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 int g = 1212; 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 { [2 x i{{[0-9]+}}]*, float*, [2 x [[S_FLOAT_TY]]]*, [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]*, float* } 25 // CHECK-DAG: [[CAP_TMAIN_TY:%.+]] = type { [2 x i{{[0-9]+}}]*, i{{[0-9]+}}*, [2 x [[S_INT_TY]]]*, [[S_INT_TY]]*, [[S_INT_TY]]*, i{{[0-9]+}}* } 26 // CHECK-DAG: [[IMPLICIT_BARRIER_LOC:@.+]] = private unnamed_addr constant %{{.+}} { i32 0, i32 66, i32 0, i32 0, i8* 27 // CHECK-DAG: [[REDUCTION_LOC:@.+]] = private unnamed_addr constant %{{.+}} { i32 0, i32 18, i32 0, i32 0, i8* 28 // CHECK-DAG: [[REDUCTION_LOCK:@.+]] = common global [8 x i32] zeroinitializer 29 30 template <typename T> 31 T tmain() { 32 T t; 33 S<T> test; 34 T t_var = T(), t_var1; 35 T vec[] = {1, 2}; 36 S<T> s_arr[] = {1, 2}; 37 S<T> var(3), var1; 38 #pragma omp parallel reduction(+:t_var) reduction(&:var) reduction(&& : var1) reduction(min: t_var1) 39 { 40 vec[0] = t_var; 41 s_arr[0] = var; 42 } 43 return T(); 44 } 45 46 int main() { 47 #ifdef LAMBDA 48 // LAMBDA: [[G:@.+]] = global i{{[0-9]+}} 1212, 49 // LAMBDA-LABEL: @main 50 // LAMBDA: call void [[OUTER_LAMBDA:@.+]]( 51 [&]() { 52 // LAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]]( 53 // LAMBDA: [[G_LOCAL_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[AGG_CAPTURED:%.+]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 54 // LAMBDA: store i{{[0-9]+}}* [[G]], i{{[0-9]+}}** [[G_LOCAL_REF]] 55 // LAMBDA: [[ARG:%.+]] = bitcast %{{.+}}* [[AGG_CAPTURED]] to i8* 56 // LAMBDA: call void {{.+}}* @__kmpc_fork_call({{.+}}, i32 1, {{.+}}* [[OMP_REGION:@.+]] to {{.+}}, i8* [[ARG]]) 57 #pragma omp parallel reduction(+:g) 58 { 59 // LAMBDA: define{{.*}} internal{{.*}} void [[OMP_REGION]](i32* %{{.+}}, i32* %{{.+}}, %{{.+}}* [[ARG:%.+]]) 60 // LAMBDA: [[G_PRIVATE_ADDR:%.+]] = alloca i{{[0-9]+}}, 61 62 // Reduction list for runtime. 63 // LAMBDA: [[RED_LIST:%.+]] = alloca [1 x i8*], 64 65 // LAMBDA: store %{{.+}}* [[ARG]], %{{.+}}** [[ARG_REF:%.+]], 66 // LAMBDA: [[ARG:%.+]] = load %{{.+}}*, %{{.+}}** [[ARG_REF]] 67 // LAMBDA: [[G_REF_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 68 // LAMBDA: [[G_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[G_REF_ADDR]] 69 // LAMBDA: store i{{[0-9]+}} 0, i{{[0-9]+}}* [[G_PRIVATE_ADDR]] 70 // LAMBDA: call i32 @__kmpc_cancel_barrier( 71 g = 1; 72 // LAMBDA: store volatile i{{[0-9]+}} 1, i{{[0-9]+}}* [[G_PRIVATE_ADDR]], 73 // LAMBDA: [[G_PRIVATE_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG:%.+]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 74 // LAMBDA: store i{{[0-9]+}}* [[G_PRIVATE_ADDR]], i{{[0-9]+}}** [[G_PRIVATE_ADDR_REF]] 75 // LAMBDA: call void [[INNER_LAMBDA:@.+]](%{{.+}}* [[ARG]]) 76 77 // LAMBDA: [[G_PRIV_REF:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[RED_LIST]], i32 0, i32 0 78 // LAMBDA: [[BITCAST:%.+]] = bitcast i32* [[G_PRIVATE_ADDR]] to i8* 79 // LAMBDA: store i8* [[BITCAST]], i8** [[G_PRIV_REF]], 80 // LAMBDA: call i32 @__kmpc_reduce_nowait( 81 // LAMBDA: switch i32 %{{.+}}, label %[[REDUCTION_DONE:.+]] [ 82 // LAMBDA: i32 1, label %[[CASE1:.+]] 83 // LAMBDA: i32 2, label %[[CASE2:.+]] 84 // LAMBDA: [[CASE1]] 85 // LAMBDA: [[G_VAL:%.+]] = load i32, i32* [[G_REF]] 86 // LAMBDA: [[G_PRIV_VAL:%.+]] = load i32, i32* [[G_PRIVATE_ADDR]] 87 // LAMBDA: [[ADD:%.+]] = add nsw i32 [[G_VAL]], [[G_PRIV_VAL]] 88 // LAMBDA: store i32 [[ADD]], i32* [[G_REF]] 89 // LAMBDA: call void @__kmpc_end_reduce_nowait( 90 // LAMBDA: br label %[[REDUCTION_DONE]] 91 // LAMBDA: [[CASE2]] 92 // LAMBDA: [[G_PRIV_VAL:%.+]] = load i32, i32* [[G_PRIVATE_ADDR]] 93 // LAMBDA: atomicrmw add i32* [[G_REF]], i32 [[G_PRIV_VAL]] monotonic 94 // LAMBDA: br label %[[REDUCTION_DONE]] 95 // LAMBDA: [[REDUCTION_DONE]] 96 // LAMBDA: ret void 97 [&]() { 98 // LAMBDA: define {{.+}} void [[INNER_LAMBDA]](%{{.+}}* [[ARG_PTR:%.+]]) 99 // LAMBDA: store %{{.+}}* [[ARG_PTR]], %{{.+}}** [[ARG_PTR_REF:%.+]], 100 g = 2; 101 // LAMBDA: [[ARG_PTR:%.+]] = load %{{.+}}*, %{{.+}}** [[ARG_PTR_REF]] 102 // LAMBDA: [[G_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 103 // LAMBDA: [[G_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[G_PTR_REF]] 104 // LAMBDA: store volatile i{{[0-9]+}} 2, i{{[0-9]+}}* [[G_REF]] 105 }(); 106 } 107 }(); 108 return 0; 109 #elif defined(BLOCKS) 110 // BLOCKS: [[G:@.+]] = global i{{[0-9]+}} 1212, 111 // BLOCKS-LABEL: @main 112 // BLOCKS: call void {{%.+}}(i8* 113 ^{ 114 // BLOCKS: define{{.*}} internal{{.*}} void {{.+}}(i8* 115 // BLOCKS: [[G_LOCAL_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[AGG_CAPTURED:%.+]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 116 // BLOCKS: store i{{[0-9]+}}* [[G]], i{{[0-9]+}}** [[G_LOCAL_REF]] 117 // BLOCKS: [[ARG:%.+]] = bitcast %{{.+}}* [[AGG_CAPTURED]] to i8* 118 // BLOCKS: call void {{.+}}* @__kmpc_fork_call({{.+}}, i32 1, {{.+}}* [[OMP_REGION:@.+]] to {{.+}}, i8* [[ARG]]) 119 #pragma omp parallel reduction(-:g) 120 { 121 // BLOCKS: define{{.*}} internal{{.*}} void [[OMP_REGION]](i32* %{{.+}}, i32* %{{.+}}, %{{.+}}* [[ARG:%.+]]) 122 // BLOCKS: [[G_PRIVATE_ADDR:%.+]] = alloca i{{[0-9]+}}, 123 124 // Reduction list for runtime. 125 // BLOCKS: [[RED_LIST:%.+]] = alloca [1 x i8*], 126 127 // BLOCKS: store %{{.+}}* [[ARG]], %{{.+}}** [[ARG_REF:%.+]], 128 // BLOCKS: [[ARG:%.+]] = load %{{.+}}*, %{{.+}}** [[ARG_REF]] 129 // BLOCKS: [[G_REF_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 130 // BLOCKS: [[G_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[G_REF_ADDR]] 131 // BLOCKS: store i{{[0-9]+}} 0, i{{[0-9]+}}* [[G_PRIVATE_ADDR]] 132 // BLOCKS: call i32 @__kmpc_cancel_barrier( 133 g = 1; 134 // BLOCKS: store volatile i{{[0-9]+}} 1, i{{[0-9]+}}* [[G_PRIVATE_ADDR]], 135 // BLOCKS-NOT: [[G]]{{[[^:word:]]}} 136 // BLOCKS: i{{[0-9]+}}* [[G_PRIVATE_ADDR]] 137 // BLOCKS-NOT: [[G]]{{[[^:word:]]}} 138 // BLOCKS: call void {{%.+}}(i8* 139 140 // BLOCKS: [[G_PRIV_REF:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[RED_LIST]], i32 0, i32 0 141 // BLOCKS: [[BITCAST:%.+]] = bitcast i32* [[G_PRIVATE_ADDR]] to i8* 142 // BLOCKS: store i8* [[BITCAST]], i8** [[G_PRIV_REF]], 143 // BLOCKS: call i32 @__kmpc_reduce_nowait( 144 // BLOCKS: switch i32 %{{.+}}, label %[[REDUCTION_DONE:.+]] [ 145 // BLOCKS: i32 1, label %[[CASE1:.+]] 146 // BLOCKS: i32 2, label %[[CASE2:.+]] 147 // BLOCKS: [[CASE1]] 148 // BLOCKS: [[G_VAL:%.+]] = load i32, i32* [[G_REF]] 149 // BLOCKS: [[G_PRIV_VAL:%.+]] = load i32, i32* [[G_PRIVATE_ADDR]] 150 // BLOCKS: [[ADD:%.+]] = add nsw i32 [[G_VAL]], [[G_PRIV_VAL]] 151 // BLOCKS: store i32 [[ADD]], i32* [[G_REF]] 152 // BLOCKS: call void @__kmpc_end_reduce_nowait( 153 // BLOCKS: br label %[[REDUCTION_DONE]] 154 // BLOCKS: [[CASE2]] 155 // BLOCKS: [[G_PRIV_VAL:%.+]] = load i32, i32* [[G_PRIVATE_ADDR]] 156 // BLOCKS: atomicrmw add i32* [[G_REF]], i32 [[G_PRIV_VAL]] monotonic 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 i{{[0-9]+}} 2, i{{[0-9]+}}* 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 reduction(+:t_var) reduction(&:var) reduction(&& : var1) reduction(min: t_var1) 178 { 179 vec[0] = t_var; 180 s_arr[0] = var; 181 } 182 return tmain<int>(); 183 #endif 184 } 185 186 // CHECK: define {{.*}}i{{[0-9]+}} @main() 187 // CHECK: [[TEST:%.+]] = alloca [[S_FLOAT_TY]], 188 // CHECK: call {{.*}} [[S_FLOAT_TY_CONSTR:@.+]]([[S_FLOAT_TY]]* [[TEST]]) 189 // CHECK: %{{.+}} = bitcast [[CAP_MAIN_TY]]* 190 // 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 191 // CHECK: = call {{.*}}i{{.+}} [[TMAIN_INT:@.+]]() 192 // CHECK: call {{.*}} [[S_FLOAT_TY_DESTR:@.+]]([[S_FLOAT_TY]]* 193 // CHECK: ret 194 // 195 // CHECK: define internal void [[MAIN_MICROTASK]](i{{[0-9]+}}* [[GTID_ADDR:%.+]], i{{[0-9]+}}* %{{.+}}, [[CAP_MAIN_TY]]* %{{.+}}) 196 // CHECK: [[T_VAR_PRIV:%.+]] = alloca float, 197 // CHECK: [[VAR_PRIV:%.+]] = alloca [[S_FLOAT_TY]], 198 // CHECK: [[VAR1_PRIV:%.+]] = alloca [[S_FLOAT_TY]], 199 // CHECK: [[T_VAR1_PRIV:%.+]] = alloca float, 200 201 // Reduction list for runtime. 202 // CHECK: [[RED_LIST:%.+]] = alloca [4 x i8*], 203 204 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]], 205 206 // CHECK: [[T_VAR_PTR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}} 207 // CHECK: [[T_VAR_REF:%.+]] = load float*, float** [[T_VAR_PTR_REF]], 208 // For + reduction operation initial value of private variable is 0. 209 // CHECK: store float 0.0{{.+}}, float* [[T_VAR_PRIV]], 210 211 // CHECK: [[VAR_PTR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}} 212 // CHECK: [[VAR_REF:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[VAR_PTR_REF:%.+]], 213 // For & reduction operation initial value of private variable is ones in all bits. 214 // CHECK: call {{.*}} [[S_FLOAT_TY_CONSTR:@.+]]([[S_FLOAT_TY]]* [[VAR_PRIV]]) 215 216 // CHECK: [[VAR1_PTR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}} 217 // CHECK: [[VAR1_REF:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[VAR_PTR_REF:%.+]], 218 // For && reduction operation initial value of private variable is 1.0. 219 // CHECK: call {{.*}} [[S_FLOAT_TY_CONSTR:@.+]]([[S_FLOAT_TY]]* [[VAR1_PRIV]]) 220 221 // CHECK: [[T_VAR1_PTR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}} 222 // CHECK: [[T_VAR1_REF:%.+]] = load float*, float** [[T_VAR1_PTR_REF]], 223 // For min reduction operation initial value of private variable is largest repesentable value. 224 // CHECK: store float 0x47EFFFFFE0000000, float* [[T_VAR1_PRIV]], 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: call i32 @__kmpc_cancel_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i{{[0-9]+}} [[GTID]]) 229 230 // Skip checks for internal operations. 231 232 // void *RedList[<n>] = {<ReductionVars>[0], ..., <ReductionVars>[<n>-1]}; 233 234 // CHECK: [[T_VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 0 235 // CHECK: [[BITCAST:%.+]] = bitcast float* [[T_VAR_PRIV]] to i8* 236 // CHECK: store i8* [[BITCAST]], i8** [[T_VAR_PRIV_REF]], 237 // CHECK: [[VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 1 238 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_PRIV]] to i8* 239 // CHECK: store i8* [[BITCAST]], i8** [[VAR_PRIV_REF]], 240 // CHECK: [[VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 2 241 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_PRIV]] to i8* 242 // CHECK: store i8* [[BITCAST]], i8** [[VAR1_PRIV_REF]], 243 // CHECK: [[T_VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 3 244 // CHECK: [[BITCAST:%.+]] = bitcast float* [[T_VAR1_PRIV]] to i8* 245 // CHECK: store i8* [[BITCAST]], i8** [[T_VAR1_PRIV_REF]], 246 247 // res = __kmpc_reduce_nowait(<loc>, <gtid>, <n>, sizeof(RedList), RedList, reduce_func, &<lock>); 248 249 // CHECK: [[BITCAST:%.+]] = bitcast [4 x i8*]* [[RED_LIST]] to i8* 250 // 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]]) 251 252 // switch(res) 253 // CHECK: switch i32 [[RES]], label %[[RED_DONE:.+]] [ 254 // CHECK: i32 1, label %[[CASE1:.+]] 255 // CHECK: i32 2, label %[[CASE2:.+]] 256 // CHECK: ] 257 258 // case 1: 259 // t_var += t_var_reduction; 260 // CHECK: [[T_VAR_VAL:%.+]] = load float, float* [[T_VAR_REF]], 261 // CHECK: [[T_VAR_PRIV_VAL:%.+]] = load float, float* [[T_VAR_PRIV]], 262 // CHECK: [[UP:%.+]] = fadd float [[T_VAR_VAL]], [[T_VAR_PRIV_VAL]] 263 // CHECK: store float [[UP]], float* [[T_VAR_REF]], 264 265 // var = var.operator &(var_reduction); 266 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_REF]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_PRIV]]) 267 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_REF]] to i8* 268 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8* 269 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 270 271 // var1 = var1.operator &&(var1_reduction); 272 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_REF]]) 273 // CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 274 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]] 275 // CHECK: [[TRUE]] 276 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_PRIV]]) 277 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 278 // CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]] 279 // CHECK: [[TRUE2]] 280 // CHECK: br label %[[END2:.+]] 281 // CHECK: [[FALSE2]] 282 // CHECK: br label %[[END2]] 283 // CHECK: [[END2]] 284 // CHECK: [[COND_LVALUE:%.+]] = phi [[S_FLOAT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ] 285 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_REF]] to i8* 286 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8* 287 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 288 289 // t_var1 = min(t_var1, t_var1_reduction); 290 // CHECK: [[T_VAR1_VAL:%.+]] = load float, float* [[T_VAR1_REF]], 291 // CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load float, float* [[T_VAR1_PRIV]], 292 // CHECK: [[CMP:%.+]] = fcmp olt float [[T_VAR1_VAL]], [[T_VAR1_PRIV_VAL]] 293 // CHECK: [[UP:%.+]] = uitofp i1 [[CMP]] to float 294 // CHECK: store float [[UP]], float* [[T_VAR1_REF]], 295 296 // __kmpc_end_reduce_nowait(<loc>, <gtid>, &<lock>); 297 // CHECK: call void @__kmpc_end_reduce_nowait(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], [8 x i32]* [[REDUCTION_LOCK]]) 298 299 // break; 300 // CHECK: br label %[[RED_DONE]] 301 302 // case 2: 303 // t_var += t_var_reduction; 304 // CHECK: load float, float* [[T_VAR_PRIV]] 305 // CHECK: [[T_VAR_REF_INT:%.+]] = bitcast float* [[T_VAR_REF]] to i32* 306 // CHECK: load atomic i32, i32* [[T_VAR_REF_INT]] monotonic, 307 // CHECK: [[OLD1:%.+]] = bitcast i32 %{{.+}} to float 308 // CHECK: br label %[[CONT:.+]] 309 // CHECK: [[CONT]] 310 // CHECK: [[ORIG_OLD:%.+]] = phi float [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %[[CONT]] ] 311 // CHECK: [[UP:%.+]] = fadd float 312 // CHECK: [[ORIG_OLD_INT:%.+]] = bitcast float [[ORIG_OLD]] to i32 313 // CHECK: [[UP_INT:%.+]] = bitcast float [[UP]] to i32 314 // CHECK: [[T_VAR_REF_INT:%.+]] = bitcast float* [[T_VAR_REF]] to i32* 315 // CHECK: [[RES:%.+]] = cmpxchg i32* [[T_VAR_REF_INT]], i32 [[ORIG_OLD_INT]], i32 [[UP_INT]] monotonic monotonic 316 // CHECK: [[OLD_VAL_INT:%.+]] = extractvalue { i32, i1 } [[RES]], 0 317 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1 318 // CHECK: [[OLD2]] = bitcast i32 [[OLD_VAL_INT]] to float 319 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[ATOMIC_DONE:.+]], label %[[CONT]] 320 // CHECK: [[ATOMIC_DONE]] 321 322 // var = var.operator &(var_reduction); 323 // CHECK: call void @__kmpc_critical( 324 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_REF]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_PRIV]]) 325 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_REF]] to i8* 326 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8* 327 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 328 // CHECK: call void @__kmpc_end_critical( 329 330 // var1 = var1.operator &&(var1_reduction); 331 // CHECK: call void @__kmpc_critical( 332 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_REF]]) 333 // CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 334 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]] 335 // CHECK: [[TRUE]] 336 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_PRIV]]) 337 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 338 // CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]] 339 // CHECK: [[TRUE2]] 340 // CHECK: br label %[[END2:.+]] 341 // CHECK: [[FALSE2]] 342 // CHECK: br label %[[END2]] 343 // CHECK: [[END2]] 344 // CHECK: [[COND_LVALUE:%.+]] = phi [[S_FLOAT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ] 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: load atomic i32, i32* [[T_VAR1_REF_INT]] monotonic, 354 // CHECK: [[OLD1:%.+]] = bitcast i32 %{{.+}} to float 355 // CHECK: br label %[[CONT:.+]] 356 // CHECK: [[CONT]] 357 // CHECK: [[ORIG_OLD:%.+]] = phi float [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %[[CONT]] ] 358 // CHECK: [[CMP:%.+]] = fcmp olt float 359 // CHECK: [[UP:%.+]] = uitofp i1 [[CMP]] to float 360 // CHECK: [[ORIG_OLD_INT:%.+]] = bitcast float [[ORIG_OLD]] to i32 361 // CHECK: [[UP_INT:%.+]] = bitcast float [[UP]] to i32 362 // CHECK: [[T_VAR1_REF_INT:%.+]] = bitcast float* [[T_VAR1_REF]] to i32* 363 // CHECK: [[RES:%.+]] = cmpxchg i32* [[T_VAR1_REF_INT]], i32 [[ORIG_OLD_INT]], i32 [[UP_INT]] monotonic monotonic 364 // CHECK: [[OLD_VAL_INT:%.+]] = extractvalue { i32, i1 } [[RES]], 0 365 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1 366 // CHECK: [[OLD2]] = bitcast i32 [[OLD_VAL_INT]] to float 367 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[ATOMIC_DONE:.+]], label %[[CONT]] 368 // CHECK: [[ATOMIC_DONE]] 369 370 // break; 371 // CHECK: br label %[[RED_DONE]] 372 // CHECK: [[RED_DONE]] 373 // CHECK: call i32 @__kmpc_cancel_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i{{[0-9]+}} [[GTID]]) 374 375 // 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: ret void 378 379 // void reduce_func(void *lhs[<n>], void *rhs[<n>]) { 380 // *(Type0*)lhs[0] = ReductionOperation0(*(Type0*)lhs[0], *(Type0*)rhs[0]); 381 // ... 382 // *(Type<n>-1*)lhs[<n>-1] = ReductionOperation<n>-1(*(Type<n>-1*)lhs[<n>-1], 383 // *(Type<n>-1*)rhs[<n>-1]); 384 // } 385 // CHECK: define internal void [[REDUCTION_FUNC]](i8*, i8*) 386 // t_var_lhs = (float*)lhs[0]; 387 // CHECK: [[T_VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS:%.+]], i32 0, i32 0 388 // CHECK: [[T_VAR_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR_RHS_REF]], 389 // CHECK: [[T_VAR_RHS:%.+]] = bitcast i8* [[T_VAR_RHS_VOID]] to float* 390 // t_var_rhs = (float*)rhs[0]; 391 // CHECK: [[T_VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS:%.+]], i32 0, i32 0 392 // CHECK: [[T_VAR_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR_LHS_REF]], 393 // CHECK: [[T_VAR_LHS:%.+]] = bitcast i8* [[T_VAR_LHS_VOID]] to float* 394 395 // var_lhs = (S<float>*)lhs[1]; 396 // CHECK: [[VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i32 0, i32 1 397 // CHECK: [[VAR_RHS_VOID:%.+]] = load i8*, i8** [[VAR_RHS_REF]], 398 // CHECK: [[VAR_RHS:%.+]] = bitcast i8* [[VAR_RHS_VOID]] to [[S_FLOAT_TY]]* 399 // var_rhs = (S<float>*)rhs[1]; 400 // CHECK: [[VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i32 0, i32 1 401 // CHECK: [[VAR_LHS_VOID:%.+]] = load i8*, i8** [[VAR_LHS_REF]], 402 // CHECK: [[VAR_LHS:%.+]] = bitcast i8* [[VAR_LHS_VOID]] to [[S_FLOAT_TY]]* 403 404 // var1_lhs = (S<float>*)lhs[2]; 405 // CHECK: [[VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i32 0, i32 2 406 // CHECK: [[VAR1_RHS_VOID:%.+]] = load i8*, i8** [[VAR1_RHS_REF]], 407 // CHECK: [[VAR1_RHS:%.+]] = bitcast i8* [[VAR1_RHS_VOID]] to [[S_FLOAT_TY]]* 408 // var1_rhs = (S<float>*)rhs[2]; 409 // CHECK: [[VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i32 0, i32 2 410 // CHECK: [[VAR1_LHS_VOID:%.+]] = load i8*, i8** [[VAR1_LHS_REF]], 411 // CHECK: [[VAR1_LHS:%.+]] = bitcast i8* [[VAR1_LHS_VOID]] to [[S_FLOAT_TY]]* 412 413 // t_var1_lhs = (float*)lhs[3]; 414 // CHECK: [[T_VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i32 0, i32 3 415 // CHECK: [[T_VAR1_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_RHS_REF]], 416 // CHECK: [[T_VAR1_RHS:%.+]] = bitcast i8* [[T_VAR1_RHS_VOID]] to float* 417 // t_var1_rhs = (float*)rhs[3]; 418 // CHECK: [[T_VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i32 0, i32 3 419 // CHECK: [[T_VAR1_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_LHS_REF]], 420 // CHECK: [[T_VAR1_LHS:%.+]] = bitcast i8* [[T_VAR1_LHS_VOID]] to float* 421 422 // t_var_lhs += t_var_rhs; 423 // CHECK: [[T_VAR_LHS_VAL:%.+]] = load float, float* [[T_VAR_LHS]], 424 // CHECK: [[T_VAR_RHS_VAL:%.+]] = load float, float* [[T_VAR_RHS]], 425 // CHECK: [[UP:%.+]] = fadd float [[T_VAR_LHS_VAL]], [[T_VAR_RHS_VAL]] 426 // CHECK: store float [[UP]], float* [[T_VAR_LHS]], 427 428 // var_lhs = var_lhs.operator &(var_rhs); 429 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_LHS]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_RHS]]) 430 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_LHS]] to i8* 431 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8* 432 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 433 434 // var1_lhs = var1_lhs.operator &&(var1_rhs); 435 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_LHS]]) 436 // CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 437 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]] 438 // CHECK: [[TRUE]] 439 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_RHS]]) 440 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 441 // CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]] 442 // CHECK: [[TRUE2]] 443 // CHECK: br label %[[END2:.+]] 444 // CHECK: [[FALSE2]] 445 // CHECK: br label %[[END2]] 446 // CHECK: [[END2]] 447 // CHECK: [[COND_LVALUE:%.+]] = phi [[S_FLOAT_TY]]* [ [[VAR1_LHS]], %[[TRUE2]] ], [ [[VAR1_RHS]], %[[FALSE2]] ] 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: [[UP:%.+]] = uitofp i1 [[CMP]] to float 457 // CHECK: store float [[UP]], float* [[T_VAR1_LHS]], 458 // CHECK: ret void 459 460 // CHECK: define {{.*}} i{{[0-9]+}} [[TMAIN_INT]]() 461 // CHECK: [[TEST:%.+]] = alloca [[S_INT_TY]], 462 // CHECK: call {{.*}} [[S_INT_TY_CONSTR:@.+]]([[S_INT_TY]]* [[TEST]]) 463 // CHECK: %{{.+}} = bitcast [[CAP_TMAIN_TY]]* 464 // 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 465 // CHECK: call {{.*}} [[S_INT_TY_DESTR:@.+]]([[S_INT_TY]]* 466 // CHECK: ret 467 // 468 // CHECK: define internal void [[TMAIN_MICROTASK]](i{{[0-9]+}}* [[GTID_ADDR:%.+]], i{{[0-9]+}}* %{{.+}}, [[CAP_TMAIN_TY]]* %{{.+}}) 469 // CHECK: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}}, 470 // CHECK: [[VAR_PRIV:%.+]] = alloca [[S_INT_TY]], 471 // CHECK: [[VAR1_PRIV:%.+]] = alloca [[S_INT_TY]], 472 // CHECK: [[T_VAR1_PRIV:%.+]] = alloca i{{[0-9]+}}, 473 474 // Reduction list for runtime. 475 // CHECK: [[RED_LIST:%.+]] = alloca [4 x i8*], 476 477 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]], 478 479 // CHECK: [[T_VAR_PTR_REF:%.+]] = getelementptr inbounds [[CAP_TMAIN_TY]], [[CAP_TMAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}} 480 // CHECK: [[T_VAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[T_VAR_PTR_REF]], 481 // For + reduction operation initial value of private variable is 0. 482 // CHECK: store i{{[0-9]+}} 0, i{{[0-9]+}}* [[T_VAR_PRIV]], 483 484 // CHECK: [[VAR_PTR_REF:%.+]] = getelementptr inbounds [[CAP_TMAIN_TY]], [[CAP_TMAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}} 485 // CHECK: [[VAR_REF:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[VAR_PTR_REF:%.+]], 486 // For & reduction operation initial value of private variable is ones in all bits. 487 // CHECK: call {{.*}} [[S_INT_TY_CONSTR:@.+]]([[S_INT_TY]]* [[VAR_PRIV]]) 488 489 // CHECK: [[VAR1_PTR_REF:%.+]] = getelementptr inbounds [[CAP_TMAIN_TY]], [[CAP_TMAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}} 490 // CHECK: [[VAR1_REF:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[VAR_PTR_REF:%.+]], 491 // For && reduction operation initial value of private variable is 1.0. 492 // CHECK: call {{.*}} [[S_INT_TY_CONSTR:@.+]]([[S_INT_TY]]* [[VAR1_PRIV]]) 493 494 // CHECK: [[T_VAR1_PTR_REF:%.+]] = getelementptr inbounds [[CAP_TMAIN_TY]], [[CAP_TMAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}} 495 // CHECK: [[T_VAR1_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[T_VAR1_PTR_REF]], 496 // For min reduction operation initial value of private variable is largest repesentable value. 497 // CHECK: store i{{[0-9]+}} 2147483647, i{{[0-9]+}}* [[T_VAR1_PRIV]], 498 499 // CHECK: [[GTID_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_ADDR]] 500 // CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_REF]] 501 // CHECK: call i32 @__kmpc_cancel_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i{{[0-9]+}} [[GTID]]) 502 503 // Skip checks for internal operations. 504 505 // void *RedList[<n>] = {<ReductionVars>[0], ..., <ReductionVars>[<n>-1]}; 506 507 // CHECK: [[T_VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 0 508 // CHECK: [[BITCAST:%.+]] = bitcast i{{[0-9]+}}* [[T_VAR_PRIV]] to i8* 509 // CHECK: store i8* [[BITCAST]], i8** [[T_VAR_PRIV_REF]], 510 // CHECK: [[VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 1 511 // CHECK: [[BITCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR_PRIV]] to i8* 512 // CHECK: store i8* [[BITCAST]], i8** [[VAR_PRIV_REF]], 513 // CHECK: [[VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 2 514 // CHECK: [[BITCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_PRIV]] to i8* 515 // CHECK: store i8* [[BITCAST]], i8** [[VAR1_PRIV_REF]], 516 // CHECK: [[T_VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 3 517 // CHECK: [[BITCAST:%.+]] = bitcast i{{[0-9]+}}* [[T_VAR1_PRIV]] to i8* 518 // CHECK: store i8* [[BITCAST]], i8** [[T_VAR1_PRIV_REF]], 519 520 // res = __kmpc_reduce_nowait(<loc>, <gtid>, <n>, sizeof(RedList), RedList, reduce_func, &<lock>); 521 522 // CHECK: [[BITCAST:%.+]] = bitcast [4 x i8*]* [[RED_LIST]] to i8* 523 // 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]]) 524 525 // switch(res) 526 // CHECK: switch i32 [[RES]], label %[[RED_DONE:.+]] [ 527 // CHECK: i32 1, label %[[CASE1:.+]] 528 // CHECK: i32 2, label %[[CASE2:.+]] 529 // CHECK: ] 530 531 // case 1: 532 // t_var += t_var_reduction; 533 // CHECK: [[T_VAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_REF]], 534 // CHECK: [[T_VAR_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_PRIV]], 535 // CHECK: [[UP:%.+]] = add nsw i{{[0-9]+}} [[T_VAR_VAL]], [[T_VAR_PRIV_VAL]] 536 // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR_REF]], 537 538 // var = var.operator &(var_reduction); 539 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_REF]], [[S_INT_TY]]* dereferenceable(4) [[VAR_PRIV]]) 540 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_REF]] to i8* 541 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* 542 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 543 544 // var1 = var1.operator &&(var1_reduction); 545 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_REF]]) 546 // CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 547 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]] 548 // CHECK: [[TRUE]] 549 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_PRIV]]) 550 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 551 // CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]] 552 // CHECK: [[TRUE2]] 553 // CHECK: br label %[[END2:.+]] 554 // CHECK: [[FALSE2]] 555 // CHECK: br label %[[END2]] 556 // CHECK: [[END2]] 557 // CHECK: [[COND_LVALUE:%.+]] = phi [[S_INT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ] 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: [[UP:%.+]] = zext i1 [[CMP]] to i{{[0-9]+}} 567 // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR1_REF]], 568 569 // __kmpc_end_reduce_nowait(<loc>, <gtid>, &<lock>); 570 // CHECK: call void @__kmpc_end_reduce_nowait(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], [8 x i32]* [[REDUCTION_LOCK]]) 571 572 // break; 573 // CHECK: br label %[[RED_DONE]] 574 575 // case 2: 576 // t_var += t_var_reduction; 577 // CHECK: [[T_VAR_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_PRIV]] 578 // CHECK: atomicrmw add i32* [[T_VAR_REF]], i32 [[T_VAR_PRIV_VAL]] monotonic 579 580 // var = var.operator &(var_reduction); 581 // CHECK: call void @__kmpc_critical( 582 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_REF]], [[S_INT_TY]]* dereferenceable(4) [[VAR_PRIV]]) 583 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_REF]] to i8* 584 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* 585 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 586 // CHECK: call void @__kmpc_end_critical( 587 588 // var1 = var1.operator &&(var1_reduction); 589 // CHECK: call void @__kmpc_critical( 590 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_REF]]) 591 // CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 592 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]] 593 // CHECK: [[TRUE]] 594 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_PRIV]]) 595 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 596 // CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]] 597 // CHECK: [[TRUE2]] 598 // CHECK: br label %[[END2:.+]] 599 // CHECK: [[FALSE2]] 600 // CHECK: br label %[[END2]] 601 // CHECK: [[END2]] 602 // CHECK: [[COND_LVALUE:%.+]] = phi [[S_INT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ] 603 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8* 604 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* 605 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 606 // CHECK: call void @__kmpc_end_critical( 607 608 // t_var1 = min(t_var1, t_var1_reduction); 609 // CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_PRIV]] 610 // CHECK: atomicrmw min i32* [[T_VAR1_REF]], i32 [[T_VAR1_PRIV_VAL]] monotonic 611 612 // break; 613 // CHECK: br label %[[RED_DONE]] 614 // CHECK: [[RED_DONE]] 615 // CHECK: call i32 @__kmpc_cancel_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i{{[0-9]+}} [[GTID]]) 616 617 // CHECK-DAG: call {{.*}} [[S_INT_TY_DESTR]]([[S_INT_TY]]* [[VAR_PRIV]]) 618 // CHECK-DAG: call {{.*}} [[S_INT_TY_DESTR]]([[S_INT_TY]]* 619 // CHECK: ret void 620 621 // void reduce_func(void *lhs[<n>], void *rhs[<n>]) { 622 // *(Type0*)lhs[0] = ReductionOperation0(*(Type0*)lhs[0], *(Type0*)rhs[0]); 623 // ... 624 // *(Type<n>-1*)lhs[<n>-1] = ReductionOperation<n>-1(*(Type<n>-1*)lhs[<n>-1], 625 // *(Type<n>-1*)rhs[<n>-1]); 626 // } 627 // CHECK: define internal void [[REDUCTION_FUNC]](i8*, i8*) 628 // t_var_lhs = (i{{[0-9]+}}*)lhs[0]; 629 // CHECK: [[T_VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS:%.+]], i32 0, i32 0 630 // CHECK: [[T_VAR_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR_RHS_REF]], 631 // CHECK: [[T_VAR_RHS:%.+]] = bitcast i8* [[T_VAR_RHS_VOID]] to i{{[0-9]+}}* 632 // t_var_rhs = (i{{[0-9]+}}*)rhs[0]; 633 // CHECK: [[T_VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS:%.+]], i32 0, i32 0 634 // CHECK: [[T_VAR_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR_LHS_REF]], 635 // CHECK: [[T_VAR_LHS:%.+]] = bitcast i8* [[T_VAR_LHS_VOID]] to i{{[0-9]+}}* 636 637 // var_lhs = (S<i{{[0-9]+}}>*)lhs[1]; 638 // CHECK: [[VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i32 0, i32 1 639 // CHECK: [[VAR_RHS_VOID:%.+]] = load i8*, i8** [[VAR_RHS_REF]], 640 // CHECK: [[VAR_RHS:%.+]] = bitcast i8* [[VAR_RHS_VOID]] to [[S_INT_TY]]* 641 // var_rhs = (S<i{{[0-9]+}}>*)rhs[1]; 642 // CHECK: [[VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i32 0, i32 1 643 // CHECK: [[VAR_LHS_VOID:%.+]] = load i8*, i8** [[VAR_LHS_REF]], 644 // CHECK: [[VAR_LHS:%.+]] = bitcast i8* [[VAR_LHS_VOID]] to [[S_INT_TY]]* 645 646 // var1_lhs = (S<i{{[0-9]+}}>*)lhs[2]; 647 // CHECK: [[VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i32 0, i32 2 648 // CHECK: [[VAR1_RHS_VOID:%.+]] = load i8*, i8** [[VAR1_RHS_REF]], 649 // CHECK: [[VAR1_RHS:%.+]] = bitcast i8* [[VAR1_RHS_VOID]] to [[S_INT_TY]]* 650 // var1_rhs = (S<i{{[0-9]+}}>*)rhs[2]; 651 // CHECK: [[VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i32 0, i32 2 652 // CHECK: [[VAR1_LHS_VOID:%.+]] = load i8*, i8** [[VAR1_LHS_REF]], 653 // CHECK: [[VAR1_LHS:%.+]] = bitcast i8* [[VAR1_LHS_VOID]] to [[S_INT_TY]]* 654 655 // t_var1_lhs = (i{{[0-9]+}}*)lhs[3]; 656 // CHECK: [[T_VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i32 0, i32 3 657 // CHECK: [[T_VAR1_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_RHS_REF]], 658 // CHECK: [[T_VAR1_RHS:%.+]] = bitcast i8* [[T_VAR1_RHS_VOID]] to i{{[0-9]+}}* 659 // t_var1_rhs = (i{{[0-9]+}}*)rhs[3]; 660 // CHECK: [[T_VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i32 0, i32 3 661 // CHECK: [[T_VAR1_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_LHS_REF]], 662 // CHECK: [[T_VAR1_LHS:%.+]] = bitcast i8* [[T_VAR1_LHS_VOID]] to i{{[0-9]+}}* 663 664 // t_var_lhs += t_var_rhs; 665 // CHECK: [[T_VAR_LHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_LHS]], 666 // CHECK: [[T_VAR_RHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_RHS]], 667 // CHECK: [[UP:%.+]] = add nsw i{{[0-9]+}} [[T_VAR_LHS_VAL]], [[T_VAR_RHS_VAL]] 668 // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR_LHS]], 669 670 // var_lhs = var_lhs.operator &(var_rhs); 671 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_LHS]], [[S_INT_TY]]* dereferenceable(4) [[VAR_RHS]]) 672 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_LHS]] to i8* 673 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* 674 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 675 676 // var1_lhs = var1_lhs.operator &&(var1_rhs); 677 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_LHS]]) 678 // CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 679 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]] 680 // CHECK: [[TRUE]] 681 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_RHS]]) 682 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 683 // CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]] 684 // CHECK: [[TRUE2]] 685 // CHECK: br label %[[END2:.+]] 686 // CHECK: [[FALSE2]] 687 // CHECK: br label %[[END2]] 688 // CHECK: [[END2]] 689 // CHECK: [[COND_LVALUE:%.+]] = phi [[S_INT_TY]]* [ [[VAR1_LHS]], %[[TRUE2]] ], [ [[VAR1_RHS]], %[[FALSE2]] ] 690 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_LHS]] to i8* 691 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* 692 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 693 694 // t_var1_lhs = min(t_var1_lhs, t_var1_rhs); 695 // CHECK: [[T_VAR1_LHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_LHS]], 696 // CHECK: [[T_VAR1_RHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_RHS]], 697 // CHECK: [[CMP:%.+]] = icmp slt i{{[0-9]+}} [[T_VAR1_LHS_VAL]], [[T_VAR1_RHS_VAL]] 698 // CHECK: [[UP:%.+]] = zext i1 [[CMP]] to i{{[0-9]+}} 699 // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR1_LHS]], 700 // CHECK: ret void 701 702 #endif 703 704