1 // RUN: %clang_cc1 -verify -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 %s -emit-llvm -o - | FileCheck %s 4 // RUN: %clang_cc1 -verify -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 -fopenmp -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s 6 // expected-no-diagnostics 7 // REQUIRES: x86-registered-target 8 #ifndef HEADER 9 #define HEADER 10 11 volatile double g, g_orig; 12 volatile double &g1 = g_orig; 13 14 template <class T> 15 struct S { 16 T f; 17 S(T a) : f(a + g) {} 18 S() : f(g) {} 19 operator T() { return T(); } 20 S &operator&(const S &) { return *this; } 21 ~S() {} 22 }; 23 24 // CHECK-DAG: [[S_FLOAT_TY:%.+]] = type { float } 25 // CHECK-DAG: [[S_INT_TY:%.+]] = type { i{{[0-9]+}} } 26 // CHECK-DAG: [[ATOMIC_REDUCE_BARRIER_LOC:@.+]] = private unnamed_addr constant %{{.+}} { i32 0, i32 18, i32 0, i32 0, i8* 27 // CHECK-DAG: [[IMPLICIT_BARRIER_LOC:@.+]] = private unnamed_addr constant %{{.+}} { i32 0, i32 66, i32 0, i32 0, i8* 28 // CHECK-DAG: [[REDUCTION_LOC:@.+]] = private unnamed_addr constant %{{.+}} { i32 0, i32 18, i32 0, i32 0, i8* 29 // CHECK-DAG: [[REDUCTION_LOCK:@.+]] = common global [8 x i32] zeroinitializer 30 31 template <typename T> 32 T tmain() { 33 T t; 34 S<T> test; 35 T t_var = T(), t_var1; 36 T vec[] = {1, 2}; 37 S<T> s_arr[] = {1, 2}; 38 S<T> &var = test; 39 S<T> var1; 40 #pragma omp parallel 41 #pragma omp for reduction(+:t_var) reduction(&:var) reduction(&& : var1) reduction(min: t_var1) nowait 42 for (int i = 0; i < 2; ++i) { 43 vec[i] = t_var; 44 s_arr[i] = var; 45 } 46 #pragma omp parallel 47 #pragma omp for reduction(&& : t_var) 48 for (int i = 0; i < 2; ++i) { 49 vec[i] = t_var; 50 s_arr[i] = var; 51 } 52 return T(); 53 } 54 55 int main() { 56 #ifdef LAMBDA 57 // LAMBDA: [[G:@.+]] = global double 58 // LAMBDA-LABEL: @main 59 // LAMBDA: call void [[OUTER_LAMBDA:@.+]]( 60 [&]() { 61 // LAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]]( 62 // LAMBDA: call void {{.+}} @__kmpc_fork_call({{.+}}, i32 0, {{.+}}* [[OMP_REGION:@.+]] to {{.+}}) 63 #pragma omp parallel 64 #pragma omp for reduction(+:g, g1) 65 for (int i = 0; i < 2; ++i) { 66 // LAMBDA: define{{.*}} internal{{.*}} void [[OMP_REGION]](i32* noalias %{{.+}}, i32* noalias %{{.+}}) 67 // LAMBDA: [[G_PRIVATE_ADDR:%.+]] = alloca double, 68 69 // Reduction list for runtime. 70 // LAMBDA: [[RED_LIST:%.+]] = alloca [2 x i8*], 71 72 // LAMBDA: store double 0.0{{.+}}, double* [[G_PRIVATE_ADDR]] 73 // LAMBDA: call void @__kmpc_for_static_init_4( 74 g = 1; 75 g1 = 1; 76 // LAMBDA: store double 1.0{{.+}}, double* [[G_PRIVATE_ADDR]], 77 // LAMBDA: [[G_PRIVATE_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG:%.+]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 78 // LAMBDA: store double* [[G_PRIVATE_ADDR]], double** [[G_PRIVATE_ADDR_REF]] 79 // LAMBDA: call void [[INNER_LAMBDA:@.+]](%{{.+}}* [[ARG]]) 80 // LAMBDA: call void @__kmpc_for_static_fini( 81 82 // LAMBDA: [[G_PRIV_REF:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[RED_LIST]], i64 0, i64 0 83 // LAMBDA: [[BITCAST:%.+]] = bitcast double* [[G_PRIVATE_ADDR]] to i8* 84 // LAMBDA: store i8* [[BITCAST]], i8** [[G_PRIV_REF]], 85 // LAMBDA: call i32 @__kmpc_reduce( 86 // LAMBDA: switch i32 %{{.+}}, label %[[REDUCTION_DONE:.+]] [ 87 // LAMBDA: i32 1, label %[[CASE1:.+]] 88 // LAMBDA: i32 2, label %[[CASE2:.+]] 89 // LAMBDA: [[CASE1]] 90 // LAMBDA: [[G_VAL:%.+]] = load double, double* [[G]] 91 // LAMBDA: [[G_PRIV_VAL:%.+]] = load double, double* [[G_PRIVATE_ADDR]] 92 // LAMBDA: [[ADD:%.+]] = fadd double [[G_VAL]], [[G_PRIV_VAL]] 93 // LAMBDA: store double [[ADD]], double* [[G]] 94 // LAMBDA: call void @__kmpc_end_reduce( 95 // LAMBDA: br label %[[REDUCTION_DONE]] 96 // LAMBDA: [[CASE2]] 97 // LAMBDA: [[G_PRIV_VAL:%.+]] = load double, double* [[G_PRIVATE_ADDR]] 98 // LAMBDA: fadd double 99 // LAMBDA: cmpxchg i64* 100 // LAMBDA: call void @__kmpc_end_reduce( 101 // LAMBDA: br label %[[REDUCTION_DONE]] 102 // LAMBDA: [[REDUCTION_DONE]] 103 // LAMBDA: ret void 104 [&]() { 105 // LAMBDA: define {{.+}} void [[INNER_LAMBDA]](%{{.+}}* [[ARG_PTR:%.+]]) 106 // LAMBDA: store %{{.+}}* [[ARG_PTR]], %{{.+}}** [[ARG_PTR_REF:%.+]], 107 g = 2; 108 g1 = 2; 109 // LAMBDA: [[ARG_PTR:%.+]] = load %{{.+}}*, %{{.+}}** [[ARG_PTR_REF]] 110 // LAMBDA: [[G_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 111 // LAMBDA: [[G_REF:%.+]] = load double*, double** [[G_PTR_REF]] 112 // LAMBDA: store double 2.0{{.+}}, double* [[G_REF]] 113 }(); 114 } 115 }(); 116 return 0; 117 #elif defined(BLOCKS) 118 // BLOCKS: [[G:@.+]] = global double 119 // BLOCKS-LABEL: @main 120 // BLOCKS: call void {{%.+}}(i8 121 ^{ 122 // BLOCKS: define{{.*}} internal{{.*}} void {{.+}}(i8* 123 // BLOCKS: call void {{.+}} @__kmpc_fork_call({{.+}}, i32 0, {{.+}}* [[OMP_REGION:@.+]] to {{.+}}) 124 #pragma omp parallel 125 #pragma omp for reduction(-:g, g1) 126 for (int i = 0; i < 2; ++i) { 127 // BLOCKS: define{{.*}} internal{{.*}} void [[OMP_REGION]](i32* noalias %{{.+}}, i32* noalias %{{.+}}) 128 // BLOCKS: [[G_PRIVATE_ADDR:%.+]] = alloca double, 129 130 // Reduction list for runtime. 131 // BLOCKS: [[RED_LIST:%.+]] = alloca [2 x i8*], 132 133 // BLOCKS: store double 0.0{{.+}}, double* [[G_PRIVATE_ADDR]] 134 g = 1; 135 g1 = 1; 136 // BLOCKS: call void @__kmpc_for_static_init_4( 137 // BLOCKS: store double 1.0{{.+}}, double* [[G_PRIVATE_ADDR]], 138 // BLOCKS-NOT: [[G]]{{[[^:word:]]}} 139 // BLOCKS: double* [[G_PRIVATE_ADDR]] 140 // BLOCKS-NOT: [[G]]{{[[^:word:]]}} 141 // BLOCKS: call void {{%.+}}(i8 142 // BLOCKS: call void @__kmpc_for_static_fini( 143 144 // BLOCKS: [[G_PRIV_REF:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[RED_LIST]], i64 0, i64 0 145 // BLOCKS: [[BITCAST:%.+]] = bitcast double* [[G_PRIVATE_ADDR]] to i8* 146 // BLOCKS: store i8* [[BITCAST]], i8** [[G_PRIV_REF]], 147 // BLOCKS: call i32 @__kmpc_reduce( 148 // BLOCKS: switch i32 %{{.+}}, label %[[REDUCTION_DONE:.+]] [ 149 // BLOCKS: i32 1, label %[[CASE1:.+]] 150 // BLOCKS: i32 2, label %[[CASE2:.+]] 151 // BLOCKS: [[CASE1]] 152 // BLOCKS: [[G_VAL:%.+]] = load double, double* [[G]] 153 // BLOCKS: [[G_PRIV_VAL:%.+]] = load double, double* [[G_PRIVATE_ADDR]] 154 // BLOCKS: [[ADD:%.+]] = fadd double [[G_VAL]], [[G_PRIV_VAL]] 155 // BLOCKS: store double [[ADD]], double* [[G]] 156 // BLOCKS: call void @__kmpc_end_reduce( 157 // BLOCKS: br label %[[REDUCTION_DONE]] 158 // BLOCKS: [[CASE2]] 159 // BLOCKS: [[G_PRIV_VAL:%.+]] = load double, double* [[G_PRIVATE_ADDR]] 160 // BLOCKS: fadd double 161 // BLOCKS: cmpxchg i64* 162 // BLOCKS: call void @__kmpc_end_reduce( 163 // BLOCKS: br label %[[REDUCTION_DONE]] 164 // BLOCKS: [[REDUCTION_DONE]] 165 // BLOCKS: ret void 166 ^{ 167 // BLOCKS: define {{.+}} void {{@.+}}(i8* 168 g = 2; 169 g1 = 2; 170 // BLOCKS-NOT: [[G]]{{[[^:word:]]}} 171 // BLOCKS: store double 2.0{{.+}}, double* 172 // BLOCKS-NOT: [[G]]{{[[^:word:]]}} 173 // BLOCKS: ret 174 }(); 175 } 176 }(); 177 return 0; 178 #else 179 S<float> test; 180 float t_var = 0, t_var1; 181 int vec[] = {1, 2}; 182 S<float> s_arr[] = {1, 2}; 183 S<float> &var = test; 184 S<float> var1; 185 #pragma omp parallel 186 #pragma omp for reduction(+:t_var) reduction(&:var) reduction(&& : var1) reduction(min: t_var1) 187 for (int i = 0; i < 2; ++i) { 188 vec[i] = t_var; 189 s_arr[i] = var; 190 } 191 return tmain<int>(); 192 #endif 193 } 194 195 // CHECK: define {{.*}}i{{[0-9]+}} @main() 196 // CHECK: [[TEST:%.+]] = alloca [[S_FLOAT_TY]], 197 // CHECK: call {{.*}} [[S_FLOAT_TY_CONSTR:@.+]]([[S_FLOAT_TY]]* [[TEST]]) 198 // CHECK: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{[0-9]+}} 6, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, float*, [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]*, float*, [2 x i32]*, [2 x [[S_FLOAT_TY]]]*)* [[MAIN_MICROTASK:@.+]] to void 199 // CHECK: = call {{.*}}i{{.+}} [[TMAIN_INT:@.+]]() 200 // CHECK: call {{.*}} [[S_FLOAT_TY_DESTR:@.+]]([[S_FLOAT_TY]]* 201 // CHECK: ret 202 // 203 // CHECK: define internal void [[MAIN_MICROTASK]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, float* dereferenceable(4) %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}}, float* dereferenceable(4) %{{.+}}, [2 x i32]* dereferenceable(8) %vec, [2 x [[S_FLOAT_TY]]]* dereferenceable(8) %{{.+}}) 204 // CHECK: [[T_VAR_PRIV:%.+]] = alloca float, 205 // CHECK: [[VAR_PRIV:%.+]] = alloca [[S_FLOAT_TY]], 206 // CHECK: [[VAR1_PRIV:%.+]] = alloca [[S_FLOAT_TY]], 207 // CHECK: [[T_VAR1_PRIV:%.+]] = alloca float, 208 209 // Reduction list for runtime. 210 // CHECK: [[RED_LIST:%.+]] = alloca [4 x i8*], 211 212 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]], 213 214 // CHECK: [[T_VAR_REF:%.+]] = load float*, float** % 215 // CHECK: [[VAR1_REF:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** % 216 // CHECK: [[T_VAR1_REF:%.+]] = load float*, float** % 217 218 // For + reduction operation initial value of private variable is 0. 219 // CHECK: store float 0.0{{.+}}, float* [[T_VAR_PRIV]], 220 221 // For & reduction operation initial value of private variable is ones in all bits. 222 // CHECK: [[VAR_REF:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** % 223 // CHECK: call {{.*}} [[S_FLOAT_TY_CONSTR:@.+]]([[S_FLOAT_TY]]* [[VAR_PRIV]]) 224 225 // For && reduction operation initial value of private variable is 1.0. 226 // CHECK: call {{.*}} [[S_FLOAT_TY_CONSTR:@.+]]([[S_FLOAT_TY]]* [[VAR1_PRIV]]) 227 228 // For min reduction operation initial value of private variable is largest repesentable value. 229 // CHECK: store float 0x47EFFFFFE0000000, float* [[T_VAR1_PRIV]], 230 231 232 // CHECK: [[GTID_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_ADDR]] 233 // CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_REF]] 234 // CHECK: call void @__kmpc_for_static_init_4( 235 // Skip checks for internal operations. 236 // CHECK: call void @__kmpc_for_static_fini( 237 238 // void *RedList[<n>] = {<ReductionVars>[0], ..., <ReductionVars>[<n>-1]}; 239 240 // CHECK: [[T_VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 0 241 // CHECK: [[BITCAST:%.+]] = bitcast float* [[T_VAR_PRIV]] to i8* 242 // CHECK: store i8* [[BITCAST]], i8** [[T_VAR_PRIV_REF]], 243 // CHECK: [[VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 1 244 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_PRIV]] to i8* 245 // CHECK: store i8* [[BITCAST]], i8** [[VAR_PRIV_REF]], 246 // CHECK: [[VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 2 247 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_PRIV]] to i8* 248 // CHECK: store i8* [[BITCAST]], i8** [[VAR1_PRIV_REF]], 249 // CHECK: [[T_VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 3 250 // CHECK: [[BITCAST:%.+]] = bitcast float* [[T_VAR1_PRIV]] to i8* 251 // CHECK: store i8* [[BITCAST]], i8** [[T_VAR1_PRIV_REF]], 252 253 // res = __kmpc_reduce(<loc>, <gtid>, <n>, sizeof(RedList), RedList, reduce_func, &<lock>); 254 255 // CHECK: [[BITCAST:%.+]] = bitcast [4 x i8*]* [[RED_LIST]] to i8* 256 // CHECK: [[RES:%.+]] = call i32 @__kmpc_reduce(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], i32 4, i64 32, i8* [[BITCAST]], void (i8*, i8*)* [[REDUCTION_FUNC:@.+]], [8 x i32]* [[REDUCTION_LOCK]]) 257 258 // switch(res) 259 // CHECK: switch i32 [[RES]], label %[[RED_DONE:.+]] [ 260 // CHECK: i32 1, label %[[CASE1:.+]] 261 // CHECK: i32 2, label %[[CASE2:.+]] 262 // CHECK: ] 263 264 // case 1: 265 // t_var += t_var_reduction; 266 // CHECK: [[T_VAR_VAL:%.+]] = load float, float* [[T_VAR_REF]], 267 // CHECK: [[T_VAR_PRIV_VAL:%.+]] = load float, float* [[T_VAR_PRIV]], 268 // CHECK: [[UP:%.+]] = fadd float [[T_VAR_VAL]], [[T_VAR_PRIV_VAL]] 269 // CHECK: store float [[UP]], float* [[T_VAR_REF]], 270 271 // var = var.operator &(var_reduction); 272 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_REF]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_PRIV]]) 273 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_REF]] to i8* 274 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8* 275 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 276 277 // var1 = var1.operator &&(var1_reduction); 278 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_REF]]) 279 // CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 280 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]] 281 // CHECK: [[TRUE]] 282 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_PRIV]]) 283 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 284 // CHECK: br label %[[END2]] 285 // CHECK: [[END2]] 286 // CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ] 287 // CHECK: [[CONV:%.+]] = uitofp i1 [[COND_LVALUE]] to float 288 // CHECK: call void @{{.+}}([[S_FLOAT_TY]]* [[COND_LVALUE:%.+]], float [[CONV]]) 289 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_REF]] to i8* 290 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8* 291 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 292 293 // t_var1 = min(t_var1, t_var1_reduction); 294 // CHECK: [[T_VAR1_VAL:%.+]] = load float, float* [[T_VAR1_REF]], 295 // CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load float, float* [[T_VAR1_PRIV]], 296 // CHECK: [[CMP:%.+]] = fcmp olt float [[T_VAR1_VAL]], [[T_VAR1_PRIV_VAL]] 297 // CHECK: br i1 [[CMP]] 298 // CHECK: [[UP:%.+]] = phi float 299 // CHECK: store float [[UP]], float* [[T_VAR1_REF]], 300 301 // __kmpc_end_reduce(<loc>, <gtid>, &<lock>); 302 // CHECK: call void @__kmpc_end_reduce(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], [8 x i32]* [[REDUCTION_LOCK]]) 303 304 // break; 305 // CHECK: br label %[[RED_DONE]] 306 307 // case 2: 308 // t_var += t_var_reduction; 309 // CHECK: load float, float* [[T_VAR_PRIV]] 310 // CHECK: [[T_VAR_REF_INT:%.+]] = bitcast float* [[T_VAR_REF]] to i32* 311 // CHECK: [[OLD1:%.+]] = load atomic i32, i32* [[T_VAR_REF_INT]] monotonic, 312 // CHECK: br label %[[CONT:.+]] 313 // CHECK: [[CONT]] 314 // CHECK: [[ORIG_OLD_INT:%.+]] = phi i32 [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %[[CONT]] ] 315 // CHECK: fadd float 316 // CHECK: [[UP_INT:%.+]] = load i32, i32* 317 // CHECK: [[T_VAR_REF_INT:%.+]] = bitcast float* [[T_VAR_REF]] to i32* 318 // CHECK: [[RES:%.+]] = cmpxchg i32* [[T_VAR_REF_INT]], i32 [[ORIG_OLD_INT]], i32 [[UP_INT]] monotonic monotonic 319 // CHECK: [[OLD2:%.+]] = extractvalue { i32, i1 } [[RES]], 0 320 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1 321 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[ATOMIC_DONE:.+]], label %[[CONT]] 322 // CHECK: [[ATOMIC_DONE]] 323 324 // var = var.operator &(var_reduction); 325 // CHECK: call void @__kmpc_critical( 326 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_REF]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_PRIV]]) 327 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_REF]] to i8* 328 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8* 329 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 330 // CHECK: call void @__kmpc_end_critical( 331 332 // var1 = var1.operator &&(var1_reduction); 333 // CHECK: call void @__kmpc_critical( 334 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_REF]]) 335 // CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 336 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]] 337 // CHECK: [[TRUE]] 338 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_PRIV]]) 339 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 340 // CHECK: br label %[[END2]] 341 // CHECK: [[END2]] 342 // CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ] 343 // CHECK: [[CONV:%.+]] = uitofp i1 [[COND_LVALUE]] to float 344 // CHECK: call void @{{.+}}([[S_FLOAT_TY]]* [[COND_LVALUE:%.+]], float [[CONV]]) 345 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_REF]] to i8* 346 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8* 347 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 348 // CHECK: call void @__kmpc_end_critical( 349 350 // t_var1 = min(t_var1, t_var1_reduction); 351 // CHECK: load float, float* [[T_VAR1_PRIV]] 352 // CHECK: [[T_VAR1_REF_INT:%.+]] = bitcast float* [[T_VAR1_REF]] to i32* 353 // CHECK: [[OLD1:%.+]] = load atomic i32, i32* [[T_VAR1_REF_INT]] monotonic, 354 // CHECK: br label %[[CONT:.+]] 355 // CHECK: [[CONT]] 356 // CHECK: [[ORIG_OLD_INT:%.+]] = phi i32 [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %{{.+}} ] 357 // CHECK: [[CMP:%.+]] = fcmp olt float 358 // CHECK: br i1 [[CMP]] 359 // CHECK: phi float 360 // CHECK: [[UP_INT:%.+]] = load i32 361 // CHECK: [[T_VAR1_REF_INT:%.+]] = bitcast float* [[T_VAR1_REF]] to i32* 362 // CHECK: [[RES:%.+]] = cmpxchg i32* [[T_VAR1_REF_INT]], i32 [[ORIG_OLD_INT]], i32 [[UP_INT]] monotonic monotonic 363 // CHECK: [[OLD2:%.+]] = extractvalue { i32, i1 } [[RES]], 0 364 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1 365 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[ATOMIC_DONE:.+]], label %[[CONT]] 366 // CHECK: [[ATOMIC_DONE]] 367 368 // __kmpc_end_reduce(<loc>, <gtid>, &<lock>); 369 // CHECK: call void @__kmpc_end_reduce(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], [8 x i32]* [[REDUCTION_LOCK]]) 370 371 // break; 372 // CHECK: br label %[[RED_DONE]] 373 // CHECK: [[RED_DONE]] 374 // CHECK-DAG: call {{.*}} [[S_FLOAT_TY_DESTR]]([[S_FLOAT_TY]]* [[VAR_PRIV]]) 375 // CHECK-DAG: call {{.*}} [[S_FLOAT_TY_DESTR]]([[S_FLOAT_TY]]* 376 // CHECK: call void @__kmpc_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i{{[0-9]+}} [[GTID]]) 377 // CHECK: call void @__kmpc_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i{{[0-9]+}} [[GTID]]) 378 379 // CHECK: ret void 380 381 // void reduce_func(void *lhs[<n>], void *rhs[<n>]) { 382 // *(Type0*)lhs[0] = ReductionOperation0(*(Type0*)lhs[0], *(Type0*)rhs[0]); 383 // ... 384 // *(Type<n>-1*)lhs[<n>-1] = ReductionOperation<n>-1(*(Type<n>-1*)lhs[<n>-1], 385 // *(Type<n>-1*)rhs[<n>-1]); 386 // } 387 // CHECK: define internal void [[REDUCTION_FUNC]](i8*, i8*) 388 // t_var_lhs = (float*)lhs[0]; 389 // CHECK: [[T_VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS:%.+]], i64 0, i64 0 390 // CHECK: [[T_VAR_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR_RHS_REF]], 391 // CHECK: [[T_VAR_RHS:%.+]] = bitcast i8* [[T_VAR_RHS_VOID]] to float* 392 // t_var_rhs = (float*)rhs[0]; 393 // CHECK: [[T_VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS:%.+]], i64 0, i64 0 394 // CHECK: [[T_VAR_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR_LHS_REF]], 395 // CHECK: [[T_VAR_LHS:%.+]] = bitcast i8* [[T_VAR_LHS_VOID]] to float* 396 397 // var_lhs = (S<float>*)lhs[1]; 398 // CHECK: [[VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 1 399 // CHECK: [[VAR_RHS_VOID:%.+]] = load i8*, i8** [[VAR_RHS_REF]], 400 // CHECK: [[VAR_RHS:%.+]] = bitcast i8* [[VAR_RHS_VOID]] to [[S_FLOAT_TY]]* 401 // var_rhs = (S<float>*)rhs[1]; 402 // CHECK: [[VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 1 403 // CHECK: [[VAR_LHS_VOID:%.+]] = load i8*, i8** [[VAR_LHS_REF]], 404 // CHECK: [[VAR_LHS:%.+]] = bitcast i8* [[VAR_LHS_VOID]] to [[S_FLOAT_TY]]* 405 406 // var1_lhs = (S<float>*)lhs[2]; 407 // CHECK: [[VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 2 408 // CHECK: [[VAR1_RHS_VOID:%.+]] = load i8*, i8** [[VAR1_RHS_REF]], 409 // CHECK: [[VAR1_RHS:%.+]] = bitcast i8* [[VAR1_RHS_VOID]] to [[S_FLOAT_TY]]* 410 // var1_rhs = (S<float>*)rhs[2]; 411 // CHECK: [[VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 2 412 // CHECK: [[VAR1_LHS_VOID:%.+]] = load i8*, i8** [[VAR1_LHS_REF]], 413 // CHECK: [[VAR1_LHS:%.+]] = bitcast i8* [[VAR1_LHS_VOID]] to [[S_FLOAT_TY]]* 414 415 // t_var1_lhs = (float*)lhs[3]; 416 // CHECK: [[T_VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 3 417 // CHECK: [[T_VAR1_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_RHS_REF]], 418 // CHECK: [[T_VAR1_RHS:%.+]] = bitcast i8* [[T_VAR1_RHS_VOID]] to float* 419 // t_var1_rhs = (float*)rhs[3]; 420 // CHECK: [[T_VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 3 421 // CHECK: [[T_VAR1_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_LHS_REF]], 422 // CHECK: [[T_VAR1_LHS:%.+]] = bitcast i8* [[T_VAR1_LHS_VOID]] to float* 423 424 // t_var_lhs += t_var_rhs; 425 // CHECK: [[T_VAR_LHS_VAL:%.+]] = load float, float* [[T_VAR_LHS]], 426 // CHECK: [[T_VAR_RHS_VAL:%.+]] = load float, float* [[T_VAR_RHS]], 427 // CHECK: [[UP:%.+]] = fadd float [[T_VAR_LHS_VAL]], [[T_VAR_RHS_VAL]] 428 // CHECK: store float [[UP]], float* [[T_VAR_LHS]], 429 430 // var_lhs = var_lhs.operator &(var_rhs); 431 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_LHS]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_RHS]]) 432 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_LHS]] to i8* 433 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8* 434 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 435 436 // var1_lhs = var1_lhs.operator &&(var1_rhs); 437 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_LHS]]) 438 // CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 439 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]] 440 // CHECK: [[TRUE]] 441 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_RHS]]) 442 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 443 // CHECK: br label %[[END2]] 444 // CHECK: [[END2]] 445 // CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ] 446 // CHECK: [[CONV:%.+]] = uitofp i1 [[COND_LVALUE]] to float 447 // CHECK: call void @{{.+}}([[S_FLOAT_TY]]* [[COND_LVALUE:%.+]], float [[CONV]]) 448 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_LHS]] to i8* 449 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8* 450 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 451 452 // t_var1_lhs = min(t_var1_lhs, t_var1_rhs); 453 // CHECK: [[T_VAR1_LHS_VAL:%.+]] = load float, float* [[T_VAR1_LHS]], 454 // CHECK: [[T_VAR1_RHS_VAL:%.+]] = load float, float* [[T_VAR1_RHS]], 455 // CHECK: [[CMP:%.+]] = fcmp olt float [[T_VAR1_LHS_VAL]], [[T_VAR1_RHS_VAL]] 456 // CHECK: br i1 [[CMP]] 457 // CHECK: [[UP:%.+]] = phi float 458 // CHECK: store float [[UP]], float* [[T_VAR1_LHS]], 459 // CHECK: ret void 460 461 // CHECK: define {{.*}} i{{[0-9]+}} [[TMAIN_INT]]() 462 // CHECK: [[TEST:%.+]] = alloca [[S_INT_TY]], 463 // CHECK: call {{.*}} [[S_INT_TY_CONSTR:@.+]]([[S_INT_TY]]* [[TEST]]) 464 // CHECK: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{[0-9]+}} 6, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, i32*, [[S_INT_TY]]*, [[S_INT_TY]]*, i32*, [2 x i32]*, [2 x [[S_INT_TY]]]*)* [[TMAIN_MICROTASK:@.+]] to void 465 // CHECK: call {{.*}} [[S_INT_TY_DESTR:@.+]]([[S_INT_TY]]* 466 // CHECK: ret 467 // 468 // CHECK: define internal void [[TMAIN_MICROTASK]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, i32* dereferenceable(4) %{{.+}}, [[S_INT_TY]]* dereferenceable(4) %{{.+}}, [[S_INT_TY]]* dereferenceable(4) %{{.+}}, i32* dereferenceable(4) %{{.+}}, [2 x i32]* dereferenceable(8) %{{.+}}, [2 x [[S_INT_TY]]]* dereferenceable(8) %{{.+}}) 469 // CHECK: alloca i{{[0-9]+}}, 470 // CHECK: alloca i{{[0-9]+}}, 471 // CHECK: alloca i{{[0-9]+}}, 472 // CHECK: alloca i{{[0-9]+}}, 473 // CHECK: alloca i{{[0-9]+}}, 474 // CHECK: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}}, 475 // CHECK: [[VAR_PRIV:%.+]] = alloca [[S_INT_TY]], 476 // CHECK: [[VAR1_PRIV:%.+]] = alloca [[S_INT_TY]], 477 // CHECK: [[T_VAR1_PRIV:%.+]] = alloca i{{[0-9]+}}, 478 479 // Reduction list for runtime. 480 // CHECK: [[RED_LIST:%.+]] = alloca [4 x i8*], 481 482 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]], 483 484 // CHECK: [[T_VAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** % 485 // CHECK: [[VAR1_REF:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** % 486 // CHECK: [[T_VAR1_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** % 487 488 // For + reduction operation initial value of private variable is 0. 489 // CHECK: store i{{[0-9]+}} 0, i{{[0-9]+}}* [[T_VAR_PRIV]], 490 491 // For & reduction operation initial value of private variable is ones in all bits. 492 // CHECK: [[VAR_REF:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** % 493 // CHECK: call {{.*}} [[S_INT_TY_CONSTR:@.+]]([[S_INT_TY]]* [[VAR_PRIV]]) 494 495 // For && reduction operation initial value of private variable is 1.0. 496 // CHECK: call {{.*}} [[S_INT_TY_CONSTR:@.+]]([[S_INT_TY]]* [[VAR1_PRIV]]) 497 498 // For min reduction operation initial value of private variable is largest repesentable value. 499 // CHECK: store i{{[0-9]+}} 2147483647, i{{[0-9]+}}* [[T_VAR1_PRIV]], 500 501 // CHECK: [[GTID_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_ADDR]] 502 // CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_REF]] 503 // CHECK: call void @__kmpc_for_static_init_4( 504 // Skip checks for internal operations. 505 // CHECK: call void @__kmpc_for_static_fini( 506 507 // void *RedList[<n>] = {<ReductionVars>[0], ..., <ReductionVars>[<n>-1]}; 508 509 // CHECK: [[T_VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 0 510 // CHECK: [[BITCAST:%.+]] = bitcast i{{[0-9]+}}* [[T_VAR_PRIV]] to i8* 511 // CHECK: store i8* [[BITCAST]], i8** [[T_VAR_PRIV_REF]], 512 // CHECK: [[VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 1 513 // CHECK: [[BITCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR_PRIV]] to i8* 514 // CHECK: store i8* [[BITCAST]], i8** [[VAR_PRIV_REF]], 515 // CHECK: [[VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 2 516 // CHECK: [[BITCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_PRIV]] to i8* 517 // CHECK: store i8* [[BITCAST]], i8** [[VAR1_PRIV_REF]], 518 // CHECK: [[T_VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 3 519 // CHECK: [[BITCAST:%.+]] = bitcast i{{[0-9]+}}* [[T_VAR1_PRIV]] to i8* 520 // CHECK: store i8* [[BITCAST]], i8** [[T_VAR1_PRIV_REF]], 521 522 // res = __kmpc_reduce_nowait(<loc>, <gtid>, <n>, sizeof(RedList), RedList, reduce_func, &<lock>); 523 524 // CHECK: [[BITCAST:%.+]] = bitcast [4 x i8*]* [[RED_LIST]] to i8* 525 // CHECK: [[RES:%.+]] = call i32 @__kmpc_reduce_nowait(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], i32 4, i64 32, i8* [[BITCAST]], void (i8*, i8*)* [[REDUCTION_FUNC:@.+]], [8 x i32]* [[REDUCTION_LOCK]]) 526 527 // switch(res) 528 // CHECK: switch i32 [[RES]], label %[[RED_DONE:.+]] [ 529 // CHECK: i32 1, label %[[CASE1:.+]] 530 // CHECK: i32 2, label %[[CASE2:.+]] 531 // CHECK: ] 532 533 // case 1: 534 // t_var += t_var_reduction; 535 // CHECK: [[T_VAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_REF]], 536 // CHECK: [[T_VAR_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_PRIV]], 537 // CHECK: [[UP:%.+]] = add nsw i{{[0-9]+}} [[T_VAR_VAL]], [[T_VAR_PRIV_VAL]] 538 // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR_REF]], 539 540 // var = var.operator &(var_reduction); 541 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_REF]], [[S_INT_TY]]* dereferenceable(4) [[VAR_PRIV]]) 542 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_REF]] to i8* 543 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* 544 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 545 546 // var1 = var1.operator &&(var1_reduction); 547 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_REF]]) 548 // CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 549 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]] 550 // CHECK: [[TRUE]] 551 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_PRIV]]) 552 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 553 // CHECK: br label %[[END2]] 554 // CHECK: [[END2]] 555 // CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ] 556 // CHECK: [[CONV:%.+]] = zext i1 [[COND_LVALUE]] to i32 557 // CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]]) 558 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8* 559 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* 560 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 561 562 // t_var1 = min(t_var1, t_var1_reduction); 563 // CHECK: [[T_VAR1_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_REF]], 564 // CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_PRIV]], 565 // CHECK: [[CMP:%.+]] = icmp slt i{{[0-9]+}} [[T_VAR1_VAL]], [[T_VAR1_PRIV_VAL]] 566 // CHECK: br i1 [[CMP]] 567 // CHECK: [[UP:%.+]] = phi i32 568 // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR1_REF]], 569 570 // __kmpc_end_reduce_nowait(<loc>, <gtid>, &<lock>); 571 // CHECK: call void @__kmpc_end_reduce_nowait(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], [8 x i32]* [[REDUCTION_LOCK]]) 572 573 // break; 574 // CHECK: br label %[[RED_DONE]] 575 576 // case 2: 577 // t_var += t_var_reduction; 578 // CHECK: [[T_VAR_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_PRIV]] 579 // CHECK: atomicrmw add i32* [[T_VAR_REF]], i32 [[T_VAR_PRIV_VAL]] monotonic 580 581 // var = var.operator &(var_reduction); 582 // CHECK: call void @__kmpc_critical( 583 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_REF]], [[S_INT_TY]]* dereferenceable(4) [[VAR_PRIV]]) 584 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_REF]] to i8* 585 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* 586 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 587 // CHECK: call void @__kmpc_end_critical( 588 589 // var1 = var1.operator &&(var1_reduction); 590 // CHECK: call void @__kmpc_critical( 591 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_REF]]) 592 // CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 593 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]] 594 // CHECK: [[TRUE]] 595 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_PRIV]]) 596 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 597 // CHECK: br label %[[END2]] 598 // CHECK: [[END2]] 599 // CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ] 600 // CHECK: [[CONV:%.+]] = zext i1 [[COND_LVALUE]] to i32 601 // CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]]) 602 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8* 603 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* 604 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 605 // CHECK: call void @__kmpc_end_critical( 606 607 // t_var1 = min(t_var1, t_var1_reduction); 608 // CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_PRIV]] 609 // CHECK: atomicrmw min i32* [[T_VAR1_REF]], i32 [[T_VAR1_PRIV_VAL]] monotonic 610 611 // break; 612 // CHECK: br label %[[RED_DONE]] 613 // CHECK: [[RED_DONE]] 614 // CHECK-DAG: call {{.*}} [[S_INT_TY_DESTR]]([[S_INT_TY]]* [[VAR_PRIV]]) 615 // CHECK-DAG: call {{.*}} [[S_INT_TY_DESTR]]([[S_INT_TY]]* 616 // CHECK: call void @__kmpc_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i{{[0-9]+}} [[GTID]]) 617 // CHECK: ret void 618 619 // void reduce_func(void *lhs[<n>], void *rhs[<n>]) { 620 // *(Type0*)lhs[0] = ReductionOperation0(*(Type0*)lhs[0], *(Type0*)rhs[0]); 621 // ... 622 // *(Type<n>-1*)lhs[<n>-1] = ReductionOperation<n>-1(*(Type<n>-1*)lhs[<n>-1], 623 // *(Type<n>-1*)rhs[<n>-1]); 624 // } 625 // CHECK: define internal void [[REDUCTION_FUNC]](i8*, i8*) 626 // t_var_lhs = (i{{[0-9]+}}*)lhs[0]; 627 // CHECK: [[T_VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS:%.+]], i64 0, i64 0 628 // CHECK: [[T_VAR_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR_RHS_REF]], 629 // CHECK: [[T_VAR_RHS:%.+]] = bitcast i8* [[T_VAR_RHS_VOID]] to i{{[0-9]+}}* 630 // t_var_rhs = (i{{[0-9]+}}*)rhs[0]; 631 // CHECK: [[T_VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS:%.+]], i64 0, i64 0 632 // CHECK: [[T_VAR_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR_LHS_REF]], 633 // CHECK: [[T_VAR_LHS:%.+]] = bitcast i8* [[T_VAR_LHS_VOID]] to i{{[0-9]+}}* 634 635 // var_lhs = (S<i{{[0-9]+}}>*)lhs[1]; 636 // CHECK: [[VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 1 637 // CHECK: [[VAR_RHS_VOID:%.+]] = load i8*, i8** [[VAR_RHS_REF]], 638 // CHECK: [[VAR_RHS:%.+]] = bitcast i8* [[VAR_RHS_VOID]] to [[S_INT_TY]]* 639 // var_rhs = (S<i{{[0-9]+}}>*)rhs[1]; 640 // CHECK: [[VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 1 641 // CHECK: [[VAR_LHS_VOID:%.+]] = load i8*, i8** [[VAR_LHS_REF]], 642 // CHECK: [[VAR_LHS:%.+]] = bitcast i8* [[VAR_LHS_VOID]] to [[S_INT_TY]]* 643 644 // var1_lhs = (S<i{{[0-9]+}}>*)lhs[2]; 645 // CHECK: [[VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 2 646 // CHECK: [[VAR1_RHS_VOID:%.+]] = load i8*, i8** [[VAR1_RHS_REF]], 647 // CHECK: [[VAR1_RHS:%.+]] = bitcast i8* [[VAR1_RHS_VOID]] to [[S_INT_TY]]* 648 // var1_rhs = (S<i{{[0-9]+}}>*)rhs[2]; 649 // CHECK: [[VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 2 650 // CHECK: [[VAR1_LHS_VOID:%.+]] = load i8*, i8** [[VAR1_LHS_REF]], 651 // CHECK: [[VAR1_LHS:%.+]] = bitcast i8* [[VAR1_LHS_VOID]] to [[S_INT_TY]]* 652 653 // t_var1_lhs = (i{{[0-9]+}}*)lhs[3]; 654 // CHECK: [[T_VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 3 655 // CHECK: [[T_VAR1_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_RHS_REF]], 656 // CHECK: [[T_VAR1_RHS:%.+]] = bitcast i8* [[T_VAR1_RHS_VOID]] to i{{[0-9]+}}* 657 // t_var1_rhs = (i{{[0-9]+}}*)rhs[3]; 658 // CHECK: [[T_VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 3 659 // CHECK: [[T_VAR1_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_LHS_REF]], 660 // CHECK: [[T_VAR1_LHS:%.+]] = bitcast i8* [[T_VAR1_LHS_VOID]] to i{{[0-9]+}}* 661 662 // t_var_lhs += t_var_rhs; 663 // CHECK: [[T_VAR_LHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_LHS]], 664 // CHECK: [[T_VAR_RHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_RHS]], 665 // CHECK: [[UP:%.+]] = add nsw i{{[0-9]+}} [[T_VAR_LHS_VAL]], [[T_VAR_RHS_VAL]] 666 // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR_LHS]], 667 668 // var_lhs = var_lhs.operator &(var_rhs); 669 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_LHS]], [[S_INT_TY]]* dereferenceable(4) [[VAR_RHS]]) 670 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_LHS]] to i8* 671 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* 672 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 673 674 // var1_lhs = var1_lhs.operator &&(var1_rhs); 675 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_LHS]]) 676 // CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 677 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]] 678 // CHECK: [[TRUE]] 679 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_RHS]]) 680 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 681 // CHECK: br label %[[END2]] 682 // CHECK: [[END2]] 683 // CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ] 684 // CHECK: [[CONV:%.+]] = zext i1 [[COND_LVALUE]] to i32 685 // CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]]) 686 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_LHS]] to i8* 687 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* 688 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 689 690 // t_var1_lhs = min(t_var1_lhs, t_var1_rhs); 691 // CHECK: [[T_VAR1_LHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_LHS]], 692 // CHECK: [[T_VAR1_RHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_RHS]], 693 // CHECK: [[CMP:%.+]] = icmp slt i{{[0-9]+}} [[T_VAR1_LHS_VAL]], [[T_VAR1_RHS_VAL]] 694 // CHECK: br i1 [[CMP]] 695 // CHECK: [[UP:%.+]] = phi i32 696 // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR1_LHS]], 697 // CHECK: ret void 698 699 #endif 700 701