Pirama Arumuga Nainar | 58878f8 | 2015-05-06 11:48:57 -0700 | [diff] [blame^] | 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 | g = 1; |
| 71 | // LAMBDA: store volatile i{{[0-9]+}} 1, i{{[0-9]+}}* [[G_PRIVATE_ADDR]], |
| 72 | // LAMBDA: [[G_PRIVATE_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG:%.+]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 |
| 73 | // LAMBDA: store i{{[0-9]+}}* [[G_PRIVATE_ADDR]], i{{[0-9]+}}** [[G_PRIVATE_ADDR_REF]] |
| 74 | // LAMBDA: call void [[INNER_LAMBDA:@.+]](%{{.+}}* [[ARG]]) |
| 75 | |
| 76 | // LAMBDA: [[G_PRIV_REF:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[RED_LIST]], i32 0, i32 0 |
| 77 | // LAMBDA: [[BITCAST:%.+]] = bitcast i32* [[G_PRIVATE_ADDR]] to i8* |
| 78 | // LAMBDA: store i8* [[BITCAST]], i8** [[G_PRIV_REF]], |
| 79 | // LAMBDA: call i32 @__kmpc_reduce_nowait( |
| 80 | // LAMBDA: switch i32 %{{.+}}, label %[[REDUCTION_DONE:.+]] [ |
| 81 | // LAMBDA: i32 1, label %[[CASE1:.+]] |
| 82 | // LAMBDA: i32 2, label %[[CASE2:.+]] |
| 83 | // LAMBDA: [[CASE1]] |
| 84 | // LAMBDA: [[G_VAL:%.+]] = load i32, i32* [[G_REF]] |
| 85 | // LAMBDA: [[G_PRIV_VAL:%.+]] = load i32, i32* [[G_PRIVATE_ADDR]] |
| 86 | // LAMBDA: [[ADD:%.+]] = add nsw i32 [[G_VAL]], [[G_PRIV_VAL]] |
| 87 | // LAMBDA: store i32 [[ADD]], i32* [[G_REF]] |
| 88 | // LAMBDA: call void @__kmpc_end_reduce_nowait( |
| 89 | // LAMBDA: br label %[[REDUCTION_DONE]] |
| 90 | // LAMBDA: [[CASE2]] |
| 91 | // LAMBDA: [[G_PRIV_VAL:%.+]] = load i32, i32* [[G_PRIVATE_ADDR]] |
| 92 | // LAMBDA: atomicrmw add i32* [[G_REF]], i32 [[G_PRIV_VAL]] monotonic |
| 93 | // LAMBDA: br label %[[REDUCTION_DONE]] |
| 94 | // LAMBDA: [[REDUCTION_DONE]] |
| 95 | // LAMBDA: ret void |
| 96 | [&]() { |
| 97 | // LAMBDA: define {{.+}} void [[INNER_LAMBDA]](%{{.+}}* [[ARG_PTR:%.+]]) |
| 98 | // LAMBDA: store %{{.+}}* [[ARG_PTR]], %{{.+}}** [[ARG_PTR_REF:%.+]], |
| 99 | g = 2; |
| 100 | // LAMBDA: [[ARG_PTR:%.+]] = load %{{.+}}*, %{{.+}}** [[ARG_PTR_REF]] |
| 101 | // LAMBDA: [[G_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 |
| 102 | // LAMBDA: [[G_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[G_PTR_REF]] |
| 103 | // LAMBDA: store volatile i{{[0-9]+}} 2, i{{[0-9]+}}* [[G_REF]] |
| 104 | }(); |
| 105 | } |
| 106 | }(); |
| 107 | return 0; |
| 108 | #elif defined(BLOCKS) |
| 109 | // BLOCKS: [[G:@.+]] = global i{{[0-9]+}} 1212, |
| 110 | // BLOCKS-LABEL: @main |
| 111 | // BLOCKS: call void {{%.+}}(i8 |
| 112 | ^{ |
| 113 | // BLOCKS: define{{.*}} internal{{.*}} void {{.+}}(i8* |
| 114 | // BLOCKS: [[G_LOCAL_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[AGG_CAPTURED:%.+]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 |
| 115 | // BLOCKS: store i{{[0-9]+}}* [[G]], i{{[0-9]+}}** [[G_LOCAL_REF]] |
| 116 | // BLOCKS: [[ARG:%.+]] = bitcast %{{.+}}* [[AGG_CAPTURED]] to i8* |
| 117 | // BLOCKS: call void {{.+}} @__kmpc_fork_call({{.+}}, i32 1, {{.+}}* [[OMP_REGION:@.+]] to {{.+}}, i8* [[ARG]]) |
| 118 | #pragma omp parallel reduction(-:g) |
| 119 | { |
| 120 | // BLOCKS: define{{.*}} internal{{.*}} void [[OMP_REGION]](i32* %{{.+}}, i32* %{{.+}}, %{{.+}}* [[ARG:%.+]]) |
| 121 | // BLOCKS: [[G_PRIVATE_ADDR:%.+]] = alloca i{{[0-9]+}}, |
| 122 | |
| 123 | // Reduction list for runtime. |
| 124 | // BLOCKS: [[RED_LIST:%.+]] = alloca [1 x i8*], |
| 125 | |
| 126 | // BLOCKS: store %{{.+}}* [[ARG]], %{{.+}}** [[ARG_REF:%.+]], |
| 127 | // BLOCKS: [[ARG:%.+]] = load %{{.+}}*, %{{.+}}** [[ARG_REF]] |
| 128 | // BLOCKS: [[G_REF_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 |
| 129 | // BLOCKS: [[G_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[G_REF_ADDR]] |
| 130 | // BLOCKS: store i{{[0-9]+}} 0, i{{[0-9]+}}* [[G_PRIVATE_ADDR]] |
| 131 | g = 1; |
| 132 | // BLOCKS: store volatile i{{[0-9]+}} 1, i{{[0-9]+}}* [[G_PRIVATE_ADDR]], |
| 133 | // BLOCKS-NOT: [[G]]{{[[^:word:]]}} |
| 134 | // BLOCKS: i{{[0-9]+}}* [[G_PRIVATE_ADDR]] |
| 135 | // BLOCKS-NOT: [[G]]{{[[^:word:]]}} |
| 136 | // BLOCKS: call void {{%.+}}(i8 |
| 137 | |
| 138 | // BLOCKS: [[G_PRIV_REF:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[RED_LIST]], i32 0, i32 0 |
| 139 | // BLOCKS: [[BITCAST:%.+]] = bitcast i32* [[G_PRIVATE_ADDR]] to i8* |
| 140 | // BLOCKS: store i8* [[BITCAST]], i8** [[G_PRIV_REF]], |
| 141 | // BLOCKS: call i32 @__kmpc_reduce_nowait( |
| 142 | // BLOCKS: switch i32 %{{.+}}, label %[[REDUCTION_DONE:.+]] [ |
| 143 | // BLOCKS: i32 1, label %[[CASE1:.+]] |
| 144 | // BLOCKS: i32 2, label %[[CASE2:.+]] |
| 145 | // BLOCKS: [[CASE1]] |
| 146 | // BLOCKS: [[G_VAL:%.+]] = load i32, i32* [[G_REF]] |
| 147 | // BLOCKS: [[G_PRIV_VAL:%.+]] = load i32, i32* [[G_PRIVATE_ADDR]] |
| 148 | // BLOCKS: [[ADD:%.+]] = add nsw i32 [[G_VAL]], [[G_PRIV_VAL]] |
| 149 | // BLOCKS: store i32 [[ADD]], i32* [[G_REF]] |
| 150 | // BLOCKS: call void @__kmpc_end_reduce_nowait( |
| 151 | // BLOCKS: br label %[[REDUCTION_DONE]] |
| 152 | // BLOCKS: [[CASE2]] |
| 153 | // BLOCKS: [[G_PRIV_VAL:%.+]] = load i32, i32* [[G_PRIVATE_ADDR]] |
| 154 | // BLOCKS: atomicrmw add i32* [[G_REF]], i32 [[G_PRIV_VAL]] monotonic |
| 155 | // BLOCKS: br label %[[REDUCTION_DONE]] |
| 156 | // BLOCKS: [[REDUCTION_DONE]] |
| 157 | // BLOCKS: ret void |
| 158 | ^{ |
| 159 | // BLOCKS: define {{.+}} void {{@.+}}(i8* |
| 160 | g = 2; |
| 161 | // BLOCKS-NOT: [[G]]{{[[^:word:]]}} |
| 162 | // BLOCKS: store volatile i{{[0-9]+}} 2, i{{[0-9]+}}* |
| 163 | // BLOCKS-NOT: [[G]]{{[[^:word:]]}} |
| 164 | // BLOCKS: ret |
| 165 | }(); |
| 166 | } |
| 167 | }(); |
| 168 | return 0; |
| 169 | #else |
| 170 | S<float> test; |
| 171 | float t_var = 0, t_var1; |
| 172 | int vec[] = {1, 2}; |
| 173 | S<float> s_arr[] = {1, 2}; |
| 174 | S<float> var(3), var1; |
| 175 | #pragma omp parallel reduction(+:t_var) reduction(&:var) reduction(&& : var1) reduction(min: t_var1) |
| 176 | { |
| 177 | vec[0] = t_var; |
| 178 | s_arr[0] = var; |
| 179 | } |
| 180 | return tmain<int>(); |
| 181 | #endif |
| 182 | } |
| 183 | |
| 184 | // CHECK: define {{.*}}i{{[0-9]+}} @main() |
| 185 | // CHECK: [[TEST:%.+]] = alloca [[S_FLOAT_TY]], |
| 186 | // CHECK: call {{.*}} [[S_FLOAT_TY_CONSTR:@.+]]([[S_FLOAT_TY]]* [[TEST]]) |
| 187 | // CHECK: %{{.+}} = bitcast [[CAP_MAIN_TY]]* |
| 188 | // 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 |
| 189 | // CHECK: = call {{.*}}i{{.+}} [[TMAIN_INT:@.+]]() |
| 190 | // CHECK: call {{.*}} [[S_FLOAT_TY_DESTR:@.+]]([[S_FLOAT_TY]]* |
| 191 | // CHECK: ret |
| 192 | // |
| 193 | // CHECK: define internal void [[MAIN_MICROTASK]](i{{[0-9]+}}* [[GTID_ADDR:%.+]], i{{[0-9]+}}* %{{.+}}, [[CAP_MAIN_TY]]* %{{.+}}) |
| 194 | // CHECK: [[T_VAR_PRIV:%.+]] = alloca float, |
| 195 | // CHECK: [[VAR_PRIV:%.+]] = alloca [[S_FLOAT_TY]], |
| 196 | // CHECK: [[VAR1_PRIV:%.+]] = alloca [[S_FLOAT_TY]], |
| 197 | // CHECK: [[T_VAR1_PRIV:%.+]] = alloca float, |
| 198 | |
| 199 | // Reduction list for runtime. |
| 200 | // CHECK: [[RED_LIST:%.+]] = alloca [4 x i8*], |
| 201 | |
| 202 | // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]], |
| 203 | |
| 204 | // CHECK: [[T_VAR_PTR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}} |
| 205 | // CHECK: [[T_VAR_REF:%.+]] = load float*, float** [[T_VAR_PTR_REF]], |
| 206 | // For + reduction operation initial value of private variable is 0. |
| 207 | // CHECK: store float 0.0{{.+}}, float* [[T_VAR_PRIV]], |
| 208 | |
| 209 | // CHECK: [[VAR_PTR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}} |
| 210 | // CHECK: [[VAR_REF:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[VAR_PTR_REF:%.+]], |
| 211 | // For & reduction operation initial value of private variable is ones in all bits. |
| 212 | // CHECK: call {{.*}} [[S_FLOAT_TY_CONSTR:@.+]]([[S_FLOAT_TY]]* [[VAR_PRIV]]) |
| 213 | |
| 214 | // CHECK: [[VAR1_PTR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}} |
| 215 | // CHECK: [[VAR1_REF:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[VAR_PTR_REF:%.+]], |
| 216 | // For && reduction operation initial value of private variable is 1.0. |
| 217 | // CHECK: call {{.*}} [[S_FLOAT_TY_CONSTR:@.+]]([[S_FLOAT_TY]]* [[VAR1_PRIV]]) |
| 218 | |
| 219 | // CHECK: [[T_VAR1_PTR_REF:%.+]] = getelementptr inbounds [[CAP_MAIN_TY]], [[CAP_MAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}} |
| 220 | // CHECK: [[T_VAR1_REF:%.+]] = load float*, float** [[T_VAR1_PTR_REF]], |
| 221 | // For min reduction operation initial value of private variable is largest repesentable value. |
| 222 | // CHECK: store float 0x47EFFFFFE0000000, float* [[T_VAR1_PRIV]], |
| 223 | |
| 224 | // Skip checks for internal operations. |
| 225 | |
| 226 | // void *RedList[<n>] = {<ReductionVars>[0], ..., <ReductionVars>[<n>-1]}; |
| 227 | |
| 228 | // CHECK: [[T_VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 0 |
| 229 | // CHECK: [[BITCAST:%.+]] = bitcast float* [[T_VAR_PRIV]] to i8* |
| 230 | // CHECK: store i8* [[BITCAST]], i8** [[T_VAR_PRIV_REF]], |
| 231 | // CHECK: [[VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 1 |
| 232 | // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_PRIV]] to i8* |
| 233 | // CHECK: store i8* [[BITCAST]], i8** [[VAR_PRIV_REF]], |
| 234 | // CHECK: [[VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 2 |
| 235 | // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_PRIV]] to i8* |
| 236 | // CHECK: store i8* [[BITCAST]], i8** [[VAR1_PRIV_REF]], |
| 237 | // CHECK: [[T_VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 3 |
| 238 | // CHECK: [[BITCAST:%.+]] = bitcast float* [[T_VAR1_PRIV]] to i8* |
| 239 | // CHECK: store i8* [[BITCAST]], i8** [[T_VAR1_PRIV_REF]], |
| 240 | |
| 241 | // res = __kmpc_reduce_nowait(<loc>, <gtid>, <n>, sizeof(RedList), RedList, reduce_func, &<lock>); |
| 242 | |
| 243 | // CHECK: [[GTID_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_ADDR]] |
| 244 | // CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_REF]] |
| 245 | // CHECK: [[BITCAST:%.+]] = bitcast [4 x i8*]* [[RED_LIST]] to i8* |
| 246 | // 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]]) |
| 247 | |
| 248 | // switch(res) |
| 249 | // CHECK: switch i32 [[RES]], label %[[RED_DONE:.+]] [ |
| 250 | // CHECK: i32 1, label %[[CASE1:.+]] |
| 251 | // CHECK: i32 2, label %[[CASE2:.+]] |
| 252 | // CHECK: ] |
| 253 | |
| 254 | // case 1: |
| 255 | // t_var += t_var_reduction; |
| 256 | // CHECK: [[T_VAR_VAL:%.+]] = load float, float* [[T_VAR_REF]], |
| 257 | // CHECK: [[T_VAR_PRIV_VAL:%.+]] = load float, float* [[T_VAR_PRIV]], |
| 258 | // CHECK: [[UP:%.+]] = fadd float [[T_VAR_VAL]], [[T_VAR_PRIV_VAL]] |
| 259 | // CHECK: store float [[UP]], float* [[T_VAR_REF]], |
| 260 | |
| 261 | // var = var.operator &(var_reduction); |
| 262 | // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_REF]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_PRIV]]) |
| 263 | // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_REF]] to i8* |
| 264 | // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8* |
| 265 | // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) |
| 266 | |
| 267 | // var1 = var1.operator &&(var1_reduction); |
| 268 | // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_REF]]) |
| 269 | // CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 |
| 270 | // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]] |
| 271 | // CHECK: [[TRUE]] |
| 272 | // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_PRIV]]) |
| 273 | // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 |
| 274 | // CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]] |
| 275 | // CHECK: [[TRUE2]] |
| 276 | // CHECK: br label %[[END2:.+]] |
| 277 | // CHECK: [[FALSE2]] |
| 278 | // CHECK: br label %[[END2]] |
| 279 | // CHECK: [[END2]] |
| 280 | // CHECK: [[COND_LVALUE:%.+]] = phi [[S_FLOAT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ] |
| 281 | // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_REF]] to i8* |
| 282 | // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8* |
| 283 | // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) |
| 284 | |
| 285 | // t_var1 = min(t_var1, t_var1_reduction); |
| 286 | // CHECK: [[T_VAR1_VAL:%.+]] = load float, float* [[T_VAR1_REF]], |
| 287 | // CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load float, float* [[T_VAR1_PRIV]], |
| 288 | // CHECK: [[CMP:%.+]] = fcmp olt float [[T_VAR1_VAL]], [[T_VAR1_PRIV_VAL]] |
| 289 | // CHECK: [[UP:%.+]] = uitofp i1 [[CMP]] to float |
| 290 | // CHECK: store float [[UP]], float* [[T_VAR1_REF]], |
| 291 | |
| 292 | // __kmpc_end_reduce_nowait(<loc>, <gtid>, &<lock>); |
| 293 | // CHECK: call void @__kmpc_end_reduce_nowait(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], [8 x i32]* [[REDUCTION_LOCK]]) |
| 294 | |
| 295 | // break; |
| 296 | // CHECK: br label %[[RED_DONE]] |
| 297 | |
| 298 | // case 2: |
| 299 | // t_var += t_var_reduction; |
| 300 | // CHECK: load float, float* [[T_VAR_PRIV]] |
| 301 | // CHECK: [[T_VAR_REF_INT:%.+]] = bitcast float* [[T_VAR_REF]] to i32* |
| 302 | // CHECK: load atomic i32, i32* [[T_VAR_REF_INT]] monotonic, |
| 303 | // CHECK: [[OLD1:%.+]] = bitcast i32 %{{.+}} to float |
| 304 | // CHECK: br label %[[CONT:.+]] |
| 305 | // CHECK: [[CONT]] |
| 306 | // CHECK: [[ORIG_OLD:%.+]] = phi float [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %[[CONT]] ] |
| 307 | // CHECK: [[UP:%.+]] = fadd float |
| 308 | // CHECK: [[ORIG_OLD_INT:%.+]] = bitcast float [[ORIG_OLD]] to i32 |
| 309 | // CHECK: [[UP_INT:%.+]] = bitcast float [[UP]] to i32 |
| 310 | // CHECK: [[T_VAR_REF_INT:%.+]] = bitcast float* [[T_VAR_REF]] to i32* |
| 311 | // CHECK: [[RES:%.+]] = cmpxchg i32* [[T_VAR_REF_INT]], i32 [[ORIG_OLD_INT]], i32 [[UP_INT]] monotonic monotonic |
| 312 | // CHECK: [[OLD_VAL_INT:%.+]] = extractvalue { i32, i1 } [[RES]], 0 |
| 313 | // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1 |
| 314 | // CHECK: [[OLD2]] = bitcast i32 [[OLD_VAL_INT]] to float |
| 315 | // CHECK: br i1 [[SUCCESS_FAIL]], label %[[ATOMIC_DONE:.+]], label %[[CONT]] |
| 316 | // CHECK: [[ATOMIC_DONE]] |
| 317 | |
| 318 | // var = var.operator &(var_reduction); |
| 319 | // CHECK: call void @__kmpc_critical( |
| 320 | // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_REF]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_PRIV]]) |
| 321 | // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_REF]] to i8* |
| 322 | // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8* |
| 323 | // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) |
| 324 | // CHECK: call void @__kmpc_end_critical( |
| 325 | |
| 326 | // var1 = var1.operator &&(var1_reduction); |
| 327 | // CHECK: call void @__kmpc_critical( |
| 328 | // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_REF]]) |
| 329 | // CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 |
| 330 | // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]] |
| 331 | // CHECK: [[TRUE]] |
| 332 | // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_PRIV]]) |
| 333 | // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 |
| 334 | // CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]] |
| 335 | // CHECK: [[TRUE2]] |
| 336 | // CHECK: br label %[[END2:.+]] |
| 337 | // CHECK: [[FALSE2]] |
| 338 | // CHECK: br label %[[END2]] |
| 339 | // CHECK: [[END2]] |
| 340 | // CHECK: [[COND_LVALUE:%.+]] = phi [[S_FLOAT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ] |
| 341 | // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_REF]] to i8* |
| 342 | // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8* |
| 343 | // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) |
| 344 | // CHECK: call void @__kmpc_end_critical( |
| 345 | |
| 346 | // t_var1 = min(t_var1, t_var1_reduction); |
| 347 | // CHECK: load float, float* [[T_VAR1_PRIV]] |
| 348 | // CHECK: [[T_VAR1_REF_INT:%.+]] = bitcast float* [[T_VAR1_REF]] to i32* |
| 349 | // CHECK: load atomic i32, i32* [[T_VAR1_REF_INT]] monotonic, |
| 350 | // CHECK: [[OLD1:%.+]] = bitcast i32 %{{.+}} to float |
| 351 | // CHECK: br label %[[CONT:.+]] |
| 352 | // CHECK: [[CONT]] |
| 353 | // CHECK: [[ORIG_OLD:%.+]] = phi float [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %[[CONT]] ] |
| 354 | // CHECK: [[CMP:%.+]] = fcmp olt float |
| 355 | // CHECK: [[UP:%.+]] = uitofp i1 [[CMP]] to float |
| 356 | // CHECK: [[ORIG_OLD_INT:%.+]] = bitcast float [[ORIG_OLD]] to i32 |
| 357 | // CHECK: [[UP_INT:%.+]] = bitcast float [[UP]] to i32 |
| 358 | // CHECK: [[T_VAR1_REF_INT:%.+]] = bitcast float* [[T_VAR1_REF]] to i32* |
| 359 | // CHECK: [[RES:%.+]] = cmpxchg i32* [[T_VAR1_REF_INT]], i32 [[ORIG_OLD_INT]], i32 [[UP_INT]] monotonic monotonic |
| 360 | // CHECK: [[OLD_VAL_INT:%.+]] = extractvalue { i32, i1 } [[RES]], 0 |
| 361 | // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1 |
| 362 | // CHECK: [[OLD2]] = bitcast i32 [[OLD_VAL_INT]] to float |
| 363 | // CHECK: br i1 [[SUCCESS_FAIL]], label %[[ATOMIC_DONE:.+]], label %[[CONT]] |
| 364 | // CHECK: [[ATOMIC_DONE]] |
| 365 | |
| 366 | // break; |
| 367 | // CHECK: br label %[[RED_DONE]] |
| 368 | // CHECK: [[RED_DONE]] |
| 369 | // CHECK: call i32 @__kmpc_cancel_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i{{[0-9]+}} [[GTID]]) |
| 370 | |
| 371 | // CHECK-DAG: call {{.*}} [[S_FLOAT_TY_DESTR]]([[S_FLOAT_TY]]* [[VAR_PRIV]]) |
| 372 | // CHECK-DAG: call {{.*}} [[S_FLOAT_TY_DESTR]]([[S_FLOAT_TY]]* |
| 373 | // CHECK: ret void |
| 374 | |
| 375 | // void reduce_func(void *lhs[<n>], void *rhs[<n>]) { |
| 376 | // *(Type0*)lhs[0] = ReductionOperation0(*(Type0*)lhs[0], *(Type0*)rhs[0]); |
| 377 | // ... |
| 378 | // *(Type<n>-1*)lhs[<n>-1] = ReductionOperation<n>-1(*(Type<n>-1*)lhs[<n>-1], |
| 379 | // *(Type<n>-1*)rhs[<n>-1]); |
| 380 | // } |
| 381 | // CHECK: define internal void [[REDUCTION_FUNC]](i8*, i8*) |
| 382 | // t_var_lhs = (float*)lhs[0]; |
| 383 | // CHECK: [[T_VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS:%.+]], i32 0, i32 0 |
| 384 | // CHECK: [[T_VAR_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR_RHS_REF]], |
| 385 | // CHECK: [[T_VAR_RHS:%.+]] = bitcast i8* [[T_VAR_RHS_VOID]] to float* |
| 386 | // t_var_rhs = (float*)rhs[0]; |
| 387 | // CHECK: [[T_VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS:%.+]], i32 0, i32 0 |
| 388 | // CHECK: [[T_VAR_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR_LHS_REF]], |
| 389 | // CHECK: [[T_VAR_LHS:%.+]] = bitcast i8* [[T_VAR_LHS_VOID]] to float* |
| 390 | |
| 391 | // var_lhs = (S<float>*)lhs[1]; |
| 392 | // CHECK: [[VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i32 0, i32 1 |
| 393 | // CHECK: [[VAR_RHS_VOID:%.+]] = load i8*, i8** [[VAR_RHS_REF]], |
| 394 | // CHECK: [[VAR_RHS:%.+]] = bitcast i8* [[VAR_RHS_VOID]] to [[S_FLOAT_TY]]* |
| 395 | // var_rhs = (S<float>*)rhs[1]; |
| 396 | // CHECK: [[VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i32 0, i32 1 |
| 397 | // CHECK: [[VAR_LHS_VOID:%.+]] = load i8*, i8** [[VAR_LHS_REF]], |
| 398 | // CHECK: [[VAR_LHS:%.+]] = bitcast i8* [[VAR_LHS_VOID]] to [[S_FLOAT_TY]]* |
| 399 | |
| 400 | // var1_lhs = (S<float>*)lhs[2]; |
| 401 | // CHECK: [[VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i32 0, i32 2 |
| 402 | // CHECK: [[VAR1_RHS_VOID:%.+]] = load i8*, i8** [[VAR1_RHS_REF]], |
| 403 | // CHECK: [[VAR1_RHS:%.+]] = bitcast i8* [[VAR1_RHS_VOID]] to [[S_FLOAT_TY]]* |
| 404 | // var1_rhs = (S<float>*)rhs[2]; |
| 405 | // CHECK: [[VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i32 0, i32 2 |
| 406 | // CHECK: [[VAR1_LHS_VOID:%.+]] = load i8*, i8** [[VAR1_LHS_REF]], |
| 407 | // CHECK: [[VAR1_LHS:%.+]] = bitcast i8* [[VAR1_LHS_VOID]] to [[S_FLOAT_TY]]* |
| 408 | |
| 409 | // t_var1_lhs = (float*)lhs[3]; |
| 410 | // CHECK: [[T_VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i32 0, i32 3 |
| 411 | // CHECK: [[T_VAR1_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_RHS_REF]], |
| 412 | // CHECK: [[T_VAR1_RHS:%.+]] = bitcast i8* [[T_VAR1_RHS_VOID]] to float* |
| 413 | // t_var1_rhs = (float*)rhs[3]; |
| 414 | // CHECK: [[T_VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i32 0, i32 3 |
| 415 | // CHECK: [[T_VAR1_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_LHS_REF]], |
| 416 | // CHECK: [[T_VAR1_LHS:%.+]] = bitcast i8* [[T_VAR1_LHS_VOID]] to float* |
| 417 | |
| 418 | // t_var_lhs += t_var_rhs; |
| 419 | // CHECK: [[T_VAR_LHS_VAL:%.+]] = load float, float* [[T_VAR_LHS]], |
| 420 | // CHECK: [[T_VAR_RHS_VAL:%.+]] = load float, float* [[T_VAR_RHS]], |
| 421 | // CHECK: [[UP:%.+]] = fadd float [[T_VAR_LHS_VAL]], [[T_VAR_RHS_VAL]] |
| 422 | // CHECK: store float [[UP]], float* [[T_VAR_LHS]], |
| 423 | |
| 424 | // var_lhs = var_lhs.operator &(var_rhs); |
| 425 | // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_LHS]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_RHS]]) |
| 426 | // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_LHS]] to i8* |
| 427 | // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8* |
| 428 | // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) |
| 429 | |
| 430 | // var1_lhs = var1_lhs.operator &&(var1_rhs); |
| 431 | // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_LHS]]) |
| 432 | // CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 |
| 433 | // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]] |
| 434 | // CHECK: [[TRUE]] |
| 435 | // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_RHS]]) |
| 436 | // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0 |
| 437 | // CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]] |
| 438 | // CHECK: [[TRUE2]] |
| 439 | // CHECK: br label %[[END2:.+]] |
| 440 | // CHECK: [[FALSE2]] |
| 441 | // CHECK: br label %[[END2]] |
| 442 | // CHECK: [[END2]] |
| 443 | // CHECK: [[COND_LVALUE:%.+]] = phi [[S_FLOAT_TY]]* [ [[VAR1_LHS]], %[[TRUE2]] ], [ [[VAR1_RHS]], %[[FALSE2]] ] |
| 444 | // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_LHS]] to i8* |
| 445 | // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8* |
| 446 | // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) |
| 447 | |
| 448 | // t_var1_lhs = min(t_var1_lhs, t_var1_rhs); |
| 449 | // CHECK: [[T_VAR1_LHS_VAL:%.+]] = load float, float* [[T_VAR1_LHS]], |
| 450 | // CHECK: [[T_VAR1_RHS_VAL:%.+]] = load float, float* [[T_VAR1_RHS]], |
| 451 | // CHECK: [[CMP:%.+]] = fcmp olt float [[T_VAR1_LHS_VAL]], [[T_VAR1_RHS_VAL]] |
| 452 | // CHECK: [[UP:%.+]] = uitofp i1 [[CMP]] to float |
| 453 | // CHECK: store float [[UP]], float* [[T_VAR1_LHS]], |
| 454 | // CHECK: ret void |
| 455 | |
| 456 | // CHECK: define {{.*}} i{{[0-9]+}} [[TMAIN_INT]]() |
| 457 | // CHECK: [[TEST:%.+]] = alloca [[S_INT_TY]], |
| 458 | // CHECK: call {{.*}} [[S_INT_TY_CONSTR:@.+]]([[S_INT_TY]]* [[TEST]]) |
| 459 | // CHECK: %{{.+}} = bitcast [[CAP_TMAIN_TY]]* |
| 460 | // 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 |
| 461 | // CHECK: call {{.*}} [[S_INT_TY_DESTR:@.+]]([[S_INT_TY]]* |
| 462 | // CHECK: ret |
| 463 | // |
| 464 | // CHECK: define internal void [[TMAIN_MICROTASK]](i{{[0-9]+}}* [[GTID_ADDR:%.+]], i{{[0-9]+}}* %{{.+}}, [[CAP_TMAIN_TY]]* %{{.+}}) |
| 465 | // CHECK: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}}, |
| 466 | // CHECK: [[VAR_PRIV:%.+]] = alloca [[S_INT_TY]], |
| 467 | // CHECK: [[VAR1_PRIV:%.+]] = alloca [[S_INT_TY]], |
| 468 | // CHECK: [[T_VAR1_PRIV:%.+]] = alloca i{{[0-9]+}}, |
| 469 | |
| 470 | // Reduction list for runtime. |
| 471 | // CHECK: [[RED_LIST:%.+]] = alloca [4 x i8*], |
| 472 | |
| 473 | // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]], |
| 474 | |
| 475 | // CHECK: [[T_VAR_PTR_REF:%.+]] = getelementptr inbounds [[CAP_TMAIN_TY]], [[CAP_TMAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}} |
| 476 | // CHECK: [[T_VAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[T_VAR_PTR_REF]], |
| 477 | // For + reduction operation initial value of private variable is 0. |
| 478 | // CHECK: store i{{[0-9]+}} 0, i{{[0-9]+}}* [[T_VAR_PRIV]], |
| 479 | |
| 480 | // CHECK: [[VAR_PTR_REF:%.+]] = getelementptr inbounds [[CAP_TMAIN_TY]], [[CAP_TMAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}} |
| 481 | // CHECK: [[VAR_REF:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[VAR_PTR_REF:%.+]], |
| 482 | // For & reduction operation initial value of private variable is ones in all bits. |
| 483 | // CHECK: call {{.*}} [[S_INT_TY_CONSTR:@.+]]([[S_INT_TY]]* [[VAR_PRIV]]) |
| 484 | |
| 485 | // CHECK: [[VAR1_PTR_REF:%.+]] = getelementptr inbounds [[CAP_TMAIN_TY]], [[CAP_TMAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}} |
| 486 | // CHECK: [[VAR1_REF:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[VAR_PTR_REF:%.+]], |
| 487 | // For && reduction operation initial value of private variable is 1.0. |
| 488 | // CHECK: call {{.*}} [[S_INT_TY_CONSTR:@.+]]([[S_INT_TY]]* [[VAR1_PRIV]]) |
| 489 | |
| 490 | // CHECK: [[T_VAR1_PTR_REF:%.+]] = getelementptr inbounds [[CAP_TMAIN_TY]], [[CAP_TMAIN_TY]]* %{{.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} {{[0-9]+}} |
| 491 | // CHECK: [[T_VAR1_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[T_VAR1_PTR_REF]], |
| 492 | // For min reduction operation initial value of private variable is largest repesentable value. |
| 493 | // CHECK: store i{{[0-9]+}} 2147483647, i{{[0-9]+}}* [[T_VAR1_PRIV]], |
| 494 | |
| 495 | // Skip checks for internal operations. |
| 496 | |
| 497 | // void *RedList[<n>] = {<ReductionVars>[0], ..., <ReductionVars>[<n>-1]}; |
| 498 | |
| 499 | // CHECK: [[T_VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 0 |
| 500 | // CHECK: [[BITCAST:%.+]] = bitcast i{{[0-9]+}}* [[T_VAR_PRIV]] to i8* |
| 501 | // CHECK: store i8* [[BITCAST]], i8** [[T_VAR_PRIV_REF]], |
| 502 | // CHECK: [[VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 1 |
| 503 | // CHECK: [[BITCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR_PRIV]] to i8* |
| 504 | // CHECK: store i8* [[BITCAST]], i8** [[VAR_PRIV_REF]], |
| 505 | // CHECK: [[VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 2 |
| 506 | // CHECK: [[BITCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_PRIV]] to i8* |
| 507 | // CHECK: store i8* [[BITCAST]], i8** [[VAR1_PRIV_REF]], |
| 508 | // CHECK: [[T_VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i32 0, i32 3 |
| 509 | // CHECK: [[BITCAST:%.+]] = bitcast i{{[0-9]+}}* [[T_VAR1_PRIV]] to i8* |
| 510 | // CHECK: store i8* [[BITCAST]], i8** [[T_VAR1_PRIV_REF]], |
| 511 | |
| 512 | // res = __kmpc_reduce_nowait(<loc>, <gtid>, <n>, sizeof(RedList), RedList, reduce_func, &<lock>); |
| 513 | |
| 514 | // CHECK: [[GTID_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_ADDR]] |
| 515 | // CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_REF]] |
| 516 | // CHECK: [[BITCAST:%.+]] = bitcast [4 x i8*]* [[RED_LIST]] to i8* |
| 517 | // 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]]) |
| 518 | |
| 519 | // switch(res) |
| 520 | // CHECK: switch i32 [[RES]], label %[[RED_DONE:.+]] [ |
| 521 | // CHECK: i32 1, label %[[CASE1:.+]] |
| 522 | // CHECK: i32 2, label %[[CASE2:.+]] |
| 523 | // CHECK: ] |
| 524 | |
| 525 | // case 1: |
| 526 | // t_var += t_var_reduction; |
| 527 | // CHECK: [[T_VAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_REF]], |
| 528 | // CHECK: [[T_VAR_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_PRIV]], |
| 529 | // CHECK: [[UP:%.+]] = add nsw i{{[0-9]+}} [[T_VAR_VAL]], [[T_VAR_PRIV_VAL]] |
| 530 | // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR_REF]], |
| 531 | |
| 532 | // var = var.operator &(var_reduction); |
| 533 | // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_REF]], [[S_INT_TY]]* dereferenceable(4) [[VAR_PRIV]]) |
| 534 | // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_REF]] to i8* |
| 535 | // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* |
| 536 | // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) |
| 537 | |
| 538 | // var1 = var1.operator &&(var1_reduction); |
| 539 | // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_REF]]) |
| 540 | // CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 |
| 541 | // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]] |
| 542 | // CHECK: [[TRUE]] |
| 543 | // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_PRIV]]) |
| 544 | // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 |
| 545 | // CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]] |
| 546 | // CHECK: [[TRUE2]] |
| 547 | // CHECK: br label %[[END2:.+]] |
| 548 | // CHECK: [[FALSE2]] |
| 549 | // CHECK: br label %[[END2]] |
| 550 | // CHECK: [[END2]] |
| 551 | // CHECK: [[COND_LVALUE:%.+]] = phi [[S_INT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ] |
| 552 | // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8* |
| 553 | // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* |
| 554 | // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) |
| 555 | |
| 556 | // t_var1 = min(t_var1, t_var1_reduction); |
| 557 | // CHECK: [[T_VAR1_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_REF]], |
| 558 | // CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_PRIV]], |
| 559 | // CHECK: [[CMP:%.+]] = icmp slt i{{[0-9]+}} [[T_VAR1_VAL]], [[T_VAR1_PRIV_VAL]] |
| 560 | // CHECK: [[UP:%.+]] = zext i1 [[CMP]] to i{{[0-9]+}} |
| 561 | // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR1_REF]], |
| 562 | |
| 563 | // __kmpc_end_reduce_nowait(<loc>, <gtid>, &<lock>); |
| 564 | // CHECK: call void @__kmpc_end_reduce_nowait(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], [8 x i32]* [[REDUCTION_LOCK]]) |
| 565 | |
| 566 | // break; |
| 567 | // CHECK: br label %[[RED_DONE]] |
| 568 | |
| 569 | // case 2: |
| 570 | // t_var += t_var_reduction; |
| 571 | // CHECK: [[T_VAR_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_PRIV]] |
| 572 | // CHECK: atomicrmw add i32* [[T_VAR_REF]], i32 [[T_VAR_PRIV_VAL]] monotonic |
| 573 | |
| 574 | // var = var.operator &(var_reduction); |
| 575 | // CHECK: call void @__kmpc_critical( |
| 576 | // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_REF]], [[S_INT_TY]]* dereferenceable(4) [[VAR_PRIV]]) |
| 577 | // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_REF]] to i8* |
| 578 | // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* |
| 579 | // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) |
| 580 | // CHECK: call void @__kmpc_end_critical( |
| 581 | |
| 582 | // var1 = var1.operator &&(var1_reduction); |
| 583 | // CHECK: call void @__kmpc_critical( |
| 584 | // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_REF]]) |
| 585 | // CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 |
| 586 | // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]] |
| 587 | // CHECK: [[TRUE]] |
| 588 | // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_PRIV]]) |
| 589 | // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 |
| 590 | // CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]] |
| 591 | // CHECK: [[TRUE2]] |
| 592 | // CHECK: br label %[[END2:.+]] |
| 593 | // CHECK: [[FALSE2]] |
| 594 | // CHECK: br label %[[END2]] |
| 595 | // CHECK: [[END2]] |
| 596 | // CHECK: [[COND_LVALUE:%.+]] = phi [[S_INT_TY]]* [ [[VAR1_REF]], %[[TRUE2]] ], [ [[VAR1_PRIV]], %[[FALSE2]] ] |
| 597 | // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8* |
| 598 | // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* |
| 599 | // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) |
| 600 | // CHECK: call void @__kmpc_end_critical( |
| 601 | |
| 602 | // t_var1 = min(t_var1, t_var1_reduction); |
| 603 | // CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_PRIV]] |
| 604 | // CHECK: atomicrmw min i32* [[T_VAR1_REF]], i32 [[T_VAR1_PRIV_VAL]] monotonic |
| 605 | |
| 606 | // break; |
| 607 | // CHECK: br label %[[RED_DONE]] |
| 608 | // CHECK: [[RED_DONE]] |
| 609 | // CHECK: call i32 @__kmpc_cancel_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i{{[0-9]+}} [[GTID]]) |
| 610 | |
| 611 | // CHECK-DAG: call {{.*}} [[S_INT_TY_DESTR]]([[S_INT_TY]]* [[VAR_PRIV]]) |
| 612 | // CHECK-DAG: call {{.*}} [[S_INT_TY_DESTR]]([[S_INT_TY]]* |
| 613 | // CHECK: ret void |
| 614 | |
| 615 | // void reduce_func(void *lhs[<n>], void *rhs[<n>]) { |
| 616 | // *(Type0*)lhs[0] = ReductionOperation0(*(Type0*)lhs[0], *(Type0*)rhs[0]); |
| 617 | // ... |
| 618 | // *(Type<n>-1*)lhs[<n>-1] = ReductionOperation<n>-1(*(Type<n>-1*)lhs[<n>-1], |
| 619 | // *(Type<n>-1*)rhs[<n>-1]); |
| 620 | // } |
| 621 | // CHECK: define internal void [[REDUCTION_FUNC]](i8*, i8*) |
| 622 | // t_var_lhs = (i{{[0-9]+}}*)lhs[0]; |
| 623 | // CHECK: [[T_VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS:%.+]], i32 0, i32 0 |
| 624 | // CHECK: [[T_VAR_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR_RHS_REF]], |
| 625 | // CHECK: [[T_VAR_RHS:%.+]] = bitcast i8* [[T_VAR_RHS_VOID]] to i{{[0-9]+}}* |
| 626 | // t_var_rhs = (i{{[0-9]+}}*)rhs[0]; |
| 627 | // CHECK: [[T_VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS:%.+]], i32 0, i32 0 |
| 628 | // CHECK: [[T_VAR_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR_LHS_REF]], |
| 629 | // CHECK: [[T_VAR_LHS:%.+]] = bitcast i8* [[T_VAR_LHS_VOID]] to i{{[0-9]+}}* |
| 630 | |
| 631 | // var_lhs = (S<i{{[0-9]+}}>*)lhs[1]; |
| 632 | // CHECK: [[VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i32 0, i32 1 |
| 633 | // CHECK: [[VAR_RHS_VOID:%.+]] = load i8*, i8** [[VAR_RHS_REF]], |
| 634 | // CHECK: [[VAR_RHS:%.+]] = bitcast i8* [[VAR_RHS_VOID]] to [[S_INT_TY]]* |
| 635 | // var_rhs = (S<i{{[0-9]+}}>*)rhs[1]; |
| 636 | // CHECK: [[VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i32 0, i32 1 |
| 637 | // CHECK: [[VAR_LHS_VOID:%.+]] = load i8*, i8** [[VAR_LHS_REF]], |
| 638 | // CHECK: [[VAR_LHS:%.+]] = bitcast i8* [[VAR_LHS_VOID]] to [[S_INT_TY]]* |
| 639 | |
| 640 | // var1_lhs = (S<i{{[0-9]+}}>*)lhs[2]; |
| 641 | // CHECK: [[VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i32 0, i32 2 |
| 642 | // CHECK: [[VAR1_RHS_VOID:%.+]] = load i8*, i8** [[VAR1_RHS_REF]], |
| 643 | // CHECK: [[VAR1_RHS:%.+]] = bitcast i8* [[VAR1_RHS_VOID]] to [[S_INT_TY]]* |
| 644 | // var1_rhs = (S<i{{[0-9]+}}>*)rhs[2]; |
| 645 | // CHECK: [[VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i32 0, i32 2 |
| 646 | // CHECK: [[VAR1_LHS_VOID:%.+]] = load i8*, i8** [[VAR1_LHS_REF]], |
| 647 | // CHECK: [[VAR1_LHS:%.+]] = bitcast i8* [[VAR1_LHS_VOID]] to [[S_INT_TY]]* |
| 648 | |
| 649 | // t_var1_lhs = (i{{[0-9]+}}*)lhs[3]; |
| 650 | // CHECK: [[T_VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i32 0, i32 3 |
| 651 | // CHECK: [[T_VAR1_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_RHS_REF]], |
| 652 | // CHECK: [[T_VAR1_RHS:%.+]] = bitcast i8* [[T_VAR1_RHS_VOID]] to i{{[0-9]+}}* |
| 653 | // t_var1_rhs = (i{{[0-9]+}}*)rhs[3]; |
| 654 | // CHECK: [[T_VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i32 0, i32 3 |
| 655 | // CHECK: [[T_VAR1_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_LHS_REF]], |
| 656 | // CHECK: [[T_VAR1_LHS:%.+]] = bitcast i8* [[T_VAR1_LHS_VOID]] to i{{[0-9]+}}* |
| 657 | |
| 658 | // t_var_lhs += t_var_rhs; |
| 659 | // CHECK: [[T_VAR_LHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_LHS]], |
| 660 | // CHECK: [[T_VAR_RHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_RHS]], |
| 661 | // CHECK: [[UP:%.+]] = add nsw i{{[0-9]+}} [[T_VAR_LHS_VAL]], [[T_VAR_RHS_VAL]] |
| 662 | // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR_LHS]], |
| 663 | |
| 664 | // var_lhs = var_lhs.operator &(var_rhs); |
| 665 | // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_LHS]], [[S_INT_TY]]* dereferenceable(4) [[VAR_RHS]]) |
| 666 | // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_LHS]] to i8* |
| 667 | // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* |
| 668 | // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) |
| 669 | |
| 670 | // var1_lhs = var1_lhs.operator &&(var1_rhs); |
| 671 | // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_LHS]]) |
| 672 | // CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 |
| 673 | // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[FALSE:.+]] |
| 674 | // CHECK: [[TRUE]] |
| 675 | // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_RHS]]) |
| 676 | // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0 |
| 677 | // CHECK: br i1 [[VAR1_REDUCTION_BOOL]], label %[[TRUE2:.+]], label %[[FALSE2:.+]] |
| 678 | // CHECK: [[TRUE2]] |
| 679 | // CHECK: br label %[[END2:.+]] |
| 680 | // CHECK: [[FALSE2]] |
| 681 | // CHECK: br label %[[END2]] |
| 682 | // CHECK: [[END2]] |
| 683 | // CHECK: [[COND_LVALUE:%.+]] = phi [[S_INT_TY]]* [ [[VAR1_LHS]], %[[TRUE2]] ], [ [[VAR1_RHS]], %[[FALSE2]] ] |
| 684 | // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_LHS]] to i8* |
| 685 | // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* |
| 686 | // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) |
| 687 | |
| 688 | // t_var1_lhs = min(t_var1_lhs, t_var1_rhs); |
| 689 | // CHECK: [[T_VAR1_LHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_LHS]], |
| 690 | // CHECK: [[T_VAR1_RHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_RHS]], |
| 691 | // CHECK: [[CMP:%.+]] = icmp slt i{{[0-9]+}} [[T_VAR1_LHS_VAL]], [[T_VAR1_RHS_VAL]] |
| 692 | // CHECK: [[UP:%.+]] = zext i1 [[CMP]] to i{{[0-9]+}} |
| 693 | // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR1_LHS]], |
| 694 | // CHECK: ret void |
| 695 | |
| 696 | #endif |
| 697 | |