1 // RUN: %clang_cc1 -verify -fopenmp=libiomp5 -x c++ -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck %s
2 // RUN: %clang_cc1 -fopenmp=libiomp5 -x c++ -std=c++11 -triple %itanium_abi_triple -emit-pch -o %t %s
3 // RUN: %clang_cc1 -fopenmp=libiomp5 -x c++ -triple %itanium_abi_triple -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 %itanium_abi_triple -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s
5 // RUN: %clang_cc1 -verify -fopenmp=libiomp5 -x c++ -fblocks -DBLOCKS -triple %itanium_abi_triple -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 #pragma omp threadprivate(g)
12 
13 template <class T>
14 struct S {
15   T f;
SS16   S(T a) : f(a + g) {}
SS17   S() : f(g) {}
operator =S18   S &operator=(const S &) { return *this; };
operator TS19   operator T() { return T(); }
~SS20   ~S() {}
21 };
22 
23 // CHECK-DAG: [[S_FLOAT_TY:%.+]] = type { float }
24 // CHECK-DAG: [[S_INT_TY:%.+]] = type { i{{[0-9]+}} }
25 // CHECK-DAG: [[IMPLICIT_BARRIER_LOC:@.+]] = private unnamed_addr constant %{{.+}} { i32 0, i32 66, i32 0, i32 0, i8*
26 
27 
28 // CHECK-DAG: [[T_VAR:@.+]] = internal global i{{[0-9]+}} 1122,
29 // CHECK-DAG: [[VEC:@.+]] = internal global [2 x i{{[0-9]+}}] [i{{[0-9]+}} 1, i{{[0-9]+}} 2],
30 // CHECK-DAG: [[S_ARR:@.+]] = internal global [2 x [[S_FLOAT_TY]]] zeroinitializer,
31 // CHECK-DAG: [[VAR:@.+]] = internal global [[S_FLOAT_TY]] zeroinitializer,
32 // CHECK-DAG: [[TMAIN_T_VAR:@.+]] = linkonce_odr global i{{[0-9]+}} 333,
33 // CHECK-DAG: [[TMAIN_VEC:@.+]] = linkonce_odr global [2 x i{{[0-9]+}}] [i{{[0-9]+}} 3, i{{[0-9]+}} 3],
34 // CHECK-DAG: [[TMAIN_S_ARR:@.+]] = linkonce_odr global [2 x [[S_INT_TY]]] zeroinitializer,
35 // CHECK-DAG: [[TMAIN_VAR:@.+]] = linkonce_odr global [[S_INT_TY]] zeroinitializer,
36 template <typename T>
tmain()37 T tmain() {
38   S<T> test;
39   test = S<T>();
40   static T t_var = 333;
41   static T vec[] = {3, 3};
42   static S<T> s_arr[] = {1, 2};
43   static S<T> var(3);
44 #pragma omp threadprivate(t_var, vec, s_arr, var)
45 #pragma omp parallel copyin(t_var, vec, s_arr, var)
46   {
47     vec[0] = t_var;
48     s_arr[0] = var;
49   }
50 #pragma omp parallel copyin(t_var)
51   {}
52   return T();
53 }
54 
main()55 int main() {
56 #ifdef LAMBDA
57   // LAMBDA: [[G:@.+]] = global i{{[0-9]+}} 1212,
58   // LAMBDA-LABEL: @main
59   // LAMBDA: call{{( x86_thiscallcc)?}} void [[OUTER_LAMBDA:@.+]](
60   [&]() {
61   // LAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]](
62   // LAMBDA: call void {{.+}} @__kmpc_fork_call({{.+}}, i32 1, {{.+}}* [[OMP_REGION:@.+]] to {{.+}}, i8*
63 #pragma omp parallel copyin(g)
64   {
65     // LAMBDA: define{{.*}} internal{{.*}} void [[OMP_REGION]](i32* %{{.+}}, i32* %{{.+}}, %{{.+}}* [[ARG:%.+]])
66 
67     // threadprivate_g = g;
68     // LAMBDA: call i8* @__kmpc_threadprivate_cached({{.+}} [[G]]
69     // LAMBDA: ptrtoint i{{[0-9]+}}* %{{.+}} to i{{[0-9]+}}
70     // LAMBDA: icmp ne i{{[0-9]+}} ptrtoint (i{{[0-9]+}}* [[G]] to i{{[0-9]+}}), %{{.+}}
71     // LAMBDA: br i1 %{{.+}}, label %[[NOT_MASTER:.+]], label %[[DONE:.+]]
72     // LAMBDA: [[NOT_MASTER]]
73     // LAMBDA: load i{{[0-9]+}}, i{{[0-9]+}}* [[G]],
74     // LAMBDA: store volatile i{{[0-9]+}} %{{.+}}, i{{[0-9]+}}* %{{.+}},
75     // LAMBDA: [[DONE]]
76 
77     // LAMBDA: call i32 @__kmpc_cancel_barrier(
78     g = 1;
79     // LAMBDA: call{{( x86_thiscallcc)?}} void [[INNER_LAMBDA:@.+]](%{{.+}}*
80     [&]() {
81       // LAMBDA: define {{.+}} void [[INNER_LAMBDA]](%{{.+}}* [[ARG_PTR:%.+]])
82       // LAMBDA: store %{{.+}}* [[ARG_PTR]], %{{.+}}** [[ARG_PTR_REF:%.+]],
83       g = 2;
84       // LAMBDA: [[ARG_PTR:%.+]] = load %{{.+}}*, %{{.+}}** [[ARG_PTR_REF]]
85     }();
86   }
87   }();
88   return 0;
89 #elif defined(BLOCKS)
90   // BLOCKS: [[G:@.+]] = global i{{[0-9]+}} 1212,
91   // BLOCKS-LABEL: @main
92   // BLOCKS: call void {{%.+}}(i8
93   ^{
94   // BLOCKS: define{{.*}} internal{{.*}} void {{.+}}(i8*
95   // BLOCKS: call void {{.+}} @__kmpc_fork_call({{.+}}, i32 1, {{.+}}* [[OMP_REGION:@.+]] to {{.+}}, i8*
96 #pragma omp parallel copyin(g)
97   {
98     // BLOCKS: define{{.*}} internal{{.*}} void [[OMP_REGION]](i32* %{{.+}}, i32* %{{.+}}, %{{.+}}* [[ARG:%.+]])
99 
100     // threadprivate_g = g;
101     // BLOCKS: call i8* @__kmpc_threadprivate_cached({{.+}} [[G]]
102     // BLOCKS: ptrtoint i{{[0-9]+}}* %{{.+}} to i{{[0-9]+}}
103     // BLOCKS: icmp ne i{{[0-9]+}} ptrtoint (i{{[0-9]+}}* [[G]] to i{{[0-9]+}}), %{{.+}}
104     // BLOCKS: br i1 %{{.+}}, label %[[NOT_MASTER:.+]], label %[[DONE:.+]]
105     // BLOCKS: [[NOT_MASTER]]
106     // BLOCKS: load i{{[0-9]+}}, i{{[0-9]+}}* [[G]],
107     // BLOCKS: store volatile i{{[0-9]+}} %{{.+}}, i{{[0-9]+}}* %{{.+}},
108     // BLOCKS: [[DONE]]
109 
110     // BLOCKS: call i32 @__kmpc_cancel_barrier(
111     g = 1;
112     // BLOCKS: store volatile i{{[0-9]+}} 1, i{{[0-9]+}}*
113     // BLOCKS-NOT: [[G]]{{[[^:word:]]}}
114     // BLOCKS: call void {{%.+}}(i8
115     ^{
116       // BLOCKS: define {{.+}} void {{@.+}}(i8*
117       g = 2;
118       // BLOCKS-NOT: [[G]]{{[[^:word:]]}}
119       // BLOCKS: call i8* @__kmpc_threadprivate_cached({{.+}} [[G]]
120       // BLOCKS: store volatile i{{[0-9]+}} 2, i{{[0-9]+}}*
121       // BLOCKS-NOT: [[G]]{{[[^:word:]]}}
122       // BLOCKS: ret
123     }();
124   }
125   }();
126   return 0;
127 #else
128   S<float> test;
129   test = S<float>();
130   static int t_var = 1122;
131   static int vec[] = {1, 2};
132   static S<float> s_arr[] = {1, 2};
133   static S<float> var(3);
134 #pragma omp threadprivate(t_var, vec, s_arr, var)
135 #pragma omp parallel copyin(t_var, vec, s_arr, var)
136   {
137     vec[0] = t_var;
138     s_arr[0] = var;
139   }
140 #pragma omp parallel copyin(t_var)
141   {}
142   return tmain<int>();
143 #endif
144 }
145 
146 // CHECK-LABEL: @main
147 // CHECK: [[TEST:%.+]] = alloca [[S_FLOAT_TY]],
148 // CHECK: call {{.*}} [[S_FLOAT_TY_COPY_ASSIGN:@.+]]([[S_FLOAT_TY]]* [[TEST]], [[S_FLOAT_TY]]*
149 // 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]+}}*, {{%.+}}*)* [[MAIN_MICROTASK:@.+]] to void (i32*, i32*, ...)*), i8* %{{.+}})
150 // 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]+}}*, {{%.+}}*)* [[MAIN_MICROTASK1:@.+]] to void (i32*, i32*, ...)*), i8* %{{.+}})
151 // CHECK: = call {{.*}}i{{.+}} [[TMAIN_INT:@.+]]()
152 // CHECK: call {{.*}} [[S_FLOAT_TY_DESTR:@.+]]([[S_FLOAT_TY]]*
153 // CHECK: ret
154 //
155 // CHECK: define internal void [[MAIN_MICROTASK]](i{{[0-9]+}}* [[GTID_ADDR:%.+]], i{{[0-9]+}}* %{{.+}}, {{%.+}}* %{{.+}})
156 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]],
157 // CHECK: [[GTID_ADDR:%.+]] = load i32*, i32** [[GTID_ADDR_ADDR]],
158 // CHECK: [[GTID:%.+]] = load i32, i32* [[GTID_ADDR]],
159 
160 // threadprivate_t_var = t_var;
161 // CHECK: call i8* @__kmpc_threadprivate_cached({{.+}} [[T_VAR]]
162 // CHECK: ptrtoint i{{[0-9]+}}* %{{.+}} to i{{[0-9]+}}
163 // CHECK: icmp ne i{{[0-9]+}} ptrtoint (i{{[0-9]+}}* [[T_VAR]] to i{{[0-9]+}}), %{{.+}}
164 // CHECK: br i1 %{{.+}}, label %[[NOT_MASTER:.+]], label %[[DONE:.+]]
165 // CHECK: [[NOT_MASTER]]
166 // CHECK: load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR]],
167 // CHECK: store i{{[0-9]+}} %{{.+}}, i{{[0-9]+}}* %{{.+}},
168 
169 // threadprivate_vec = vec;
170 // CHECK: call i8* @__kmpc_threadprivate_cached({{.+}} [[VEC]]
171 // CHECK: call void @llvm.memcpy{{.*}}(i8* %{{.+}}, i8* bitcast ([2 x i{{[0-9]+}}]* [[VEC]] to i8*),
172 
173 // threadprivate_s_arr = s_arr;
174 // CHECK: call i8* @__kmpc_threadprivate_cached({{.+}} [[S_ARR]]
175 // CHECK: [[S_ARR_PRIV_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* {{%.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 0
176 // CHECK: [[S_ARR_PRIV_END:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_PRIV_BEGIN]], i{{[0-9]+}} 2
177 // CHECK: [[IS_EMPTY:%.+]] = icmp eq [[S_FLOAT_TY]]* [[S_ARR_PRIV_BEGIN]], [[S_ARR_PRIV_END]]
178 // CHECK: br i1 [[IS_EMPTY]], label %[[S_ARR_BODY_DONE:.+]], label %[[S_ARR_BODY:.+]]
179 // CHECK: [[S_ARR_BODY]]
180 // CHECK: call {{.*}} [[S_FLOAT_TY_COPY_ASSIGN]]([[S_FLOAT_TY]]* {{.+}}, [[S_FLOAT_TY]]* {{.+}})
181 // CHECK: br i1 {{.+}}, label %{{.+}}, label %[[S_ARR_BODY]]
182 
183 // threadprivate_var = var;
184 // CHECK: call i8* @__kmpc_threadprivate_cached({{.+}} [[VAR]]
185 // CHECK: call {{.*}} [[S_FLOAT_TY_COPY_ASSIGN]]([[S_FLOAT_TY]]* {{%.+}}, [[S_FLOAT_TY]]* {{.*}}[[VAR]])
186 // CHECK: [[DONE]]
187 
188 // CHECK: call i32 @__kmpc_cancel_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i32 [[GTID]])
189 // CHECK: ret void
190 
191 // CHECK: define internal void [[MAIN_MICROTASK1]](i{{[0-9]+}}* [[GTID_ADDR:%.+]], i{{[0-9]+}}* %{{.+}}, {{%.+}}* %{{.+}})
192 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]],
193 // CHECK: [[GTID_ADDR:%.+]] = load i32*, i32** [[GTID_ADDR_ADDR]],
194 // CHECK: [[GTID:%.+]] = load i32, i32* [[GTID_ADDR]],
195 
196 // threadprivate_t_var = t_var;
197 // CHECK: call i8* @__kmpc_threadprivate_cached({{.+}} [[T_VAR]]
198 // CHECK: ptrtoint i{{[0-9]+}}* %{{.+}} to i{{[0-9]+}}
199 // CHECK: icmp ne i{{[0-9]+}} ptrtoint (i{{[0-9]+}}* [[T_VAR]] to i{{[0-9]+}}), %{{.+}}
200 // CHECK: br i1 %{{.+}}, label %[[NOT_MASTER:.+]], label %[[DONE:.+]]
201 // CHECK: [[NOT_MASTER]]
202 // CHECK: load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR]],
203 // CHECK: store i{{[0-9]+}} %{{.+}}, i{{[0-9]+}}* %{{.+}},
204 // CHECK: [[DONE]]
205 
206 // CHECK: call i32 @__kmpc_cancel_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i32 [[GTID]])
207 // CHECK: ret void
208 
209 // CHECK: define {{.*}} i{{[0-9]+}} [[TMAIN_INT]]()
210 // CHECK: [[TEST:%.+]] = alloca [[S_INT_TY]],
211 // CHECK: call {{.*}} [[S_INT_TY_COPY_ASSIGN:@.+]]([[S_INT_TY]]* [[TEST]], [[S_INT_TY]]*
212 // 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]+}}*, {{%.+}}*)* [[TMAIN_MICROTASK:@.+]] to void (i32*, i32*, ...)*), i8* %{{.+}})
213 // 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]+}}*, {{%.+}}*)* [[TMAIN_MICROTASK1:@.+]] to void (i32*, i32*, ...)*), i8* %{{.+}})
214 // CHECK: call {{.*}} [[S_INT_TY_DESTR:@.+]]([[S_INT_TY]]*
215 // CHECK: ret
216 //
217 // CHECK: define internal void [[TMAIN_MICROTASK]](i{{[0-9]+}}* [[GTID_ADDR:%.+]], i{{[0-9]+}}* %{{.+}}, {{%.+}}* %{{.+}})
218 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]],
219 // CHECK: [[GTID_ADDR:%.+]] = load i32*, i32** [[GTID_ADDR_ADDR]],
220 // CHECK: [[GTID:%.+]] = load i32, i32* [[GTID_ADDR]],
221 
222 // threadprivate_t_var = t_var;
223 // CHECK: call i8* @__kmpc_threadprivate_cached({{.+}} [[TMAIN_T_VAR]]
224 // CHECK: ptrtoint i{{[0-9]+}}* %{{.+}} to i{{[0-9]+}}
225 // CHECK: icmp ne i{{[0-9]+}} ptrtoint (i{{[0-9]+}}* [[TMAIN_T_VAR]] to i{{[0-9]+}}), %{{.+}}
226 // CHECK: br i1 %{{.+}}, label %[[NOT_MASTER:.+]], label %[[DONE:.+]]
227 // CHECK: [[NOT_MASTER]]
228 // CHECK: load i{{[0-9]+}}, i{{[0-9]+}}* [[TMAIN_T_VAR]],
229 // CHECK: store i{{[0-9]+}} %{{.+}}, i{{[0-9]+}}* %{{.+}},
230 
231 // threadprivate_vec = vec;
232 // CHECK: call i8* @__kmpc_threadprivate_cached({{.+}} [[TMAIN_VEC]]
233 // CHECK: call void @llvm.memcpy{{.*}}(i8* %{{.+}}, i8* bitcast ([2 x i{{[0-9]+}}]* [[TMAIN_VEC]] to i8*),
234 
235 // threadprivate_s_arr = s_arr;
236 // CHECK: call i8* @__kmpc_threadprivate_cached({{.+}} [[TMAIN_S_ARR]]
237 // CHECK: [[S_ARR_PRIV_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* {{%.+}}, i{{[0-9]+}} 0, i{{[0-9]+}} 0
238 // CHECK: [[S_ARR_PRIV_END:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_PRIV_BEGIN]], i{{[0-9]+}} 2
239 // CHECK: [[IS_EMPTY:%.+]] = icmp eq [[S_INT_TY]]* [[S_ARR_PRIV_BEGIN]], [[S_ARR_PRIV_END]]
240 // CHECK: br i1 [[IS_EMPTY]], label %[[S_ARR_BODY_DONE:.+]], label %[[S_ARR_BODY:.+]]
241 // CHECK: [[S_ARR_BODY]]
242 // CHECK: call {{.*}} [[S_INT_TY_COPY_ASSIGN]]([[S_INT_TY]]* {{.+}}, [[S_INT_TY]]* {{.+}})
243 // CHECK: br i1 {{.+}}, label %{{.+}}, label %[[S_ARR_BODY]]
244 
245 // threadprivate_var = var;
246 // CHECK: call i8* @__kmpc_threadprivate_cached({{.+}} [[TMAIN_VAR]]
247 // CHECK: call {{.*}} [[S_INT_TY_COPY_ASSIGN]]([[S_INT_TY]]* {{%.+}}, [[S_INT_TY]]* {{.*}}[[TMAIN_VAR]])
248 // CHECK: [[DONE]]
249 
250 // CHECK: call i32 @__kmpc_cancel_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i32 [[GTID]])
251 // CHECK: ret void
252 
253 // CHECK: define internal void [[TMAIN_MICROTASK1]](i{{[0-9]+}}* [[GTID_ADDR:%.+]], i{{[0-9]+}}* %{{.+}}, {{%.+}}* %{{.+}})
254 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]],
255 // CHECK: [[GTID_ADDR:%.+]] = load i32*, i32** [[GTID_ADDR_ADDR]],
256 // CHECK: [[GTID:%.+]] = load i32, i32* [[GTID_ADDR]],
257 
258 // threadprivate_t_var = t_var;
259 // CHECK: call i8* @__kmpc_threadprivate_cached({{.+}} [[TMAIN_T_VAR]]
260 // CHECK: ptrtoint i{{[0-9]+}}* %{{.+}} to i{{[0-9]+}}
261 // CHECK: icmp ne i{{[0-9]+}} ptrtoint (i{{[0-9]+}}* [[TMAIN_T_VAR]] to i{{[0-9]+}}), %{{.+}}
262 // CHECK: br i1 %{{.+}}, label %[[NOT_MASTER:.+]], label %[[DONE:.+]]
263 // CHECK: [[NOT_MASTER]]
264 // CHECK: load i{{[0-9]+}}, i{{[0-9]+}}* [[TMAIN_T_VAR]],
265 // CHECK: store i{{[0-9]+}} %{{.+}}, i{{[0-9]+}}* %{{.+}},
266 // CHECK: [[DONE]]
267 
268 // CHECK: call i32 @__kmpc_cancel_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i32 [[GTID]])
269 // CHECK: ret void
270 
271 #endif
272 
273