clang/test/OpenMP/master_taskloop_simd_lastprivate_codegen.cpp

   1 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck %s
   2 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s
   3 // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s
   4 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s
   5 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s
   6 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=ARRAY %s
   7
   8 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
   9 // RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s
  10 // RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
  11 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
  12 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
  13 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
  14 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
  15 // expected-no-diagnostics
  16
  17 #ifndef ARRAY
  18 #ifndef HEADER
  19 #define HEADER
  20
  21 template <class T>
  22 struct S {
  23   T f;
  24   S(T a) : f(a) {}
  25   S() : f() {}
  26   S(const S &s, T t = T()) : f(s.f + t) {}
  27   operator T() { return T(); }
  28   ~S() {}
  29 };
  30
  31 volatile double g;
  32
  33 // CHECK-DAG: [[KMP_TASK_T_TY:%.+]] = type { ptr, ptr, i32, %union{{.+}}, %union{{.+}}, i64, i64, i64, i32, ptr }
  34 // CHECK-DAG: [[S_DOUBLE_TY:%.+]] = type { double }
  35 // CHECK-DAG: [[PRIVATES_MAIN_TY:%.+]] = type {{.?}}{ [2 x [[S_DOUBLE_TY]]], [[S_DOUBLE_TY]], i32, [2 x i32]
  36 // CHECK-DAG: [[CAP_MAIN_TY:%.+]] = type { ptr, ptr, ptr, ptr, ptr }
  37 // CHECK-DAG: [[KMP_TASK_MAIN_TY:%.+]] = type { [[KMP_TASK_T_TY]], [[PRIVATES_MAIN_TY]] }
  38 // CHECK-DAG: [[S_INT_TY:%.+]] = type { i32 }
  39 // CHECK-DAG: [[CAP_TMAIN_TY:%.+]] = type { ptr, ptr, ptr, ptr }
  40 // CHECK-DAG: [[PRIVATES_TMAIN_TY:%.+]] = type { i32, [2 x i32], [2 x [[S_INT_TY]]], [[S_INT_TY]], [104 x i8] }
  41 // CHECK-DAG: [[KMP_TASK_TMAIN_TY:%.+]] = type { [[KMP_TASK_T_TY]], [{{[0-9]+}} x i8], [[PRIVATES_TMAIN_TY]] }
  42 template <typename T>
  43 T tmain() {
  44   S<T> ttt;
  45   S<T> test;
  46   T t_var __attribute__((aligned(128))) = T();
  47   T vec[] = {1, 2};
  48   S<T> s_arr[] = {1, 2};
  49   S<T> var(3);
  50 #pragma omp master taskloop simd lastprivate(t_var, vec, s_arr, s_arr, var, var)
  51   for (int i = 0; i < 10; ++i) {
  52     vec[0] = t_var;
  53     s_arr[0] = var;
  54   }
  55   return T();
  56 }
  57
  58 int main() {
  59   static int sivar;
  60 #ifdef LAMBDA
  61   // LAMBDA: [[G:@.+]] ={{.*}} global double
  62   // LAMBDA: [[SIVAR:@.+]] = internal global i{{[0-9]+}} 0,
  63   // LAMBDA-LABEL: @main
  64   // LAMBDA: call{{( x86_thiscallcc)?}} void [[OUTER_LAMBDA:@.+]](
  65   [&]() {
  66   // LAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]](
  67   // LAMBDA: [[RES:%.+]] = call ptr @__kmpc_omp_task_alloc(ptr @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 96, i64 16, ptr [[TASK_ENTRY:@[^ ]+]])
  68 // LAMBDA: [[PRIVATES:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i{{.+}} 0, i{{.+}} 1
  69
  70 // LAMBDA: call void @__kmpc_taskloop(ptr @{{.+}}, i32 %{{.+}}, ptr [[RES]], i32 1, ptr %{{.+}}, ptr %{{.+}}, i64 %{{.+}}, i32 1, i32 0, i64 0, ptr [[MAIN_DUP:@.+]])
  71 // LAMBDA: ret
  72 #pragma omp master taskloop simd lastprivate(g, sivar)
  73   for (int i = 0; i < 10; ++i) {
  74     // LAMBDA: define {{.+}} void [[INNER_LAMBDA:@.+]]({{.+}} [[ARG_PTR:%.+]])
  75     // LAMBDA: store ptr [[ARG_PTR]], ptr [[ARG_PTR_REF:%.+]],
  76     // LAMBDA: [[ARG_PTR:%.+]] = load ptr, ptr [[ARG_PTR_REF]]
  77     // LAMBDA: [[G_PTR_REF:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
  78     // LAMBDA: [[G_REF:%.+]] = load ptr, ptr [[G_PTR_REF]]
  79     // LAMBDA: store double 2.0{{.+}}, ptr [[G_REF]]
  80
  81     // LAMBDA: store ptr %{{.+}}, ptr %{{.+}},
  82     // LAMBDA: define internal noundef i32 [[TASK_ENTRY]](i32 noundef %0, ptr noalias noundef %1)
  83     g = 1;
  84     sivar = 11;
  85     // LAMBDA: store double 1.0{{.+}}, ptr %{{.+}},
  86     // LAMBDA: store i{{[0-9]+}} 11, ptr %{{.+}},
  87     // LAMBDA: call void [[INNER_LAMBDA]]({{.+}}
  88     // LAMBDA: icmp ne i32 %{{.+}}, 0
  89     // LAMBDA: br i1
  90     // LAMBDA: load double, ptr %
  91     // LAMBDA: store volatile double %
  92     // LAMBDA: load i32, ptr %
  93     // LAMBDA: store i32 %
  94     // LAMBDA: ret
  95     [&]() {
  96       g = 2;
  97       sivar = 22;
  98     }();
  99   }
 100   }();
 101   return 0;
 102 #elif defined(BLOCKS)
 103   // BLOCKS: [[G:@.+]] ={{.*}} global double
 104   // BLOCKS: [[SIVAR:@.+]] = internal global i{{[0-9]+}} 0,
 105   // BLOCKS-LABEL: @main
 106   // BLOCKS: call void {{%.+}}(ptr
 107   ^{
 108   // BLOCKS: define{{.*}} internal{{.*}} void {{.+}}(ptr
 109   // BLOCKS: [[RES:%.+]] = call ptr @__kmpc_omp_task_alloc(ptr @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 96, i64 16, ptr [[TASK_ENTRY:@[^ ]+]])
 110   // BLOCKS: [[PRIVATES:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i{{.+}} 0, i{{.+}} 1
 111   // BLOCKS: call void @__kmpc_taskloop(ptr @{{.+}}, i32 %{{.+}}, ptr [[RES]], i32 1, ptr %{{.+}}, ptr %{{.+}}, i64 %{{.+}}, i32 1, i32 0, i64 0, ptr [[MAIN_DUP:@.+]])
 112   // BLOCKS: ret
 113 #pragma omp master taskloop simd lastprivate(g, sivar)
 114   for (int i = 0; i < 10; ++i) {
 115     // BLOCKS: define {{.+}} void {{@.+}}(ptr
 116     // BLOCKS-NOT: [[G]]{{[[^:word:]]}}
 117     // BLOCKS: store double 2.0{{.+}}, ptr
 118     // BLOCKS-NOT: [[G]]{{[[^:word:]]}}
 119     // BLOCKS-NOT: [[SIVAR]]{{[[^:word:]]}}
 120     // BLOCKS: store i{{[0-9]+}} 22, ptr
 121     // BLOCKS-NOT: [[SIVAR]]{{[[^:word:]]}}
 122     // BLOCKS: ret
 123
 124     // BLOCKS: store ptr %{{.+}}, ptr %{{.+}},
 125     // BLOCKS: store ptr %{{.+}}, ptr %{{.+}},
 126     // BLOCKS: define internal noundef i32 [[TASK_ENTRY]](i32 noundef %0, ptr noalias noundef %1)
 127     g = 1;
 128     sivar = 11;
 129     // BLOCKS: store double 1.0{{.+}}, ptr %{{.+}},
 130     // BLOCKS-NOT: [[G]]{{[[^:word:]]}}
 131     // BLOCKS: store i{{[0-9]+}} 11, ptr %{{.+}},
 132     // BLOCKS-NOT: [[SIVAR]]{{[[^:word:]]}}
 133     // BLOCKS: call void {{%.+}}(ptr
 134     // BLOCKS: icmp ne i32 %{{.+}}, 0
 135     // BLOCKS: br i1
 136     // BLOCKS: load double, ptr %
 137     // BLOCKS: store volatile double %
 138     // BLOCKS: load i32, ptr %
 139     // BLOCKS: store i32 %
 140     ^{
 141       g = 2;
 142       sivar = 22;
 143     }();
 144   }
 145   }();
 146   return 0;
 147 #else
 148   S<double> ttt;
 149   S<double> test;
 150   int t_var = 0;
 151   int vec[] = {1, 2};
 152   S<double> s_arr[] = {1, 2};
 153   S<double> var(3);
 154 #pragma omp master taskloop simd lastprivate(var, t_var, s_arr, vec, s_arr, var, sivar)
 155   for (int i = 0; i < 10; ++i) {
 156     vec[0] = t_var;
 157     s_arr[0] = var;
 158     sivar = 33;
 159   }
 160   return tmain<int>();
 161 #endif
 162 }
 163
 164 // CHECK: [[SIVAR:.+]] = internal global i{{[0-9]+}} 0,
 165 // CHECK: define{{.*}} i{{[0-9]+}} @main()
 166 // CHECK: alloca [[S_DOUBLE_TY]],
 167 // CHECK: [[TEST:%.+]] = alloca [[S_DOUBLE_TY]],
 168 // CHECK: [[T_VAR_ADDR:%.+]] = alloca i32,
 169 // CHECK: [[VEC_ADDR:%.+]] = alloca [2 x i32],
 170 // CHECK: [[S_ARR_ADDR:%.+]] = alloca [2 x [[S_DOUBLE_TY]]],
 171 // CHECK: [[VAR_ADDR:%.+]] = alloca [[S_DOUBLE_TY]],
 172 // CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(ptr [[LOC:@.+]])
 173
 174 // CHECK: call {{.*}} [[S_DOUBLE_TY_CONSTR:@.+]](ptr {{[^,]*}} [[TEST]])
 175
 176 // Store original variables in capture struct.
 177 // CHECK: [[VEC_REF:%.+]] = getelementptr inbounds nuw [[CAP_MAIN_TY]], ptr %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 0
 178 // CHECK: store ptr [[VEC_ADDR]], ptr [[VEC_REF]],
 179 // CHECK: [[T_VAR_REF:%.+]] = getelementptr inbounds nuw [[CAP_MAIN_TY]], ptr %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 1
 180 // CHECK: store ptr [[T_VAR_ADDR]], ptr [[T_VAR_REF]],
 181 // CHECK: [[S_ARR_REF:%.+]] = getelementptr inbounds nuw [[CAP_MAIN_TY]], ptr %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 2
 182 // CHECK: store ptr [[S_ARR_ADDR]], ptr [[S_ARR_REF]],
 183 // CHECK: [[VAR_REF:%.+]] = getelementptr inbounds nuw [[CAP_MAIN_TY]], ptr %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 3
 184 // CHECK: store ptr [[VAR_ADDR]], ptr [[VAR_REF]],
 185 // CHECK: [[SIVAR_REF:%.+]] = getelementptr inbounds nuw [[CAP_MAIN_TY]], ptr %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 4
 186 // CHECK: store ptr [[SIVAR]], ptr [[SIVAR_REF]],
 187
 188 // Allocate task.
 189 // Returns struct kmp_task_t {
 190 //         [[KMP_TASK_T]] task_data;
 191 //         [[KMP_TASK_MAIN_TY]] privates;
 192 //       };
 193 // CHECK: [[RES:%.+]] = call ptr @__kmpc_omp_task_alloc(ptr [[LOC]], i32 [[GTID]], i32 9, i64 120, i64 40, ptr [[TASK_ENTRY:@[^ ]+]])
 194
 195 // Fill kmp_task_t->shareds by copying from original capture argument.
 196 // CHECK: [[TASK:%.+]] = getelementptr inbounds nuw [[KMP_TASK_MAIN_TY]], ptr [[RES]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
 197 // CHECK: [[SHAREDS_REF_ADDR:%.+]] = getelementptr inbounds nuw [[KMP_TASK_T_TY]], ptr [[TASK]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
 198 // CHECK: [[SHAREDS_REF:%.+]] = load ptr, ptr [[SHAREDS_REF_ADDR]],
 199 // CHECK: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[SHAREDS_REF]], ptr align 8 %{{.+}}, i64 40, i1 false)
 200
 201 // Initialize kmp_task_t->privates with default values (no init for simple types, default constructors for classes).
 202 // Also copy address of private copy to the corresponding shareds reference.
 203 // CHECK: [[PRIVATES:%.+]] = getelementptr inbounds nuw [[KMP_TASK_MAIN_TY]], ptr [[RES]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
 204
 205 // Constructors for s_arr and var.
 206 // s_arr;
 207 // CHECK: [[PRIVATE_S_ARR_REF:%.+]] = getelementptr inbounds nuw [[PRIVATES_MAIN_TY]], ptr [[PRIVATES]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
 208 // CHECK: call {{.*}} [[S_DOUBLE_TY_CONSTR]](ptr {{[^,]*}} [[S_ARR_CUR:%[^,]+]])
 209 // CHECK: getelementptr inbounds [[S_DOUBLE_TY]], ptr [[S_ARR_CUR]], i{{.+}} 1
 210 // CHECK: icmp eq
 211 // CHECK: br i1
 212
 213 // var;
 214 // CHECK: [[PRIVATE_VAR_REF:%.+]] = getelementptr inbounds nuw [[PRIVATES_MAIN_TY]], ptr [[PRIVATES]], i{{.+}} 0, i{{.+}} 1
 215 // CHECK: call {{.*}} [[S_DOUBLE_TY_CONSTR]](ptr {{[^,]*}} [[PRIVATE_VAR_REF]])
 216
 217 // t_var;
 218 // vec;
 219 // sivar;
 220
 221 // Provide pointer to destructor function, which will destroy private variables at the end of the task.
 222 // CHECK: [[DESTRUCTORS_REF:%.+]] = getelementptr inbounds nuw [[KMP_TASK_T_TY]], ptr [[TASK]], i{{.+}} 0, i{{.+}} 3
 223 // CHECK: store ptr [[DESTRUCTORS:@.+]], ptr [[DESTRUCTORS_REF]],
 224
 225 // Start task.
 226 // CHECK: call void @__kmpc_taskloop(ptr [[LOC]], i32 [[GTID]], ptr [[RES]], i32 1, ptr %{{.+}}, ptr %{{.+}}, i64 %{{.+}}, i32 1, i32 0, i64 0, ptr [[MAIN_DUP:@.+]])
 227
 228 // CHECK: = call noundef i{{.+}} [[TMAIN_INT:@.+]]()
 229
 230 // No destructors must be called for private copies of s_arr and var.
 231 // CHECK-NOT: getelementptr inbounds nuw [[PRIVATES_MAIN_TY]], ptr [[PRIVATES]], i{{.+}} 0, i{{.+}} 2
 232 // CHECK-NOT: getelementptr inbounds nuw [[PRIVATES_MAIN_TY]], ptr [[PRIVATES]], i{{.+}} 0, i{{.+}} 3
 233 // CHECK: call void [[S_DOUBLE_TY_DESTR:@.+]](ptr
 234 // CHECK-NOT: getelementptr inbounds nuw [[PRIVATES_MAIN_TY]], ptr [[PRIVATES]], i{{.+}} 0, i{{.+}} 2
 235 // CHECK-NOT: getelementptr inbounds nuw [[PRIVATES_MAIN_TY]], ptr [[PRIVATES]], i{{.+}} 0, i{{.+}} 3
 236 // CHECK: ret
 237 //
 238
 239 // CHECK: define internal void [[PRIVATES_MAP_FN:@.+]](ptr noalias noundef %0, ptr noalias noundef %1, ptr noalias noundef %2, ptr noalias noundef %3, ptr noalias noundef %4, ptr noalias noundef %5)
 240 // CHECK: [[PRIVATES:%.+]] = load ptr, ptr
 241 // CHECK: [[PRIV_S_VAR:%.+]] = getelementptr inbounds nuw [[PRIVATES_MAIN_TY]], ptr [[PRIVATES]], i32 0, i32 0
 242 // CHECK: [[ARG3:%.+]] = load ptr, ptr %{{.+}},
 243 // CHECK: store ptr [[PRIV_S_VAR]], ptr [[ARG3]],
 244 // CHECK: [[PRIV_VAR:%.+]] = getelementptr inbounds nuw [[PRIVATES_MAIN_TY]], ptr [[PRIVATES]], i32 0, i32 1
 245 // CHECK: [[ARG1:%.+]] = load ptr, ptr {{.+}},
 246 // CHECK: store ptr [[PRIV_VAR]], ptr [[ARG1]],
 247 // CHECK: [[PRIV_T_VAR:%.+]] = getelementptr inbounds nuw [[PRIVATES_MAIN_TY]], ptr [[PRIVATES]], i32 0, i32 2
 248 // CHECK: [[ARG2:%.+]] = load ptr, ptr %{{.+}},
 249 // CHECK: store ptr [[PRIV_T_VAR]], ptr [[ARG2]],
 250 // CHECK: [[PRIV_VEC:%.+]] = getelementptr inbounds nuw [[PRIVATES_MAIN_TY]], ptr [[PRIVATES]], i32 0, i32 3
 251 // CHECK: [[ARG4:%.+]] = load ptr, ptr %{{.+}},
 252 // CHECK: store ptr [[PRIV_VEC]], ptr [[ARG4]],
 253 // CHECK: [[PRIV_SIVAR:%.+]] = getelementptr inbounds nuw [[PRIVATES_MAIN_TY]], ptr [[PRIVATES]], i32 0, i32 4
 254 // CHECK: [[ARG5:%.+]] = load ptr, ptr %{{.+}},
 255 // CHECK: store ptr [[PRIV_SIVAR]], ptr [[ARG5]],
 256 // CHECK: ret void
 257
 258 // CHECK: define internal noundef i32 [[TASK_ENTRY]](i32 noundef %0, ptr noalias noundef %1)
 259
 260 // CHECK: %__context
 261 // CHECK: [[PRIV_VAR_ADDR:%.+]] = alloca ptr,
 262 // CHECK: [[PRIV_T_VAR_ADDR:%.+]] = alloca ptr,
 263 // CHECK: [[PRIV_S_ARR_ADDR:%.+]] = alloca ptr,
 264 // CHECK: [[PRIV_VEC_ADDR:%.+]] = alloca ptr,
 265 // CHECK: [[PRIV_SIVAR_ADDR:%.+]] = alloca ptr,
 266 // CHECK: store ptr [[PRIVATES_MAP_FN]], ptr [[MAP_FN_ADDR:%.+]],
 267 // CHECK: [[MAP_FN:%.+]] = load ptr, ptr [[MAP_FN_ADDR]],
 268
 269 // CHECK: call void [[MAP_FN]](ptr %{{.+}}, ptr [[PRIV_VAR_ADDR]], ptr [[PRIV_T_VAR_ADDR]], ptr [[PRIV_S_ARR_ADDR]], ptr [[PRIV_VEC_ADDR]], ptr [[PRIV_SIVAR_ADDR]])
 270
 271 // CHECK: [[PRIV_VAR:%.+]] = load ptr, ptr [[PRIV_VAR_ADDR]],
 272 // CHECK: [[PRIV_T_VAR:%.+]] = load ptr, ptr [[PRIV_T_VAR_ADDR]],
 273 // CHECK: [[PRIV_S_ARR:%.+]] = load ptr, ptr [[PRIV_S_ARR_ADDR]],
 274 // CHECK: [[PRIV_VEC:%.+]] = load ptr, ptr [[PRIV_VEC_ADDR]],
 275 // CHECK: [[PRIV_SIVAR:%.+]] = load ptr, ptr [[PRIV_SIVAR_ADDR]],
 276
 277 // Privates actually are used.
 278 // CHECK-DAG: [[PRIV_VAR]]
 279 // CHECK-DAG: [[PRIV_T_VAR]]
 280 // CHECK-DAG: [[PRIV_S_ARR]]
 281 // CHECK-DAG: [[PRIV_VEC]]
 282 // CHECK-DAG: [[PRIV_SIVAR]]
 283
 284 // CHECK:     icmp ne i32 %{{.+}}, 0
 285 // CHECK-NEXT: br i1
 286 // CHECK: call void @llvm.memcpy.p0.p0.i64(ptr align {{[0-9]+}} %
 287 // CHECK: load i32, ptr %
 288 // CHECK: store i32 %{{.+}}, ptr %
 289 // CHECK: phi ptr
 290 // CHECK: phi ptr
 291 // CHECK: call void @llvm.memcpy.p0.p0.i64(ptr align {{[0-9]+}} %
 292 // CHECK: icmp eq ptr %
 293 // CHECK-NEXT: br i1
 294 // CHECK: call void @llvm.memcpy.p0.p0.i64(ptr align {{[0-9]+}} %
 295 // CHECK: load i32, ptr %
 296 // CHECK: store i32 %{{.+}}, ptr %
 297 // CHECK: br label
 298 // CHECK: ret
 299
 300 // CHECK: define internal void [[MAIN_DUP]](ptr noundef %0, ptr noundef %1, i32 noundef %2)
 301 // CHECK: getelementptr inbounds nuw [[KMP_TASK_MAIN_TY]], ptr %{{.+}}, i32 0, i32 0
 302 // CHECK: getelementptr inbounds nuw [[KMP_TASK_T_TY]], ptr %{{.+}}, i32 0, i32 8
 303 // CHECK: load i32, ptr %
 304 // CHECK: store i32 %{{.+}}, ptr %
 305 // CHECK: getelementptr inbounds nuw [[KMP_TASK_MAIN_TY]], ptr %{{.+}}, i32 0, i32 1
 306 // CHECK: getelementptr inbounds nuw [[PRIVATES_MAIN_TY]], ptr %{{.+}}, i32 0, i32 0
 307 // CHECK: getelementptr inbounds [2 x [[S_DOUBLE_TY]]], ptr %{{.+}}, i32 0, i32 0
 308 // CHECK: getelementptr inbounds [[S_DOUBLE_TY]], ptr %{{.+}}, i64 2
 309 // CHECK: br label %
 310
 311 // CHECK: phi ptr
 312 // CHECK: call {{.*}} [[S_DOUBLE_TY_CONSTR]](ptr
 313 // CHECK: getelementptr inbounds [[S_DOUBLE_TY]], ptr %{{.+}}, i64 1
 314 // CHECK: icmp eq ptr %
 315 // CHECK: br i1 %
 316
 317 // CHECK: getelementptr inbounds nuw [[PRIVATES_MAIN_TY]], ptr %{{.+}}, i32 0, i32 1
 318 // CHECK: call {{.*}} [[S_DOUBLE_TY_CONSTR]](ptr
 319 // CHECK: ret void
 320
 321 // CHECK: define internal noundef i32 [[DESTRUCTORS]](i32 noundef %0, ptr noalias noundef %1)
 322 // CHECK: [[PRIVATES:%.+]] = getelementptr inbounds nuw [[KMP_TASK_MAIN_TY]], ptr [[RES_KMP_TASK:%.+]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
 323 // CHECK: [[PRIVATE_S_ARR_REF:%.+]] = getelementptr inbounds nuw [[PRIVATES_MAIN_TY]], ptr [[PRIVATES]], i{{.+}} 0, i{{.+}} 0
 324 // CHECK: [[PRIVATE_VAR_REF:%.+]] = getelementptr inbounds nuw [[PRIVATES_MAIN_TY]], ptr [[PRIVATES]], i{{.+}} 0, i{{.+}} 1
 325 // CHECK: call {{.*}} [[S_DOUBLE_TY_DESTR]](ptr {{[^,]*}} [[PRIVATE_VAR_REF]])
 326 // CHECK: getelementptr inbounds [2 x [[S_DOUBLE_TY]]], ptr [[PRIVATE_S_ARR_REF]], i{{.+}} 0, i{{.+}} 0
 327 // CHECK: getelementptr inbounds [[S_DOUBLE_TY]], ptr %{{.+}}, i{{.+}} 2
 328 // CHECK: [[PRIVATE_S_ARR_ELEM_REF:%.+]] = getelementptr inbounds [[S_DOUBLE_TY]], ptr %{{.+}}, i{{.+}} -1
 329 // CHECK: call {{.*}} [[S_DOUBLE_TY_DESTR]](ptr {{[^,]*}} [[PRIVATE_S_ARR_ELEM_REF]])
 330 // CHECK: icmp eq
 331 // CHECK: br i1
 332 // CHECK: ret i32
 333
 334 // CHECK: define {{.*}} i{{[0-9]+}} [[TMAIN_INT]]()
 335 // CHECK: alloca [[S_INT_TY]],
 336 // CHECK: [[TEST:%.+]] = alloca [[S_INT_TY]],
 337 // CHECK: [[T_VAR_ADDR:%.+]] = alloca i32, align 128
 338 // CHECK: [[VEC_ADDR:%.+]] = alloca [2 x i32],
 339 // CHECK: [[S_ARR_ADDR:%.+]] = alloca [2 x [[S_INT_TY]]],
 340 // CHECK: [[VAR_ADDR:%.+]] = alloca [[S_INT_TY]],
 341 // CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(ptr [[LOC:@.+]])
 342
 343 // CHECK: call {{.*}} [[S_INT_TY_CONSTR:@.+]](ptr {{[^,]*}} [[TEST]])
 344
 345 // Store original variables in capture struct.
 346 // CHECK: [[VEC_REF:%.+]] = getelementptr inbounds nuw [[CAP_TMAIN_TY]], ptr %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 0
 347 // CHECK: store ptr [[VEC_ADDR]], ptr [[VEC_REF]],
 348 // CHECK: [[T_VAR_REF:%.+]] = getelementptr inbounds nuw [[CAP_TMAIN_TY]], ptr %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 1
 349 // CHECK: store ptr [[T_VAR_ADDR]], ptr [[T_VAR_REF]],
 350 // CHECK: [[S_ARR_REF:%.+]] = getelementptr inbounds nuw [[CAP_TMAIN_TY]], ptr %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 2
 351 // CHECK: store ptr [[S_ARR_ADDR]], ptr [[S_ARR_REF]],
 352 // CHECK: [[VAR_REF:%.+]] = getelementptr inbounds nuw [[CAP_TMAIN_TY]], ptr %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 3
 353 // CHECK: store ptr [[VAR_ADDR]], ptr [[VAR_REF]],
 354
 355 // Allocate task.
 356 // Returns struct kmp_task_t {
 357 //         [[KMP_TASK_T_TY]] task_data;
 358 //         [[KMP_TASK_TMAIN_TY]] privates;
 359 //       };
 360 // CHECK: [[RES:%.+]] = call ptr @__kmpc_omp_task_alloc(ptr [[LOC]], i32 [[GTID]], i32 9, i64 256, i64 32, ptr [[TASK_ENTRY:@[^ ]+]])
 361
 362 // Fill kmp_task_t->shareds by copying from original capture argument.
 363 // CHECK: [[TASK:%.+]] = getelementptr inbounds nuw [[KMP_TASK_TMAIN_TY]], ptr [[RES]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
 364 // CHECK: [[SHAREDS_REF_ADDR:%.+]] = getelementptr inbounds nuw [[KMP_TASK_T_TY]], ptr [[TASK]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
 365 // CHECK: [[SHAREDS_REF:%.+]] = load ptr, ptr [[SHAREDS_REF_ADDR]],
 366 // CHECK: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[SHAREDS_REF]], ptr align 8 %{{.+}}, i64 32, i1 false)
 367
 368 // Initialize kmp_task_t->privates with default values (no init for simple types, default constructors for classes).
 369 // CHECK: [[PRIVATES:%.+]] = getelementptr inbounds nuw [[KMP_TASK_TMAIN_TY]], ptr [[RES]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
 370
 371 // t_var;
 372 // vec;
 373
 374 // Constructors for s_arr and var.
 375 // a_arr;
 376 // CHECK: [[PRIVATE_S_ARR_REF:%.+]] = getelementptr inbounds nuw [[PRIVATES_TMAIN_TY]], ptr [[PRIVATES]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
 377 // CHECK: getelementptr inbounds [2 x [[S_INT_TY]]], ptr [[PRIVATE_S_ARR_REF]], i{{.+}} 0, i{{.+}} 0
 378 // CHECK: getelementptr inbounds [[S_INT_TY]], ptr %{{.+}}, i{{.+}} 2
 379 // CHECK: call {{.*}} [[S_INT_TY_CONSTR]](ptr {{[^,]*}} [[S_ARR_CUR:%[^,]+]])
 380 // CHECK: getelementptr inbounds [[S_INT_TY]], ptr [[S_ARR_CUR]], i{{.+}} 1
 381 // CHECK: icmp eq
 382 // CHECK: br i1
 383
 384 // var;
 385 // CHECK: [[PRIVATE_VAR_REF:%.+]] = getelementptr inbounds nuw [[PRIVATES_TMAIN_TY]], ptr [[PRIVATES]], i{{.+}} 0, i{{.+}} 3
 386 // CHECK: call {{.*}} [[S_INT_TY_CONSTR]](ptr {{[^,]*}} [[PRIVATE_VAR_REF]])
 387
 388 // Provide pointer to destructor function, which will destroy private variables at the end of the task.
 389 // CHECK: [[DESTRUCTORS_REF:%.+]] = getelementptr inbounds nuw [[KMP_TASK_T_TY]], ptr [[TASK]], i{{.+}} 0, i{{.+}} 3
 390 // CHECK: store ptr [[DESTRUCTORS:@.+]], ptr [[DESTRUCTORS_REF]],
 391
 392 // Start task.
 393 // CHECK: call void @__kmpc_taskloop(ptr [[LOC]], i32 [[GTID]], ptr [[RES]], i32 1, ptr %{{.+}}, ptr %{{.+}}, i64 %{{.+}}, i32 1, i32 0, i64 0, ptr [[TMAIN_DUP:@.+]])
 394
 395 // No destructors must be called for private copies of s_arr and var.
 396 // CHECK-NOT: getelementptr inbounds nuw [[PRIVATES_TMAIN_TY]], ptr [[PRIVATES]], i{{.+}} 0, i{{.+}} 2
 397 // CHECK-NOT: getelementptr inbounds nuw [[PRIVATES_TMAIN_TY]], ptr [[PRIVATES]], i{{.+}} 0, i{{.+}} 3
 398 // CHECK: call void [[S_INT_TY_DESTR:@.+]](ptr noundef
 399 // CHECK-NOT: getelementptr inbounds nuw [[PRIVATES_TMAIN_TY]], ptr [[PRIVATES]], i{{.+}} 0, i{{.+}} 2
 400 // CHECK-NOT: getelementptr inbounds nuw [[PRIVATES_TMAIN_TY]], ptr [[PRIVATES]], i{{.+}} 0, i{{.+}} 3
 401 // CHECK: ret
 402 //
 403
 404 // CHECK: define internal void [[PRIVATES_MAP_FN:@.+]](ptr noalias noundef %0, ptr noalias noundef %1, ptr noalias noundef %2, ptr noalias noundef %3, ptr noalias noundef %4)
 405 // CHECK: [[PRIVATES:%.+]] = load ptr, ptr
 406 // CHECK: [[PRIV_T_VAR:%.+]] = getelementptr inbounds nuw [[PRIVATES_TMAIN_TY]], ptr [[PRIVATES]], i32 0, i32 0
 407 // CHECK: [[ARG1:%.+]] = load ptr, ptr %{{.+}},
 408 // CHECK: store ptr [[PRIV_T_VAR]], ptr [[ARG1]],
 409 // CHECK: [[PRIV_VEC:%.+]] = getelementptr inbounds nuw [[PRIVATES_TMAIN_TY]], ptr [[PRIVATES]], i32 0, i32 1
 410 // CHECK: [[ARG2:%.+]] = load ptr, ptr %{{.+}},
 411 // CHECK: store ptr [[PRIV_VEC]], ptr [[ARG2]],
 412 // CHECK: [[PRIV_S_VAR:%.+]] = getelementptr inbounds nuw [[PRIVATES_TMAIN_TY]], ptr [[PRIVATES]], i32 0, i32 2
 413 // CHECK: [[ARG3:%.+]] = load ptr, ptr %{{.+}},
 414 // CHECK: store ptr [[PRIV_S_VAR]], ptr [[ARG3]],
 415 // CHECK: [[PRIV_VAR:%.+]] = getelementptr inbounds nuw [[PRIVATES_TMAIN_TY]], ptr [[PRIVATES]], i32 0, i32 3
 416 // CHECK: [[ARG4:%.+]] = load ptr, ptr {{.+}},
 417 // CHECK: store ptr [[PRIV_VAR]], ptr [[ARG4]],
 418 // CHECK: ret void
 419
 420 // CHECK: define internal noundef i32 [[TASK_ENTRY]](i32 noundef %0, ptr noalias noundef %1)
 421 // CHECK: %__context
 422 // CHECK-DAG: [[PRIV_T_VAR_ADDR:%.+]] = alloca ptr,
 423 // CHECK-DAG: [[PRIV_VEC_ADDR:%.+]] = alloca ptr,
 424 // CHECK-DAG: [[PRIV_S_ARR_ADDR:%.+]] = alloca ptr,
 425 // CHECK-DAG: [[PRIV_VAR_ADDR:%.+]] = alloca ptr,
 426 // CHECK: store ptr [[PRIVATES_MAP_FN]], ptr [[MAP_FN_ADDR:%.+]],
 427 // CHECK: [[MAP_FN:%.+]] = load ptr, ptr [[MAP_FN_ADDR]],
 428 // CHECK: call void [[MAP_FN]](ptr %{{.+}}, ptr [[PRIV_T_VAR_ADDR]], ptr [[PRIV_VEC_ADDR]], ptr [[PRIV_S_ARR_ADDR]], ptr [[PRIV_VAR_ADDR]])
 429 // CHECK: [[PRIV_T_VAR:%.+]] = load ptr, ptr [[PRIV_T_VAR_ADDR]],
 430 // CHECK: [[PRIV_VEC:%.+]] = load ptr, ptr [[PRIV_VEC_ADDR]],
 431 // CHECK: [[PRIV_S_ARR:%.+]] = load ptr, ptr [[PRIV_S_ARR_ADDR]],
 432 // CHECK: [[PRIV_VAR:%.+]] = load ptr, ptr [[PRIV_VAR_ADDR]],
 433
 434 // Privates actually are used.
 435 // CHECK-DAG: [[PRIV_VAR]]
 436 // CHECK-DAG: [[PRIV_T_VAR]]
 437 // CHECK-DAG: [[PRIV_S_ARR]]
 438 // CHECK-DAG: [[PRIV_VEC]]
 439
 440 // CHECK:     icmp ne i32 %{{.+}}, 0
 441 // CHECK-NEXT: br i1
 442 // CHECK: load i32, ptr %
 443 // CHECK: store i32 %{{.+}}, ptr %
 444 // CHECK: call void @llvm.memcpy.p0.p0.i64(ptr align {{[0-9]+}} %
 445 // CHECK: phi ptr
 446 // CHECK: phi ptr
 447 // CHECK: call void @llvm.memcpy.p0.p0.i64(ptr align {{[0-9]+}} %
 448 // CHECK: icmp eq ptr %
 449 // CHECK-NEXT: br i1
 450 // CHECK: call void @llvm.memcpy.p0.p0.i64(ptr align {{[0-9]+}} %
 451 // CHECK: br label
 452 // CHECK: ret
 453
 454 // CHECK: define internal void [[TMAIN_DUP]](ptr noundef %0, ptr noundef %1, i32 noundef %2)
 455 // CHECK: getelementptr inbounds nuw [[KMP_TASK_TMAIN_TY]], ptr %{{.+}}, i32 0, i32 0
 456 // CHECK: getelementptr inbounds nuw [[KMP_TASK_T_TY]], ptr %{{.+}}, i32 0, i32 8
 457 // CHECK: load i32, ptr %
 458 // CHECK: store i32 %{{.+}}, ptr %
 459 // CHECK: getelementptr inbounds nuw [[KMP_TASK_TMAIN_TY]], ptr %{{.+}}, i32 0, i32 2
 460 // CHECK: getelementptr inbounds nuw [[PRIVATES_TMAIN_TY]], ptr %{{.+}}, i32 0, i32 2
 461 // CHECK: getelementptr inbounds [2 x [[S_INT_TY]]], ptr %{{.+}}, i32 0, i32 0
 462 // CHECK: getelementptr inbounds [[S_INT_TY]], ptr %{{.+}}, i64 2
 463 // CHECK: br label %
 464
 465 // CHECK: phi ptr
 466 // CHECK: call {{.*}} [[S_INT_TY_CONSTR]](ptr
 467 // CHECK: getelementptr inbounds [[S_INT_TY]], ptr %{{.+}}, i64 1
 468 // CHECK: icmp eq ptr %
 469 // CHECK: br i1 %
 470
 471 // CHECK: getelementptr inbounds nuw [[PRIVATES_TMAIN_TY]], ptr %{{.+}}, i32 0, i32 3
 472 // CHECK: call {{.*}} [[S_INT_TY_CONSTR]](ptr
 473 // CHECK: ret void
 474
 475 // CHECK: define internal noundef i32 [[DESTRUCTORS]](i32 noundef %0, ptr noalias noundef %1)
 476 // CHECK: [[PRIVATES:%.+]] = getelementptr inbounds nuw [[KMP_TASK_TMAIN_TY]], ptr [[RES_KMP_TASK:%.+]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
 477 // CHECK: [[PRIVATE_S_ARR_REF:%.+]] = getelementptr inbounds nuw [[PRIVATES_TMAIN_TY]], ptr [[PRIVATES]], i{{.+}} 0, i{{.+}} 2
 478 // CHECK: [[PRIVATE_VAR_REF:%.+]] = getelementptr inbounds nuw [[PRIVATES_TMAIN_TY]], ptr [[PRIVATES]], i{{.+}} 0, i{{.+}} 3
 479 // CHECK: call void [[S_INT_TY_DESTR]](ptr {{[^,]*}} [[PRIVATE_VAR_REF]])
 480 // CHECK: getelementptr inbounds [2 x [[S_INT_TY]]], ptr [[PRIVATE_S_ARR_REF]], i{{.+}} 0, i{{.+}} 0
 481 // CHECK: getelementptr inbounds [[S_INT_TY]], ptr %{{.+}}, i{{.+}} 2
 482 // CHECK: [[PRIVATE_S_ARR_ELEM_REF:%.+]] = getelementptr inbounds [[S_INT_TY]], ptr %{{.+}}, i{{.+}} -1
 483 // CHECK: call void [[S_INT_TY_DESTR]](ptr {{[^,]*}} [[PRIVATE_S_ARR_ELEM_REF]])
 484 // CHECK: icmp eq
 485 // CHECK: br i1
 486 // CHECK: ret i32
 487
 488 #endif
 489 #else
 490 // ARRAY-LABEL: array_func
 491 struct St {
 492   int a, b;
 493   St() : a(0), b(0) {}
 494   St(const St &) {}
 495   ~St() {}
 496 };
 497
 498 void array_func(int n, float a[n], St s[2]) {
 499 // ARRAY: call ptr @__kmpc_omp_task_alloc(
 500 // ARRAY: call void @__kmpc_taskloop(
 501 // ARRAY: store ptr %{{.+}}, ptr %{{.+}},
 502 // ARRAY: store ptr %{{.+}}, ptr %{{.+}},
 503 // ARRAY: icmp ne i32 %{{.+}}, 0
 504 // ARRAY: store ptr %{{.+}}, ptr %{{.+}},
 505 // ARRAY: store ptr %{{.+}}, ptr %{{.+}},
 506 #pragma omp master taskloop simd lastprivate(a, s)
 507   for (int i = 0; i < 10; ++i)
 508     ;
 509 }
 510 #endif
 511