1 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck %s 2 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s 3 // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s 4 // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s 5 // RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s 6 // expected-no-diagnostics 7 #ifndef HEADER 8 #define HEADER 9 10 volatile double g, g_orig; 11 volatile double &g1 = g_orig; 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: [[ATOMIC_REDUCE_BARRIER_LOC:@.+]] = private unnamed_addr constant %{{.+}} { i32 0, i32 18, i32 0, i32 0, i8* 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 = test; 38 S<T> 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 extern S<float> **foo(); 55 56 int main() { 57 #ifdef LAMBDA 58 // LAMBDA: [[G:@.+]] = global double 59 // LAMBDA-LABEL: @main 60 // LAMBDA: call void [[OUTER_LAMBDA:@.+]]( 61 [&]() { 62 // LAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]]( 63 // LAMBDA: call void {{.+}} @__kmpc_fork_call({{.+}}, i32 0, {{.+}}* [[OMP_REGION:@.+]] to {{.+}}) 64 #pragma omp parallel 65 #pragma omp for reduction(+:g, g1) 66 for (int i = 0; i < 2; ++i) { 67 // LAMBDA: define{{.*}} internal{{.*}} void [[OMP_REGION]](i32* noalias %{{.+}}, i32* noalias %{{.+}}) 68 // LAMBDA: [[G_PRIVATE_ADDR:%.+]] = alloca double, 69 70 // Reduction list for runtime. 71 // LAMBDA: [[RED_LIST:%.+]] = alloca [2 x i8*], 72 73 // LAMBDA: store double 0.0{{.+}}, double* [[G_PRIVATE_ADDR]] 74 // LAMBDA: call void @__kmpc_for_static_init_4( 75 g = 1; 76 g1 = 1; 77 // LAMBDA: store double 1.0{{.+}}, double* [[G_PRIVATE_ADDR]], 78 // LAMBDA: [[G_PRIVATE_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG:%.+]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 79 // LAMBDA: store double* [[G_PRIVATE_ADDR]], double** [[G_PRIVATE_ADDR_REF]] 80 // LAMBDA: call void [[INNER_LAMBDA:@.+]](%{{.+}}* [[ARG]]) 81 // LAMBDA: call void @__kmpc_for_static_fini( 82 83 // LAMBDA: [[G_PRIV_REF:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[RED_LIST]], i64 0, i64 0 84 // LAMBDA: [[BITCAST:%.+]] = bitcast double* [[G_PRIVATE_ADDR]] to i8* 85 // LAMBDA: store i8* [[BITCAST]], i8** [[G_PRIV_REF]], 86 // LAMBDA: call i32 @__kmpc_reduce( 87 // LAMBDA: switch i32 %{{.+}}, label %[[REDUCTION_DONE:.+]] [ 88 // LAMBDA: i32 1, label %[[CASE1:.+]] 89 // LAMBDA: i32 2, label %[[CASE2:.+]] 90 // LAMBDA: [[CASE1]] 91 // LAMBDA: [[G_VAL:%.+]] = load double, double* [[G]] 92 // LAMBDA: [[G_PRIV_VAL:%.+]] = load double, double* [[G_PRIVATE_ADDR]] 93 // LAMBDA: [[ADD:%.+]] = fadd double [[G_VAL]], [[G_PRIV_VAL]] 94 // LAMBDA: store double [[ADD]], double* [[G]] 95 // LAMBDA: call void @__kmpc_end_reduce( 96 // LAMBDA: br label %[[REDUCTION_DONE]] 97 // LAMBDA: [[CASE2]] 98 // LAMBDA: [[G_PRIV_VAL:%.+]] = load double, double* [[G_PRIVATE_ADDR]] 99 // LAMBDA: fadd double 100 // LAMBDA: cmpxchg i64* 101 // LAMBDA: call void @__kmpc_end_reduce( 102 // LAMBDA: br label %[[REDUCTION_DONE]] 103 // LAMBDA: [[REDUCTION_DONE]] 104 // LAMBDA: ret void 105 [&]() { 106 // LAMBDA: define {{.+}} void [[INNER_LAMBDA]](%{{.+}}* [[ARG_PTR:%.+]]) 107 // LAMBDA: store %{{.+}}* [[ARG_PTR]], %{{.+}}** [[ARG_PTR_REF:%.+]], 108 g = 2; 109 g1 = 2; 110 // LAMBDA: [[ARG_PTR:%.+]] = load %{{.+}}*, %{{.+}}** [[ARG_PTR_REF]] 111 // LAMBDA: [[G_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 112 // LAMBDA: [[G_REF:%.+]] = load double*, double** [[G_PTR_REF]] 113 // LAMBDA: store double 2.0{{.+}}, double* [[G_REF]] 114 }(); 115 } 116 }(); 117 return 0; 118 #elif defined(BLOCKS) 119 // BLOCKS: [[G:@.+]] = global double 120 // BLOCKS-LABEL: @main 121 // BLOCKS: call void {{%.+}}(i8 122 ^{ 123 // BLOCKS: define{{.*}} internal{{.*}} void {{.+}}(i8* 124 // BLOCKS: call void {{.+}} @__kmpc_fork_call({{.+}}, i32 0, {{.+}}* [[OMP_REGION:@.+]] to {{.+}}) 125 #pragma omp parallel 126 #pragma omp for reduction(-:g, g1) 127 for (int i = 0; i < 2; ++i) { 128 // BLOCKS: define{{.*}} internal{{.*}} void [[OMP_REGION]](i32* noalias %{{.+}}, i32* noalias %{{.+}}) 129 // BLOCKS: [[G_PRIVATE_ADDR:%.+]] = alloca double, 130 131 // Reduction list for runtime. 132 // BLOCKS: [[RED_LIST:%.+]] = alloca [2 x i8*], 133 134 // BLOCKS: store double 0.0{{.+}}, double* [[G_PRIVATE_ADDR]] 135 g = 1; 136 g1 = 1; 137 // BLOCKS: call void @__kmpc_for_static_init_4( 138 // BLOCKS: store double 1.0{{.+}}, double* [[G_PRIVATE_ADDR]], 139 // BLOCKS-NOT: [[G]]{{[[^:word:]]}} 140 // BLOCKS: double* [[G_PRIVATE_ADDR]] 141 // BLOCKS-NOT: [[G]]{{[[^:word:]]}} 142 // BLOCKS: call void {{%.+}}(i8 143 // BLOCKS: call void @__kmpc_for_static_fini( 144 145 // BLOCKS: [[G_PRIV_REF:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[RED_LIST]], i64 0, i64 0 146 // BLOCKS: [[BITCAST:%.+]] = bitcast double* [[G_PRIVATE_ADDR]] to i8* 147 // BLOCKS: store i8* [[BITCAST]], i8** [[G_PRIV_REF]], 148 // BLOCKS: call i32 @__kmpc_reduce( 149 // BLOCKS: switch i32 %{{.+}}, label %[[REDUCTION_DONE:.+]] [ 150 // BLOCKS: i32 1, label %[[CASE1:.+]] 151 // BLOCKS: i32 2, label %[[CASE2:.+]] 152 // BLOCKS: [[CASE1]] 153 // BLOCKS: [[G_VAL:%.+]] = load double, double* [[G]] 154 // BLOCKS: [[G_PRIV_VAL:%.+]] = load double, double* [[G_PRIVATE_ADDR]] 155 // BLOCKS: [[ADD:%.+]] = fadd double [[G_VAL]], [[G_PRIV_VAL]] 156 // BLOCKS: store double [[ADD]], double* [[G]] 157 // BLOCKS: call void @__kmpc_end_reduce( 158 // BLOCKS: br label %[[REDUCTION_DONE]] 159 // BLOCKS: [[CASE2]] 160 // BLOCKS: [[G_PRIV_VAL:%.+]] = load double, double* [[G_PRIVATE_ADDR]] 161 // BLOCKS: fadd double 162 // BLOCKS: cmpxchg i64* 163 // BLOCKS: call void @__kmpc_end_reduce( 164 // BLOCKS: br label %[[REDUCTION_DONE]] 165 // BLOCKS: [[REDUCTION_DONE]] 166 // BLOCKS: ret void 167 ^{ 168 // BLOCKS: define {{.+}} void {{@.+}}(i8* 169 g = 2; 170 g1 = 2; 171 // BLOCKS-NOT: [[G]]{{[[^:word:]]}} 172 // BLOCKS: store double 2.0{{.+}}, double* 173 // BLOCKS-NOT: [[G]]{{[[^:word:]]}} 174 // BLOCKS: ret 175 }(); 176 } 177 }(); 178 return 0; 179 #else 180 S<float> test; 181 float t_var = 0, t_var1; 182 int vec[] = {1, 2}; 183 S<float> s_arr[] = {1, 2}; 184 S<float> &var = test; 185 S<float> var1, arrs[10][4]; 186 S<float> **var2 = foo(); 187 S<float> vvar2[2]; 188 S<float> (&var3)[2] = s_arr; 189 #pragma omp parallel 190 #pragma omp for reduction(+:t_var) reduction(&:var) reduction(&& : var1) reduction(min: t_var1) 191 for (int i = 0; i < 2; ++i) { 192 vec[i] = t_var; 193 s_arr[i] = var; 194 } 195 int arr[10][vec[1]]; 196 #pragma omp parallel for reduction(+:arr[1][:vec[1]]) reduction(&:arrs[1:vec[1]][1:2]) 197 for (int i = 0; i < 10; ++i) 198 ++arr[1][i]; 199 #pragma omp parallel 200 #pragma omp for reduction(+:arr) reduction(&:arrs) 201 for (int i = 0; i < 10; ++i) 202 ++arr[1][i]; 203 #pragma omp parallel 204 #pragma omp for reduction(& : var2[0 : 5][1 : 6]) 205 for (int i = 0; i < 10; ++i) 206 ; 207 #pragma omp parallel 208 #pragma omp for reduction(& : vvar2[0 : 5]) 209 for (int i = 0; i < 10; ++i) 210 ; 211 #pragma omp parallel 212 #pragma omp for reduction(& : var3[1 : 2]) 213 for (int i = 0; i < 10; ++i) 214 ; 215 #pragma omp parallel 216 #pragma omp for reduction(& : var3) 217 for (int i = 0; i < 10; ++i) 218 ; 219 return tmain<int>(); 220 #endif 221 } 222 223 // CHECK: define {{.*}}i{{[0-9]+}} @main() 224 // CHECK: [[TEST:%.+]] = alloca [[S_FLOAT_TY]], 225 // CHECK: call {{.*}} [[S_FLOAT_TY_CONSTR:@.+]]([[S_FLOAT_TY]]* [[TEST]]) 226 // CHECK: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{[0-9]+}} 6, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, float*, [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]*, float*, [2 x i32]*, [2 x [[S_FLOAT_TY]]]*)* [[MAIN_MICROTASK:@.+]] to void 227 // CHECK: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{[0-9]+}} 5, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, i64, i64, i32*, [2 x i32]*, [10 x [4 x [[S_FLOAT_TY]]]]*)* [[MAIN_MICROTASK1:@.+]] to void 228 // CHECK: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{[0-9]+}} 4, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, i64, i64, i32*, [10 x [4 x [[S_FLOAT_TY]]]]*)* [[MAIN_MICROTASK2:@.+]] to void 229 // 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]+}}*, [[S_FLOAT_TY]]***)* [[MAIN_MICROTASK3:@.+]] to void 230 // 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]+}}*, [2 x [[S_FLOAT_TY]]]*)* [[MAIN_MICROTASK4:@.+]] to void 231 // 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]+}}*, [2 x [[S_FLOAT_TY]]]*)* [[MAIN_MICROTASK5:@.+]] to void 232 // 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]+}}*, [2 x [[S_FLOAT_TY]]]*)* [[MAIN_MICROTASK6:@.+]] to void 233 // CHECK: = call {{.*}}i{{.+}} [[TMAIN_INT:@.+]]() 234 // CHECK: call {{.*}} [[S_FLOAT_TY_DESTR:@.+]]([[S_FLOAT_TY]]* 235 // CHECK: ret 236 // 237 // CHECK: define internal void [[MAIN_MICROTASK]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, float* dereferenceable(4) %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}}, float* dereferenceable(4) %{{.+}}, [2 x i32]* dereferenceable(8) %vec, [2 x [[S_FLOAT_TY]]]* dereferenceable(8) %{{.+}}) 238 // CHECK: [[T_VAR_PRIV:%.+]] = alloca float, 239 // CHECK: [[VAR_PRIV:%.+]] = alloca [[S_FLOAT_TY]], 240 // CHECK: [[VAR1_PRIV:%.+]] = alloca [[S_FLOAT_TY]], 241 // CHECK: [[T_VAR1_PRIV:%.+]] = alloca float, 242 243 // Reduction list for runtime. 244 // CHECK: [[RED_LIST:%.+]] = alloca [4 x i8*], 245 246 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]], 247 248 // CHECK: [[T_VAR_REF:%.+]] = load float*, float** % 249 // CHECK: [[VAR1_REF:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** % 250 // CHECK: [[T_VAR1_REF:%.+]] = load float*, float** % 251 252 // For + reduction operation initial value of private variable is 0. 253 // CHECK: store float 0.0{{.+}}, float* [[T_VAR_PRIV]], 254 255 // For & reduction operation initial value of private variable is ones in all bits. 256 // CHECK: [[VAR_REF:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** % 257 // CHECK: call {{.*}} [[S_FLOAT_TY_CONSTR:@.+]]([[S_FLOAT_TY]]* [[VAR_PRIV]]) 258 259 // For && reduction operation initial value of private variable is 1.0. 260 // CHECK: call {{.*}} [[S_FLOAT_TY_CONSTR:@.+]]([[S_FLOAT_TY]]* [[VAR1_PRIV]]) 261 262 // For min reduction operation initial value of private variable is largest repesentable value. 263 // CHECK: store float 0x47EFFFFFE0000000, float* [[T_VAR1_PRIV]], 264 265 266 // CHECK: [[GTID_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_ADDR]] 267 // CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_REF]] 268 // CHECK: call void @__kmpc_for_static_init_4( 269 // Skip checks for internal operations. 270 // CHECK: call void @__kmpc_for_static_fini( 271 272 // void *RedList[<n>] = {<ReductionVars>[0], ..., <ReductionVars>[<n>-1]}; 273 274 // CHECK: [[T_VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 0 275 // CHECK: [[BITCAST:%.+]] = bitcast float* [[T_VAR_PRIV]] to i8* 276 // CHECK: store i8* [[BITCAST]], i8** [[T_VAR_PRIV_REF]], 277 // CHECK: [[VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 1 278 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_PRIV]] to i8* 279 // CHECK: store i8* [[BITCAST]], i8** [[VAR_PRIV_REF]], 280 // CHECK: [[VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 2 281 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_PRIV]] to i8* 282 // CHECK: store i8* [[BITCAST]], i8** [[VAR1_PRIV_REF]], 283 // CHECK: [[T_VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 3 284 // CHECK: [[BITCAST:%.+]] = bitcast float* [[T_VAR1_PRIV]] to i8* 285 // CHECK: store i8* [[BITCAST]], i8** [[T_VAR1_PRIV_REF]], 286 287 // res = __kmpc_reduce(<loc>, <gtid>, <n>, sizeof(RedList), RedList, reduce_func, &<lock>); 288 289 // CHECK: [[BITCAST:%.+]] = bitcast [4 x i8*]* [[RED_LIST]] to i8* 290 // 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]]) 291 292 // switch(res) 293 // CHECK: switch i32 [[RES]], label %[[RED_DONE:.+]] [ 294 // CHECK: i32 1, label %[[CASE1:.+]] 295 // CHECK: i32 2, label %[[CASE2:.+]] 296 // CHECK: ] 297 298 // case 1: 299 // t_var += t_var_reduction; 300 // CHECK: [[T_VAR_VAL:%.+]] = load float, float* [[T_VAR_REF]], 301 // CHECK: [[T_VAR_PRIV_VAL:%.+]] = load float, float* [[T_VAR_PRIV]], 302 // CHECK: [[UP:%.+]] = fadd float [[T_VAR_VAL]], [[T_VAR_PRIV_VAL]] 303 // CHECK: store float [[UP]], float* [[T_VAR_REF]], 304 305 // var = var.operator &(var_reduction); 306 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_REF]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_PRIV]]) 307 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_REF]] to i8* 308 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8* 309 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 310 311 // var1 = var1.operator &&(var1_reduction); 312 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_REF]]) 313 // CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 314 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]] 315 // CHECK: [[TRUE]] 316 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_PRIV]]) 317 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 318 // CHECK: br label %[[END2]] 319 // CHECK: [[END2]] 320 // CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ] 321 // CHECK: [[CONV:%.+]] = uitofp i1 [[COND_LVALUE]] to float 322 // CHECK: call void @{{.+}}([[S_FLOAT_TY]]* [[COND_LVALUE:%.+]], float [[CONV]]) 323 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_REF]] to i8* 324 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8* 325 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 326 327 // t_var1 = min(t_var1, t_var1_reduction); 328 // CHECK: [[T_VAR1_VAL:%.+]] = load float, float* [[T_VAR1_REF]], 329 // CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load float, float* [[T_VAR1_PRIV]], 330 // CHECK: [[CMP:%.+]] = fcmp olt float [[T_VAR1_VAL]], [[T_VAR1_PRIV_VAL]] 331 // CHECK: br i1 [[CMP]] 332 // CHECK: [[UP:%.+]] = phi float 333 // CHECK: store float [[UP]], float* [[T_VAR1_REF]], 334 335 // __kmpc_end_reduce(<loc>, <gtid>, &<lock>); 336 // CHECK: call void @__kmpc_end_reduce(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], [8 x i32]* [[REDUCTION_LOCK]]) 337 338 // break; 339 // CHECK: br label %[[RED_DONE]] 340 341 // case 2: 342 // t_var += t_var_reduction; 343 // CHECK: load float, float* [[T_VAR_PRIV]] 344 // CHECK: [[T_VAR_REF_INT:%.+]] = bitcast float* [[T_VAR_REF]] to i32* 345 // CHECK: [[OLD1:%.+]] = load atomic i32, i32* [[T_VAR_REF_INT]] monotonic, 346 // CHECK: br label %[[CONT:.+]] 347 // CHECK: [[CONT]] 348 // CHECK: [[ORIG_OLD_INT:%.+]] = phi i32 [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %[[CONT]] ] 349 // CHECK: fadd float 350 // CHECK: [[UP_INT:%.+]] = load i32, i32* 351 // CHECK: [[T_VAR_REF_INT:%.+]] = bitcast float* [[T_VAR_REF]] to i32* 352 // CHECK: [[RES:%.+]] = cmpxchg i32* [[T_VAR_REF_INT]], i32 [[ORIG_OLD_INT]], i32 [[UP_INT]] monotonic monotonic 353 // CHECK: [[OLD2:%.+]] = extractvalue { i32, i1 } [[RES]], 0 354 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1 355 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[ATOMIC_DONE:.+]], label %[[CONT]] 356 // CHECK: [[ATOMIC_DONE]] 357 358 // var = var.operator &(var_reduction); 359 // CHECK: call void @__kmpc_critical( 360 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_REF]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_PRIV]]) 361 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_REF]] to i8* 362 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8* 363 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 364 // CHECK: call void @__kmpc_end_critical( 365 366 // var1 = var1.operator &&(var1_reduction); 367 // CHECK: call void @__kmpc_critical( 368 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_REF]]) 369 // CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 370 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]] 371 // CHECK: [[TRUE]] 372 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_PRIV]]) 373 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 374 // CHECK: br label %[[END2]] 375 // CHECK: [[END2]] 376 // CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ] 377 // CHECK: [[CONV:%.+]] = uitofp i1 [[COND_LVALUE]] to float 378 // CHECK: call void @{{.+}}([[S_FLOAT_TY]]* [[COND_LVALUE:%.+]], float [[CONV]]) 379 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_REF]] to i8* 380 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8* 381 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 382 // CHECK: call void @__kmpc_end_critical( 383 384 // t_var1 = min(t_var1, t_var1_reduction); 385 // CHECK: load float, float* [[T_VAR1_PRIV]] 386 // CHECK: [[T_VAR1_REF_INT:%.+]] = bitcast float* [[T_VAR1_REF]] to i32* 387 // CHECK: [[OLD1:%.+]] = load atomic i32, i32* [[T_VAR1_REF_INT]] monotonic, 388 // CHECK: br label %[[CONT:.+]] 389 // CHECK: [[CONT]] 390 // CHECK: [[ORIG_OLD_INT:%.+]] = phi i32 [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %{{.+}} ] 391 // CHECK: [[CMP:%.+]] = fcmp olt float 392 // CHECK: br i1 [[CMP]] 393 // CHECK: phi float 394 // CHECK: [[UP_INT:%.+]] = load i32 395 // CHECK: [[T_VAR1_REF_INT:%.+]] = bitcast float* [[T_VAR1_REF]] to i32* 396 // CHECK: [[RES:%.+]] = cmpxchg i32* [[T_VAR1_REF_INT]], i32 [[ORIG_OLD_INT]], i32 [[UP_INT]] monotonic monotonic 397 // CHECK: [[OLD2:%.+]] = extractvalue { i32, i1 } [[RES]], 0 398 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1 399 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[ATOMIC_DONE:.+]], label %[[CONT]] 400 // CHECK: [[ATOMIC_DONE]] 401 402 // __kmpc_end_reduce(<loc>, <gtid>, &<lock>); 403 // CHECK: call void @__kmpc_end_reduce(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], [8 x i32]* [[REDUCTION_LOCK]]) 404 405 // break; 406 // CHECK: br label %[[RED_DONE]] 407 // CHECK: [[RED_DONE]] 408 // CHECK-DAG: call {{.*}} [[S_FLOAT_TY_DESTR]]([[S_FLOAT_TY]]* [[VAR_PRIV]]) 409 // CHECK-DAG: call {{.*}} [[S_FLOAT_TY_DESTR]]([[S_FLOAT_TY]]* 410 // CHECK: call void @__kmpc_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i{{[0-9]+}} [[GTID]]) 411 412 // CHECK: ret void 413 414 // void reduce_func(void *lhs[<n>], void *rhs[<n>]) { 415 // *(Type0*)lhs[0] = ReductionOperation0(*(Type0*)lhs[0], *(Type0*)rhs[0]); 416 // ... 417 // *(Type<n>-1*)lhs[<n>-1] = ReductionOperation<n>-1(*(Type<n>-1*)lhs[<n>-1], 418 // *(Type<n>-1*)rhs[<n>-1]); 419 // } 420 // CHECK: define internal void [[REDUCTION_FUNC]](i8*, i8*) 421 // t_var_lhs = (float*)lhs[0]; 422 // CHECK: [[T_VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS:%.+]], i64 0, i64 0 423 // CHECK: [[T_VAR_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR_RHS_REF]], 424 // CHECK: [[T_VAR_RHS:%.+]] = bitcast i8* [[T_VAR_RHS_VOID]] to float* 425 // t_var_rhs = (float*)rhs[0]; 426 // CHECK: [[T_VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS:%.+]], i64 0, i64 0 427 // CHECK: [[T_VAR_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR_LHS_REF]], 428 // CHECK: [[T_VAR_LHS:%.+]] = bitcast i8* [[T_VAR_LHS_VOID]] to float* 429 430 // var_lhs = (S<float>*)lhs[1]; 431 // CHECK: [[VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 1 432 // CHECK: [[VAR_RHS_VOID:%.+]] = load i8*, i8** [[VAR_RHS_REF]], 433 // CHECK: [[VAR_RHS:%.+]] = bitcast i8* [[VAR_RHS_VOID]] to [[S_FLOAT_TY]]* 434 // var_rhs = (S<float>*)rhs[1]; 435 // CHECK: [[VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 1 436 // CHECK: [[VAR_LHS_VOID:%.+]] = load i8*, i8** [[VAR_LHS_REF]], 437 // CHECK: [[VAR_LHS:%.+]] = bitcast i8* [[VAR_LHS_VOID]] to [[S_FLOAT_TY]]* 438 439 // var1_lhs = (S<float>*)lhs[2]; 440 // CHECK: [[VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 2 441 // CHECK: [[VAR1_RHS_VOID:%.+]] = load i8*, i8** [[VAR1_RHS_REF]], 442 // CHECK: [[VAR1_RHS:%.+]] = bitcast i8* [[VAR1_RHS_VOID]] to [[S_FLOAT_TY]]* 443 // var1_rhs = (S<float>*)rhs[2]; 444 // CHECK: [[VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 2 445 // CHECK: [[VAR1_LHS_VOID:%.+]] = load i8*, i8** [[VAR1_LHS_REF]], 446 // CHECK: [[VAR1_LHS:%.+]] = bitcast i8* [[VAR1_LHS_VOID]] to [[S_FLOAT_TY]]* 447 448 // t_var1_lhs = (float*)lhs[3]; 449 // CHECK: [[T_VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 3 450 // CHECK: [[T_VAR1_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_RHS_REF]], 451 // CHECK: [[T_VAR1_RHS:%.+]] = bitcast i8* [[T_VAR1_RHS_VOID]] to float* 452 // t_var1_rhs = (float*)rhs[3]; 453 // CHECK: [[T_VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 3 454 // CHECK: [[T_VAR1_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_LHS_REF]], 455 // CHECK: [[T_VAR1_LHS:%.+]] = bitcast i8* [[T_VAR1_LHS_VOID]] to float* 456 457 // t_var_lhs += t_var_rhs; 458 // CHECK: [[T_VAR_LHS_VAL:%.+]] = load float, float* [[T_VAR_LHS]], 459 // CHECK: [[T_VAR_RHS_VAL:%.+]] = load float, float* [[T_VAR_RHS]], 460 // CHECK: [[UP:%.+]] = fadd float [[T_VAR_LHS_VAL]], [[T_VAR_RHS_VAL]] 461 // CHECK: store float [[UP]], float* [[T_VAR_LHS]], 462 463 // var_lhs = var_lhs.operator &(var_rhs); 464 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_LHS]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_RHS]]) 465 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_LHS]] to i8* 466 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8* 467 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 468 469 // var1_lhs = var1_lhs.operator &&(var1_rhs); 470 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_LHS]]) 471 // CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 472 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]] 473 // CHECK: [[TRUE]] 474 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_RHS]]) 475 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 476 // CHECK: br label %[[END2]] 477 // CHECK: [[END2]] 478 // CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ] 479 // CHECK: [[CONV:%.+]] = uitofp i1 [[COND_LVALUE]] to float 480 // CHECK: call void @{{.+}}([[S_FLOAT_TY]]* [[COND_LVALUE:%.+]], float [[CONV]]) 481 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_LHS]] to i8* 482 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8* 483 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 484 485 // t_var1_lhs = min(t_var1_lhs, t_var1_rhs); 486 // CHECK: [[T_VAR1_LHS_VAL:%.+]] = load float, float* [[T_VAR1_LHS]], 487 // CHECK: [[T_VAR1_RHS_VAL:%.+]] = load float, float* [[T_VAR1_RHS]], 488 // CHECK: [[CMP:%.+]] = fcmp olt float [[T_VAR1_LHS_VAL]], [[T_VAR1_RHS_VAL]] 489 // CHECK: br i1 [[CMP]] 490 // CHECK: [[UP:%.+]] = phi float 491 // CHECK: store float [[UP]], float* [[T_VAR1_LHS]], 492 // CHECK: ret void 493 494 // CHECK: define internal void [[MAIN_MICROTASK1]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, i64 %{{.+}}, i64 %{{.+}}, i32* %{{.+}}, [2 x i32]* dereferenceable(8) %{{.+}}, [10 x [4 x [[S_FLOAT_TY]]]]* dereferenceable(160) %{{.+}}) 495 496 // Reduction list for runtime. 497 // CHECK: [[RED_LIST:%.+]] = alloca [4 x i8*], 498 499 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]], 500 501 // CHECK: [[IDX1:%.+]] = mul nsw i64 1, %{{.+}} 502 // CHECK: [[LB1:%.+]] = getelementptr inbounds i32, i32* %{{.+}}, i64 [[IDX1]] 503 // CHECK: [[LB1_0:%.+]] = getelementptr inbounds i32, i32* [[LB1]], i64 0 504 // CHECK: [[IDX1:%.+]] = mul nsw i64 1, %{{.+}} 505 // CHECK: [[UB1:%.+]] = getelementptr inbounds i32, i32* %{{.+}}, i64 [[IDX1]] 506 // CHECK: [[UB1_UP:%.+]] = getelementptr inbounds i32, i32* [[UB1]], i64 % 507 // CHECK: [[UB_CAST:%.+]] = ptrtoint i32* [[UB1_UP]] to i64 508 // CHECK: [[LB_CAST:%.+]] = ptrtoint i32* [[LB1_0]] to i64 509 // CHECK: [[DIFF:%.+]] = sub i64 [[UB_CAST]], [[LB_CAST]] 510 // CHECK: [[SIZE_1:%.+]] = sdiv exact i64 [[DIFF]], ptrtoint (i32* getelementptr (i32, i32* null, i32 1) to i64) 511 // CHECK: [[ARR_SIZE:%.+]] = add nuw i64 [[SIZE_1]], 1 512 // CHECK: call i8* @llvm.stacksave() 513 // CHECK: [[ARR_PRIV:%.+]] = alloca i32, i64 [[ARR_SIZE]], 514 515 // Check initialization of private copy. 516 // CHECK: [[END:%.+]] = getelementptr i32, i32* [[ARR_PRIV]], i64 [[ARR_SIZE]] 517 // CHECK: [[ISEMPTY:%.+]] = icmp eq i32* [[ARR_PRIV]], [[END]] 518 // CHECK: br i1 [[ISEMPTY]], 519 // CHECK: phi i32* 520 // CHECK: store i32 0, i32* % 521 // CHECK: [[DONE:%.+]] = icmp eq i32* %{{.+}}, [[END]] 522 // CHECK: br i1 [[DONE]], 523 524 // CHECK: [[ARRS_PRIV:%.+]] = alloca [[S_FLOAT_TY]], i64 [[ARRS_SIZE:%.+]], 525 526 // Check initialization of private copy. 527 // CHECK: [[END:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[ARRS_PRIV]], i64 [[ARRS_SIZE]] 528 // CHECK: [[ISEMPTY:%.+]] = icmp eq [[S_FLOAT_TY]]* [[ARRS_PRIV]], [[END]] 529 // CHECK: br i1 [[ISEMPTY]], 530 // CHECK: phi [[S_FLOAT_TY]]* 531 // CHECK: call void @_ZN1SIfEC1Ev([[S_FLOAT_TY]]* % 532 // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[END]] 533 // CHECK: br i1 [[DONE]], 534 535 // CHECK: [[GTID_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_ADDR]] 536 // CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_REF]] 537 // CHECK: call void @__kmpc_for_static_init_4( 538 // Skip checks for internal operations. 539 // CHECK: call void @__kmpc_for_static_fini( 540 541 // void *RedList[<n>] = {<ReductionVars>[0], ..., <ReductionVars>[<n>-1]}; 542 543 // CHECK: [[ARR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 0 544 // CHECK: [[BITCAST:%.+]] = bitcast i32* [[ARR_PRIV]] to i8* 545 // CHECK: store i8* [[BITCAST]], i8** [[ARR_PRIV_REF]], 546 // CHECK: [[ARR_SIZE_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 1 547 // CHECK: [[BITCAST:%.+]] = inttoptr i64 [[ARR_SIZE]] to i8* 548 // CHECK: store i8* [[BITCAST]], i8** [[ARR_SIZE_REF]], 549 // CHECK: [[ARRS_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 2 550 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[ARRS_PRIV]] to i8* 551 // CHECK: store i8* [[BITCAST]], i8** [[ARRS_PRIV_REF]], 552 // CHECK: [[ARRS_SIZE_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 3 553 // CHECK: [[BITCAST:%.+]] = inttoptr i64 [[ARRS_SIZE]] to i8* 554 // CHECK: store i8* [[BITCAST]], i8** [[ARRS_SIZE_REF]], 555 556 // res = __kmpc_reduce(<loc>, <gtid>, <n>, sizeof(RedList), RedList, reduce_func, &<lock>); 557 558 // CHECK: [[GTID_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_ADDR]] 559 // CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_REF]] 560 // CHECK: [[BITCAST:%.+]] = bitcast [4 x i8*]* [[RED_LIST]] to i8* 561 // CHECK: [[RES:%.+]] = call i32 @__kmpc_reduce_nowait(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], i32 2, i64 32, i8* [[BITCAST]], void (i8*, i8*)* [[REDUCTION_FUNC:@.+]], [8 x i32]* [[REDUCTION_LOCK]]) 562 563 // switch(res) 564 // CHECK: switch i32 [[RES]], label %[[RED_DONE:.+]] [ 565 // CHECK: i32 1, label %[[CASE1:.+]] 566 // CHECK: i32 2, label %[[CASE2:.+]] 567 // CHECK: ] 568 569 // case 1: 570 // CHECK: [[CASE1]] 571 572 // arr[:] += arr_reduction[:]; 573 // CHECK: [[END:%.+]] = getelementptr i32, i32* [[LB1_0]], i64 [[ARR_SIZE]] 574 // CHECK: [[ISEMPTY:%.+]] = icmp eq i32* [[LB1_0]], [[END]] 575 // CHECK: br i1 [[ISEMPTY]], 576 // CHECK: phi i32* 577 // CHECK: [[ADD:%.+]] = add nsw i32 % 578 // CHECK: store i32 [[ADD]], i32* % 579 // CHECK: [[DONE:%.+]] = icmp eq i32* %{{.+}}, [[END]] 580 // CHECK: br i1 [[DONE]], 581 582 // arrs[:] = var.operator &(arrs_reduction[:]); 583 // CHECK: [[END:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[ARRS_LB:%.+]], i64 [[ARRS_SIZE]] 584 // CHECK: [[ISEMPTY:%.+]] = icmp eq [[S_FLOAT_TY]]* [[ARRS_LB]], [[END]] 585 // CHECK: br i1 [[ISEMPTY]], 586 // CHECK: phi [[S_FLOAT_TY]]* 587 // CHECK: [[AND:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @_ZN1SIfEanERKS0_([[S_FLOAT_TY]]* %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}}) 588 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[AND]] to i8* 589 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* [[BITCAST]], i64 4, i32 4, i1 false) 590 // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[END]] 591 // CHECK: br i1 [[DONE]], 592 593 // __kmpc_end_reduce(<loc>, <gtid>, &<lock>); 594 // CHECK: call void @__kmpc_end_reduce_nowait(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], [8 x i32]* [[REDUCTION_LOCK]]) 595 596 // break; 597 // CHECK: br label %[[RED_DONE]] 598 599 // case 2: 600 // CHECK: [[CASE2]] 601 602 // arr[:] += arr_reduction[:]; 603 // CHECK: [[END:%.+]] = getelementptr i32, i32* [[LB1_0]], i64 [[ARR_SIZE]] 604 // CHECK: [[ISEMPTY:%.+]] = icmp eq i32* [[LB1_0]], [[END]] 605 // CHECK: br i1 [[ISEMPTY]], 606 // CHECK: phi i32* 607 // CHECK: atomicrmw add i32* %{{.+}}, i32 %{{.+}} monotonic 608 // CHECK: [[DONE:%.+]] = icmp eq i32* %{{.+}}, [[END]] 609 // CHECK: br i1 [[DONE]], 610 611 // arrs[:] = var.operator &(arrs_reduction[:]); 612 // CHECK: [[END:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[ARRS_LB:%.+]], i64 [[ARRS_SIZE]] 613 // CHECK: [[ISEMPTY:%.+]] = icmp eq [[S_FLOAT_TY]]* [[ARRS_LB]], [[END]] 614 // CHECK: br i1 [[ISEMPTY]], 615 // CHECK: phi [[S_FLOAT_TY]]* 616 // CHECK: call void @__kmpc_critical( 617 // CHECK: [[AND:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @_ZN1SIfEanERKS0_([[S_FLOAT_TY]]* %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}}) 618 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[AND]] to i8* 619 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* [[BITCAST]], i64 4, i32 4, i1 false) 620 // CHECK: call void @__kmpc_end_critical( 621 // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[END]] 622 // CHECK: br i1 [[DONE]], 623 624 // break; 625 // CHECK: br label %[[RED_DONE]] 626 // CHECK: [[RED_DONE]] 627 628 // Check destruction of private copy. 629 // CHECK: [[END:%.+]] = getelementptr inbounds [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[ARRS_PRIV]], i64 [[ARRS_SIZE]] 630 // CHECK: [[ISEMPTY:%.+]] = icmp eq [[S_FLOAT_TY]]* [[ARRS_PRIV]], [[END]] 631 // CHECK: br i1 [[ISEMPTY]], 632 // CHECK: phi [[S_FLOAT_TY]]* 633 // CHECK: call void @_ZN1SIfED1Ev([[S_FLOAT_TY]]* % 634 // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[ARRS_PRIV]] 635 // CHECK: br i1 [[DONE]], 636 // CHECK: call void @llvm.stackrestore(i8* 637 638 // CHECK: ret void 639 640 // void reduce_func(void *lhs[<n>], void *rhs[<n>]) { 641 // *(Type0*)lhs[0] = ReductionOperation0(*(Type0*)lhs[0], *(Type0*)rhs[0]); 642 // ... 643 // *(Type<n>-1*)lhs[<n>-1] = ReductionOperation<n>-1(*(Type<n>-1*)lhs[<n>-1], 644 // *(Type<n>-1*)rhs[<n>-1]); 645 // } 646 // CHECK: define internal void [[REDUCTION_FUNC]](i8*, i8*) 647 // arr_rhs = (int*)rhs[0]; 648 // CHECK: [[ARR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS:%.+]], i64 0, i64 0 649 // CHECK: [[ARR_RHS_VOID:%.+]] = load i8*, i8** [[ARR_RHS_REF]], 650 // CHECK: [[ARR_RHS:%.+]] = bitcast i8* [[ARR_RHS_VOID]] to i32* 651 // arr_lhs = (int*)lhs[0]; 652 // CHECK: [[ARR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS:%.+]], i64 0, i64 0 653 // CHECK: [[ARR_LHS_VOID:%.+]] = load i8*, i8** [[ARR_LHS_REF]], 654 // CHECK: [[ARR_LHS:%.+]] = bitcast i8* [[ARR_LHS_VOID]] to i32* 655 656 // arr_size = (size_t)lhs[1]; 657 // CHECK: [[ARR_SIZE_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 1 658 // CHECK: [[ARR_SIZE_VOID:%.+]] = load i8*, i8** [[ARR_SIZE_REF]], 659 // CHECK: [[ARR_SIZE:%.+]] = ptrtoint i8* [[ARR_SIZE_VOID]] to i64 660 661 // arrs_rhs = (S<float>*)rhs[2]; 662 // CHECK: [[ARRS_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 2 663 // CHECK: [[ARRS_RHS_VOID:%.+]] = load i8*, i8** [[ARRS_RHS_REF]], 664 // CHECK: [[ARRS_RHS:%.+]] = bitcast i8* [[ARRS_RHS_VOID]] to [[S_FLOAT_TY]]* 665 // arrs_lhs = (S<float>*)lhs[2]; 666 // CHECK: [[ARRS_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 2 667 // CHECK: [[ARRS_LHS_VOID:%.+]] = load i8*, i8** [[ARRS_LHS_REF]], 668 // CHECK: [[ARRS_LHS:%.+]] = bitcast i8* [[ARRS_LHS_VOID]] to [[S_FLOAT_TY]]* 669 670 // arrs_size = (size_t)lhs[3]; 671 // CHECK: [[ARRS_SIZE_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 3 672 // CHECK: [[ARRS_SIZE_VOID:%.+]] = load i8*, i8** [[ARRS_SIZE_REF]], 673 // CHECK: [[ARRS_SIZE:%.+]] = ptrtoint i8* [[ARRS_SIZE_VOID]] to i64 674 675 // arr_lhs[:] += arr_rhs[:]; 676 // CHECK: [[END:%.+]] = getelementptr i32, i32* [[ARR_LHS]], i64 [[ARR_SIZE]] 677 // CHECK: [[ISEMPTY:%.+]] = icmp eq i32* [[ARR_LHS]], [[END]] 678 // CHECK: br i1 [[ISEMPTY]], 679 // CHECK: phi i32* 680 // CHECK: [[ADD:%.+]] = add nsw i32 % 681 // CHECK: store i32 [[ADD]], i32* % 682 // CHECK: [[DONE:%.+]] = icmp eq i32* %{{.+}}, [[END]] 683 // CHECK: br i1 [[DONE]], 684 685 // arrs_lhs = arrs_lhs.operator &(arrs_rhs); 686 // CHECK: [[END:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[ARRS_LB:%.+]], i64 [[ARRS_SIZE]] 687 // CHECK: [[ISEMPTY:%.+]] = icmp eq [[S_FLOAT_TY]]* [[ARRS_LB]], [[END]] 688 // CHECK: br i1 [[ISEMPTY]], 689 // CHECK: phi [[S_FLOAT_TY]]* 690 // CHECK: [[AND:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @_ZN1SIfEanERKS0_([[S_FLOAT_TY]]* %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}}) 691 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[AND]] to i8* 692 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* [[BITCAST]], i64 4, i32 4, i1 false) 693 // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[END]] 694 // CHECK: br i1 [[DONE]], 695 696 // CHECK: ret void 697 698 // CHECK: define internal void [[MAIN_MICROTASK2]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, i64 %{{.+}}, i64 %{{.+}}, i32* %{{.+}}, [10 x [4 x [[S_FLOAT_TY]]]]* dereferenceable(160) %{{.+}}) 699 700 // CHECK: [[ARRS_PRIV:%.+]] = alloca [10 x [4 x [[S_FLOAT_TY]]]], 701 702 // Reduction list for runtime. 703 // CHECK: [[RED_LIST:%.+]] = alloca [3 x i8*], 704 705 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]], 706 707 // CHECK: [[ARR_SIZE:%.+]] = mul nuw i64 %{{.+}}, 4 708 // CHECK: call i8* @llvm.stacksave() 709 // CHECK: [[ARR_PRIV:%.+]] = alloca i32, i64 [[ARR_SIZE]], 710 711 // Check initialization of private copy. 712 // CHECK: [[END:%.+]] = getelementptr i32, i32* [[ARR_PRIV]], i64 [[ARR_SIZE]] 713 // CHECK: [[ISEMPTY:%.+]] = icmp eq i32* [[ARR_PRIV]], [[END]] 714 // CHECK: br i1 [[ISEMPTY]], 715 // CHECK: phi i32* 716 // CHECK: store i32 0, i32* % 717 // CHECK: [[DONE:%.+]] = icmp eq i32* %{{.+}}, [[END]] 718 // CHECK: br i1 [[DONE]], 719 720 // Check initialization of private copy. 721 // CHECK: [[LHS_BEGIN:%.+]] = bitcast [10 x [4 x [[S_FLOAT_TY]]]]* %{{.+}} to [[S_FLOAT_TY]]* 722 // CHECK: [[BEGIN:%.+]] = getelementptr inbounds [10 x [4 x [[S_FLOAT_TY]]]], [10 x [4 x [[S_FLOAT_TY]]]]* [[ARRS_PRIV]], i32 0, i32 0, i32 0 723 // CHECK: [[END:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[BEGIN]], i64 40 724 // CHECK: [[ISEMPTY:%.+]] = icmp eq [[S_FLOAT_TY]]* [[BEGIN]], [[END]] 725 // CHECK: br i1 [[ISEMPTY]], 726 // CHECK: phi [[S_FLOAT_TY]]* 727 // CHECK: call void @_ZN1SIfEC1Ev([[S_FLOAT_TY]]* % 728 // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[END]] 729 // CHECK: br i1 [[DONE]], 730 // CHECK: [[ARRS_PRIV_BEGIN:%.+]] = bitcast [10 x [4 x [[S_FLOAT_TY]]]]* [[ARRS_PRIV]] to [[S_FLOAT_TY]]* 731 732 // CHECK: [[GTID_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_ADDR]] 733 // CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_REF]] 734 // CHECK: call void @__kmpc_for_static_init_4( 735 // Skip checks for internal operations. 736 // CHECK: call void @__kmpc_for_static_fini( 737 738 // void *RedList[<n>] = {<ReductionVars>[0], ..., <ReductionVars>[<n>-1]}; 739 740 // CHECK: [[ARR_PRIV_REF:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[RED_LIST]], i64 0, i64 0 741 // CHECK: [[BITCAST:%.+]] = bitcast i32* [[ARR_PRIV]] to i8* 742 // CHECK: store i8* [[BITCAST]], i8** [[ARR_PRIV_REF]], 743 // CHECK: [[ARR_SIZE_REF:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[RED_LIST]], i64 0, i64 1 744 // CHECK: [[BITCAST:%.+]] = inttoptr i64 [[ARR_SIZE]] to i8* 745 // CHECK: store i8* [[BITCAST]], i8** [[ARR_SIZE_REF]], 746 // CHECK: [[ARRS_PRIV_REF:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[RED_LIST]], i64 0, i64 2 747 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[ARRS_PRIV_BEGIN]] to i8* 748 // CHECK: store i8* [[BITCAST]], i8** [[ARRS_PRIV_REF]], 749 750 // res = __kmpc_reduce(<loc>, <gtid>, <n>, sizeof(RedList), RedList, reduce_func, &<lock>); 751 752 // CHECK: [[GTID_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_ADDR]] 753 // CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_REF]] 754 // CHECK: [[BITCAST:%.+]] = bitcast [3 x i8*]* [[RED_LIST]] to i8* 755 // CHECK: [[RES:%.+]] = call i32 @__kmpc_reduce(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], i32 2, i64 24, i8* [[BITCAST]], void (i8*, i8*)* [[REDUCTION_FUNC:@.+]], [8 x i32]* [[REDUCTION_LOCK]]) 756 757 // switch(res) 758 // CHECK: switch i32 [[RES]], label %[[RED_DONE:.+]] [ 759 // CHECK: i32 1, label %[[CASE1:.+]] 760 // CHECK: i32 2, label %[[CASE2:.+]] 761 // CHECK: ] 762 763 // case 1: 764 // CHECK: [[CASE1]] 765 766 // arr[:] += arr_reduction[:]; 767 // CHECK: [[END:%.+]] = getelementptr i32, i32* [[LB1_0:%.+]], i64 [[ARR_SIZE]] 768 // CHECK: [[ISEMPTY:%.+]] = icmp eq i32* [[LB1_0]], [[END]] 769 // CHECK: br i1 [[ISEMPTY]], 770 // CHECK: phi i32* 771 // CHECK: [[ADD:%.+]] = add nsw i32 % 772 // CHECK: store i32 [[ADD]], i32* % 773 // CHECK: [[DONE:%.+]] = icmp eq i32* %{{.+}}, [[END]] 774 // CHECK: br i1 [[DONE]], 775 776 // arrs[:] = var.operator &(arrs_reduction[:]); 777 // CHECK: [[END:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[LHS_BEGIN]], i64 40 778 // CHECK: [[ISEMPTY:%.+]] = icmp eq [[S_FLOAT_TY]]* [[LHS_BEGIN]], [[END]] 779 // CHECK: br i1 [[ISEMPTY]], 780 // CHECK: phi [[S_FLOAT_TY]]* 781 // CHECK: [[AND:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @_ZN1SIfEanERKS0_([[S_FLOAT_TY]]* %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}}) 782 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[AND]] to i8* 783 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* [[BITCAST]], i64 4, i32 4, i1 false) 784 // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[END]] 785 // CHECK: br i1 [[DONE]], 786 787 // __kmpc_end_reduce(<loc>, <gtid>, &<lock>); 788 // CHECK: call void @__kmpc_end_reduce(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], [8 x i32]* [[REDUCTION_LOCK]]) 789 790 // break; 791 // CHECK: br label %[[RED_DONE]] 792 793 // case 2: 794 // CHECK: [[CASE2]] 795 796 // arr[:] += arr_reduction[:]; 797 // CHECK: [[END:%.+]] = getelementptr i32, i32* [[LB1_0]], i64 [[ARR_SIZE]] 798 // CHECK: [[ISEMPTY:%.+]] = icmp eq i32* [[LB1_0]], [[END]] 799 // CHECK: br i1 [[ISEMPTY]], 800 // CHECK: phi i32* 801 // CHECK: atomicrmw add i32* %{{.+}}, i32 %{{.+}} monotonic 802 // CHECK: [[DONE:%.+]] = icmp eq i32* %{{.+}}, [[END]] 803 // CHECK: br i1 [[DONE]], 804 805 // arrs[:] = var.operator &(arrs_reduction[:]); 806 // CHECK: [[END:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[LHS_BEGIN]], i64 40 807 // CHECK: [[ISEMPTY:%.+]] = icmp eq [[S_FLOAT_TY]]* [[LHS_BEGIN]], [[END]] 808 // CHECK: br i1 [[ISEMPTY]], 809 // CHECK: phi [[S_FLOAT_TY]]* 810 // CHECK: call void @__kmpc_critical( 811 // CHECK: [[AND:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @_ZN1SIfEanERKS0_([[S_FLOAT_TY]]* %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}}) 812 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[AND]] to i8* 813 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* [[BITCAST]], i64 4, i32 4, i1 false) 814 // CHECK: call void @__kmpc_end_critical( 815 // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[END]] 816 // CHECK: br i1 [[DONE]], 817 818 // break; 819 // CHECK: br label %[[RED_DONE]] 820 // CHECK: [[RED_DONE]] 821 822 // Check destruction of private copy. 823 // CHECK: [[BEGIN:%.+]] = getelementptr inbounds [10 x [4 x [[S_FLOAT_TY]]]], [10 x [4 x [[S_FLOAT_TY]]]]* [[ARRS_PRIV]], i32 0, i32 0, i32 0 824 // CHECK: [[END:%.+]] = getelementptr inbounds [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[BEGIN]], i64 40 825 // CHECK: br 826 // CHECK: phi [[S_FLOAT_TY]]* 827 // CHECK: call void @_ZN1SIfED1Ev([[S_FLOAT_TY]]* % 828 // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[BEGIN]] 829 // CHECK: br i1 [[DONE]], 830 // CHECK: call void @llvm.stackrestore(i8* 831 // CHECK: call void @__kmpc_barrier( 832 833 // CHECK: ret void 834 835 // void reduce_func(void *lhs[<n>], void *rhs[<n>]) { 836 // *(Type0*)lhs[0] = ReductionOperation0(*(Type0*)lhs[0], *(Type0*)rhs[0]); 837 // ... 838 // *(Type<n>-1*)lhs[<n>-1] = ReductionOperation<n>-1(*(Type<n>-1*)lhs[<n>-1], 839 // *(Type<n>-1*)rhs[<n>-1]); 840 // } 841 // CHECK: define internal void [[REDUCTION_FUNC]](i8*, i8*) 842 // arr_rhs = (int*)rhs[0]; 843 // CHECK: [[ARR_RHS_REF:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[RED_LIST_RHS:%.+]], i64 0, i64 0 844 // CHECK: [[ARR_RHS_VOID:%.+]] = load i8*, i8** [[ARR_RHS_REF]], 845 // CHECK: [[ARR_RHS:%.+]] = bitcast i8* [[ARR_RHS_VOID]] to i32* 846 // arr_lhs = (int*)lhs[0]; 847 // CHECK: [[ARR_LHS_REF:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[RED_LIST_LHS:%.+]], i64 0, i64 0 848 // CHECK: [[ARR_LHS_VOID:%.+]] = load i8*, i8** [[ARR_LHS_REF]], 849 // CHECK: [[ARR_LHS:%.+]] = bitcast i8* [[ARR_LHS_VOID]] to i32* 850 851 // arr_size = (size_t)lhs[1]; 852 // CHECK: [[ARR_SIZE_REF:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[RED_LIST_LHS]], i64 0, i64 1 853 // CHECK: [[ARR_SIZE_VOID:%.+]] = load i8*, i8** [[ARR_SIZE_REF]], 854 // CHECK: [[ARR_SIZE:%.+]] = ptrtoint i8* [[ARR_SIZE_VOID]] to i64 855 856 // arrs_rhs = (S<float>*)rhs[2]; 857 // CHECK: [[ARRS_RHS_REF:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[RED_LIST_RHS]], i64 0, i64 2 858 // CHECK: [[ARRS_RHS_VOID:%.+]] = load i8*, i8** [[ARRS_RHS_REF]], 859 // CHECK: [[ARRS_RHS:%.+]] = bitcast i8* [[ARRS_RHS_VOID]] to [[S_FLOAT_TY]]* 860 // arrs_lhs = (S<float>*)lhs[2]; 861 // CHECK: [[ARRS_LHS_REF:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[RED_LIST_LHS]], i64 0, i64 2 862 // CHECK: [[ARRS_LHS_VOID:%.+]] = load i8*, i8** [[ARRS_LHS_REF]], 863 // CHECK: [[ARRS_LHS:%.+]] = bitcast i8* [[ARRS_LHS_VOID]] to [[S_FLOAT_TY]]* 864 865 // arr_lhs[:] += arr_rhs[:]; 866 // CHECK: [[END:%.+]] = getelementptr i32, i32* [[ARR_LHS]], i64 [[ARR_SIZE]] 867 // CHECK: [[ISEMPTY:%.+]] = icmp eq i32* [[ARR_LHS]], [[END]] 868 // CHECK: br i1 [[ISEMPTY]], 869 // CHECK: phi i32* 870 // CHECK: [[ADD:%.+]] = add nsw i32 % 871 // CHECK: store i32 [[ADD]], i32* % 872 // CHECK: [[DONE:%.+]] = icmp eq i32* %{{.+}}, [[END]] 873 // CHECK: br i1 [[DONE]], 874 875 // arrs_lhs = arrs_lhs.operator &(arrs_rhs); 876 // CHECK: [[END:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[ARRS_LB:%.+]], i64 40 877 // CHECK: [[ISEMPTY:%.+]] = icmp eq [[S_FLOAT_TY]]* [[ARRS_LB]], [[END]] 878 // CHECK: br i1 [[ISEMPTY]], 879 // CHECK: phi [[S_FLOAT_TY]]* 880 // CHECK: [[AND:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @_ZN1SIfEanERKS0_([[S_FLOAT_TY]]* %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}}) 881 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[AND]] to i8* 882 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* [[BITCAST]], i64 4, i32 4, i1 false) 883 // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[END]] 884 // CHECK: br i1 [[DONE]], 885 886 // CHECK: ret void 887 888 // CHECK: define internal void [[MAIN_MICROTASK3]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [[S_FLOAT_TY]]*** dereferenceable(8) %{{.+}}) 889 890 // CHECK: [[VAR2_ORIG_ADDR:%.+]] = alloca [[S_FLOAT_TY]]***, 891 892 // Reduction list for runtime. 893 // CHECK: [[RED_LIST:%.+]] = alloca [2 x i8*], 894 895 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]], 896 // CHECK: [[VAR2_ORIG:%.+]] = load [[S_FLOAT_TY]]***, [[S_FLOAT_TY]]**** [[VAR2_ORIG_ADDR]], 897 898 // CHECK: load [[S_FLOAT_TY]]**, [[S_FLOAT_TY]]*** [[VAR2_ORIG]], 899 // CHECK: getelementptr inbounds [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** %{{.+}}, i64 0 900 // CHECK: load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** % 901 // CHECK: [[LOW:%.+]] = getelementptr inbounds [[S_FLOAT_TY]], [[S_FLOAT_TY]]* %{{.+}}, i64 1 902 // CHECK: load [[S_FLOAT_TY]]**, [[S_FLOAT_TY]]*** [[VAR2_ORIG]], 903 // CHECK: getelementptr inbounds [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** %{{.+}}, i64 4 904 // CHECK: load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** % 905 // CHECK: getelementptr inbounds [[S_FLOAT_TY]], [[S_FLOAT_TY]]* %{{.+}}, i64 6 906 // CHECK: [[LD:%.+]] = load [[S_FLOAT_TY]]**, [[S_FLOAT_TY]]*** [[VAR2_ORIG]], 907 // CHECK: [[ORIG_START:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[LD]], 908 // CHECK: [[LAST:%.+]] = ptrtoint [[S_FLOAT_TY]]* %{{.+}} to i64 909 // CHECK: [[FIRST:%.+]] = ptrtoint [[S_FLOAT_TY]]* [[LOW]] to i64 910 // CHECK: [[BYTE_DIF:%.+]] = sub i64 [[LAST]], [[FIRST]] 911 // CHECK: [[DIF:%.+]] = sdiv exact i64 [[BYTE_DIF]], ptrtoint (float* getelementptr (float, float* null, i32 1) to i64) 912 // CHECK: [[SIZE:%.+]] = add nuw i64 [[DIF]], 1 913 // CHECK: call i8* @llvm.stacksave() 914 // CHECK: [[VAR2_PRIV:%.+]] = alloca [[S_FLOAT_TY]], i64 [[SIZE]], 915 // CHECK: [[START:%.+]] = ptrtoint [[S_FLOAT_TY]]* [[ORIG_START]] to i64 916 // CHECK: [[LOW_BOUND:%.+]] = ptrtoint [[S_FLOAT_TY]]* [[LOW]] to i64 917 // CHECK: [[OFFSET_BYTES:%.+]] = sub i64 [[START]], [[LOW_BOUND]] 918 // CHECK: [[OFFSET:%.+]] = sdiv exact i64 [[OFFSET_BYTES]], ptrtoint (float* getelementptr (float, float* null, i32 1) to i64) 919 // CHECK: [[PSEUDO_VAR2_PRIV:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[VAR2_PRIV]], i64 [[OFFSET]] 920 // CHECK: store [[S_FLOAT_TY]]** [[REF:.+]], [[S_FLOAT_TY]]*** % 921 // CHECK: store [[S_FLOAT_TY]]* [[PSEUDO_VAR2_PRIV]], [[S_FLOAT_TY]]** [[REF]] 922 // CHECK: ret void 923 924 // CHECK: define internal void [[MAIN_MICROTASK4]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [2 x [[S_FLOAT_TY]]]* dereferenceable(8) %{{.+}}) 925 926 // CHECK: [[VVAR2_ORIG_ADDR:%.+]] = alloca [2 x [[S_FLOAT_TY]]]*, 927 928 // Reduction list for runtime. 929 // CHECK: [[RED_LIST:%.+]] = alloca [2 x i8*], 930 931 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]], 932 // CHECK: [[VVAR2_ORIG:%.+]] = load [2 x [[S_FLOAT_TY]]]*, [2 x [[S_FLOAT_TY]]]** [[VVAR2_ORIG_ADDR]], 933 934 // CHECK: [[LOW:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[VVAR2_ORIG]], i64 0, i64 0 935 // CHECK: getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[VVAR2_ORIG]], i64 0, i64 4 936 // CHECK: [[ORIG_START:%.+]] = bitcast [2 x [[S_FLOAT_TY]]]* [[VVAR2_ORIG]] to [[S_FLOAT_TY]]* 937 // CHECK: [[LAST:%.+]] = ptrtoint [[S_FLOAT_TY]]* %{{.+}} to i64 938 // CHECK: [[FIRST:%.+]] = ptrtoint [[S_FLOAT_TY]]* [[LOW]] to i64 939 // CHECK: [[BYTE_DIF:%.+]] = sub i64 [[LAST]], [[FIRST]] 940 // CHECK: [[DIF:%.+]] = sdiv exact i64 [[BYTE_DIF]], ptrtoint (float* getelementptr (float, float* null, i32 1) to i64) 941 // CHECK: [[SIZE:%.+]] = add nuw i64 [[DIF]], 1 942 // CHECK: call i8* @llvm.stacksave() 943 // CHECK: [[VVAR2_PRIV:%.+]] = alloca [[S_FLOAT_TY]], i64 [[SIZE]], 944 // CHECK: [[START:%.+]] = ptrtoint [[S_FLOAT_TY]]* [[ORIG_START]] to i64 945 // CHECK: [[LOW_BOUND:%.+]] = ptrtoint [[S_FLOAT_TY]]* [[LOW]] to i64 946 // CHECK: [[OFFSET_BYTES:%.+]] = sub i64 [[START]], [[LOW_BOUND]] 947 // CHECK: [[OFFSET:%.+]] = sdiv exact i64 [[OFFSET_BYTES]], ptrtoint (float* getelementptr (float, float* null, i32 1) to i64) 948 // CHECK: [[PSEUDO_VVAR2_PRIV:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[VVAR2_PRIV]], i64 [[OFFSET]] 949 // CHECK: [[VVAR2_PRIV:%.+]] = bitcast [[S_FLOAT_TY]]* [[PSEUDO_VVAR2_PRIV]] to [2 x [[S_FLOAT_TY]]]* 950 // CHECK: ret void 951 952 // CHECK: define internal void [[MAIN_MICROTASK5]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [2 x [[S_FLOAT_TY]]]* dereferenceable(8) %{{.+}}) 953 954 // CHECK: [[VAR3_ORIG_ADDR:%.+]] = alloca [2 x [[S_FLOAT_TY]]]*, 955 956 // Reduction list for runtime. 957 // CHECK: [[RED_LIST:%.+]] = alloca [2 x i8*], 958 959 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]], 960 961 // CHECK: [[VAR3_ORIG:%.+]] = load [2 x [[S_FLOAT_TY]]]*, [2 x [[S_FLOAT_TY]]]** [[VAR3_ORIG_ADDR]], 962 // CHECK: store [2 x [[S_FLOAT_TY]]]* [[VAR3_ORIG]], [2 x [[S_FLOAT_TY]]]** [[VAR3_ORIG_ADDR:%.+]], 963 // CHECK: [[VAR3_ORIG:%.+]] = load [2 x [[S_FLOAT_TY]]]*, [2 x [[S_FLOAT_TY]]]** [[VAR3_ORIG_ADDR]], 964 // CHECK: [[LOW:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[VAR3_ORIG]], i64 0, i64 1 965 // CHECK: [[VAR3_ORIG:%.+]] = load [2 x [[S_FLOAT_TY]]]*, [2 x [[S_FLOAT_TY]]]** [[VAR3_ORIG_ADDR]], 966 // CHECK: getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[VAR3_ORIG]], i64 0, i64 2 967 // CHECK: [[VAR3_ORIG:%.+]] = load [2 x [[S_FLOAT_TY]]]*, [2 x [[S_FLOAT_TY]]]** [[VAR3_ORIG_ADDR]], 968 // CHECK: [[ORIG_START:%.+]] = bitcast [2 x [[S_FLOAT_TY]]]* [[VAR3_ORIG]] to [[S_FLOAT_TY]]* 969 // CHECK: [[LAST:%.+]] = ptrtoint [[S_FLOAT_TY]]* %{{.+}} to i64 970 // CHECK: [[FIRST:%.+]] = ptrtoint [[S_FLOAT_TY]]* [[LOW]] to i64 971 // CHECK: [[BYTE_DIF:%.+]] = sub i64 [[LAST]], [[FIRST]] 972 // CHECK: [[DIF:%.+]] = sdiv exact i64 [[BYTE_DIF]], ptrtoint (float* getelementptr (float, float* null, i32 1) to i64) 973 // CHECK: [[SIZE:%.+]] = add nuw i64 [[DIF]], 1 974 // CHECK: call i8* @llvm.stacksave() 975 // CHECK: [[VAR3_PRIV:%.+]] = alloca [[S_FLOAT_TY]], i64 [[SIZE]], 976 // CHECK: [[START:%.+]] = ptrtoint [[S_FLOAT_TY]]* [[ORIG_START]] to i64 977 // CHECK: [[LOW_BOUND:%.+]] = ptrtoint [[S_FLOAT_TY]]* [[LOW]] to i64 978 // CHECK: [[OFFSET_BYTES:%.+]] = sub i64 [[START]], [[LOW_BOUND]] 979 // CHECK: [[OFFSET:%.+]] = sdiv exact i64 [[OFFSET_BYTES]], ptrtoint (float* getelementptr (float, float* null, i32 1) to i64) 980 // CHECK: [[PSEUDO_VAR3_PRIV:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[VAR3_PRIV]], i64 [[OFFSET]] 981 // CHECK: [[VAR3_PRIV:%.+]] = bitcast [[S_FLOAT_TY]]* [[PSEUDO_VAR3_PRIV]] to [2 x [[S_FLOAT_TY]]]* 982 983 // CHECK: store [2 x [[S_FLOAT_TY]]]* [[VAR3_PRIV]], [2 x [[S_FLOAT_TY]]]** % 984 985 // CHECK: ret void 986 987 // CHECK: define internal void [[MAIN_MICROTASK6]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [2 x [[S_FLOAT_TY]]]* dereferenceable(8) %{{.+}}) 988 989 // CHECK: [[VAR3_ORIG_ADDR:%.+]] = alloca [2 x [[S_FLOAT_TY]]]*, 990 // CHECK: [[VAR3_PRIV:%.+]] = alloca [2 x [[S_FLOAT_TY]]], 991 992 // Reduction list for runtime. 993 // CHECK: [[RED_LIST:%.+]] = alloca [1 x i8*], 994 995 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]], 996 997 // CHECK: [[VAR3_ORIG:%.+]] = load [2 x [[S_FLOAT_TY]]]*, [2 x [[S_FLOAT_TY]]]** [[VAR3_ORIG_ADDR]], 998 // CHECK: store [2 x [[S_FLOAT_TY]]]* [[VAR3_ORIG]], [2 x [[S_FLOAT_TY]]]** [[VAR3_ORIG_ADDR:%.+]], 999 // CHECK: [[VAR3_ORIG:%.+]] = load [2 x [[S_FLOAT_TY]]]*, [2 x [[S_FLOAT_TY]]]** [[VAR3_ORIG_ADDR]], 1000 // CHECK: bitcast [2 x [[S_FLOAT_TY]]]* [[VAR3_ORIG]] to [[S_FLOAT_TY]]* 1001 // CHECK: getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[VAR3_PRIV]], i32 0, i32 0 1002 // CHECK: getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* %{{.+}}, i64 2 1003 1004 // CHECK: store [2 x [[S_FLOAT_TY]]]* [[VAR3_PRIV]], [2 x [[S_FLOAT_TY]]]** % 1005 1006 // CHECK: ret void 1007 1008 // CHECK: define {{.*}} i{{[0-9]+}} [[TMAIN_INT]]() 1009 // CHECK: [[TEST:%.+]] = alloca [[S_INT_TY]], 1010 // CHECK: call {{.*}} [[S_INT_TY_CONSTR:@.+]]([[S_INT_TY]]* [[TEST]]) 1011 // CHECK: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{[0-9]+}} 6, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)* bitcast (void (i{{[0-9]+}}*, i{{[0-9]+}}*, i32*, [[S_INT_TY]]*, [[S_INT_TY]]*, i32*, [2 x i32]*, [2 x [[S_INT_TY]]]*)* [[TMAIN_MICROTASK:@.+]] to void 1012 // CHECK: call {{.*}} [[S_INT_TY_DESTR:@.+]]([[S_INT_TY]]* 1013 // CHECK: ret 1014 // 1015 // CHECK: define internal void [[TMAIN_MICROTASK]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, i32* dereferenceable(4) %{{.+}}, [[S_INT_TY]]* dereferenceable(4) %{{.+}}, [[S_INT_TY]]* dereferenceable(4) %{{.+}}, i32* dereferenceable(4) %{{.+}}, [2 x i32]* dereferenceable(8) %{{.+}}, [2 x [[S_INT_TY]]]* dereferenceable(8) %{{.+}}) 1016 // CHECK: alloca i{{[0-9]+}}, 1017 // CHECK: alloca i{{[0-9]+}}, 1018 // CHECK: alloca i{{[0-9]+}}, 1019 // CHECK: alloca i{{[0-9]+}}, 1020 // CHECK: alloca i{{[0-9]+}}, 1021 // CHECK: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}}, 1022 // CHECK: [[VAR_PRIV:%.+]] = alloca [[S_INT_TY]], 1023 // CHECK: [[VAR1_PRIV:%.+]] = alloca [[S_INT_TY]], 1024 // CHECK: [[T_VAR1_PRIV:%.+]] = alloca i{{[0-9]+}}, 1025 1026 // Reduction list for runtime. 1027 // CHECK: [[RED_LIST:%.+]] = alloca [4 x i8*], 1028 1029 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]], 1030 1031 // CHECK: [[T_VAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** % 1032 // CHECK: [[VAR1_REF:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** % 1033 // CHECK: [[T_VAR1_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** % 1034 1035 // For + reduction operation initial value of private variable is 0. 1036 // CHECK: store i{{[0-9]+}} 0, i{{[0-9]+}}* [[T_VAR_PRIV]], 1037 1038 // For & reduction operation initial value of private variable is ones in all bits. 1039 // CHECK: [[VAR_REF:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** % 1040 // CHECK: call {{.*}} [[S_INT_TY_CONSTR:@.+]]([[S_INT_TY]]* [[VAR_PRIV]]) 1041 1042 // For && reduction operation initial value of private variable is 1.0. 1043 // CHECK: call {{.*}} [[S_INT_TY_CONSTR:@.+]]([[S_INT_TY]]* [[VAR1_PRIV]]) 1044 1045 // For min reduction operation initial value of private variable is largest repesentable value. 1046 // CHECK: store i{{[0-9]+}} 2147483647, i{{[0-9]+}}* [[T_VAR1_PRIV]], 1047 1048 // CHECK: [[GTID_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_ADDR]] 1049 // CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_REF]] 1050 // CHECK: call void @__kmpc_for_static_init_4( 1051 // Skip checks for internal operations. 1052 // CHECK: call void @__kmpc_for_static_fini( 1053 1054 // void *RedList[<n>] = {<ReductionVars>[0], ..., <ReductionVars>[<n>-1]}; 1055 1056 // CHECK: [[T_VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 0 1057 // CHECK: [[BITCAST:%.+]] = bitcast i{{[0-9]+}}* [[T_VAR_PRIV]] to i8* 1058 // CHECK: store i8* [[BITCAST]], i8** [[T_VAR_PRIV_REF]], 1059 // CHECK: [[VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 1 1060 // CHECK: [[BITCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR_PRIV]] to i8* 1061 // CHECK: store i8* [[BITCAST]], i8** [[VAR_PRIV_REF]], 1062 // CHECK: [[VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 2 1063 // CHECK: [[BITCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_PRIV]] to i8* 1064 // CHECK: store i8* [[BITCAST]], i8** [[VAR1_PRIV_REF]], 1065 // CHECK: [[T_VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 3 1066 // CHECK: [[BITCAST:%.+]] = bitcast i{{[0-9]+}}* [[T_VAR1_PRIV]] to i8* 1067 // CHECK: store i8* [[BITCAST]], i8** [[T_VAR1_PRIV_REF]], 1068 1069 // res = __kmpc_reduce_nowait(<loc>, <gtid>, <n>, sizeof(RedList), RedList, reduce_func, &<lock>); 1070 1071 // CHECK: [[BITCAST:%.+]] = bitcast [4 x i8*]* [[RED_LIST]] to i8* 1072 // 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]]) 1073 1074 // switch(res) 1075 // CHECK: switch i32 [[RES]], label %[[RED_DONE:.+]] [ 1076 // CHECK: i32 1, label %[[CASE1:.+]] 1077 // CHECK: i32 2, label %[[CASE2:.+]] 1078 // CHECK: ] 1079 1080 // case 1: 1081 // t_var += t_var_reduction; 1082 // CHECK: [[T_VAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_REF]], 1083 // CHECK: [[T_VAR_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_PRIV]], 1084 // CHECK: [[UP:%.+]] = add nsw i{{[0-9]+}} [[T_VAR_VAL]], [[T_VAR_PRIV_VAL]] 1085 // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR_REF]], 1086 1087 // var = var.operator &(var_reduction); 1088 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_REF]], [[S_INT_TY]]* dereferenceable(4) [[VAR_PRIV]]) 1089 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_REF]] to i8* 1090 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* 1091 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 1092 1093 // var1 = var1.operator &&(var1_reduction); 1094 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_REF]]) 1095 // CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 1096 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]] 1097 // CHECK: [[TRUE]] 1098 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_PRIV]]) 1099 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 1100 // CHECK: br label %[[END2]] 1101 // CHECK: [[END2]] 1102 // CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ] 1103 // CHECK: [[CONV:%.+]] = zext i1 [[COND_LVALUE]] to i32 1104 // CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]]) 1105 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8* 1106 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* 1107 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 1108 1109 // t_var1 = min(t_var1, t_var1_reduction); 1110 // CHECK: [[T_VAR1_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_REF]], 1111 // CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_PRIV]], 1112 // CHECK: [[CMP:%.+]] = icmp slt i{{[0-9]+}} [[T_VAR1_VAL]], [[T_VAR1_PRIV_VAL]] 1113 // CHECK: br i1 [[CMP]] 1114 // CHECK: [[UP:%.+]] = phi i32 1115 // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR1_REF]], 1116 1117 // __kmpc_end_reduce_nowait(<loc>, <gtid>, &<lock>); 1118 // CHECK: call void @__kmpc_end_reduce_nowait(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], [8 x i32]* [[REDUCTION_LOCK]]) 1119 1120 // break; 1121 // CHECK: br label %[[RED_DONE]] 1122 1123 // case 2: 1124 // t_var += t_var_reduction; 1125 // CHECK: [[T_VAR_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_PRIV]] 1126 // CHECK: atomicrmw add i32* [[T_VAR_REF]], i32 [[T_VAR_PRIV_VAL]] monotonic 1127 1128 // var = var.operator &(var_reduction); 1129 // CHECK: call void @__kmpc_critical( 1130 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_REF]], [[S_INT_TY]]* dereferenceable(4) [[VAR_PRIV]]) 1131 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_REF]] to i8* 1132 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* 1133 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 1134 // CHECK: call void @__kmpc_end_critical( 1135 1136 // var1 = var1.operator &&(var1_reduction); 1137 // CHECK: call void @__kmpc_critical( 1138 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_REF]]) 1139 // CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 1140 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]] 1141 // CHECK: [[TRUE]] 1142 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_PRIV]]) 1143 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 1144 // CHECK: br label %[[END2]] 1145 // CHECK: [[END2]] 1146 // CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ] 1147 // CHECK: [[CONV:%.+]] = zext i1 [[COND_LVALUE]] to i32 1148 // CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]]) 1149 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8* 1150 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* 1151 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 1152 // CHECK: call void @__kmpc_end_critical( 1153 1154 // t_var1 = min(t_var1, t_var1_reduction); 1155 // CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_PRIV]] 1156 // CHECK: atomicrmw min i32* [[T_VAR1_REF]], i32 [[T_VAR1_PRIV_VAL]] monotonic 1157 1158 // break; 1159 // CHECK: br label %[[RED_DONE]] 1160 // CHECK: [[RED_DONE]] 1161 // CHECK-DAG: call {{.*}} [[S_INT_TY_DESTR]]([[S_INT_TY]]* [[VAR_PRIV]]) 1162 // CHECK-DAG: call {{.*}} [[S_INT_TY_DESTR]]([[S_INT_TY]]* 1163 // CHECK: ret void 1164 1165 // void reduce_func(void *lhs[<n>], void *rhs[<n>]) { 1166 // *(Type0*)lhs[0] = ReductionOperation0(*(Type0*)lhs[0], *(Type0*)rhs[0]); 1167 // ... 1168 // *(Type<n>-1*)lhs[<n>-1] = ReductionOperation<n>-1(*(Type<n>-1*)lhs[<n>-1], 1169 // *(Type<n>-1*)rhs[<n>-1]); 1170 // } 1171 // CHECK: define internal void [[REDUCTION_FUNC]](i8*, i8*) 1172 // t_var_lhs = (i{{[0-9]+}}*)lhs[0]; 1173 // CHECK: [[T_VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS:%.+]], i64 0, i64 0 1174 // CHECK: [[T_VAR_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR_RHS_REF]], 1175 // CHECK: [[T_VAR_RHS:%.+]] = bitcast i8* [[T_VAR_RHS_VOID]] to i{{[0-9]+}}* 1176 // t_var_rhs = (i{{[0-9]+}}*)rhs[0]; 1177 // CHECK: [[T_VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS:%.+]], i64 0, i64 0 1178 // CHECK: [[T_VAR_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR_LHS_REF]], 1179 // CHECK: [[T_VAR_LHS:%.+]] = bitcast i8* [[T_VAR_LHS_VOID]] to i{{[0-9]+}}* 1180 1181 // var_lhs = (S<i{{[0-9]+}}>*)lhs[1]; 1182 // CHECK: [[VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 1 1183 // CHECK: [[VAR_RHS_VOID:%.+]] = load i8*, i8** [[VAR_RHS_REF]], 1184 // CHECK: [[VAR_RHS:%.+]] = bitcast i8* [[VAR_RHS_VOID]] to [[S_INT_TY]]* 1185 // var_rhs = (S<i{{[0-9]+}}>*)rhs[1]; 1186 // CHECK: [[VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 1 1187 // CHECK: [[VAR_LHS_VOID:%.+]] = load i8*, i8** [[VAR_LHS_REF]], 1188 // CHECK: [[VAR_LHS:%.+]] = bitcast i8* [[VAR_LHS_VOID]] to [[S_INT_TY]]* 1189 1190 // var1_lhs = (S<i{{[0-9]+}}>*)lhs[2]; 1191 // CHECK: [[VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 2 1192 // CHECK: [[VAR1_RHS_VOID:%.+]] = load i8*, i8** [[VAR1_RHS_REF]], 1193 // CHECK: [[VAR1_RHS:%.+]] = bitcast i8* [[VAR1_RHS_VOID]] to [[S_INT_TY]]* 1194 // var1_rhs = (S<i{{[0-9]+}}>*)rhs[2]; 1195 // CHECK: [[VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 2 1196 // CHECK: [[VAR1_LHS_VOID:%.+]] = load i8*, i8** [[VAR1_LHS_REF]], 1197 // CHECK: [[VAR1_LHS:%.+]] = bitcast i8* [[VAR1_LHS_VOID]] to [[S_INT_TY]]* 1198 1199 // t_var1_lhs = (i{{[0-9]+}}*)lhs[3]; 1200 // CHECK: [[T_VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 3 1201 // CHECK: [[T_VAR1_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_RHS_REF]], 1202 // CHECK: [[T_VAR1_RHS:%.+]] = bitcast i8* [[T_VAR1_RHS_VOID]] to i{{[0-9]+}}* 1203 // t_var1_rhs = (i{{[0-9]+}}*)rhs[3]; 1204 // CHECK: [[T_VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 3 1205 // CHECK: [[T_VAR1_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_LHS_REF]], 1206 // CHECK: [[T_VAR1_LHS:%.+]] = bitcast i8* [[T_VAR1_LHS_VOID]] to i{{[0-9]+}}* 1207 1208 // t_var_lhs += t_var_rhs; 1209 // CHECK: [[T_VAR_LHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_LHS]], 1210 // CHECK: [[T_VAR_RHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_RHS]], 1211 // CHECK: [[UP:%.+]] = add nsw i{{[0-9]+}} [[T_VAR_LHS_VAL]], [[T_VAR_RHS_VAL]] 1212 // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR_LHS]], 1213 1214 // var_lhs = var_lhs.operator &(var_rhs); 1215 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_LHS]], [[S_INT_TY]]* dereferenceable(4) [[VAR_RHS]]) 1216 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_LHS]] to i8* 1217 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* 1218 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 1219 1220 // var1_lhs = var1_lhs.operator &&(var1_rhs); 1221 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_LHS]]) 1222 // CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 1223 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]] 1224 // CHECK: [[TRUE]] 1225 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_RHS]]) 1226 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 1227 // CHECK: br label %[[END2]] 1228 // CHECK: [[END2]] 1229 // CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ] 1230 // CHECK: [[CONV:%.+]] = zext i1 [[COND_LVALUE]] to i32 1231 // CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]]) 1232 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_LHS]] to i8* 1233 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* 1234 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) 1235 1236 // t_var1_lhs = min(t_var1_lhs, t_var1_rhs); 1237 // CHECK: [[T_VAR1_LHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_LHS]], 1238 // CHECK: [[T_VAR1_RHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_RHS]], 1239 // CHECK: [[CMP:%.+]] = icmp slt i{{[0-9]+}} [[T_VAR1_LHS_VAL]], [[T_VAR1_RHS_VAL]] 1240 // CHECK: br i1 [[CMP]] 1241 // CHECK: [[UP:%.+]] = phi i32 1242 // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR1_LHS]], 1243 // CHECK: ret void 1244 1245 #endif 1246 1247