1 // RUN: %clang_cc1 -verify -fopenmp=libiomp5 -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck %s 2 // RUN: %clang_cc1 -fopenmp=libiomp5 -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s 3 // RUN: %clang_cc1 -fopenmp=libiomp5 -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=libiomp5 -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=libiomp5 -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s 6 // expected-no-diagnostics 7 #ifndef HEADER 8 #define HEADER 9 10 volatile double g; 11 12 template <class T> 13 struct S { 14 T f; 15 S(T a) : f(a + g) {} 16 S() : f(g) {} 17 operator T() { return T(); } 18 S &operator&(const S &) { return *this; } 19 ~S() {} 20 }; 21 22 // CHECK-DAG: [[S_FLOAT_TY:%.+]] = type { float } 23 // CHECK-DAG: [[S_INT_TY:%.+]] = type { i{{[0-9]+}} } 24 // CHECK-DAG: [[CAP_MAIN_TY:%.+]] = type { float*, [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]*, float*, [2 x i{{[0-9]+}}]*, [2 x [[S_FLOAT_TY]]]* } 25 // CHECK-DAG: [[CAP_TMAIN_TY:%.+]] = type { i{{[0-9]+}}*, [[S_INT_TY]]*, [[S_INT_TY]]*, i{{[0-9]+}}*, [2 x i{{[0-9]+}}]*, [2 x [[S_INT_TY]]]* } 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(3), var1; 39 #pragma omp parallel 40 #pragma omp for reduction(+:t_var) reduction(&:var) reduction(&& : var1) reduction(min: t_var1) nowait 41 for (int i = 0; i < 2; ++i) { 42 vec[i] = t_var; 43 s_arr[i] = var; 44 } 45 #pragma omp parallel 46 #pragma omp for reduction(&& : t_var) 47 for (int i = 0; i < 2; ++i) { 48 vec[i] = t_var; 49 s_arr[i] = var; 50 } 51 return T(); 52 } 53 54 int main() { 55 #ifdef LAMBDA 56 // LAMBDA: [[G:@.+]] = global double 57 // LAMBDA-LABEL: @main 58 // LAMBDA: call void [[OUTER_LAMBDA:@.+]]( 59 [&]() { 60 // LAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]]( 61 // LAMBDA: call void {{.+}} @__kmpc_fork_call({{.+}}, i32 1, {{.+}}* [[OMP_REGION:@.+]] to {{.+}}, i8* %{{.+}}) 62 #pragma omp parallel 63 #pragma omp for reduction(+:g) 64 for (int i = 0; i < 2; ++i) { 65 // LAMBDA: define{{.*}} internal{{.*}} void [[OMP_REGION]](i32* %{{.+}}, i32* %{{.+}}, %{{.+}}* %{{.+}}) 66 // LAMBDA: [[G_PRIVATE_ADDR:%.+]] = alloca double, 67 68 // Reduction list for runtime. 69 // LAMBDA: [[RED_LIST:%.+]] = alloca [1 x i8*], 70 71 // LAMBDA: store double 0.0{{.+}}, double* [[G_PRIVATE_ADDR]] 72 // LAMBDA: call void @__kmpc_for_static_init_4( 73 g = 1; 74 // LAMBDA: store volatile double 1.0{{.+}}, double* [[G_PRIVATE_ADDR]], 75 // LAMBDA: [[G_PRIVATE_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG:%.+]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 76 // LAMBDA: store double* [[G_PRIVATE_ADDR]], double** [[G_PRIVATE_ADDR_REF]] 77 // LAMBDA: call void [[INNER_LAMBDA:@.+]](%{{.+}}* [[ARG]]) 78 // LAMBDA: call void @__kmpc_for_static_fini( 79 80 // LAMBDA: [[G_PRIV_REF:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[RED_LIST]], i32 0, i32 0 81 // LAMBDA: [[BITCAST:%.+]] = bitcast double* [[G_PRIVATE_ADDR]] to i8* 82 // LAMBDA: store i8* [[BITCAST]], i8** [[G_PRIV_REF]], 83 // LAMBDA: call i32 @__kmpc_reduce( 84 // LAMBDA: switch i32 %{{.+}}, label %[[REDUCTION_DONE:.+]] [ 85 // LAMBDA: i32 1, label %[[CASE1:.+]] 86 // LAMBDA: i32 2, label %[[CASE2:.+]] 87 // LAMBDA: [[CASE1]] 88 // LAMBDA: [[G_VAL:%.+]] = load double, double* [[G]] 89 // LAMBDA: [[G_PRIV_VAL:%.+]] = load double, double* [[G_PRIVATE_ADDR]] 90 // LAMBDA: [[ADD:%.+]] = fadd double [[G_VAL]], [[G_PRIV_VAL]] 91 // LAMBDA: store double [[ADD]], double* [[G]] 92 // LAMBDA: call void @__kmpc_end_reduce( 93 // LAMBDA: br label %[[REDUCTION_DONE]] 94 // LAMBDA: [[CASE2]] 95 // LAMBDA: [[G_PRIV_VAL:%.+]] = load double, double* [[G_PRIVATE_ADDR]] 96 // LAMBDA: fadd double 97 // LAMBDA: cmpxchg i64* 98 // LAMBDA: br label %[[REDUCTION_DONE]] 99 // LAMBDA: [[REDUCTION_DONE]] 100 // LAMBDA: ret void 101 [&]() { 102 // LAMBDA: define {{.+}} void [[INNER_LAMBDA]](%{{.+}}* [[ARG_PTR:%.+]]) 103 // LAMBDA: store %{{.+}}* [[ARG_PTR]], %{{.+}}** [[ARG_PTR_REF:%.+]], 104 g = 2; 105 // LAMBDA: [[ARG_PTR:%.+]] = load %{{.+}}*, %{{.+}}** [[ARG_PTR_REF]] 106 // LAMBDA: [[G_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 107 // LAMBDA: [[G_REF:%.+]] = load double*, double** [[G_PTR_REF]] 108 // LAMBDA: store volatile double 2.0{{.+}}, double* [[G_REF]] 109 }(); 110 } 111 }(); 112 return 0; 113 #elif defined(BLOCKS) 114 // BLOCKS: [[G:@.+]] = global double 115 // BLOCKS-LABEL: @main 116 // BLOCKS: call void {{%.+}}(i8 117 ^{ 118 // BLOCKS: define{{.*}} internal{{.*}} void {{.+}}(i8* 119 // BLOCKS: call void {{.+}} @__kmpc_fork_call({{.+}}, i32 1, {{.+}}* [[OMP_REGION:@.+]] to {{.+}}, i8* %{{.+}}) 120 #pragma omp parallel 121 #pragma omp for reduction(-:g) 122 for (int i = 0; i < 2; ++i) { 123 // BLOCKS: define{{.*}} internal{{.*}} void [[OMP_REGION]](i32* %{{.+}}, i32* %{{.+}}, %{{.+}}* %{{.+}}) 124 // BLOCKS: [[G_PRIVATE_ADDR:%.+]] = alloca double, 125 126 // Reduction list for runtime. 127 // BLOCKS: [[RED_LIST:%.+]] = alloca [1 x i8*], 128 129 // BLOCKS: store double 0.0{{.+}}, double* [[G_PRIVATE_ADDR]] 130 g = 1; 131 // BLOCKS: call void @__kmpc_for_static_init_4( 132 // BLOCKS: store volatile double 1.0{{.+}}, double* [[G_PRIVATE_ADDR]], 133 // BLOCKS-NOT: [[G]]{{[[^:word:]]}} 134 // BLOCKS: double* [[G_PRIVATE_ADDR]] 135 // BLOCKS-NOT: [[G]]{{[[^:word:]]}} 136 // BLOCKS: call void {{%.+}}(i8 137 // BLOCKS: call void @__kmpc_for_static_fini( 138 139 // BLOCKS: [[G_PRIV_REF:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[RED_LIST]], i32 0, i32 0 140 // BLOCKS: [[BITCAST:%.+]] = bitcast double* [[G_PRIVATE_ADDR]] to i8* 141 // BLOCKS: store i8* [[BITCAST]], i8** [[G_PRIV_REF]], 142 // BLOCKS: call i32 @__kmpc_reduce( 143 // BLOCKS: switch i32 %{{.+}}, label %[[REDUCTION_DONE:.+]] [ 144 // BLOCKS: i32 1, label %[[CASE1:.+]] 145 // BLOCKS: i32 2, label %[[CASE2:.+]] 146 // BLOCKS: [[CASE1]] 147 // BLOCKS: [[G_VAL:%.+]] = load double, double* [[G]] 148 // BLOCKS: [[G_PRIV_VAL:%.+]] = load double, double* [[G_PRIVATE_ADDR]] 149 // BLOCKS: [[ADD:%.+]] = fadd double [[G_VAL]], [[G_PRIV_VAL]] 150 // BLOCKS: store double [[ADD]], double* [[G]] 151 // BLOCKS: call void @__kmpc_end_reduce( 152 // BLOCKS: br label %[[REDUCTION_DONE]] 153 // BLOCKS: [[CASE2]] 154 // BLOCKS: [[G_PRIV_VAL:%.+]] = load double, double* [[G_PRIVATE_ADDR]] 155 // BLOCKS: fadd double 156 // BLOCKS: cmpxchg i64* 157 // BLOCKS: br label %[[REDUCTION_DONE]] 158 // BLOCKS: [[REDUCTION_DONE]] 159 // BLOCKS: ret void 160 ^{ 161 // BLOCKS: define {{.+}} void {{@.+}}(i8* 162 g = 2; 163 // BLOCKS-NOT: [[G]]{{[[^:word:]]}} 164 // BLOCKS: store volatile double 2.0{{.+}}, double* 165 // BLOCKS-NOT: [[G]]{{[[^:word:]]}} 166 // BLOCKS: ret 167 }(); 168 } 169 }(); 170 return 0; 171 #else 172 S<float> test; 173 float t_var = 0, t_var1; 174 int vec[] = {1, 2}; 175 S<float> s_arr[] = {1, 2}; 176 S<float> var(3), var1; 177 #pragma omp parallel 178 #pragma omp for reduction(+:t_var) reduction(&:var) reduction(&& : var1) reduction(min: t_var1) 179 for (int i = 0; i < 2; ++i) { 180 vec[i] = t_var; 181 s_arr[i] = var; 182 } 183 return tmain<int>(); 184 #endif 185 } 186 187 // CHECK: define {{.*}}i{{[0-9]+}} @main() 188 // CHECK: [[TEST:%.+]] = alloca [[S_FLOAT_TY]], 189 // CHECK: call {{.*}} [[S_FLOAT_TY_CONSTR:@.+]]([[S_FLOAT_TY]]* [[TEST]]) 190 // CHECK: %{{.+}} = bitcast [[CAP_MAIN_TY]]* 191 // CHECK: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{[0-9]+}} 1, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, [[CAP_MAIN_TY]]*)* [[MAIN_MICROTASK:@.+]] to void 192 // CHECK: = call {{.*}}i{{.+}} [[TMAIN_INT:@.+]]() 193 // CHECK: call {{.*}} [[S_FLOAT_TY_DESTR:@.+]]([[S_FLOAT_TY]]* 194 // CHECK: ret 195 // 196 // CHECK: define internal void [[MAIN_MICROTASK]](i{{[0-9]+}}* [[GTID_ADDR:%.+]], i{{[0-9]+}}* %{{.+}}, [[CAP_MAIN_TY]]* %{{.+}}) 197 // CHECK: [[T_VAR_PRIV:%.+]] = alloca float, 198 // CHECK: [[VAR_PRIV:%.+]] = alloca [[S_FLOAT_TY]], 199 // CHECK: [[VAR1_PRIV:%.+]] = alloca [[S_FLOAT_TY]], 200 // CHECK: [[T_VAR1_PRIV:%.+]] = alloca float, 201 202 // Reduction list for runtime. 203 // CHECK: [[RED_LIST:%.+]] = alloca [4 x i8*], 204 205 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]], 206 207 // CHECK: [[T_VAR_PTR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}} 208 // CHECK: [[T_VAR_REF:%.+]] = load float*, float** [[T_VAR_PTR_REF]], 209 // For + reduction operation initial value of private variable is 0. 210 // CHECK: store float 0.0{{.+}}, float* [[T_VAR_PRIV]], 211 212 // CHECK: [[VAR_PTR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}} 213 // CHECK: [[VAR_REF:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[VAR_PTR_REF:%.+]], 214 // For & reduction operation initial value of private variable is ones in all bits. 215 // CHECK: call {{.*}} [[S_FLOAT_TY_CONSTR:@.+]]([[S_FLOAT_TY]]* [[VAR_PRIV]]) 216 217 // CHECK: [[VAR1_PTR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}} 218 // CHECK: [[VAR1_REF:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[VAR_PTR_REF:%.+]], 219 // For && reduction operation initial value of private variable is 1.0. 220 // CHECK: call {{.*}} [[S_FLOAT_TY_CONSTR:@.+]]([[S_FLOAT_TY]]* [[VAR1_PRIV]]) 221 222 // CHECK: [[T_VAR1_PTR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}} 223 // CHECK: [[T_VAR1_REF:%.+]] = load float*, float** [[T_VAR1_PTR_REF]], 224 // For min reduction operation initial value of private variable is largest repesentable value. 225 // CHECK: store float 0x47EFFFFFE0000000, float* [[T_VAR1_PRIV]], 226 227 228 // CHECK: [[GTID_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_ADDR]] 229 // CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_REF]] 230 // CHECK: call void @__kmpc_for_static_init_4( 231 // Skip checks for internal operations. 232 // CHECK: call void @__kmpc_for_static_fini( 233 234 // void *RedList[<n>] = {<ReductionVars>[0], ..., <ReductionVars>[<n>-1]}; 235 236 // CHECK: [[T_VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 0 237 // CHECK: [[BITCAST:%.+]] = bitcast float* [[T_VAR_PRIV]] to i8* 238 // CHECK: store i8* [[BITCAST]], i8** [[T_VAR_PRIV_REF]], 239 // CHECK: [[VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 1 240 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_PRIV]] to i8* 241 // CHECK: store i8* [[BITCAST]], i8** [[VAR_PRIV_REF]], 242 // CHECK: [[VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 2 243 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_PRIV]] to i8* 244 // CHECK: store i8* [[BITCAST]], i8** [[VAR1_PRIV_REF]], 245 // CHECK: [[T_VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 3 246 // CHECK: [[BITCAST:%.+]] = bitcast float* [[T_VAR1_PRIV]] to i8* 247 // CHECK: store i8* [[BITCAST]], i8** [[T_VAR1_PRIV_REF]], 248 249 // res = __kmpc_reduce(<loc>, <gtid>, <n>, sizeof(RedList), RedList, reduce_func, &<lock>); 250 251 // CHECK: [[BITCAST:%.+]] = bitcast [4 x i8*]* [[RED_LIST]] to i8* 252 // 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]]) 253 254 // switch(res) 255 // CHECK: switch i32 [[RES]], label %[[RED_DONE:.+]] [ 256 // CHECK: i32 1, label %[[CASE1:.+]] 257 // CHECK: i32 2, label %[[CASE2:.+]] 258 // CHECK: ] 259 260 // case 1: 261 // t_var += t_var_reduction; 262 // CHECK: [[T_VAR_VAL:%.+]] = load float, float* [[T_VAR_REF]], 263 // CHECK: [[T_VAR_PRIV_VAL:%.+]] = load float, float* [[T_VAR_PRIV]], 264 // CHECK: [[UP:%.+]] = fadd float [[T_VAR_VAL]], [[T_VAR_PRIV_VAL]] 265 // CHECK: store float [[UP]], float* [[T_VAR_REF]], 266 267 // var = var.operator &(var_reduction); 268 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_REF]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_PRIV]]) 269 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_REF]] to i8* 270 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8* 271 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 272 273 // var1 = var1.operator &&(var1_reduction); 274 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_REF]]) 275 // CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 276 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]] 277 // CHECK: [[TRUE]] 278 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_PRIV]]) 279 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 280 // CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]] 281 // CHECK: [[TRUE2]] 282 // CHECK: br label %[[END2:.+]] 283 // CHECK: [[FALSE2]] 284 // CHECK: br label %[[END2]] 285 // CHECK: [[END2]] 286 // CHECK: [[COND_LVALUE:%.+]] = phi [[S_FLOAT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ] 287 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_REF]] to i8* 288 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8* 289 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 290 291 // t_var1 = min(t_var1, t_var1_reduction); 292 // CHECK: [[T_VAR1_VAL:%.+]] = load float, float* [[T_VAR1_REF]], 293 // CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load float, float* [[T_VAR1_PRIV]], 294 // CHECK: [[CMP:%.+]] = fcmp olt float [[T_VAR1_VAL]], [[T_VAR1_PRIV_VAL]] 295 // CHECK: [[UP:%.+]] = uitofp i1 [[CMP]] to float 296 // CHECK: store float [[UP]], float* [[T_VAR1_REF]], 297 298 // __kmpc_end_reduce(<loc>, <gtid>, &<lock>); 299 // CHECK: call void @__kmpc_end_reduce(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], [8 x i32]* [[REDUCTION_LOCK]]) 300 301 // break; 302 // CHECK: br label %[[RED_DONE]] 303 304 // case 2: 305 // t_var += t_var_reduction; 306 // CHECK: load float, float* [[T_VAR_PRIV]] 307 // CHECK: [[T_VAR_REF_INT:%.+]] = bitcast float* [[T_VAR_REF]] to i32* 308 // CHECK: load atomic i32, i32* [[T_VAR_REF_INT]] monotonic, 309 // CHECK: [[OLD1:%.+]] = bitcast i32 %{{.+}} to float 310 // CHECK: br label %[[CONT:.+]] 311 // CHECK: [[CONT]] 312 // CHECK: [[ORIG_OLD:%.+]] = phi float [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %[[CONT]] ] 313 // CHECK: [[UP:%.+]] = fadd float 314 // CHECK: [[ORIG_OLD_INT:%.+]] = bitcast float [[ORIG_OLD]] to i32 315 // CHECK: [[UP_INT:%.+]] = bitcast float [[UP]] to i32 316 // CHECK: [[T_VAR_REF_INT:%.+]] = bitcast float* [[T_VAR_REF]] to i32* 317 // CHECK: [[RES:%.+]] = cmpxchg i32* [[T_VAR_REF_INT]], i32 [[ORIG_OLD_INT]], i32 [[UP_INT]] monotonic monotonic 318 // CHECK: [[OLD_VAL_INT:%.+]] = extractvalue { i32, i1 } [[RES]], 0 319 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1 320 // CHECK: [[OLD2]] = bitcast i32 [[OLD_VAL_INT]] to float 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 %[[FALSE:.+]] 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 i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]] 341 // CHECK: [[TRUE2]] 342 // CHECK: br label %[[END2:.+]] 343 // CHECK: [[FALSE2]] 344 // CHECK: br label %[[END2]] 345 // CHECK: [[END2]] 346 // CHECK: [[COND_LVALUE:%.+]] = phi [[S_FLOAT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ] 347 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_REF]] to i8* 348 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8* 349 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 350 // CHECK: call void @__kmpc_end_critical( 351 352 // t_var1 = min(t_var1, t_var1_reduction); 353 // CHECK: load float, float* [[T_VAR1_PRIV]] 354 // CHECK: [[T_VAR1_REF_INT:%.+]] = bitcast float* [[T_VAR1_REF]] to i32* 355 // CHECK: load atomic i32, i32* [[T_VAR1_REF_INT]] monotonic, 356 // CHECK: [[OLD1:%.+]] = bitcast i32 %{{.+}} to float 357 // CHECK: br label %[[CONT:.+]] 358 // CHECK: [[CONT]] 359 // CHECK: [[ORIG_OLD:%.+]] = phi float [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %[[CONT]] ] 360 // CHECK: [[CMP:%.+]] = fcmp olt float 361 // CHECK: [[UP:%.+]] = uitofp i1 [[CMP]] to float 362 // CHECK: [[ORIG_OLD_INT:%.+]] = bitcast float [[ORIG_OLD]] to i32 363 // CHECK: [[UP_INT:%.+]] = bitcast float [[UP]] to i32 364 // CHECK: [[T_VAR1_REF_INT:%.+]] = bitcast float* [[T_VAR1_REF]] to i32* 365 // CHECK: [[RES:%.+]] = cmpxchg i32* [[T_VAR1_REF_INT]], i32 [[ORIG_OLD_INT]], i32 [[UP_INT]] monotonic monotonic 366 // CHECK: [[OLD_VAL_INT:%.+]] = extractvalue { i32, i1 } [[RES]], 0 367 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1 368 // CHECK: [[OLD2]] = bitcast i32 [[OLD_VAL_INT]] to float 369 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[ATOMIC_DONE:.+]], label %[[CONT]] 370 // CHECK: [[ATOMIC_DONE]] 371 372 // break; 373 // CHECK: br label %[[RED_DONE]] 374 // CHECK: [[RED_DONE]] 375 // CHECK-DAG: call {{.*}} [[S_FLOAT_TY_DESTR]]([[S_FLOAT_TY]]* [[VAR_PRIV]]) 376 // CHECK-DAG: call {{.*}} [[S_FLOAT_TY_DESTR]]([[S_FLOAT_TY]]* 377 // CHECK: call i32 @__kmpc_cancel_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i{{[0-9]+}} [[GTID]]) 378 // CHECK: call i32 @__kmpc_cancel_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i{{[0-9]+}} [[GTID]]) 379 380 // CHECK: ret void 381 382 // void reduce_func(void *lhs[<n>], void *rhs[<n>]) { 383 // *(Type0*)lhs[0] = ReductionOperation0(*(Type0*)lhs[0], *(Type0*)rhs[0]); 384 // ... 385 // *(Type<n>-1*)lhs[<n>-1] = ReductionOperation<n>-1(*(Type<n>-1*)lhs[<n>-1], 386 // *(Type<n>-1*)rhs[<n>-1]); 387 // } 388 // CHECK: define internal void [[REDUCTION_FUNC]](i8*, i8*) 389 // t_var_lhs = (float*)lhs[0]; 390 // CHECK: [[T_VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS:%.+]], i32 0, i32 0 391 // CHECK: [[T_VAR_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR_RHS_REF]], 392 // CHECK: [[T_VAR_RHS:%.+]] = bitcast i8* [[T_VAR_RHS_VOID]] to float* 393 // t_var_rhs = (float*)rhs[0]; 394 // CHECK: [[T_VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS:%.+]], i32 0, i32 0 395 // CHECK: [[T_VAR_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR_LHS_REF]], 396 // CHECK: [[T_VAR_LHS:%.+]] = bitcast i8* [[T_VAR_LHS_VOID]] to float* 397 398 // var_lhs = (S<float>*)lhs[1]; 399 // CHECK: [[VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i32 0, i32 1 400 // CHECK: [[VAR_RHS_VOID:%.+]] = load i8*, i8** [[VAR_RHS_REF]], 401 // CHECK: [[VAR_RHS:%.+]] = bitcast i8* [[VAR_RHS_VOID]] to [[S_FLOAT_TY]]* 402 // var_rhs = (S<float>*)rhs[1]; 403 // CHECK: [[VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i32 0, i32 1 404 // CHECK: [[VAR_LHS_VOID:%.+]] = load i8*, i8** [[VAR_LHS_REF]], 405 // CHECK: [[VAR_LHS:%.+]] = bitcast i8* [[VAR_LHS_VOID]] to [[S_FLOAT_TY]]* 406 407 // var1_lhs = (S<float>*)lhs[2]; 408 // CHECK: [[VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i32 0, i32 2 409 // CHECK: [[VAR1_RHS_VOID:%.+]] = load i8*, i8** [[VAR1_RHS_REF]], 410 // CHECK: [[VAR1_RHS:%.+]] = bitcast i8* [[VAR1_RHS_VOID]] to [[S_FLOAT_TY]]* 411 // var1_rhs = (S<float>*)rhs[2]; 412 // CHECK: [[VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i32 0, i32 2 413 // CHECK: [[VAR1_LHS_VOID:%.+]] = load i8*, i8** [[VAR1_LHS_REF]], 414 // CHECK: [[VAR1_LHS:%.+]] = bitcast i8* [[VAR1_LHS_VOID]] to [[S_FLOAT_TY]]* 415 416 // t_var1_lhs = (float*)lhs[3]; 417 // CHECK: [[T_VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i32 0, i32 3 418 // CHECK: [[T_VAR1_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_RHS_REF]], 419 // CHECK: [[T_VAR1_RHS:%.+]] = bitcast i8* [[T_VAR1_RHS_VOID]] to float* 420 // t_var1_rhs = (float*)rhs[3]; 421 // CHECK: [[T_VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i32 0, i32 3 422 // CHECK: [[T_VAR1_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_LHS_REF]], 423 // CHECK: [[T_VAR1_LHS:%.+]] = bitcast i8* [[T_VAR1_LHS_VOID]] to float* 424 425 // t_var_lhs += t_var_rhs; 426 // CHECK: [[T_VAR_LHS_VAL:%.+]] = load float, float* [[T_VAR_LHS]], 427 // CHECK: [[T_VAR_RHS_VAL:%.+]] = load float, float* [[T_VAR_RHS]], 428 // CHECK: [[UP:%.+]] = fadd float [[T_VAR_LHS_VAL]], [[T_VAR_RHS_VAL]] 429 // CHECK: store float [[UP]], float* [[T_VAR_LHS]], 430 431 // var_lhs = var_lhs.operator &(var_rhs); 432 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_LHS]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_RHS]]) 433 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_LHS]] to i8* 434 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8* 435 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 436 437 // var1_lhs = var1_lhs.operator &&(var1_rhs); 438 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_LHS]]) 439 // CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 440 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]] 441 // CHECK: [[TRUE]] 442 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_RHS]]) 443 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 444 // CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]] 445 // CHECK: [[TRUE2]] 446 // CHECK: br label %[[END2:.+]] 447 // CHECK: [[FALSE2]] 448 // CHECK: br label %[[END2]] 449 // CHECK: [[END2]] 450 // CHECK: [[COND_LVALUE:%.+]] = phi [[S_FLOAT_TY]]* [ [[VAR1_LHS]], %[[TRUE2]] ], [ [[VAR1_RHS]], %[[FALSE2]] ] 451 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_LHS]] to i8* 452 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8* 453 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 454 455 // t_var1_lhs = min(t_var1_lhs, t_var1_rhs); 456 // CHECK: [[T_VAR1_LHS_VAL:%.+]] = load float, float* [[T_VAR1_LHS]], 457 // CHECK: [[T_VAR1_RHS_VAL:%.+]] = load float, float* [[T_VAR1_RHS]], 458 // CHECK: [[CMP:%.+]] = fcmp olt float [[T_VAR1_LHS_VAL]], [[T_VAR1_RHS_VAL]] 459 // CHECK: [[UP:%.+]] = uitofp i1 [[CMP]] to float 460 // CHECK: store float [[UP]], float* [[T_VAR1_LHS]], 461 // CHECK: ret void 462 463 // CHECK: define {{.*}} i{{[0-9]+}} [[TMAIN_INT]]() 464 // CHECK: [[TEST:%.+]] = alloca [[S_INT_TY]], 465 // CHECK: call {{.*}} [[S_INT_TY_CONSTR:@.+]]([[S_INT_TY]]* [[TEST]]) 466 // CHECK: %{{.+}} = bitcast [[CAP_TMAIN_TY]]* 467 // CHECK: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{[0-9]+}} 1, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, [[CAP_TMAIN_TY]]*)* [[TMAIN_MICROTASK:@.+]] to void 468 // CHECK: call {{.*}} [[S_INT_TY_DESTR:@.+]]([[S_INT_TY]]* 469 // CHECK: ret 470 // 471 // CHECK: define internal void [[TMAIN_MICROTASK]](i{{[0-9]+}}* [[GTID_ADDR:%.+]], i{{[0-9]+}}* %{{.+}}, [[CAP_TMAIN_TY]]* %{{.+}}) 472 // CHECK: alloca i{{[0-9]+}}, 473 // CHECK: alloca i{{[0-9]+}}, 474 // CHECK: alloca i{{[0-9]+}}, 475 // CHECK: alloca i{{[0-9]+}}, 476 // CHECK: alloca i{{[0-9]+}}, 477 // CHECK: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}}, 478 // CHECK: [[VAR_PRIV:%.+]] = alloca [[S_INT_TY]], 479 // CHECK: [[VAR1_PRIV:%.+]] = alloca [[S_INT_TY]], 480 // CHECK: [[T_VAR1_PRIV:%.+]] = alloca i{{[0-9]+}}, 481 482 // Reduction list for runtime. 483 // CHECK: [[RED_LIST:%.+]] = alloca [4 x i8*], 484 485 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]], 486 487 // CHECK: [[T_VAR_PTR_REF:%.+]] = getelementptr inbounds [[CAP_TMAIN_TY]], [[CAP_TMAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}} 488 // CHECK: [[T_VAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[T_VAR_PTR_REF]], 489 // For + reduction operation initial value of private variable is 0. 490 // CHECK: store i{{[0-9]+}} 0, i{{[0-9]+}}* [[T_VAR_PRIV]], 491 492 // CHECK: [[VAR_PTR_REF:%.+]] = getelementptr inbounds [[CAP_TMAIN_TY]], [[CAP_TMAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}} 493 // CHECK: [[VAR_REF:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[VAR_PTR_REF:%.+]], 494 // For & reduction operation initial value of private variable is ones in all bits. 495 // CHECK: call {{.*}} [[S_INT_TY_CONSTR:@.+]]([[S_INT_TY]]* [[VAR_PRIV]]) 496 497 // CHECK: [[VAR1_PTR_REF:%.+]] = getelementptr inbounds [[CAP_TMAIN_TY]], [[CAP_TMAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}} 498 // CHECK: [[VAR1_REF:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[VAR_PTR_REF:%.+]], 499 // For && reduction operation initial value of private variable is 1.0. 500 // CHECK: call {{.*}} [[S_INT_TY_CONSTR:@.+]]([[S_INT_TY]]* [[VAR1_PRIV]]) 501 502 // CHECK: [[T_VAR1_PTR_REF:%.+]] = getelementptr inbounds [[CAP_TMAIN_TY]], [[CAP_TMAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}} 503 // CHECK: [[T_VAR1_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[T_VAR1_PTR_REF]], 504 // For min reduction operation initial value of private variable is largest repesentable value. 505 // CHECK: store i{{[0-9]+}} 2147483647, i{{[0-9]+}}* [[T_VAR1_PRIV]], 506 507 // CHECK: [[GTID_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_ADDR]] 508 // CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_REF]] 509 // CHECK: call void @__kmpc_for_static_init_4( 510 // Skip checks for internal operations. 511 // CHECK: call void @__kmpc_for_static_fini( 512 513 // void *RedList[<n>] = {<ReductionVars>[0], ..., <ReductionVars>[<n>-1]}; 514 515 // CHECK: [[T_VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 0 516 // CHECK: [[BITCAST:%.+]] = bitcast i{{[0-9]+}}* [[T_VAR_PRIV]] to i8* 517 // CHECK: store i8* [[BITCAST]], i8** [[T_VAR_PRIV_REF]], 518 // CHECK: [[VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 1 519 // CHECK: [[BITCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR_PRIV]] to i8* 520 // CHECK: store i8* [[BITCAST]], i8** [[VAR_PRIV_REF]], 521 // CHECK: [[VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 2 522 // CHECK: [[BITCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_PRIV]] to i8* 523 // CHECK: store i8* [[BITCAST]], i8** [[VAR1_PRIV_REF]], 524 // CHECK: [[T_VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 3 525 // CHECK: [[BITCAST:%.+]] = bitcast i{{[0-9]+}}* [[T_VAR1_PRIV]] to i8* 526 // CHECK: store i8* [[BITCAST]], i8** [[T_VAR1_PRIV_REF]], 527 528 // res = __kmpc_reduce_nowait(<loc>, <gtid>, <n>, sizeof(RedList), RedList, reduce_func, &<lock>); 529 530 // CHECK: [[BITCAST:%.+]] = bitcast [4 x i8*]* [[RED_LIST]] to i8* 531 // 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]]) 532 533 // switch(res) 534 // CHECK: switch i32 [[RES]], label %[[RED_DONE:.+]] [ 535 // CHECK: i32 1, label %[[CASE1:.+]] 536 // CHECK: i32 2, label %[[CASE2:.+]] 537 // CHECK: ] 538 539 // case 1: 540 // t_var += t_var_reduction; 541 // CHECK: [[T_VAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_REF]], 542 // CHECK: [[T_VAR_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_PRIV]], 543 // CHECK: [[UP:%.+]] = add nsw i{{[0-9]+}} [[T_VAR_VAL]], [[T_VAR_PRIV_VAL]] 544 // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR_REF]], 545 546 // var = var.operator &(var_reduction); 547 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_REF]], [[S_INT_TY]]* dereferenceable(4) [[VAR_PRIV]]) 548 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_REF]] to i8* 549 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* 550 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 551 552 // var1 = var1.operator &&(var1_reduction); 553 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_REF]]) 554 // CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 555 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]] 556 // CHECK: [[TRUE]] 557 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_PRIV]]) 558 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 559 // CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]] 560 // CHECK: [[TRUE2]] 561 // CHECK: br label %[[END2:.+]] 562 // CHECK: [[FALSE2]] 563 // CHECK: br label %[[END2]] 564 // CHECK: [[END2]] 565 // CHECK: [[COND_LVALUE:%.+]] = phi [[S_INT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ] 566 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8* 567 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* 568 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 569 570 // t_var1 = min(t_var1, t_var1_reduction); 571 // CHECK: [[T_VAR1_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_REF]], 572 // CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_PRIV]], 573 // CHECK: [[CMP:%.+]] = icmp slt i{{[0-9]+}} [[T_VAR1_VAL]], [[T_VAR1_PRIV_VAL]] 574 // CHECK: [[UP:%.+]] = zext i1 [[CMP]] to i{{[0-9]+}} 575 // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR1_REF]], 576 577 // __kmpc_end_reduce_nowait(<loc>, <gtid>, &<lock>); 578 // CHECK: call void @__kmpc_end_reduce_nowait(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], [8 x i32]* [[REDUCTION_LOCK]]) 579 580 // break; 581 // CHECK: br label %[[RED_DONE]] 582 583 // case 2: 584 // t_var += t_var_reduction; 585 // CHECK: [[T_VAR_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_PRIV]] 586 // CHECK: atomicrmw add i32* [[T_VAR_REF]], i32 [[T_VAR_PRIV_VAL]] monotonic 587 588 // var = var.operator &(var_reduction); 589 // CHECK: call void @__kmpc_critical( 590 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_REF]], [[S_INT_TY]]* dereferenceable(4) [[VAR_PRIV]]) 591 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_REF]] to i8* 592 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* 593 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 594 // CHECK: call void @__kmpc_end_critical( 595 596 // var1 = var1.operator &&(var1_reduction); 597 // CHECK: call void @__kmpc_critical( 598 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_REF]]) 599 // CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 600 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]] 601 // CHECK: [[TRUE]] 602 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_PRIV]]) 603 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 604 // CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]] 605 // CHECK: [[TRUE2]] 606 // CHECK: br label %[[END2:.+]] 607 // CHECK: [[FALSE2]] 608 // CHECK: br label %[[END2]] 609 // CHECK: [[END2]] 610 // CHECK: [[COND_LVALUE:%.+]] = phi [[S_INT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ] 611 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8* 612 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* 613 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 614 // CHECK: call void @__kmpc_end_critical( 615 616 // t_var1 = min(t_var1, t_var1_reduction); 617 // CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_PRIV]] 618 // CHECK: atomicrmw min i32* [[T_VAR1_REF]], i32 [[T_VAR1_PRIV_VAL]] monotonic 619 620 // break; 621 // CHECK: br label %[[RED_DONE]] 622 // CHECK: [[RED_DONE]] 623 // CHECK-DAG: call {{.*}} [[S_INT_TY_DESTR]]([[S_INT_TY]]* [[VAR_PRIV]]) 624 // CHECK-DAG: call {{.*}} [[S_INT_TY_DESTR]]([[S_INT_TY]]* 625 // CHECK: call i32 @__kmpc_cancel_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i{{[0-9]+}} [[GTID]]) 626 // CHECK: ret void 627 628 // void reduce_func(void *lhs[<n>], void *rhs[<n>]) { 629 // *(Type0*)lhs[0] = ReductionOperation0(*(Type0*)lhs[0], *(Type0*)rhs[0]); 630 // ... 631 // *(Type<n>-1*)lhs[<n>-1] = ReductionOperation<n>-1(*(Type<n>-1*)lhs[<n>-1], 632 // *(Type<n>-1*)rhs[<n>-1]); 633 // } 634 // CHECK: define internal void [[REDUCTION_FUNC]](i8*, i8*) 635 // t_var_lhs = (i{{[0-9]+}}*)lhs[0]; 636 // CHECK: [[T_VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS:%.+]], i32 0, i32 0 637 // CHECK: [[T_VAR_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR_RHS_REF]], 638 // CHECK: [[T_VAR_RHS:%.+]] = bitcast i8* [[T_VAR_RHS_VOID]] to i{{[0-9]+}}* 639 // t_var_rhs = (i{{[0-9]+}}*)rhs[0]; 640 // CHECK: [[T_VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS:%.+]], i32 0, i32 0 641 // CHECK: [[T_VAR_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR_LHS_REF]], 642 // CHECK: [[T_VAR_LHS:%.+]] = bitcast i8* [[T_VAR_LHS_VOID]] to i{{[0-9]+}}* 643 644 // var_lhs = (S<i{{[0-9]+}}>*)lhs[1]; 645 // CHECK: [[VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i32 0, i32 1 646 // CHECK: [[VAR_RHS_VOID:%.+]] = load i8*, i8** [[VAR_RHS_REF]], 647 // CHECK: [[VAR_RHS:%.+]] = bitcast i8* [[VAR_RHS_VOID]] to [[S_INT_TY]]* 648 // var_rhs = (S<i{{[0-9]+}}>*)rhs[1]; 649 // CHECK: [[VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i32 0, i32 1 650 // CHECK: [[VAR_LHS_VOID:%.+]] = load i8*, i8** [[VAR_LHS_REF]], 651 // CHECK: [[VAR_LHS:%.+]] = bitcast i8* [[VAR_LHS_VOID]] to [[S_INT_TY]]* 652 653 // var1_lhs = (S<i{{[0-9]+}}>*)lhs[2]; 654 // CHECK: [[VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i32 0, i32 2 655 // CHECK: [[VAR1_RHS_VOID:%.+]] = load i8*, i8** [[VAR1_RHS_REF]], 656 // CHECK: [[VAR1_RHS:%.+]] = bitcast i8* [[VAR1_RHS_VOID]] to [[S_INT_TY]]* 657 // var1_rhs = (S<i{{[0-9]+}}>*)rhs[2]; 658 // CHECK: [[VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i32 0, i32 2 659 // CHECK: [[VAR1_LHS_VOID:%.+]] = load i8*, i8** [[VAR1_LHS_REF]], 660 // CHECK: [[VAR1_LHS:%.+]] = bitcast i8* [[VAR1_LHS_VOID]] to [[S_INT_TY]]* 661 662 // t_var1_lhs = (i{{[0-9]+}}*)lhs[3]; 663 // CHECK: [[T_VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i32 0, i32 3 664 // CHECK: [[T_VAR1_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_RHS_REF]], 665 // CHECK: [[T_VAR1_RHS:%.+]] = bitcast i8* [[T_VAR1_RHS_VOID]] to i{{[0-9]+}}* 666 // t_var1_rhs = (i{{[0-9]+}}*)rhs[3]; 667 // CHECK: [[T_VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i32 0, i32 3 668 // CHECK: [[T_VAR1_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_LHS_REF]], 669 // CHECK: [[T_VAR1_LHS:%.+]] = bitcast i8* [[T_VAR1_LHS_VOID]] to i{{[0-9]+}}* 670 671 // t_var_lhs += t_var_rhs; 672 // CHECK: [[T_VAR_LHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_LHS]], 673 // CHECK: [[T_VAR_RHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_RHS]], 674 // CHECK: [[UP:%.+]] = add nsw i{{[0-9]+}} [[T_VAR_LHS_VAL]], [[T_VAR_RHS_VAL]] 675 // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR_LHS]], 676 677 // var_lhs = var_lhs.operator &(var_rhs); 678 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_LHS]], [[S_INT_TY]]* dereferenceable(4) [[VAR_RHS]]) 679 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_LHS]] to i8* 680 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* 681 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 682 683 // var1_lhs = var1_lhs.operator &&(var1_rhs); 684 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_LHS]]) 685 // CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 686 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]] 687 // CHECK: [[TRUE]] 688 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_RHS]]) 689 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 690 // CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]] 691 // CHECK: [[TRUE2]] 692 // CHECK: br label %[[END2:.+]] 693 // CHECK: [[FALSE2]] 694 // CHECK: br label %[[END2]] 695 // CHECK: [[END2]] 696 // CHECK: [[COND_LVALUE:%.+]] = phi [[S_INT_TY]]* [ [[VAR1_LHS]], %[[TRUE2]] ], [ [[VAR1_RHS]], %[[FALSE2]] ] 697 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_LHS]] to i8* 698 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* 699 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 700 701 // t_var1_lhs = min(t_var1_lhs, t_var1_rhs); 702 // CHECK: [[T_VAR1_LHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_LHS]], 703 // CHECK: [[T_VAR1_RHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_RHS]], 704 // CHECK: [[CMP:%.+]] = icmp slt i{{[0-9]+}} [[T_VAR1_LHS_VAL]], [[T_VAR1_RHS_VAL]] 705 // CHECK: [[UP:%.+]] = zext i1 [[CMP]] to i{{[0-9]+}} 706 // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR1_LHS]], 707 // CHECK: ret void 708 709 #endif 710 711