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 // REQUIRES: x86-registered-target 7 // expected-no-diagnostics 8 #ifndef HEADER 9 #define HEADER 10 11 volatile int g __attribute__((aligned(128))) = 1212; 12 13 template <class T> 14 struct S { 15 T f; 16 S(T a) : f(a + g) {} 17 S() : f(g) {} 18 operator T() { return T(); } 19 S &operator&(const S &) { return *this; } 20 ~S() {} 21 }; 22 23 // CHECK-DAG: [[S_FLOAT_TY:%.+]] = type { float } 24 // CHECK-DAG: [[S_INT_TY:%.+]] = type { i{{[0-9]+}} } 25 // CHECK-DAG: [[IMPLICIT_BARRIER_LOC:@.+]] = private unnamed_addr constant %{{.+}} { i32 0, i32 66, i32 0, i32 0, i8* 26 // CHECK-DAG: [[REDUCTION_LOC:@.+]] = private unnamed_addr constant %{{.+}} { i32 0, i32 18, i32 0, i32 0, i8* 27 // CHECK-DAG: [[REDUCTION_LOCK:@.+]] = common global [8 x i32] zeroinitializer 28 29 template <typename T> 30 T tmain() { 31 T t; 32 S<T> test; 33 T t_var __attribute__((aligned(128))) = T(), t_var1 __attribute__((aligned(128))); 34 T vec[] = {1, 2}; 35 S<T> s_arr[] = {1, 2}; 36 S<T> var __attribute__((aligned(128))) (3), var1 __attribute__((aligned(128))); 37 #pragma omp parallel reduction(+:t_var) reduction(&:var) reduction(&& : var1) reduction(min: t_var1) 38 { 39 vec[0] = t_var; 40 s_arr[0] = var; 41 } 42 return T(); 43 } 44 45 int main() { 46 #ifdef LAMBDA 47 // LAMBDA: [[G:@.+]] = global i{{[0-9]+}} 1212, 48 // LAMBDA-LABEL: @main 49 // LAMBDA: call void [[OUTER_LAMBDA:@.+]]( 50 [&]() { 51 // LAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]]( 52 // LAMBDA: call void {{.+}} @__kmpc_fork_call({{.+}}, i32 1, {{.+}}* [[OMP_REGION:@.+]] to {{.+}}, i32* [[G]]) 53 #pragma omp parallel reduction(+:g) 54 { 55 // LAMBDA: define{{.*}} internal{{.*}} void [[OMP_REGION]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* dereferenceable(4) %{{.+}}) 56 // LAMBDA: [[G_PRIVATE_ADDR:%.+]] = alloca i{{[0-9]+}}, 57 58 // Reduction list for runtime. 59 // LAMBDA: [[RED_LIST:%.+]] = alloca [1 x i8*], 60 61 // LAMBDA: [[G_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[G_REF_ADDR:%.+]] 62 // LAMBDA: store i{{[0-9]+}} 0, i{{[0-9]+}}* [[G_PRIVATE_ADDR]], align 128 63 g = 1; 64 // LAMBDA: store i{{[0-9]+}} 1, i{{[0-9]+}}* [[G_PRIVATE_ADDR]], align 128 65 // LAMBDA: [[G_PRIVATE_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG:%.+]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 66 // LAMBDA: store i{{[0-9]+}}* [[G_PRIVATE_ADDR]], i{{[0-9]+}}** [[G_PRIVATE_ADDR_REF]] 67 // LAMBDA: call void [[INNER_LAMBDA:@.+]](%{{.+}}* [[ARG]]) 68 69 // LAMBDA: [[G_PRIV_REF:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[RED_LIST]], i64 0, i64 0 70 // LAMBDA: [[BITCAST:%.+]] = bitcast i32* [[G_PRIVATE_ADDR]] to i8* 71 // LAMBDA: store i8* [[BITCAST]], i8** [[G_PRIV_REF]], 72 // LAMBDA: call i32 @__kmpc_reduce_nowait( 73 // LAMBDA: switch i32 %{{.+}}, label %[[REDUCTION_DONE:.+]] [ 74 // LAMBDA: i32 1, label %[[CASE1:.+]] 75 // LAMBDA: i32 2, label %[[CASE2:.+]] 76 // LAMBDA: [[CASE1]] 77 // LAMBDA: [[G_VAL:%.+]] = load i32, i32* [[G_REF]] 78 // LAMBDA: [[G_PRIV_VAL:%.+]] = load i32, i32* [[G_PRIVATE_ADDR]] 79 // LAMBDA: [[ADD:%.+]] = add nsw i32 [[G_VAL]], [[G_PRIV_VAL]] 80 // LAMBDA: store i32 [[ADD]], i32* [[G_REF]] 81 // LAMBDA: call void @__kmpc_end_reduce_nowait( 82 // LAMBDA: br label %[[REDUCTION_DONE]] 83 // LAMBDA: [[CASE2]] 84 // LAMBDA: [[G_PRIV_VAL:%.+]] = load i32, i32* [[G_PRIVATE_ADDR]] 85 // LAMBDA: atomicrmw add i32* [[G_REF]], i32 [[G_PRIV_VAL]] monotonic 86 // LAMBDA: br label %[[REDUCTION_DONE]] 87 // LAMBDA: [[REDUCTION_DONE]] 88 // LAMBDA: ret void 89 [&]() { 90 // LAMBDA: define {{.+}} void [[INNER_LAMBDA]](%{{.+}}* [[ARG_PTR:%.+]]) 91 // LAMBDA: store %{{.+}}* [[ARG_PTR]], %{{.+}}** [[ARG_PTR_REF:%.+]], 92 g = 2; 93 // LAMBDA: [[ARG_PTR:%.+]] = load %{{.+}}*, %{{.+}}** [[ARG_PTR_REF]] 94 // LAMBDA: [[G_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 95 // LAMBDA: [[G_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[G_PTR_REF]] 96 // LAMBDA: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[G_REF]] 97 }(); 98 } 99 }(); 100 return 0; 101 #elif defined(BLOCKS) 102 // BLOCKS: [[G:@.+]] = global i{{[0-9]+}} 1212, 103 // BLOCKS-LABEL: @main 104 // BLOCKS: call void {{%.+}}(i8 105 ^{ 106 // BLOCKS: define{{.*}} internal{{.*}} void {{.+}}(i8* 107 // BLOCKS: call void {{.+}} @__kmpc_fork_call({{.+}}, i32 1, {{.+}}* [[OMP_REGION:@.+]] to {{.+}}, i32* [[G]]) 108 #pragma omp parallel reduction(-:g) 109 { 110 // BLOCKS: define{{.*}} internal{{.*}} void [[OMP_REGION]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* dereferenceable(4) %{{.+}}) 111 // BLOCKS: [[G_PRIVATE_ADDR:%.+]] = alloca i{{[0-9]+}}, 112 113 // Reduction list for runtime. 114 // BLOCKS: [[RED_LIST:%.+]] = alloca [1 x i8*], 115 116 // BLOCKS: [[G_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[G_REF_ADDR:%.+]] 117 // BLOCKS: store i{{[0-9]+}} 0, i{{[0-9]+}}* [[G_PRIVATE_ADDR]], align 128 118 g = 1; 119 // BLOCKS: store i{{[0-9]+}} 1, i{{[0-9]+}}* [[G_PRIVATE_ADDR]], align 128 120 // BLOCKS-NOT: [[G]]{{[[^:word:]]}} 121 // BLOCKS: i{{[0-9]+}}* [[G_PRIVATE_ADDR]] 122 // BLOCKS-NOT: [[G]]{{[[^:word:]]}} 123 // BLOCKS: call void {{%.+}}(i8 124 125 // BLOCKS: [[G_PRIV_REF:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[RED_LIST]], i64 0, i64 0 126 // BLOCKS: [[BITCAST:%.+]] = bitcast i32* [[G_PRIVATE_ADDR]] to i8* 127 // BLOCKS: store i8* [[BITCAST]], i8** [[G_PRIV_REF]], 128 // BLOCKS: call i32 @__kmpc_reduce_nowait( 129 // BLOCKS: switch i32 %{{.+}}, label %[[REDUCTION_DONE:.+]] [ 130 // BLOCKS: i32 1, label %[[CASE1:.+]] 131 // BLOCKS: i32 2, label %[[CASE2:.+]] 132 // BLOCKS: [[CASE1]] 133 // BLOCKS: [[G_VAL:%.+]] = load i32, i32* [[G_REF]] 134 // BLOCKS: [[G_PRIV_VAL:%.+]] = load i32, i32* [[G_PRIVATE_ADDR]] 135 // BLOCKS: [[ADD:%.+]] = add nsw i32 [[G_VAL]], [[G_PRIV_VAL]] 136 // BLOCKS: store i32 [[ADD]], i32* [[G_REF]] 137 // BLOCKS: call void @__kmpc_end_reduce_nowait( 138 // BLOCKS: br label %[[REDUCTION_DONE]] 139 // BLOCKS: [[CASE2]] 140 // BLOCKS: [[G_PRIV_VAL:%.+]] = load i32, i32* [[G_PRIVATE_ADDR]] 141 // BLOCKS: atomicrmw add i32* [[G_REF]], i32 [[G_PRIV_VAL]] monotonic 142 // BLOCKS: br label %[[REDUCTION_DONE]] 143 // BLOCKS: [[REDUCTION_DONE]] 144 // BLOCKS: ret void 145 ^{ 146 // BLOCKS: define {{.+}} void {{@.+}}(i8* 147 g = 2; 148 // BLOCKS-NOT: [[G]]{{[[^:word:]]}} 149 // BLOCKS: store i{{[0-9]+}} 2, i{{[0-9]+}}* 150 // BLOCKS-NOT: [[G]]{{[[^:word:]]}} 151 // BLOCKS: ret 152 }(); 153 } 154 }(); 155 return 0; 156 #else 157 S<float> test; 158 float t_var = 0, t_var1; 159 int vec[] = {1, 2}; 160 S<float> s_arr[] = {1, 2}; 161 S<float> var(3), var1; 162 #pragma omp parallel reduction(+:t_var) reduction(&:var) reduction(&& : var1) reduction(min: t_var1) 163 { 164 vec[0] = t_var; 165 s_arr[0] = var; 166 } 167 return tmain<int>(); 168 #endif 169 } 170 171 // CHECK: define {{.*}}i{{[0-9]+}} @main() 172 // CHECK: [[TEST:%.+]] = alloca [[S_FLOAT_TY]], 173 // CHECK: call {{.*}} [[S_FLOAT_TY_CONSTR:@.+]]([[S_FLOAT_TY]]* [[TEST]]) 174 // 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]+}}*, [2 x i32]*, float*, [2 x [[S_FLOAT_TY]]]*, [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]*, float*)* [[MAIN_MICROTASK:@.+]] to void 175 // CHECK: = call {{.*}}i{{.+}} [[TMAIN_INT:@.+]]() 176 // CHECK: call {{.*}} [[S_FLOAT_TY_DESTR:@.+]]([[S_FLOAT_TY]]* 177 // CHECK: ret 178 // 179 // CHECK: define internal void [[MAIN_MICROTASK]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, 180 // CHECK: [[T_VAR_PRIV:%.+]] = alloca float, 181 // CHECK: [[VAR_PRIV:%.+]] = alloca [[S_FLOAT_TY]], 182 // CHECK: [[VAR1_PRIV:%.+]] = alloca [[S_FLOAT_TY]], 183 // CHECK: [[T_VAR1_PRIV:%.+]] = alloca float, 184 185 // Reduction list for runtime. 186 // CHECK: [[RED_LIST:%.+]] = alloca [4 x i8*], 187 188 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]], 189 190 // CHECK: [[T_VAR_REF:%.+]] = load float*, float** % 191 // CHECK: [[VAR_REF:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** % 192 // CHECK: [[VAR1_REF:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** % 193 // CHECK: [[T_VAR1_REF:%.+]] = load float*, float** % 194 195 // For + reduction operation initial value of private variable is 0. 196 // CHECK: store float 0.0{{.+}}, float* [[T_VAR_PRIV]], 197 198 // For & reduction operation initial value of private variable is ones in all bits. 199 // CHECK: call {{.*}} [[S_FLOAT_TY_CONSTR:@.+]]([[S_FLOAT_TY]]* [[VAR_PRIV]]) 200 201 // For && reduction operation initial value of private variable is 1.0. 202 // CHECK: call {{.*}} [[S_FLOAT_TY_CONSTR:@.+]]([[S_FLOAT_TY]]* [[VAR1_PRIV]]) 203 204 // For min reduction operation initial value of private variable is largest repesentable value. 205 // CHECK: store float 0x47EFFFFFE0000000, float* [[T_VAR1_PRIV]], 206 207 // Skip checks for internal operations. 208 209 // void *RedList[<n>] = {<ReductionVars>[0], ..., <ReductionVars>[<n>-1]}; 210 211 // CHECK: [[T_VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 0 212 // CHECK: [[BITCAST:%.+]] = bitcast float* [[T_VAR_PRIV]] to i8* 213 // CHECK: store i8* [[BITCAST]], i8** [[T_VAR_PRIV_REF]], 214 // CHECK: [[VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 1 215 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_PRIV]] to i8* 216 // CHECK: store i8* [[BITCAST]], i8** [[VAR_PRIV_REF]], 217 // CHECK: [[VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 2 218 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_PRIV]] to i8* 219 // CHECK: store i8* [[BITCAST]], i8** [[VAR1_PRIV_REF]], 220 // CHECK: [[T_VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 3 221 // CHECK: [[BITCAST:%.+]] = bitcast float* [[T_VAR1_PRIV]] to i8* 222 // CHECK: store i8* [[BITCAST]], i8** [[T_VAR1_PRIV_REF]], 223 224 // res = __kmpc_reduce_nowait(<loc>, <gtid>, <n>, sizeof(RedList), RedList, reduce_func, &<lock>); 225 226 // CHECK: [[GTID_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_ADDR]] 227 // CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_REF]] 228 // CHECK: [[BITCAST:%.+]] = bitcast [4 x i8*]* [[RED_LIST]] to i8* 229 // 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]]) 230 231 // switch(res) 232 // CHECK: switch i32 [[RES]], label %[[RED_DONE:.+]] [ 233 // CHECK: i32 1, label %[[CASE1:.+]] 234 // CHECK: i32 2, label %[[CASE2:.+]] 235 // CHECK: ] 236 237 // case 1: 238 // t_var += t_var_reduction; 239 // CHECK: [[T_VAR_VAL:%.+]] = load float, float* [[T_VAR_REF]], 240 // CHECK: [[T_VAR_PRIV_VAL:%.+]] = load float, float* [[T_VAR_PRIV]], 241 // CHECK: [[UP:%.+]] = fadd float [[T_VAR_VAL]], [[T_VAR_PRIV_VAL]] 242 // CHECK: store float [[UP]], float* [[T_VAR_REF]], 243 244 // var = var.operator &(var_reduction); 245 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_REF]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_PRIV]]) 246 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_REF]] to i8* 247 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8* 248 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 249 250 // var1 = var1.operator &&(var1_reduction); 251 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_REF]]) 252 // CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 253 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]] 254 // CHECK: [[TRUE]] 255 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_PRIV]]) 256 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 257 // CHECK: br label %[[END2]] 258 // CHECK: [[END2]] 259 // CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ] 260 // CHECK: [[CONV:%.+]] = uitofp i1 [[COND_LVALUE]] to float 261 // CHECK: call void @{{.+}}([[S_FLOAT_TY]]* [[COND_LVALUE:%.+]], float [[CONV]]) 262 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_REF]] to i8* 263 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8* 264 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 265 266 // t_var1 = min(t_var1, t_var1_reduction); 267 // CHECK: [[T_VAR1_VAL:%.+]] = load float, float* [[T_VAR1_REF]], 268 // CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load float, float* [[T_VAR1_PRIV]], 269 // CHECK: [[CMP:%.+]] = fcmp olt float [[T_VAR1_VAL]], [[T_VAR1_PRIV_VAL]] 270 // CHECK: br i1 [[CMP]] 271 // CHECK: [[UP:%.+]] = phi float 272 // CHECK: store float [[UP]], float* [[T_VAR1_REF]], 273 274 // __kmpc_end_reduce_nowait(<loc>, <gtid>, &<lock>); 275 // CHECK: call void @__kmpc_end_reduce_nowait(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], [8 x i32]* [[REDUCTION_LOCK]]) 276 277 // break; 278 // CHECK: br label %[[RED_DONE]] 279 280 // case 2: 281 // t_var += t_var_reduction; 282 // CHECK: load float, float* [[T_VAR_PRIV]] 283 // CHECK: [[T_VAR_REF_INT:%.+]] = bitcast float* [[T_VAR_REF]] to i32* 284 // CHECK: [[OLD1:%.+]] = load atomic i32, i32* [[T_VAR_REF_INT]] monotonic, 285 // CHECK: br label %[[CONT:.+]] 286 // CHECK: [[CONT]] 287 // CHECK: [[ORIG_OLD_INT:%.+]] = phi i32 [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %[[CONT]] ] 288 // CHECK: fadd float 289 // CHECK: [[UP_INT:%.+]] = load i32 290 // CHECK: [[T_VAR_REF_INT:%.+]] = bitcast float* [[T_VAR_REF]] to i32* 291 // CHECK: [[RES:%.+]] = cmpxchg i32* [[T_VAR_REF_INT]], i32 [[ORIG_OLD_INT]], i32 [[UP_INT]] monotonic monotonic 292 // CHECK: [[OLD2:%.+]] = extractvalue { i32, i1 } [[RES]], 0 293 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1 294 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[ATOMIC_DONE:.+]], label %[[CONT]] 295 // CHECK: [[ATOMIC_DONE]] 296 297 // var = var.operator &(var_reduction); 298 // CHECK: call void @__kmpc_critical( 299 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_REF]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_PRIV]]) 300 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_REF]] to i8* 301 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8* 302 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 303 // CHECK: call void @__kmpc_end_critical( 304 305 // var1 = var1.operator &&(var1_reduction); 306 // CHECK: call void @__kmpc_critical( 307 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_REF]]) 308 // CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 309 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]] 310 // CHECK: [[TRUE]] 311 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_PRIV]]) 312 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 313 // CHECK: br label %[[END2]] 314 // CHECK: [[END2]] 315 // CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ] 316 // CHECK: [[CONV:%.+]] = uitofp i1 [[COND_LVALUE]] to float 317 // CHECK: call void @{{.+}}([[S_FLOAT_TY]]* [[COND_LVALUE:%.+]], float [[CONV]]) 318 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_REF]] to i8* 319 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8* 320 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 321 // CHECK: call void @__kmpc_end_critical( 322 323 // t_var1 = min(t_var1, t_var1_reduction); 324 // CHECK: load float, float* [[T_VAR1_PRIV]] 325 // CHECK: [[T_VAR1_REF_INT:%.+]] = bitcast float* [[T_VAR1_REF]] to i32* 326 // CHECK: [[OLD1:%.+]] = load atomic i32, i32* [[T_VAR1_REF_INT]] monotonic, 327 // CHECK: br label %[[CONT:.+]] 328 // CHECK: [[CONT]] 329 // CHECK: [[ORIG_OLD_INT:%.+]] = phi i32 [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %{{.+}} ] 330 // CHECK: [[CMP:%.+]] = fcmp olt float 331 // CHECK: br i1 [[CMP]] 332 // CHECK: [[UP:%.+]] = phi float 333 // CHECK: [[UP_INT:%.+]] = load i32 334 // CHECK: [[T_VAR1_REF_INT:%.+]] = bitcast float* [[T_VAR1_REF]] to i32* 335 // CHECK: [[RES:%.+]] = cmpxchg i32* [[T_VAR1_REF_INT]], i32 [[ORIG_OLD_INT]], i32 [[UP_INT]] monotonic monotonic 336 // CHECK: [[OLD2:%.+]] = extractvalue { i32, i1 } [[RES]], 0 337 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1 338 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[ATOMIC_DONE:.+]], label %[[CONT]] 339 // CHECK: [[ATOMIC_DONE]] 340 341 // break; 342 // CHECK: br label %[[RED_DONE]] 343 // CHECK: [[RED_DONE]] 344 // CHECK: call void @__kmpc_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i{{[0-9]+}} [[GTID]]) 345 346 // CHECK-DAG: call {{.*}} [[S_FLOAT_TY_DESTR]]([[S_FLOAT_TY]]* [[VAR_PRIV]]) 347 // CHECK-DAG: call {{.*}} [[S_FLOAT_TY_DESTR]]([[S_FLOAT_TY]]* 348 // CHECK: ret void 349 350 // void reduce_func(void *lhs[<n>], void *rhs[<n>]) { 351 // *(Type0*)lhs[0] = ReductionOperation0(*(Type0*)lhs[0], *(Type0*)rhs[0]); 352 // ... 353 // *(Type<n>-1*)lhs[<n>-1] = ReductionOperation<n>-1(*(Type<n>-1*)lhs[<n>-1], 354 // *(Type<n>-1*)rhs[<n>-1]); 355 // } 356 // CHECK: define internal void [[REDUCTION_FUNC]](i8*, i8*) 357 // t_var_lhs = (float*)lhs[0]; 358 // CHECK: [[T_VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS:%.+]], i64 0, i64 0 359 // CHECK: [[T_VAR_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR_RHS_REF]], 360 // CHECK: [[T_VAR_RHS:%.+]] = bitcast i8* [[T_VAR_RHS_VOID]] to float* 361 // t_var_rhs = (float*)rhs[0]; 362 // CHECK: [[T_VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS:%.+]], i64 0, i64 0 363 // CHECK: [[T_VAR_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR_LHS_REF]], 364 // CHECK: [[T_VAR_LHS:%.+]] = bitcast i8* [[T_VAR_LHS_VOID]] to float* 365 366 // var_lhs = (S<float>*)lhs[1]; 367 // CHECK: [[VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 1 368 // CHECK: [[VAR_RHS_VOID:%.+]] = load i8*, i8** [[VAR_RHS_REF]], 369 // CHECK: [[VAR_RHS:%.+]] = bitcast i8* [[VAR_RHS_VOID]] to [[S_FLOAT_TY]]* 370 // var_rhs = (S<float>*)rhs[1]; 371 // CHECK: [[VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 1 372 // CHECK: [[VAR_LHS_VOID:%.+]] = load i8*, i8** [[VAR_LHS_REF]], 373 // CHECK: [[VAR_LHS:%.+]] = bitcast i8* [[VAR_LHS_VOID]] to [[S_FLOAT_TY]]* 374 375 // var1_lhs = (S<float>*)lhs[2]; 376 // CHECK: [[VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 2 377 // CHECK: [[VAR1_RHS_VOID:%.+]] = load i8*, i8** [[VAR1_RHS_REF]], 378 // CHECK: [[VAR1_RHS:%.+]] = bitcast i8* [[VAR1_RHS_VOID]] to [[S_FLOAT_TY]]* 379 // var1_rhs = (S<float>*)rhs[2]; 380 // CHECK: [[VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 2 381 // CHECK: [[VAR1_LHS_VOID:%.+]] = load i8*, i8** [[VAR1_LHS_REF]], 382 // CHECK: [[VAR1_LHS:%.+]] = bitcast i8* [[VAR1_LHS_VOID]] to [[S_FLOAT_TY]]* 383 384 // t_var1_lhs = (float*)lhs[3]; 385 // CHECK: [[T_VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 3 386 // CHECK: [[T_VAR1_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_RHS_REF]], 387 // CHECK: [[T_VAR1_RHS:%.+]] = bitcast i8* [[T_VAR1_RHS_VOID]] to float* 388 // t_var1_rhs = (float*)rhs[3]; 389 // CHECK: [[T_VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 3 390 // CHECK: [[T_VAR1_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_LHS_REF]], 391 // CHECK: [[T_VAR1_LHS:%.+]] = bitcast i8* [[T_VAR1_LHS_VOID]] to float* 392 393 // t_var_lhs += t_var_rhs; 394 // CHECK: [[T_VAR_LHS_VAL:%.+]] = load float, float* [[T_VAR_LHS]], 395 // CHECK: [[T_VAR_RHS_VAL:%.+]] = load float, float* [[T_VAR_RHS]], 396 // CHECK: [[UP:%.+]] = fadd float [[T_VAR_LHS_VAL]], [[T_VAR_RHS_VAL]] 397 // CHECK: store float [[UP]], float* [[T_VAR_LHS]], 398 399 // var_lhs = var_lhs.operator &(var_rhs); 400 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_LHS]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_RHS]]) 401 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_LHS]] to i8* 402 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8* 403 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 404 405 // var1_lhs = var1_lhs.operator &&(var1_rhs); 406 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_LHS]]) 407 // CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 408 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]] 409 // CHECK: [[TRUE]] 410 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_RHS]]) 411 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 412 // CHECK: br label %[[END2]] 413 // CHECK: [[END2]] 414 // CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ] 415 // CHECK: [[CONV:%.+]] = uitofp i1 [[COND_LVALUE]] to float 416 // CHECK: call void @{{.+}}([[S_FLOAT_TY]]* [[COND_LVALUE:%.+]], float [[CONV]]) 417 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_LHS]] to i8* 418 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8* 419 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 420 421 // t_var1_lhs = min(t_var1_lhs, t_var1_rhs); 422 // CHECK: [[T_VAR1_LHS_VAL:%.+]] = load float, float* [[T_VAR1_LHS]], 423 // CHECK: [[T_VAR1_RHS_VAL:%.+]] = load float, float* [[T_VAR1_RHS]], 424 // CHECK: [[CMP:%.+]] = fcmp olt float [[T_VAR1_LHS_VAL]], [[T_VAR1_RHS_VAL]] 425 // CHECK: br i1 [[CMP]] 426 // CHECK: [[UP:%.+]] = phi float 427 // CHECK: store float [[UP]], float* [[T_VAR1_LHS]], 428 // CHECK: ret void 429 430 // CHECK: define {{.*}} i{{[0-9]+}} [[TMAIN_INT]]() 431 // CHECK: [[TEST:%.+]] = alloca [[S_INT_TY]], 432 // CHECK: call {{.*}} [[S_INT_TY_CONSTR:@.+]]([[S_INT_TY]]* [[TEST]]) 433 // 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]+}}*, [2 x i32]*, i32*, [2 x [[S_INT_TY]]]*, [[S_INT_TY]]*, [[S_INT_TY]]*, i32*)* [[TMAIN_MICROTASK:@.+]] to void 434 // CHECK: call {{.*}} [[S_INT_TY_DESTR:@.+]]([[S_INT_TY]]* 435 // CHECK: ret 436 // 437 // CHECK: define internal void [[TMAIN_MICROTASK]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, 438 // CHECK: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}}, align 128 439 // CHECK: [[VAR_PRIV:%.+]] = alloca [[S_INT_TY]], align 128 440 // CHECK: [[VAR1_PRIV:%.+]] = alloca [[S_INT_TY]], align 128 441 // CHECK: [[T_VAR1_PRIV:%.+]] = alloca i{{[0-9]+}}, align 128 442 443 // Reduction list for runtime. 444 // CHECK: [[RED_LIST:%.+]] = alloca [4 x i8*], 445 446 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]], 447 448 // CHECK: [[T_VAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** % 449 // CHECK: [[VAR_REF:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** % 450 // CHECK: [[VAR1_REF:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** % 451 // CHECK: [[T_VAR1_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** % 452 453 // For + reduction operation initial value of private variable is 0. 454 // CHECK: store i{{[0-9]+}} 0, i{{[0-9]+}}* [[T_VAR_PRIV]], 455 456 // For & reduction operation initial value of private variable is ones in all bits. 457 // CHECK: call {{.*}} [[S_INT_TY_CONSTR:@.+]]([[S_INT_TY]]* [[VAR_PRIV]]) 458 459 // For && reduction operation initial value of private variable is 1.0. 460 // CHECK: call {{.*}} [[S_INT_TY_CONSTR:@.+]]([[S_INT_TY]]* [[VAR1_PRIV]]) 461 462 // For min reduction operation initial value of private variable is largest repesentable value. 463 // CHECK: store i{{[0-9]+}} 2147483647, i{{[0-9]+}}* [[T_VAR1_PRIV]], 464 465 // Skip checks for internal operations. 466 467 // void *RedList[<n>] = {<ReductionVars>[0], ..., <ReductionVars>[<n>-1]}; 468 469 // CHECK: [[T_VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 0 470 // CHECK: [[BITCAST:%.+]] = bitcast i{{[0-9]+}}* [[T_VAR_PRIV]] to i8* 471 // CHECK: store i8* [[BITCAST]], i8** [[T_VAR_PRIV_REF]], 472 // CHECK: [[VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 1 473 // CHECK: [[BITCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR_PRIV]] to i8* 474 // CHECK: store i8* [[BITCAST]], i8** [[VAR_PRIV_REF]], 475 // CHECK: [[VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 2 476 // CHECK: [[BITCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_PRIV]] to i8* 477 // CHECK: store i8* [[BITCAST]], i8** [[VAR1_PRIV_REF]], 478 // CHECK: [[T_VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 3 479 // CHECK: [[BITCAST:%.+]] = bitcast i{{[0-9]+}}* [[T_VAR1_PRIV]] to i8* 480 // CHECK: store i8* [[BITCAST]], i8** [[T_VAR1_PRIV_REF]], 481 482 // res = __kmpc_reduce_nowait(<loc>, <gtid>, <n>, sizeof(RedList), RedList, reduce_func, &<lock>); 483 484 // CHECK: [[GTID_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_ADDR]] 485 // CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_REF]] 486 // CHECK: [[BITCAST:%.+]] = bitcast [4 x i8*]* [[RED_LIST]] to i8* 487 // 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]]) 488 489 // switch(res) 490 // CHECK: switch i32 [[RES]], label %[[RED_DONE:.+]] [ 491 // CHECK: i32 1, label %[[CASE1:.+]] 492 // CHECK: i32 2, label %[[CASE2:.+]] 493 // CHECK: ] 494 495 // case 1: 496 // t_var += t_var_reduction; 497 // CHECK: [[T_VAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_REF]], 498 // CHECK: [[T_VAR_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_PRIV]], 499 // CHECK: [[UP:%.+]] = add nsw i{{[0-9]+}} [[T_VAR_VAL]], [[T_VAR_PRIV_VAL]] 500 // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR_REF]], 501 502 // var = var.operator &(var_reduction); 503 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_REF]], [[S_INT_TY]]* dereferenceable(4) [[VAR_PRIV]]) 504 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_REF]] to i8* 505 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* 506 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 507 508 // var1 = var1.operator &&(var1_reduction); 509 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_REF]]) 510 // CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 511 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]] 512 // CHECK: [[TRUE]] 513 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_PRIV]]) 514 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 515 // CHECK: [[END2]] 516 // CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ] 517 // CHECK: [[CONV:%.+]] = zext i1 [[COND_LVALUE]] to i32 518 // CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]]) 519 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8* 520 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* 521 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 522 523 // t_var1 = min(t_var1, t_var1_reduction); 524 // CHECK: [[T_VAR1_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_REF]], 525 // CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_PRIV]], 526 // CHECK: [[CMP:%.+]] = icmp slt i{{[0-9]+}} [[T_VAR1_VAL]], [[T_VAR1_PRIV_VAL]] 527 // CHECK: br i1 [[CMP]] 528 // CHECK: [[UP:%.+]] = phi i32 529 // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR1_REF]], 530 531 // __kmpc_end_reduce_nowait(<loc>, <gtid>, &<lock>); 532 // CHECK: call void @__kmpc_end_reduce_nowait(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], [8 x i32]* [[REDUCTION_LOCK]]) 533 534 // break; 535 // CHECK: br label %[[RED_DONE]] 536 537 // case 2: 538 // t_var += t_var_reduction; 539 // CHECK: [[T_VAR_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_PRIV]] 540 // CHECK: atomicrmw add i32* [[T_VAR_REF]], i32 [[T_VAR_PRIV_VAL]] monotonic 541 542 // var = var.operator &(var_reduction); 543 // CHECK: call void @__kmpc_critical( 544 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_REF]], [[S_INT_TY]]* dereferenceable(4) [[VAR_PRIV]]) 545 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_REF]] to i8* 546 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* 547 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 548 // CHECK: call void @__kmpc_end_critical( 549 550 // var1 = var1.operator &&(var1_reduction); 551 // CHECK: call void @__kmpc_critical( 552 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_REF]]) 553 // CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 554 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]] 555 // CHECK: [[TRUE]] 556 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_PRIV]]) 557 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 558 // CHECK: br label %[[END2]] 559 // CHECK: [[END2]] 560 // CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ] 561 // CHECK: [[CONV:%.+]] = zext i1 [[COND_LVALUE]] to i32 562 // CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]]) 563 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8* 564 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* 565 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 566 // CHECK: call void @__kmpc_end_critical( 567 568 // t_var1 = min(t_var1, t_var1_reduction); 569 // CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_PRIV]] 570 // CHECK: atomicrmw min i32* [[T_VAR1_REF]], i32 [[T_VAR1_PRIV_VAL]] monotonic 571 572 // break; 573 // CHECK: br label %[[RED_DONE]] 574 // CHECK: [[RED_DONE]] 575 // CHECK: call void @__kmpc_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i{{[0-9]+}} [[GTID]]) 576 577 // CHECK-DAG: call {{.*}} [[S_INT_TY_DESTR]]([[S_INT_TY]]* [[VAR_PRIV]]) 578 // CHECK-DAG: call {{.*}} [[S_INT_TY_DESTR]]([[S_INT_TY]]* 579 // CHECK: ret void 580 581 // void reduce_func(void *lhs[<n>], void *rhs[<n>]) { 582 // *(Type0*)lhs[0] = ReductionOperation0(*(Type0*)lhs[0], *(Type0*)rhs[0]); 583 // ... 584 // *(Type<n>-1*)lhs[<n>-1] = ReductionOperation<n>-1(*(Type<n>-1*)lhs[<n>-1], 585 // *(Type<n>-1*)rhs[<n>-1]); 586 // } 587 // CHECK: define internal void [[REDUCTION_FUNC]](i8*, i8*) 588 // t_var_lhs = (i{{[0-9]+}}*)lhs[0]; 589 // CHECK: [[T_VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS:%.+]], i64 0, i64 0 590 // CHECK: [[T_VAR_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR_RHS_REF]], 591 // CHECK: [[T_VAR_RHS:%.+]] = bitcast i8* [[T_VAR_RHS_VOID]] to i{{[0-9]+}}* 592 // t_var_rhs = (i{{[0-9]+}}*)rhs[0]; 593 // CHECK: [[T_VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS:%.+]], i64 0, i64 0 594 // CHECK: [[T_VAR_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR_LHS_REF]], 595 // CHECK: [[T_VAR_LHS:%.+]] = bitcast i8* [[T_VAR_LHS_VOID]] to i{{[0-9]+}}* 596 597 // var_lhs = (S<i{{[0-9]+}}>*)lhs[1]; 598 // CHECK: [[VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 1 599 // CHECK: [[VAR_RHS_VOID:%.+]] = load i8*, i8** [[VAR_RHS_REF]], 600 // CHECK: [[VAR_RHS:%.+]] = bitcast i8* [[VAR_RHS_VOID]] to [[S_INT_TY]]* 601 // var_rhs = (S<i{{[0-9]+}}>*)rhs[1]; 602 // CHECK: [[VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 1 603 // CHECK: [[VAR_LHS_VOID:%.+]] = load i8*, i8** [[VAR_LHS_REF]], 604 // CHECK: [[VAR_LHS:%.+]] = bitcast i8* [[VAR_LHS_VOID]] to [[S_INT_TY]]* 605 606 // var1_lhs = (S<i{{[0-9]+}}>*)lhs[2]; 607 // CHECK: [[VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 2 608 // CHECK: [[VAR1_RHS_VOID:%.+]] = load i8*, i8** [[VAR1_RHS_REF]], 609 // CHECK: [[VAR1_RHS:%.+]] = bitcast i8* [[VAR1_RHS_VOID]] to [[S_INT_TY]]* 610 // var1_rhs = (S<i{{[0-9]+}}>*)rhs[2]; 611 // CHECK: [[VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 2 612 // CHECK: [[VAR1_LHS_VOID:%.+]] = load i8*, i8** [[VAR1_LHS_REF]], 613 // CHECK: [[VAR1_LHS:%.+]] = bitcast i8* [[VAR1_LHS_VOID]] to [[S_INT_TY]]* 614 615 // t_var1_lhs = (i{{[0-9]+}}*)lhs[3]; 616 // CHECK: [[T_VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 3 617 // CHECK: [[T_VAR1_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_RHS_REF]], 618 // CHECK: [[T_VAR1_RHS:%.+]] = bitcast i8* [[T_VAR1_RHS_VOID]] to i{{[0-9]+}}* 619 // t_var1_rhs = (i{{[0-9]+}}*)rhs[3]; 620 // CHECK: [[T_VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 3 621 // CHECK: [[T_VAR1_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_LHS_REF]], 622 // CHECK: [[T_VAR1_LHS:%.+]] = bitcast i8* [[T_VAR1_LHS_VOID]] to i{{[0-9]+}}* 623 624 // t_var_lhs += t_var_rhs; 625 // CHECK: [[T_VAR_LHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_LHS]], 626 // CHECK: [[T_VAR_RHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_RHS]], 627 // CHECK: [[UP:%.+]] = add nsw i{{[0-9]+}} [[T_VAR_LHS_VAL]], [[T_VAR_RHS_VAL]] 628 // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR_LHS]], 629 630 // var_lhs = var_lhs.operator &(var_rhs); 631 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_LHS]], [[S_INT_TY]]* dereferenceable(4) [[VAR_RHS]]) 632 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_LHS]] to i8* 633 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* 634 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 635 636 // var1_lhs = var1_lhs.operator &&(var1_rhs); 637 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_LHS]]) 638 // CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 639 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]] 640 // CHECK: [[TRUE]] 641 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_RHS]]) 642 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 643 // CHECK: br label %[[END2]] 644 // CHECK: [[END2]] 645 // CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ] 646 // CHECK: [[CONV:%.+]] = zext i1 [[COND_LVALUE]] to i32 647 // CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]]) 648 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_LHS]] to i8* 649 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* 650 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 651 652 // t_var1_lhs = min(t_var1_lhs, t_var1_rhs); 653 // CHECK: [[T_VAR1_LHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_LHS]], 654 // CHECK: [[T_VAR1_RHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_RHS]], 655 // CHECK: [[CMP:%.+]] = icmp slt i{{[0-9]+}} [[T_VAR1_LHS_VAL]], [[T_VAR1_RHS_VAL]] 656 // CHECK: br i1 [[CMP]] 657 // CHECK: [[UP:%.+]] = phi i32 658 // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR1_LHS]], 659 // CHECK: ret void 660 661 #endif 662 663