1 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-unknown-linux -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s
2 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-linux -fexceptions -fcxx-exceptions -emit-pch -o %t %s
3 // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-linux -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s
4 
5 // RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY0 %s
6 // RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s
7 // RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
8 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
9 // expected-no-diagnostics
10 #ifndef HEADER
11 #define HEADER
12 
13 // CHECK: @main
main(int argc,char ** argv)14 int main(int argc, char **argv) {
15 #pragma omp parallel master reduction(task, +: argc, argv[0:10][0:argc])
16   {
17 #pragma omp task in_reduction(+: argc, argv[0:10][0:argc])
18     ;
19   }
20 }
21 
22 // CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @{{.+}}, i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, i8**)* [[OUTLINED:@.+]] to void (i32*, i32*, ...)*), i32* %{{.+}}, i8** %{{.+}})
23 
24 // CHECK: define internal void [[OUTLINED]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* {{.+}}, i8** {{.+}})
25 // CHECK: [[ARGC_FP_ADDR:%.+]] = alloca i32,
26 // CHECK: [[TR:%.+]] = alloca [2 x %struct.kmp_taskred_input_t],
27 // CHECK: [[TG:%.+]] = alloca i8*,
28 
29 // Init firstprivate copy of argc
30 // CHECK: store i32 0, i32* [[ARGC_FP_ADDR]],
31 // CHECK: [[ARGV_FP_ADDR:%.+]] = alloca i8, i64 [[SIZE:%.+]],
32 // CHECK: store i64 [[SIZE]], i64* [[SIZE_ADDR:%.+]],
33 
34 // Init firstprivate copy of argv[0:10][0:argc]
35 // CHECK: [[END:%.+]] = getelementptr i8, i8* [[ARGV_FP_ADDR]], i64 [[SIZE]]
36 // CHECK: [[EMPTY:%.+]] = icmp eq i8* [[ARGV_FP_ADDR]], [[END]]
37 // CHECK: br i1 [[EMPTY]], label %[[DONE:.+]], label %[[INIT:.+]]
38 // CHECK: [[INIT]]:
39 // CHECK: [[EL:%.+]] = phi i8* [ [[ARGV_FP_ADDR]], %{{.+}} ], [ [[NEXT_EL:%.+]], %[[INIT]] ]
40 // CHECK: store i8 0, i8* [[EL]],
41 // CHECK: [[NEXT_EL:%.+]] = getelementptr i8, i8* [[EL]], i32 1
42 // CHECK: [[FINISHED:%.+]] = icmp eq i8* [[NEXT_EL]], [[END]]
43 // CHECK: br i1 [[FINISHED]], label %[[DONE]], label %[[INIT]]
44 // CHECK: [[DONE]]:
45 
46 // Register task reduction.
47 // CHECK: [[TR0_ADDR:%.+]] = getelementptr inbounds [2 x %struct.kmp_taskred_input_t], [2 x %struct.kmp_taskred_input_t]* [[TR]], i64 0, i64 0
48 // CHECK: [[TR0_SHARED_ADDR:%.+]] = getelementptr inbounds %struct.kmp_taskred_input_t, %struct.kmp_taskred_input_t* [[TR0_ADDR]], i32 0, i32 0
49 // CHECK: [[BC:%.+]] = bitcast i32* [[ARGC_FP_ADDR]] to i8*
50 // CHECK: store i8* [[BC]], i8** [[TR0_SHARED_ADDR]],
51 // CHECK: [[TR0_ORIG_ADDR:%.+]] = getelementptr inbounds %struct.kmp_taskred_input_t, %struct.kmp_taskred_input_t* [[TR0_ADDR]], i32 0, i32 1
52 // CHECK: [[BC:%.+]] = bitcast i32* %{{.+}} to i8*
53 // CHECK: store i8* [[BC]], i8** [[TR0_ORIG_ADDR]],
54 // CHECK: [[TR0_SIZE_ADDR:%.+]] = getelementptr inbounds %struct.kmp_taskred_input_t, %struct.kmp_taskred_input_t* [[TR0_ADDR]], i32 0, i32 2
55 // CHECK: store i64 4, i64* [[TR0_SIZE_ADDR]],
56 // CHECK: [[TR0_INIT_ADDR:%.+]] = getelementptr inbounds %struct.kmp_taskred_input_t, %struct.kmp_taskred_input_t* [[TR0_ADDR]], i32 0, i32 3
57 // CHECK: store i8* bitcast (void (i8*, i8*)* [[ARGC_INIT:@.+]] to i8*), i8** [[TR0_INIT_ADDR]],
58 // CHECK: [[TR0_FINI_ADDR:%.+]] = getelementptr inbounds %struct.kmp_taskred_input_t, %struct.kmp_taskred_input_t* [[TR0_ADDR]], i32 0, i32 4
59 // CHECK: store i8* null, i8** [[TR0_FINI_ADDR]],
60 // CHECK: [[TR0_COMB_ADDR:%.+]] = getelementptr inbounds %struct.kmp_taskred_input_t, %struct.kmp_taskred_input_t* [[TR0_ADDR]], i32 0, i32 5
61 // CHECK: store i8* bitcast (void (i8*, i8*)* [[ARGC_COMB:@.+]] to i8*), i8** [[TR0_COMB_ADDR]],
62 // CHECK: [[TR0_FLAGS_ADDR:%.+]] = getelementptr inbounds %struct.kmp_taskred_input_t, %struct.kmp_taskred_input_t* [[TR0_ADDR]], i32 0, i32 6
63 // CHECK: [[BC:%.+]] = bitcast i32* [[TR0_FLAGS_ADDR]] to i8*
64 // CHECK: call void @llvm.memset.p0i8.i64(i8* {{.*}}[[BC]], i8 0, i64 4, i1 false)
65 // CHECK: [[TR1_ADDR:%.+]] = getelementptr inbounds [2 x %struct.kmp_taskred_input_t], [2 x %struct.kmp_taskred_input_t]* [[TR]], i64 0, i64 1
66 // CHECK: [[TR1_SHARED_ADDR:%.+]] = getelementptr inbounds %struct.kmp_taskred_input_t, %struct.kmp_taskred_input_t* [[TR1_ADDR]], i32 0, i32 0
67 // CHECK: store i8* [[ARGV_FP_ADDR]], i8** [[TR1_SHARED_ADDR]],
68 // CHECK: [[TR1_ORIG_ADDR:%.+]] = getelementptr inbounds %struct.kmp_taskred_input_t, %struct.kmp_taskred_input_t* [[TR1_ADDR]], i32 0, i32 1
69 // CHECK: store i8* %{{.+}}, i8** [[TR1_ORIG_ADDR]],
70 // CHECK: [[TR1_SIZE_ADDR:%.+]] = getelementptr inbounds %struct.kmp_taskred_input_t, %struct.kmp_taskred_input_t* [[TR1_ADDR]], i32 0, i32 2
71 // CHECK: store i64 %{{.+}}, i64* [[TR1_SIZE_ADDR]],
72 // CHECK: [[TR1_INIT_ADDR:%.+]] = getelementptr inbounds %struct.kmp_taskred_input_t, %struct.kmp_taskred_input_t* [[TR1_ADDR]], i32 0, i32 3
73 // CHECK: store i8* bitcast (void (i8*, i8*)* [[ARGV_INIT:@.+]] to i8*), i8** [[TR1_INIT_ADDR]],
74 // CHECK: [[TR1_FINI_ADDR:%.+]] = getelementptr inbounds %struct.kmp_taskred_input_t, %struct.kmp_taskred_input_t* [[TR1_ADDR]], i32 0, i32 4
75 // CHECK: store i8* null, i8** [[TR1_FINI_ADDR]],
76 // CHECK: [[TR1_COMB_ADDR:%.+]] = getelementptr inbounds %struct.kmp_taskred_input_t, %struct.kmp_taskred_input_t* [[TR1_ADDR]], i32 0, i32 5
77 // CHECK: store i8* bitcast (void (i8*, i8*)* [[ARGV_COMB:@.+]] to i8*), i8** [[TR1_COMB_ADDR]],
78 // CHECK: [[TR1_FLAGS_ADDR:%.+]] = getelementptr inbounds %struct.kmp_taskred_input_t, %struct.kmp_taskred_input_t* [[TR1_ADDR]], i32 0, i32 6
79 // CHECK: store i32 1, i32* [[TR1_FLAGS_ADDR]],
80 // CHECK: [[BC:%.+]] = bitcast [2 x %struct.kmp_taskred_input_t]* [[TR]] to i8*
81 // CHECK: [[TG_VAL:%.+]] = call i8* @__kmpc_taskred_modifier_init(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 0, i32 2, i8* [[BC]])
82 // CHECK: store i8* [[TG_VAL]], i8** [[TG]],
83 
84 // CHECK: [[PTR:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 1, i64 48, i64 24, i32 (i32, i8*)* bitcast (i32 (i32, [[TASK_TY:%.+]]*)* [[TASK:@.+]] to i32 (i32, i8*)*))
85 // CHECK: [[TASK_DATA_ADDR:%.+]] = bitcast i8* [[PTR]] to [[TASK_TY]]*
86 // CHECK: [[PRIVATES_ADDR:%.+]] = getelementptr inbounds [[TASK_TY]], [[TASK_TY]]* [[TASK_DATA_ADDR]], i32 0, i32 1
87 // CHECK: [[TG_PRIV_ADDR:%.+]] = getelementptr inbounds [[TASK_PRIVATES_TY:%.+]], %{{.+}}* [[PRIVATES_ADDR]], i32 0, i32 0
88 // CHECK: [[TG_VAL:%.+]] = load i8*, i8** [[TG]],
89 // CHECK: store i8* [[TG_VAL]], i8** [[TG_PRIV_ADDR]],
90 
91 // CHECK: call i32 @__kmpc_omp_task(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[PTR]])
92 
93 // CHECK: call void @__kmpc_task_reduction_modifier_fini(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 0)
94 // CHECK: call i32 @__kmpc_reduce_nowait(
95 
96 // CHECK: define internal void [[ARGC_INIT]](i8* noalias %{{.+}}, i8* noalias %{{.+}})
97 // CHECK: store i32 0, i32* %{{.+}},
98 
99 // CHECK: define internal void [[ARGC_COMB]](i8* %{{.+}}, i8* %{{.+}})
100 // CHECK: [[ADD:%.+]] = add nsw i32 %{{.+}}, %{{.+}}
101 // CHECK: store i32 [[ADD]], i32* %{{.+}},
102 
103 // CHECK: define internal void [[ARGV_INIT]](i8* noalias %{{.+}}, i8* noalias %{{.+}})
104 // CHECK: phi i8*
105 // CHECK: store i8 0, i8* [[EL:%.+]],
106 // CHECK: getelementptr i8, i8* [[EL]], i32 1
107 
108 // CHECK: define internal void [[ARGV_COMB]](i8* %{{.+}}, i8* %{{.+}})
109 // CHECK: phi i8*
110 // CHECK: [[ADD:%.+]] = add nsw i32 %{{.+}}, %{{.+}}
111 // CHECK: [[CONV:%.+]] = trunc i32 [[ADD]] to i8
112 // CHECK: store i8 [[CONV]], i8* [[EL:%.+]],
113 // CHECK: getelementptr i8, i8* [[EL]], i32 1
114 
115 // CHECK: define internal {{.*}}i32 [[TASK]](i32 {{.+}}, [[TASK_TY]]* {{.+}})
116 // CHECK-DAG: call i8* @__kmpc_task_reduction_get_th_data(i32 %{{.+}}, i8* [[TG:%.+]], i8* [[ARGC_REF:%.+]])
117 // CHECK_DAG: [[TG]] = load i8*, i8** [[TG_ADDR:%.+]],
118 // CHECK-DAG: [[ARGC_REF]] = bitcast i32* [[ARGC_ADDR:%.+]] to i8*
119 // CHECK-DAG: [[ARGC_ADDR]] = load i32*, i32** [[ARGC_ADDR_REF:%.+]],
120 // CHECK-DAG: [[ARGC_ADDR_REF]] = getelementptr inbounds [[CAPS_TY:%.+]], %{{.+}}* [[CAP:%.+]], i32 0, i32 1
121 // CHECK-DAG: call i8* @__kmpc_task_reduction_get_th_data(i32 %{{.+}}, i8* [[TG:%.+]], i8* [[ARGV_REF:%.+]])
122 // CHECK_DAG: [[TG]] = load i8*, i8** [[TG_ADDR]],
123 // CHECK-DAG: [[ARGV_REF]] = load i8*, i8** [[ARGV_ADDR:%.+]],
124 // CHECK-DAG: [[ARGV_ADDR]] = load i8**, i8*** [[ARGV_ADDR_REF:%.+]],
125 // CHECK-DAG: [[ARGV_ADDR_REF]] = getelementptr inbounds [[CAPS_TY]], [[CAPS_TY]]* [[CAP]], i32 0, i32 2
126 
127 #endif
128