  [[ENTRY:.*:]]
 207 // SIMD-ONLY0-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
 208 // SIMD-ONLY0-NEXT:    [[A:%.*]] = alloca i32, align 4
 209 // SIMD-ONLY0-NEXT:    [[A1:%.*]] = alloca i32, align 4
 210 // SIMD-ONLY0-NEXT:    [[B:%.*]] = alloca double, align 8
 211 // SIMD-ONLY0-NEXT:    [[TEMP:%.*]] = alloca i32, align 4
 212 // SIMD-ONLY0-NEXT:    store i32 0, ptr [[RETVAL]], align 4
 213 // SIMD-ONLY0-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A]], align 4
 214 // SIMD-ONLY0-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP0]], 1
 215 // SIMD-ONLY0-NEXT:    store i32 [[INC]], ptr [[A]], align 4
 216 // SIMD-ONLY0-NEXT:    store i32 2, ptr [[A1]], align 4
 217 // SIMD-ONLY0-NEXT:    store double 3.000000e+00, ptr [[B]], align 8
 218 // SIMD-ONLY0-NEXT:    [[CALL:%.*]] = call noundef i32 @_Z3fooIiL22omp_allocator_handle_t6EET_v()
 219 // SIMD-ONLY0-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TEMP]], align 4
 220 // SIMD-ONLY0-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP1]], [[CALL]]
 221 // SIMD-ONLY0-NEXT:    store i32 [[ADD]], ptr [[TEMP]], align 4
 222 // SIMD-ONLY0-NEXT:    [[TMP2:%.*]] = load i32, ptr @_ZZ4mainE4temp, align 4
 223 // SIMD-ONLY0-NEXT:    [[TMP3:%.*]] = load i32, ptr @_ZN2ns1aE, align 4
 224 // SIMD-ONLY0-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP2]], [[TMP3]]
 225 // SIMD-ONLY0-NEXT:    ret i32 [[ADD2]]
 226 //
 227 //
 228 // SIMD-ONLY0-LABEL: define linkonce_odr noundef i32 @_Z3fooIiL22omp_allocator_handle_t6EET_v(
 229 // SIMD-ONLY0-SAME: ) #[[ATTR1:[0-9]+]] comdat {
 230 // SIMD-ONLY0-NEXT:  [[ENTRY:.*:]]
 231 // SIMD-ONLY0-NEXT:    [[V:%.*]] = alloca i32, align 4
 232 // SIMD-ONLY0-NEXT:    [[V1:%.*]] = alloca i32, align 4
 233 // SIMD-ONLY0-NEXT:    [[TMP0:%.*]] = load i32, ptr @_ZN2STIiE1mE, align 4
 234 // SIMD-ONLY0-NEXT:    store i32 [[TMP0]], ptr [[V1]], align 4
 235 // SIMD-ONLY0-NEXT:    [[TMP1:%.*]] = load i32, ptr [[V]], align 4
 236 // SIMD-ONLY0-NEXT:    ret i32 [[TMP1]]
 237 //
 238 //
 239 // SIMD-ONLY0-LABEL: define dso_local void @_Z3bariRf(
 240 // SIMD-ONLY0-SAME: i32 noundef [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[Z:%.*]]) #[[ATTR1]] {
 241 // SIMD-ONLY0-NEXT:  [[ENTRY:.*:]]
 242 // SIMD-ONLY0-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
 243 // SIMD-ONLY0-NEXT:    [[Z_ADDR:%.*]] = alloca ptr, align 8
 244 // SIMD-ONLY0-NEXT:    [[A1:%.*]] = alloca i32, align 4
 245 // SIMD-ONLY0-NEXT:    [[Z2:%.*]] = alloca float, align 4
 246 // SIMD-ONLY0-NEXT:    [[TMP:%.*]] = alloca ptr, align 8
 247 // SIMD-ONLY0-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
 248 // SIMD-ONLY0-NEXT:    store ptr [[Z]], ptr [[Z_ADDR]], align 8
 249 // SIMD-ONLY0-NEXT:    store ptr [[Z2]], ptr [[TMP]], align 8
 250 // SIMD-ONLY0-NEXT:    [[TMP0:%.*]] = load i32, ptr @b, align 4
 251 // SIMD-ONLY0-NEXT:    [[TMP1:%.*]] = load i32, ptr [[A1]], align 4
 252 // SIMD-ONLY0-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP1]], [[TMP0]]
 253 // SIMD-ONLY0-NEXT:    store i32 [[ADD]], ptr [[A1]], align 4
 254 // SIMD-ONLY0-NEXT:    ret void
 255 //