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 // REQUIRES: x86-registered-target
8 #ifndef HEADER
9 #define HEADER
10 
11 volatile double g, g_orig;
12 volatile double &g1 = g_orig;
13 
14 template <class T>
15 struct S {
16   T f;
17   S(T a) : f(a + g) {}
18   S() : f(g) {}
19   operator T() { return T(); }
20   S &operator&(const S &) { return *this; }
21   ~S() {}
22 };
23 
24 // CHECK-DAG: [[S_FLOAT_TY:%.+]] = type { float }
25 // CHECK-DAG: [[S_INT_TY:%.+]] = type { i{{[0-9]+}} }
26 // CHECK-DAG: [[ATOMIC_REDUCE_BARRIER_LOC:@.+]] = private unnamed_addr constant %{{.+}} { i32 0, i32 18, i32 0, i32 0, i8*
27 // CHECK-DAG: [[IMPLICIT_BARRIER_LOC:@.+]] = private unnamed_addr constant %{{.+}} { i32 0, i32 66, i32 0, i32 0, i8*
28 // CHECK-DAG: [[REDUCTION_LOC:@.+]] = private unnamed_addr constant %{{.+}} { i32 0, i32 18, i32 0, i32 0, i8*
29 // CHECK-DAG: [[REDUCTION_LOCK:@.+]] = common global [8 x i32] zeroinitializer
30 
31 template <typename T>
32 T tmain() {
33   T t;
34   S<T> test;
35   T t_var = T(), t_var1;
36   T vec[] = {1, 2};
37   S<T> s_arr[] = {1, 2};
38   S<T> &var = test;
39   S<T> var1;
40 #pragma omp parallel
41 #pragma omp for reduction(+:t_var) reduction(&:var) reduction(&& : var1) reduction(min: t_var1) nowait
42   for (int i = 0; i < 2; ++i) {
43     vec[i] = t_var;
44     s_arr[i] = var;
45   }
46 #pragma omp parallel
47 #pragma omp for reduction(&& : t_var)
48   for (int i = 0; i < 2; ++i) {
49     vec[i] = t_var;
50     s_arr[i] = var;
51   }
52   return T();
53 }
54 
55 extern S<float> **foo();
56 
57 int main() {
58 #ifdef LAMBDA
59   // LAMBDA: [[G:@.+]] = global double
60   // LAMBDA-LABEL: @main
61   // LAMBDA: call void [[OUTER_LAMBDA:@.+]](
62   [&]() {
63   // LAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]](
64   // LAMBDA: call void {{.+}} @__kmpc_fork_call({{.+}}, i32 0, {{.+}}* [[OMP_REGION:@.+]] to {{.+}})
65 #pragma omp parallel
66 #pragma omp for reduction(+:g, g1)
67     for (int i = 0; i < 2; ++i) {
68     // LAMBDA: define{{.*}} internal{{.*}} void [[OMP_REGION]](i32* noalias %{{.+}}, i32* noalias %{{.+}})
69     // LAMBDA: [[G_PRIVATE_ADDR:%.+]] = alloca double,
70 
71     // Reduction list for runtime.
72     // LAMBDA: [[RED_LIST:%.+]] = alloca [2 x i8*],
73 
74     // LAMBDA: store double 0.0{{.+}}, double* [[G_PRIVATE_ADDR]]
75     // LAMBDA: call void @__kmpc_for_static_init_4(
76     g = 1;
77     g1 = 1;
78     // LAMBDA: store double 1.0{{.+}}, double* [[G_PRIVATE_ADDR]],
79     // LAMBDA: [[G_PRIVATE_ADDR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG:%.+]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
80     // LAMBDA: store double* [[G_PRIVATE_ADDR]], double** [[G_PRIVATE_ADDR_REF]]
81     // LAMBDA: call void [[INNER_LAMBDA:@.+]](%{{.+}}* [[ARG]])
82     // LAMBDA: call void @__kmpc_for_static_fini(
83 
84     // LAMBDA: [[G_PRIV_REF:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[RED_LIST]], i64 0, i64 0
85     // LAMBDA: [[BITCAST:%.+]] = bitcast double* [[G_PRIVATE_ADDR]] to i8*
86     // LAMBDA: store i8* [[BITCAST]], i8** [[G_PRIV_REF]],
87     // LAMBDA: call i32 @__kmpc_reduce(
88     // LAMBDA: switch i32 %{{.+}}, label %[[REDUCTION_DONE:.+]] [
89     // LAMBDA: i32 1, label %[[CASE1:.+]]
90     // LAMBDA: i32 2, label %[[CASE2:.+]]
91     // LAMBDA: [[CASE1]]
92     // LAMBDA: [[G_VAL:%.+]] = load double, double* [[G]]
93     // LAMBDA: [[G_PRIV_VAL:%.+]] = load double, double* [[G_PRIVATE_ADDR]]
94     // LAMBDA: [[ADD:%.+]] = fadd double [[G_VAL]], [[G_PRIV_VAL]]
95     // LAMBDA: store double [[ADD]], double* [[G]]
96     // LAMBDA: call void @__kmpc_end_reduce(
97     // LAMBDA: br label %[[REDUCTION_DONE]]
98     // LAMBDA: [[CASE2]]
99     // LAMBDA: [[G_PRIV_VAL:%.+]] = load double, double* [[G_PRIVATE_ADDR]]
100     // LAMBDA: fadd double
101     // LAMBDA: cmpxchg i64*
102     // LAMBDA: call void @__kmpc_end_reduce(
103     // LAMBDA: br label %[[REDUCTION_DONE]]
104     // LAMBDA: [[REDUCTION_DONE]]
105     // LAMBDA: ret void
106     [&]() {
107       // LAMBDA: define {{.+}} void [[INNER_LAMBDA]](%{{.+}}* [[ARG_PTR:%.+]])
108       // LAMBDA: store %{{.+}}* [[ARG_PTR]], %{{.+}}** [[ARG_PTR_REF:%.+]],
109       g = 2;
110       g1 = 2;
111       // LAMBDA: [[ARG_PTR:%.+]] = load %{{.+}}*, %{{.+}}** [[ARG_PTR_REF]]
112       // LAMBDA: [[G_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
113       // LAMBDA: [[G_REF:%.+]] = load double*, double** [[G_PTR_REF]]
114       // LAMBDA: store double 2.0{{.+}}, double* [[G_REF]]
115     }();
116   }
117   }();
118   return 0;
119 #elif defined(BLOCKS)
120   // BLOCKS: [[G:@.+]] = global double
121   // BLOCKS-LABEL: @main
122   // BLOCKS: call void {{%.+}}(i8
123   ^{
124   // BLOCKS: define{{.*}} internal{{.*}} void {{.+}}(i8*
125   // BLOCKS: call void {{.+}} @__kmpc_fork_call({{.+}}, i32 0, {{.+}}* [[OMP_REGION:@.+]] to {{.+}})
126 #pragma omp parallel
127 #pragma omp for reduction(-:g, g1)
128     for (int i = 0; i < 2; ++i)  {
129     // BLOCKS: define{{.*}} internal{{.*}} void [[OMP_REGION]](i32* noalias %{{.+}}, i32* noalias %{{.+}})
130     // BLOCKS: [[G_PRIVATE_ADDR:%.+]] = alloca double,
131 
132     // Reduction list for runtime.
133     // BLOCKS: [[RED_LIST:%.+]] = alloca [2 x i8*],
134 
135     // BLOCKS: store double 0.0{{.+}}, double* [[G_PRIVATE_ADDR]]
136     g = 1;
137     g1 = 1;
138     // BLOCKS: call void @__kmpc_for_static_init_4(
139     // BLOCKS: store double 1.0{{.+}}, double* [[G_PRIVATE_ADDR]],
140     // BLOCKS-NOT: [[G]]{{[[^:word:]]}}
141     // BLOCKS: double* [[G_PRIVATE_ADDR]]
142     // BLOCKS-NOT: [[G]]{{[[^:word:]]}}
143     // BLOCKS: call void {{%.+}}(i8
144     // BLOCKS: call void @__kmpc_for_static_fini(
145 
146     // BLOCKS: [[G_PRIV_REF:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[RED_LIST]], i64 0, i64 0
147     // BLOCKS: [[BITCAST:%.+]] = bitcast double* [[G_PRIVATE_ADDR]] to i8*
148     // BLOCKS: store i8* [[BITCAST]], i8** [[G_PRIV_REF]],
149     // BLOCKS: call i32 @__kmpc_reduce(
150     // BLOCKS: switch i32 %{{.+}}, label %[[REDUCTION_DONE:.+]] [
151     // BLOCKS: i32 1, label %[[CASE1:.+]]
152     // BLOCKS: i32 2, label %[[CASE2:.+]]
153     // BLOCKS: [[CASE1]]
154     // BLOCKS: [[G_VAL:%.+]] = load double, double* [[G]]
155     // BLOCKS: [[G_PRIV_VAL:%.+]] = load double, double* [[G_PRIVATE_ADDR]]
156     // BLOCKS: [[ADD:%.+]] = fadd double [[G_VAL]], [[G_PRIV_VAL]]
157     // BLOCKS: store double [[ADD]], double* [[G]]
158     // BLOCKS: call void @__kmpc_end_reduce(
159     // BLOCKS: br label %[[REDUCTION_DONE]]
160     // BLOCKS: [[CASE2]]
161     // BLOCKS: [[G_PRIV_VAL:%.+]] = load double, double* [[G_PRIVATE_ADDR]]
162     // BLOCKS: fadd double
163     // BLOCKS: cmpxchg i64*
164     // BLOCKS: call void @__kmpc_end_reduce(
165     // BLOCKS: br label %[[REDUCTION_DONE]]
166     // BLOCKS: [[REDUCTION_DONE]]
167     // BLOCKS: ret void
168     ^{
169       // BLOCKS: define {{.+}} void {{@.+}}(i8*
170       g = 2;
171       g1 = 2;
172       // BLOCKS-NOT: [[G]]{{[[^:word:]]}}
173       // BLOCKS: store double 2.0{{.+}}, double*
174       // BLOCKS-NOT: [[G]]{{[[^:word:]]}}
175       // BLOCKS: ret
176     }();
177   }
178   }();
179   return 0;
180 #else
181   S<float> test;
182   float t_var = 0, t_var1;
183   int vec[] = {1, 2};
184   S<float> s_arr[] = {1, 2};
185   S<float> &var = test;
186   S<float> var1, arrs[10][4];
187   S<float> **var2 = foo();
188   S<float> vvar2[2];
189   S<float> (&var3)[2] = s_arr;
190 #pragma omp parallel
191 #pragma omp for reduction(+:t_var) reduction(&:var) reduction(&& : var1) reduction(min: t_var1)
192   for (int i = 0; i < 2; ++i) {
193     vec[i] = t_var;
194     s_arr[i] = var;
195   }
196   int arr[10][vec[1]];
197 #pragma omp parallel for reduction(+:arr[1][:vec[1]]) reduction(&:arrs[1:vec[1]][1:2])
198   for (int i = 0; i < 10; ++i)
199     ++arr[1][i];
200 #pragma omp parallel
201 #pragma omp for reduction(+:arr) reduction(&:arrs)
202   for (int i = 0; i < 10; ++i)
203     ++arr[1][i];
204 #pragma omp parallel
205 #pragma omp for reduction(& : var2[0 : 5][1 : 6])
206   for (int i = 0; i < 10; ++i)
207     ;
208 #pragma omp parallel
209 #pragma omp for reduction(& : vvar2[0 : 5])
210   for (int i = 0; i < 10; ++i)
211     ;
212 #pragma omp parallel
213 #pragma omp for reduction(& : var3[1 : 2])
214   for (int i = 0; i < 10; ++i)
215     ;
216 #pragma omp parallel
217 #pragma omp for reduction(& : var3)
218   for (int i = 0; i < 10; ++i)
219     ;
220   return tmain<int>();
221 #endif
222 }
223 
224 // CHECK: define {{.*}}i{{[0-9]+}} @main()
225 // CHECK: [[TEST:%.+]] = alloca [[S_FLOAT_TY]],
226 // CHECK: call {{.*}} [[S_FLOAT_TY_CONSTR:@.+]]([[S_FLOAT_TY]]* [[TEST]])
227 // 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
228 // 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
229 // 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
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]+}}*, [[S_FLOAT_TY]]***)* [[MAIN_MICROTASK3:@.+]] 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_MICROTASK4:@.+]] 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_MICROTASK5:@.+]] to void
233 // 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
234 // CHECK: = call {{.*}}i{{.+}} [[TMAIN_INT:@.+]]()
235 // CHECK: call {{.*}} [[S_FLOAT_TY_DESTR:@.+]]([[S_FLOAT_TY]]*
236 // CHECK: ret
237 //
238 // 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) %{{.+}})
239 // CHECK: [[T_VAR_PRIV:%.+]] = alloca float,
240 // CHECK: [[VAR_PRIV:%.+]] = alloca [[S_FLOAT_TY]],
241 // CHECK: [[VAR1_PRIV:%.+]] = alloca [[S_FLOAT_TY]],
242 // CHECK: [[T_VAR1_PRIV:%.+]] = alloca float,
243 
244 // Reduction list for runtime.
245 // CHECK: [[RED_LIST:%.+]] = alloca [4 x i8*],
246 
247 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]],
248 
249 // CHECK: [[T_VAR_REF:%.+]] = load float*, float** %
250 // CHECK: [[VAR1_REF:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** %
251 // CHECK: [[T_VAR1_REF:%.+]] = load float*, float** %
252 
253 // For + reduction operation initial value of private variable is 0.
254 // CHECK: store float 0.0{{.+}}, float* [[T_VAR_PRIV]],
255 
256 // For & reduction operation initial value of private variable is ones in all bits.
257 // CHECK: [[VAR_REF:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** %
258 // CHECK: call {{.*}} [[S_FLOAT_TY_CONSTR:@.+]]([[S_FLOAT_TY]]* [[VAR_PRIV]])
259 
260 // For && reduction operation initial value of private variable is 1.0.
261 // CHECK: call {{.*}} [[S_FLOAT_TY_CONSTR:@.+]]([[S_FLOAT_TY]]* [[VAR1_PRIV]])
262 
263 // For min reduction operation initial value of private variable is largest repesentable value.
264 // CHECK: store float 0x47EFFFFFE0000000, float* [[T_VAR1_PRIV]],
265 
266 
267 // CHECK: [[GTID_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_ADDR]]
268 // CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_REF]]
269 // CHECK: call void @__kmpc_for_static_init_4(
270 // Skip checks for internal operations.
271 // CHECK: call void @__kmpc_for_static_fini(
272 
273 // void *RedList[<n>] = {<ReductionVars>[0], ..., <ReductionVars>[<n>-1]};
274 
275 // CHECK: [[T_VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 0
276 // CHECK: [[BITCAST:%.+]] = bitcast float* [[T_VAR_PRIV]] to i8*
277 // CHECK: store i8* [[BITCAST]], i8** [[T_VAR_PRIV_REF]],
278 // CHECK: [[VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 1
279 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_PRIV]] to i8*
280 // CHECK: store i8* [[BITCAST]], i8** [[VAR_PRIV_REF]],
281 // CHECK: [[VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 2
282 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_PRIV]] to i8*
283 // CHECK: store i8* [[BITCAST]], i8** [[VAR1_PRIV_REF]],
284 // CHECK: [[T_VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 3
285 // CHECK: [[BITCAST:%.+]] = bitcast float* [[T_VAR1_PRIV]] to i8*
286 // CHECK: store i8* [[BITCAST]], i8** [[T_VAR1_PRIV_REF]],
287 
288 // res = __kmpc_reduce(<loc>, <gtid>, <n>, sizeof(RedList), RedList, reduce_func, &<lock>);
289 
290 // CHECK: [[BITCAST:%.+]] = bitcast [4 x i8*]* [[RED_LIST]] to i8*
291 // 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]])
292 
293 // switch(res)
294 // CHECK: switch i32 [[RES]], label %[[RED_DONE:.+]] [
295 // CHECK: i32 1, label %[[CASE1:.+]]
296 // CHECK: i32 2, label %[[CASE2:.+]]
297 // CHECK: ]
298 
299 // case 1:
300 // t_var += t_var_reduction;
301 // CHECK: [[T_VAR_VAL:%.+]] = load float, float* [[T_VAR_REF]],
302 // CHECK: [[T_VAR_PRIV_VAL:%.+]] = load float, float* [[T_VAR_PRIV]],
303 // CHECK: [[UP:%.+]] = fadd float [[T_VAR_VAL]], [[T_VAR_PRIV_VAL]]
304 // CHECK: store float [[UP]], float* [[T_VAR_REF]],
305 
306 // var = var.operator &(var_reduction);
307 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_REF]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_PRIV]])
308 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_REF]] to i8*
309 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8*
310 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
311 
312 // var1 = var1.operator &&(var1_reduction);
313 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_REF]])
314 // CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0
315 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]]
316 // CHECK: [[TRUE]]
317 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_PRIV]])
318 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0
319 // CHECK: br label %[[END2]]
320 // CHECK: [[END2]]
321 // CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ]
322 // CHECK: [[CONV:%.+]] = uitofp i1 [[COND_LVALUE]] to float
323 // CHECK:  call void @{{.+}}([[S_FLOAT_TY]]* [[COND_LVALUE:%.+]], float [[CONV]])
324 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_REF]] to i8*
325 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8*
326 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
327 
328 // t_var1 = min(t_var1, t_var1_reduction);
329 // CHECK: [[T_VAR1_VAL:%.+]] = load float, float* [[T_VAR1_REF]],
330 // CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load float, float* [[T_VAR1_PRIV]],
331 // CHECK: [[CMP:%.+]] = fcmp olt float [[T_VAR1_VAL]], [[T_VAR1_PRIV_VAL]]
332 // CHECK: br i1 [[CMP]]
333 // CHECK: [[UP:%.+]] = phi float
334 // CHECK: store float [[UP]], float* [[T_VAR1_REF]],
335 
336 // __kmpc_end_reduce(<loc>, <gtid>, &<lock>);
337 // CHECK: call void @__kmpc_end_reduce(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], [8 x i32]* [[REDUCTION_LOCK]])
338 
339 // break;
340 // CHECK: br label %[[RED_DONE]]
341 
342 // case 2:
343 // t_var += t_var_reduction;
344 // CHECK: load float, float* [[T_VAR_PRIV]]
345 // CHECK: [[T_VAR_REF_INT:%.+]] = bitcast float* [[T_VAR_REF]] to i32*
346 // CHECK: [[OLD1:%.+]] = load atomic i32, i32* [[T_VAR_REF_INT]] monotonic,
347 // CHECK: br label %[[CONT:.+]]
348 // CHECK: [[CONT]]
349 // CHECK: [[ORIG_OLD_INT:%.+]] = phi i32 [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %[[CONT]] ]
350 // CHECK: fadd float
351 // CHECK: [[UP_INT:%.+]] = load i32, i32*
352 // CHECK: [[T_VAR_REF_INT:%.+]] = bitcast float* [[T_VAR_REF]] to i32*
353 // CHECK: [[RES:%.+]] = cmpxchg i32* [[T_VAR_REF_INT]], i32 [[ORIG_OLD_INT]], i32 [[UP_INT]] monotonic monotonic
354 // CHECK: [[OLD2:%.+]] = extractvalue { i32, i1 } [[RES]], 0
355 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1
356 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[ATOMIC_DONE:.+]], label %[[CONT]]
357 // CHECK: [[ATOMIC_DONE]]
358 
359 // var = var.operator &(var_reduction);
360 // CHECK: call void @__kmpc_critical(
361 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_REF]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_PRIV]])
362 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_REF]] to i8*
363 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8*
364 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
365 // CHECK: call void @__kmpc_end_critical(
366 
367 // var1 = var1.operator &&(var1_reduction);
368 // CHECK: call void @__kmpc_critical(
369 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_REF]])
370 // CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0
371 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]]
372 // CHECK: [[TRUE]]
373 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_PRIV]])
374 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0
375 // CHECK: br label %[[END2]]
376 // CHECK: [[END2]]
377 // CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ]
378 // CHECK: [[CONV:%.+]] = uitofp i1 [[COND_LVALUE]] to float
379 // CHECK:  call void @{{.+}}([[S_FLOAT_TY]]* [[COND_LVALUE:%.+]], float [[CONV]])
380 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_REF]] to i8*
381 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8*
382 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
383 // CHECK: call void @__kmpc_end_critical(
384 
385 // t_var1 = min(t_var1, t_var1_reduction);
386 // CHECK: load float, float* [[T_VAR1_PRIV]]
387 // CHECK: [[T_VAR1_REF_INT:%.+]] = bitcast float* [[T_VAR1_REF]] to i32*
388 // CHECK: [[OLD1:%.+]] = load atomic i32, i32* [[T_VAR1_REF_INT]] monotonic,
389 // CHECK: br label %[[CONT:.+]]
390 // CHECK: [[CONT]]
391 // CHECK: [[ORIG_OLD_INT:%.+]] = phi i32 [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %{{.+}} ]
392 // CHECK: [[CMP:%.+]] = fcmp olt float
393 // CHECK: br i1 [[CMP]]
394 // CHECK: phi float
395 // CHECK: [[UP_INT:%.+]] = load i32
396 // CHECK: [[T_VAR1_REF_INT:%.+]] = bitcast float* [[T_VAR1_REF]] to i32*
397 // CHECK: [[RES:%.+]] = cmpxchg i32* [[T_VAR1_REF_INT]], i32 [[ORIG_OLD_INT]], i32 [[UP_INT]] monotonic monotonic
398 // CHECK: [[OLD2:%.+]] = extractvalue { i32, i1 } [[RES]], 0
399 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1
400 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[ATOMIC_DONE:.+]], label %[[CONT]]
401 // CHECK: [[ATOMIC_DONE]]
402 
403 // __kmpc_end_reduce(<loc>, <gtid>, &<lock>);
404 // CHECK: call void @__kmpc_end_reduce(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], [8 x i32]* [[REDUCTION_LOCK]])
405 
406 // break;
407 // CHECK: br label %[[RED_DONE]]
408 // CHECK: [[RED_DONE]]
409 // CHECK-DAG: call {{.*}} [[S_FLOAT_TY_DESTR]]([[S_FLOAT_TY]]* [[VAR_PRIV]])
410 // CHECK-DAG: call {{.*}} [[S_FLOAT_TY_DESTR]]([[S_FLOAT_TY]]*
411 // CHECK: call void @__kmpc_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]], i{{[0-9]+}} [[GTID]])
412 
413 // CHECK: ret void
414 
415 // void reduce_func(void *lhs[<n>], void *rhs[<n>]) {
416 //  *(Type0*)lhs[0] = ReductionOperation0(*(Type0*)lhs[0], *(Type0*)rhs[0]);
417 //  ...
418 //  *(Type<n>-1*)lhs[<n>-1] = ReductionOperation<n>-1(*(Type<n>-1*)lhs[<n>-1],
419 //  *(Type<n>-1*)rhs[<n>-1]);
420 // }
421 // CHECK: define internal void [[REDUCTION_FUNC]](i8*, i8*)
422 // t_var_lhs = (float*)lhs[0];
423 // CHECK: [[T_VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS:%.+]], i64 0, i64 0
424 // CHECK: [[T_VAR_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR_RHS_REF]],
425 // CHECK: [[T_VAR_RHS:%.+]] = bitcast i8* [[T_VAR_RHS_VOID]] to float*
426 // t_var_rhs = (float*)rhs[0];
427 // CHECK: [[T_VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS:%.+]], i64 0, i64 0
428 // CHECK: [[T_VAR_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR_LHS_REF]],
429 // CHECK: [[T_VAR_LHS:%.+]] = bitcast i8* [[T_VAR_LHS_VOID]] to float*
430 
431 // var_lhs = (S<float>*)lhs[1];
432 // CHECK: [[VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 1
433 // CHECK: [[VAR_RHS_VOID:%.+]] = load i8*, i8** [[VAR_RHS_REF]],
434 // CHECK: [[VAR_RHS:%.+]] = bitcast i8* [[VAR_RHS_VOID]] to [[S_FLOAT_TY]]*
435 // var_rhs = (S<float>*)rhs[1];
436 // CHECK: [[VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 1
437 // CHECK: [[VAR_LHS_VOID:%.+]] = load i8*, i8** [[VAR_LHS_REF]],
438 // CHECK: [[VAR_LHS:%.+]] = bitcast i8* [[VAR_LHS_VOID]] to [[S_FLOAT_TY]]*
439 
440 // var1_lhs = (S<float>*)lhs[2];
441 // CHECK: [[VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 2
442 // CHECK: [[VAR1_RHS_VOID:%.+]] = load i8*, i8** [[VAR1_RHS_REF]],
443 // CHECK: [[VAR1_RHS:%.+]] = bitcast i8* [[VAR1_RHS_VOID]] to [[S_FLOAT_TY]]*
444 // var1_rhs = (S<float>*)rhs[2];
445 // CHECK: [[VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 2
446 // CHECK: [[VAR1_LHS_VOID:%.+]] = load i8*, i8** [[VAR1_LHS_REF]],
447 // CHECK: [[VAR1_LHS:%.+]] = bitcast i8* [[VAR1_LHS_VOID]] to [[S_FLOAT_TY]]*
448 
449 // t_var1_lhs = (float*)lhs[3];
450 // CHECK: [[T_VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 3
451 // CHECK: [[T_VAR1_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_RHS_REF]],
452 // CHECK: [[T_VAR1_RHS:%.+]] = bitcast i8* [[T_VAR1_RHS_VOID]] to float*
453 // t_var1_rhs = (float*)rhs[3];
454 // CHECK: [[T_VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 3
455 // CHECK: [[T_VAR1_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_LHS_REF]],
456 // CHECK: [[T_VAR1_LHS:%.+]] = bitcast i8* [[T_VAR1_LHS_VOID]] to float*
457 
458 // t_var_lhs += t_var_rhs;
459 // CHECK: [[T_VAR_LHS_VAL:%.+]] = load float, float* [[T_VAR_LHS]],
460 // CHECK: [[T_VAR_RHS_VAL:%.+]] = load float, float* [[T_VAR_RHS]],
461 // CHECK: [[UP:%.+]] = fadd float [[T_VAR_LHS_VAL]], [[T_VAR_RHS_VAL]]
462 // CHECK: store float [[UP]], float* [[T_VAR_LHS]],
463 
464 // var_lhs = var_lhs.operator &(var_rhs);
465 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_LHS]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_RHS]])
466 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_LHS]] to i8*
467 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8*
468 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
469 
470 // var1_lhs = var1_lhs.operator &&(var1_rhs);
471 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_LHS]])
472 // CHECK: [[VAR1_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0
473 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]]
474 // CHECK: [[TRUE]]
475 // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_RHS]])
476 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = fcmp une float [[TO_FLOAT]], 0.0
477 // CHECK: br label %[[END2]]
478 // CHECK: [[END2]]
479 // CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ]
480 // CHECK: [[CONV:%.+]] = uitofp i1 [[COND_LVALUE]] to float
481 // CHECK:  call void @{{.+}}([[S_FLOAT_TY]]* [[COND_LVALUE:%.+]], float [[CONV]])
482 // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_LHS]] to i8*
483 // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8*
484 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
485 
486 // t_var1_lhs = min(t_var1_lhs, t_var1_rhs);
487 // CHECK: [[T_VAR1_LHS_VAL:%.+]] = load float, float* [[T_VAR1_LHS]],
488 // CHECK: [[T_VAR1_RHS_VAL:%.+]] = load float, float* [[T_VAR1_RHS]],
489 // CHECK: [[CMP:%.+]] = fcmp olt float [[T_VAR1_LHS_VAL]], [[T_VAR1_RHS_VAL]]
490 // CHECK: br i1 [[CMP]]
491 // CHECK: [[UP:%.+]] = phi float
492 // CHECK: store float [[UP]], float* [[T_VAR1_LHS]],
493 // CHECK: ret void
494 
495 // CHECK: define internal void [[MAIN_MICROTASK1]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, i64 %{{.+}}, i64 %{{.+}}, i32* nonnull %{{.+}}, [2 x i32]* dereferenceable(8) %{{.+}}, [10 x [4 x [[S_FLOAT_TY]]]]* dereferenceable(160) %{{.+}})
496 
497 // Reduction list for runtime.
498 // CHECK: [[RED_LIST:%.+]] = alloca [4 x i8*],
499 
500 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]],
501 
502 // CHECK: [[IDX1:%.+]] = mul nsw i64 1, %{{.+}}
503 // CHECK: [[LB1:%.+]] = getelementptr inbounds i32, i32* %{{.+}}, i64 [[IDX1]]
504 // CHECK: [[LB1_0:%.+]] = getelementptr inbounds i32, i32* [[LB1]], i64 0
505 // CHECK: [[IDX1:%.+]] = mul nsw i64 1, %{{.+}}
506 // CHECK: [[UB1:%.+]] = getelementptr inbounds i32, i32* %{{.+}}, i64 [[IDX1]]
507 // CHECK: [[UB1_UP:%.+]] = getelementptr inbounds i32, i32* [[UB1]], i64 %
508 // CHECK: [[UB_CAST:%.+]] = ptrtoint i32* [[UB1_UP]] to i64
509 // CHECK: [[LB_CAST:%.+]] = ptrtoint i32* [[LB1_0]] to i64
510 // CHECK: [[DIFF:%.+]] = sub i64 [[UB_CAST]], [[LB_CAST]]
511 // CHECK: [[SIZE_1:%.+]] = sdiv exact i64 [[DIFF]], ptrtoint (i32* getelementptr (i32, i32* null, i32 1) to i64)
512 // CHECK: [[ARR_SIZE:%.+]] = add nuw i64 [[SIZE_1]], 1
513 // CHECK: call i8* @llvm.stacksave()
514 // CHECK: [[ARR_PRIV:%.+]] = alloca i32, i64 [[ARR_SIZE]],
515 
516 // Check initialization of private copy.
517 // CHECK: [[END:%.+]] = getelementptr i32, i32* [[ARR_PRIV]], i64 [[ARR_SIZE]]
518 // CHECK: [[ISEMPTY:%.+]] = icmp eq i32* [[ARR_PRIV]], [[END]]
519 // CHECK: br i1 [[ISEMPTY]],
520 // CHECK: phi i32*
521 // CHECK: store i32 0, i32* %
522 // CHECK: [[DONE:%.+]] = icmp eq i32* %{{.+}}, [[END]]
523 // CHECK: br i1 [[DONE]],
524 
525 // CHECK: [[ARRS_PRIV:%.+]] = alloca [[S_FLOAT_TY]], i64 [[ARRS_SIZE:%.+]],
526 
527 // Check initialization of private copy.
528 // CHECK: [[END:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[ARRS_PRIV]], i64 [[ARRS_SIZE]]
529 // CHECK: [[ISEMPTY:%.+]] = icmp eq [[S_FLOAT_TY]]* [[ARRS_PRIV]], [[END]]
530 // CHECK: br i1 [[ISEMPTY]],
531 // CHECK: phi [[S_FLOAT_TY]]*
532 // CHECK: call void @_ZN1SIfEC1Ev([[S_FLOAT_TY]]* %
533 // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[END]]
534 // CHECK: br i1 [[DONE]],
535 
536 // CHECK: [[GTID_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_ADDR]]
537 // CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_REF]]
538 // CHECK: call void @__kmpc_for_static_init_4(
539 // Skip checks for internal operations.
540 // CHECK: call void @__kmpc_for_static_fini(
541 
542 // void *RedList[<n>] = {<ReductionVars>[0], ..., <ReductionVars>[<n>-1]};
543 
544 // CHECK: [[ARR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 0
545 // CHECK: [[BITCAST:%.+]] = bitcast i32* [[ARR_PRIV]] to i8*
546 // CHECK: store i8* [[BITCAST]], i8** [[ARR_PRIV_REF]],
547 // CHECK: [[ARR_SIZE_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 1
548 // CHECK: [[BITCAST:%.+]] = inttoptr i64 [[ARR_SIZE]] to i8*
549 // CHECK: store i8* [[BITCAST]], i8** [[ARR_SIZE_REF]],
550 // CHECK: [[ARRS_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 2
551 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[ARRS_PRIV]] to i8*
552 // CHECK: store i8* [[BITCAST]], i8** [[ARRS_PRIV_REF]],
553 // CHECK: [[ARRS_SIZE_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 3
554 // CHECK: [[BITCAST:%.+]] = inttoptr i64 [[ARRS_SIZE]] to i8*
555 // CHECK: store i8* [[BITCAST]], i8** [[ARRS_SIZE_REF]],
556 
557 // res = __kmpc_reduce(<loc>, <gtid>, <n>, sizeof(RedList), RedList, reduce_func, &<lock>);
558 
559 // CHECK: [[GTID_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_ADDR]]
560 // CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_REF]]
561 // CHECK: [[BITCAST:%.+]] = bitcast [4 x i8*]* [[RED_LIST]] to i8*
562 // 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]])
563 
564 // switch(res)
565 // CHECK: switch i32 [[RES]], label %[[RED_DONE:.+]] [
566 // CHECK: i32 1, label %[[CASE1:.+]]
567 // CHECK: i32 2, label %[[CASE2:.+]]
568 // CHECK: ]
569 
570 // case 1:
571 // CHECK: [[CASE1]]
572 
573 // arr[:] += arr_reduction[:];
574 // CHECK: [[END:%.+]] = getelementptr i32, i32* [[LB1_0]], i64 [[ARR_SIZE]]
575 // CHECK: [[ISEMPTY:%.+]] = icmp eq i32* [[LB1_0]], [[END]]
576 // CHECK: br i1 [[ISEMPTY]],
577 // CHECK: phi i32*
578 // CHECK: [[ADD:%.+]] = add nsw i32 %
579 // CHECK: store i32 [[ADD]], i32* %
580 // CHECK: [[DONE:%.+]] = icmp eq i32* %{{.+}}, [[END]]
581 // CHECK: br i1 [[DONE]],
582 
583 // arrs[:] = var.operator &(arrs_reduction[:]);
584 // CHECK: [[END:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[ARRS_LB:%.+]], i64 [[ARRS_SIZE]]
585 // CHECK: [[ISEMPTY:%.+]] = icmp eq [[S_FLOAT_TY]]* [[ARRS_LB]], [[END]]
586 // CHECK: br i1 [[ISEMPTY]],
587 // CHECK: phi [[S_FLOAT_TY]]*
588 // CHECK: [[AND:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @_ZN1SIfEanERKS0_([[S_FLOAT_TY]]* %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}})
589 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[AND]] to i8*
590 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* [[BITCAST]], i64 4, i32 4, i1 false)
591 // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[END]]
592 // CHECK: br i1 [[DONE]],
593 
594 // __kmpc_end_reduce(<loc>, <gtid>, &<lock>);
595 // CHECK: call void @__kmpc_end_reduce_nowait(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], [8 x i32]* [[REDUCTION_LOCK]])
596 
597 // break;
598 // CHECK: br label %[[RED_DONE]]
599 
600 // case 2:
601 // CHECK: [[CASE2]]
602 
603 // arr[:] += arr_reduction[:];
604 // CHECK: [[END:%.+]] = getelementptr i32, i32* [[LB1_0]], i64 [[ARR_SIZE]]
605 // CHECK: [[ISEMPTY:%.+]] = icmp eq i32* [[LB1_0]], [[END]]
606 // CHECK: br i1 [[ISEMPTY]],
607 // CHECK: phi i32*
608 // CHECK: atomicrmw add i32* %{{.+}}, i32 %{{.+}} monotonic
609 // CHECK: [[DONE:%.+]] = icmp eq i32* %{{.+}}, [[END]]
610 // CHECK: br i1 [[DONE]],
611 
612 // arrs[:] = var.operator &(arrs_reduction[:]);
613 // CHECK: [[END:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[ARRS_LB:%.+]], i64 [[ARRS_SIZE]]
614 // CHECK: [[ISEMPTY:%.+]] = icmp eq [[S_FLOAT_TY]]* [[ARRS_LB]], [[END]]
615 // CHECK: br i1 [[ISEMPTY]],
616 // CHECK: phi [[S_FLOAT_TY]]*
617 // CHECK: call void @__kmpc_critical(
618 // CHECK: [[AND:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @_ZN1SIfEanERKS0_([[S_FLOAT_TY]]* %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}})
619 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[AND]] to i8*
620 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* [[BITCAST]], i64 4, i32 4, i1 false)
621 // CHECK: call void @__kmpc_end_critical(
622 // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[END]]
623 // CHECK: br i1 [[DONE]],
624 
625 // break;
626 // CHECK: br label %[[RED_DONE]]
627 // CHECK: [[RED_DONE]]
628 
629 // Check destruction of private copy.
630 // CHECK: [[END:%.+]] = getelementptr inbounds [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[ARRS_PRIV]], i64 [[ARRS_SIZE]]
631 // CHECK: [[ISEMPTY:%.+]] = icmp eq [[S_FLOAT_TY]]* [[ARRS_PRIV]], [[END]]
632 // CHECK: br i1 [[ISEMPTY]],
633 // CHECK: phi [[S_FLOAT_TY]]*
634 // CHECK: call void @_ZN1SIfED1Ev([[S_FLOAT_TY]]* %
635 // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[ARRS_PRIV]]
636 // CHECK: br i1 [[DONE]],
637 // CHECK: call void @llvm.stackrestore(i8*
638 
639 // CHECK: ret void
640 
641 // void reduce_func(void *lhs[<n>], void *rhs[<n>]) {
642 //  *(Type0*)lhs[0] = ReductionOperation0(*(Type0*)lhs[0], *(Type0*)rhs[0]);
643 //  ...
644 //  *(Type<n>-1*)lhs[<n>-1] = ReductionOperation<n>-1(*(Type<n>-1*)lhs[<n>-1],
645 //  *(Type<n>-1*)rhs[<n>-1]);
646 // }
647 // CHECK: define internal void [[REDUCTION_FUNC]](i8*, i8*)
648 // arr_rhs = (int*)rhs[0];
649 // CHECK: [[ARR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS:%.+]], i64 0, i64 0
650 // CHECK: [[ARR_RHS_VOID:%.+]] = load i8*, i8** [[ARR_RHS_REF]],
651 // CHECK: [[ARR_RHS:%.+]] = bitcast i8* [[ARR_RHS_VOID]] to i32*
652 // arr_lhs = (int*)lhs[0];
653 // CHECK: [[ARR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS:%.+]], i64 0, i64 0
654 // CHECK: [[ARR_LHS_VOID:%.+]] = load i8*, i8** [[ARR_LHS_REF]],
655 // CHECK: [[ARR_LHS:%.+]] = bitcast i8* [[ARR_LHS_VOID]] to i32*
656 
657 // arr_size = (size_t)lhs[1];
658 // CHECK: [[ARR_SIZE_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 1
659 // CHECK: [[ARR_SIZE_VOID:%.+]] = load i8*, i8** [[ARR_SIZE_REF]],
660 // CHECK: [[ARR_SIZE:%.+]] = ptrtoint i8* [[ARR_SIZE_VOID]] to i64
661 
662 // arrs_rhs = (S<float>*)rhs[2];
663 // CHECK: [[ARRS_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 2
664 // CHECK: [[ARRS_RHS_VOID:%.+]] = load i8*, i8** [[ARRS_RHS_REF]],
665 // CHECK: [[ARRS_RHS:%.+]] = bitcast i8* [[ARRS_RHS_VOID]] to [[S_FLOAT_TY]]*
666 // arrs_lhs = (S<float>*)lhs[2];
667 // CHECK: [[ARRS_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 2
668 // CHECK: [[ARRS_LHS_VOID:%.+]] = load i8*, i8** [[ARRS_LHS_REF]],
669 // CHECK: [[ARRS_LHS:%.+]] = bitcast i8* [[ARRS_LHS_VOID]] to [[S_FLOAT_TY]]*
670 
671 // arrs_size = (size_t)lhs[3];
672 // CHECK: [[ARRS_SIZE_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 3
673 // CHECK: [[ARRS_SIZE_VOID:%.+]] = load i8*, i8** [[ARRS_SIZE_REF]],
674 // CHECK: [[ARRS_SIZE:%.+]] = ptrtoint i8* [[ARRS_SIZE_VOID]] to i64
675 
676 // arr_lhs[:] += arr_rhs[:];
677 // CHECK: [[END:%.+]] = getelementptr i32, i32* [[ARR_LHS]], i64 [[ARR_SIZE]]
678 // CHECK: [[ISEMPTY:%.+]] = icmp eq i32* [[ARR_LHS]], [[END]]
679 // CHECK: br i1 [[ISEMPTY]],
680 // CHECK: phi i32*
681 // CHECK: [[ADD:%.+]] = add nsw i32 %
682 // CHECK: store i32 [[ADD]], i32* %
683 // CHECK: [[DONE:%.+]] = icmp eq i32* %{{.+}}, [[END]]
684 // CHECK: br i1 [[DONE]],
685 
686 // arrs_lhs = arrs_lhs.operator &(arrs_rhs);
687 // CHECK: [[END:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[ARRS_LB:%.+]], i64 [[ARRS_SIZE]]
688 // CHECK: [[ISEMPTY:%.+]] = icmp eq [[S_FLOAT_TY]]* [[ARRS_LB]], [[END]]
689 // CHECK: br i1 [[ISEMPTY]],
690 // CHECK: phi [[S_FLOAT_TY]]*
691 // CHECK: [[AND:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @_ZN1SIfEanERKS0_([[S_FLOAT_TY]]* %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}})
692 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[AND]] to i8*
693 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* [[BITCAST]], i64 4, i32 4, i1 false)
694 // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[END]]
695 // CHECK: br i1 [[DONE]],
696 
697 // CHECK: ret void
698 
699 // CHECK: define internal void [[MAIN_MICROTASK2]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, i64 %{{.+}}, i64 %{{.+}}, i32* nonnull %{{.+}}, [10 x [4 x [[S_FLOAT_TY]]]]* dereferenceable(160) %{{.+}})
700 
701 // CHECK: [[ARRS_PRIV:%.+]] = alloca [10 x [4 x [[S_FLOAT_TY]]]],
702 
703 // Reduction list for runtime.
704 // CHECK: [[RED_LIST:%.+]] = alloca [3 x i8*],
705 
706 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]],
707 
708 // CHECK: [[ARR_SIZE:%.+]] = mul nuw i64 %{{.+}}, 4
709 // CHECK: call i8* @llvm.stacksave()
710 // CHECK: [[ARR_PRIV:%.+]] = alloca i32, i64 [[ARR_SIZE]],
711 
712 // Check initialization of private copy.
713 // CHECK: [[END:%.+]] = getelementptr i32, i32* [[ARR_PRIV]], i64 [[ARR_SIZE]]
714 // CHECK: [[ISEMPTY:%.+]] = icmp eq i32* [[ARR_PRIV]], [[END]]
715 // CHECK: br i1 [[ISEMPTY]],
716 // CHECK: phi i32*
717 // CHECK: store i32 0, i32* %
718 // CHECK: [[DONE:%.+]] = icmp eq i32* %{{.+}}, [[END]]
719 // CHECK: br i1 [[DONE]],
720 
721 // Check initialization of private copy.
722 // CHECK: [[LHS_BEGIN:%.+]] = bitcast [10 x [4 x [[S_FLOAT_TY]]]]* %{{.+}} to [[S_FLOAT_TY]]*
723 // 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
724 // CHECK: [[END:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[BEGIN]], i64 40
725 // CHECK: [[ISEMPTY:%.+]] = icmp eq [[S_FLOAT_TY]]* [[BEGIN]], [[END]]
726 // CHECK: br i1 [[ISEMPTY]],
727 // CHECK: phi [[S_FLOAT_TY]]*
728 // CHECK: call void @_ZN1SIfEC1Ev([[S_FLOAT_TY]]* %
729 // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[END]]
730 // CHECK: br i1 [[DONE]],
731 // CHECK: [[ARRS_PRIV_BEGIN:%.+]] = bitcast [10 x [4 x [[S_FLOAT_TY]]]]* [[ARRS_PRIV]] to [[S_FLOAT_TY]]*
732 
733 // CHECK: [[GTID_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_ADDR]]
734 // CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_REF]]
735 // CHECK: call void @__kmpc_for_static_init_4(
736 // Skip checks for internal operations.
737 // CHECK: call void @__kmpc_for_static_fini(
738 
739 // void *RedList[<n>] = {<ReductionVars>[0], ..., <ReductionVars>[<n>-1]};
740 
741 // CHECK: [[ARR_PRIV_REF:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[RED_LIST]], i64 0, i64 0
742 // CHECK: [[BITCAST:%.+]] = bitcast i32* [[ARR_PRIV]] to i8*
743 // CHECK: store i8* [[BITCAST]], i8** [[ARR_PRIV_REF]],
744 // CHECK: [[ARR_SIZE_REF:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[RED_LIST]], i64 0, i64 1
745 // CHECK: [[BITCAST:%.+]] = inttoptr i64 [[ARR_SIZE]] to i8*
746 // CHECK: store i8* [[BITCAST]], i8** [[ARR_SIZE_REF]],
747 // CHECK: [[ARRS_PRIV_REF:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[RED_LIST]], i64 0, i64 2
748 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[ARRS_PRIV_BEGIN]] to i8*
749 // CHECK: store i8* [[BITCAST]], i8** [[ARRS_PRIV_REF]],
750 
751 // res = __kmpc_reduce(<loc>, <gtid>, <n>, sizeof(RedList), RedList, reduce_func, &<lock>);
752 
753 // CHECK: [[GTID_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_ADDR]]
754 // CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_REF]]
755 // CHECK: [[BITCAST:%.+]] = bitcast [3 x i8*]* [[RED_LIST]] to i8*
756 // 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]])
757 
758 // switch(res)
759 // CHECK: switch i32 [[RES]], label %[[RED_DONE:.+]] [
760 // CHECK: i32 1, label %[[CASE1:.+]]
761 // CHECK: i32 2, label %[[CASE2:.+]]
762 // CHECK: ]
763 
764 // case 1:
765 // CHECK: [[CASE1]]
766 
767 // arr[:] += arr_reduction[:];
768 // CHECK: [[END:%.+]] = getelementptr i32, i32* [[LB1_0:%.+]], i64 [[ARR_SIZE]]
769 // CHECK: [[ISEMPTY:%.+]] = icmp eq i32* [[LB1_0]], [[END]]
770 // CHECK: br i1 [[ISEMPTY]],
771 // CHECK: phi i32*
772 // CHECK: [[ADD:%.+]] = add nsw i32 %
773 // CHECK: store i32 [[ADD]], i32* %
774 // CHECK: [[DONE:%.+]] = icmp eq i32* %{{.+}}, [[END]]
775 // CHECK: br i1 [[DONE]],
776 
777 // arrs[:] = var.operator &(arrs_reduction[:]);
778 // CHECK: [[END:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[LHS_BEGIN]], i64 40
779 // CHECK: [[ISEMPTY:%.+]] = icmp eq [[S_FLOAT_TY]]* [[LHS_BEGIN]], [[END]]
780 // CHECK: br i1 [[ISEMPTY]],
781 // CHECK: phi [[S_FLOAT_TY]]*
782 // CHECK: [[AND:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @_ZN1SIfEanERKS0_([[S_FLOAT_TY]]* %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}})
783 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[AND]] to i8*
784 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* [[BITCAST]], i64 4, i32 4, i1 false)
785 // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[END]]
786 // CHECK: br i1 [[DONE]],
787 
788 // __kmpc_end_reduce(<loc>, <gtid>, &<lock>);
789 // CHECK: call void @__kmpc_end_reduce(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], [8 x i32]* [[REDUCTION_LOCK]])
790 
791 // break;
792 // CHECK: br label %[[RED_DONE]]
793 
794 // case 2:
795 // CHECK: [[CASE2]]
796 
797 // arr[:] += arr_reduction[:];
798 // CHECK: [[END:%.+]] = getelementptr i32, i32* [[LB1_0]], i64 [[ARR_SIZE]]
799 // CHECK: [[ISEMPTY:%.+]] = icmp eq i32* [[LB1_0]], [[END]]
800 // CHECK: br i1 [[ISEMPTY]],
801 // CHECK: phi i32*
802 // CHECK: atomicrmw add i32* %{{.+}}, i32 %{{.+}} monotonic
803 // CHECK: [[DONE:%.+]] = icmp eq i32* %{{.+}}, [[END]]
804 // CHECK: br i1 [[DONE]],
805 
806 // arrs[:] = var.operator &(arrs_reduction[:]);
807 // CHECK: [[END:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[LHS_BEGIN]], i64 40
808 // CHECK: [[ISEMPTY:%.+]] = icmp eq [[S_FLOAT_TY]]* [[LHS_BEGIN]], [[END]]
809 // CHECK: br i1 [[ISEMPTY]],
810 // CHECK: phi [[S_FLOAT_TY]]*
811 // CHECK: call void @__kmpc_critical(
812 // CHECK: [[AND:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @_ZN1SIfEanERKS0_([[S_FLOAT_TY]]* %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}})
813 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[AND]] to i8*
814 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* [[BITCAST]], i64 4, i32 4, i1 false)
815 // CHECK: call void @__kmpc_end_critical(
816 // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[END]]
817 // CHECK: br i1 [[DONE]],
818 
819 // break;
820 // CHECK: br label %[[RED_DONE]]
821 // CHECK: [[RED_DONE]]
822 
823 // Check destruction of private copy.
824 // 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
825 // CHECK: [[END:%.+]] = getelementptr inbounds [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[BEGIN]], i64 40
826 // CHECK: br
827 // CHECK: phi [[S_FLOAT_TY]]*
828 // CHECK: call void @_ZN1SIfED1Ev([[S_FLOAT_TY]]* %
829 // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[BEGIN]]
830 // CHECK: br i1 [[DONE]],
831 // CHECK: call void @llvm.stackrestore(i8*
832 // CHECK: call void @__kmpc_barrier(
833 
834 // CHECK: ret void
835 
836 // void reduce_func(void *lhs[<n>], void *rhs[<n>]) {
837 //  *(Type0*)lhs[0] = ReductionOperation0(*(Type0*)lhs[0], *(Type0*)rhs[0]);
838 //  ...
839 //  *(Type<n>-1*)lhs[<n>-1] = ReductionOperation<n>-1(*(Type<n>-1*)lhs[<n>-1],
840 //  *(Type<n>-1*)rhs[<n>-1]);
841 // }
842 // CHECK: define internal void [[REDUCTION_FUNC]](i8*, i8*)
843 // arr_rhs = (int*)rhs[0];
844 // CHECK: [[ARR_RHS_REF:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[RED_LIST_RHS:%.+]], i64 0, i64 0
845 // CHECK: [[ARR_RHS_VOID:%.+]] = load i8*, i8** [[ARR_RHS_REF]],
846 // CHECK: [[ARR_RHS:%.+]] = bitcast i8* [[ARR_RHS_VOID]] to i32*
847 // arr_lhs = (int*)lhs[0];
848 // CHECK: [[ARR_LHS_REF:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[RED_LIST_LHS:%.+]], i64 0, i64 0
849 // CHECK: [[ARR_LHS_VOID:%.+]] = load i8*, i8** [[ARR_LHS_REF]],
850 // CHECK: [[ARR_LHS:%.+]] = bitcast i8* [[ARR_LHS_VOID]] to i32*
851 
852 // arr_size = (size_t)lhs[1];
853 // CHECK: [[ARR_SIZE_REF:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[RED_LIST_LHS]], i64 0, i64 1
854 // CHECK: [[ARR_SIZE_VOID:%.+]] = load i8*, i8** [[ARR_SIZE_REF]],
855 // CHECK: [[ARR_SIZE:%.+]] = ptrtoint i8* [[ARR_SIZE_VOID]] to i64
856 
857 // arrs_rhs = (S<float>*)rhs[2];
858 // CHECK: [[ARRS_RHS_REF:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[RED_LIST_RHS]], i64 0, i64 2
859 // CHECK: [[ARRS_RHS_VOID:%.+]] = load i8*, i8** [[ARRS_RHS_REF]],
860 // CHECK: [[ARRS_RHS:%.+]] = bitcast i8* [[ARRS_RHS_VOID]] to [[S_FLOAT_TY]]*
861 // arrs_lhs = (S<float>*)lhs[2];
862 // CHECK: [[ARRS_LHS_REF:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[RED_LIST_LHS]], i64 0, i64 2
863 // CHECK: [[ARRS_LHS_VOID:%.+]] = load i8*, i8** [[ARRS_LHS_REF]],
864 // CHECK: [[ARRS_LHS:%.+]] = bitcast i8* [[ARRS_LHS_VOID]] to [[S_FLOAT_TY]]*
865 
866 // arr_lhs[:] += arr_rhs[:];
867 // CHECK: [[END:%.+]] = getelementptr i32, i32* [[ARR_LHS]], i64 [[ARR_SIZE]]
868 // CHECK: [[ISEMPTY:%.+]] = icmp eq i32* [[ARR_LHS]], [[END]]
869 // CHECK: br i1 [[ISEMPTY]],
870 // CHECK: phi i32*
871 // CHECK: [[ADD:%.+]] = add nsw i32 %
872 // CHECK: store i32 [[ADD]], i32* %
873 // CHECK: [[DONE:%.+]] = icmp eq i32* %{{.+}}, [[END]]
874 // CHECK: br i1 [[DONE]],
875 
876 // arrs_lhs = arrs_lhs.operator &(arrs_rhs);
877 // CHECK: [[END:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[ARRS_LB:%.+]], i64 40
878 // CHECK: [[ISEMPTY:%.+]] = icmp eq [[S_FLOAT_TY]]* [[ARRS_LB]], [[END]]
879 // CHECK: br i1 [[ISEMPTY]],
880 // CHECK: phi [[S_FLOAT_TY]]*
881 // CHECK: [[AND:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @_ZN1SIfEanERKS0_([[S_FLOAT_TY]]* %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}})
882 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[AND]] to i8*
883 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* [[BITCAST]], i64 4, i32 4, i1 false)
884 // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[END]]
885 // CHECK: br i1 [[DONE]],
886 
887 // CHECK: ret void
888 
889 // CHECK: define internal void [[MAIN_MICROTASK3]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [[S_FLOAT_TY]]*** dereferenceable(8) %{{.+}})
890 
891 // CHECK: [[VAR2_ORIG_ADDR:%.+]] = alloca [[S_FLOAT_TY]]***,
892 
893 // Reduction list for runtime.
894 // CHECK: [[RED_LIST:%.+]] = alloca [2 x i8*],
895 
896 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]],
897 // CHECK: [[VAR2_ORIG:%.+]] = load [[S_FLOAT_TY]]***, [[S_FLOAT_TY]]**** [[VAR2_ORIG_ADDR]],
898 
899 // CHECK: load [[S_FLOAT_TY]]**, [[S_FLOAT_TY]]*** [[VAR2_ORIG]],
900 // CHECK: getelementptr inbounds [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** %{{.+}}, i64 0
901 // CHECK: load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** %
902 // CHECK: [[LOW:%.+]] = getelementptr inbounds [[S_FLOAT_TY]], [[S_FLOAT_TY]]* %{{.+}}, i64 1
903 // CHECK: load [[S_FLOAT_TY]]**, [[S_FLOAT_TY]]*** [[VAR2_ORIG]],
904 // CHECK: getelementptr inbounds [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** %{{.+}}, i64 4
905 // CHECK: load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** %
906 // CHECK: getelementptr inbounds [[S_FLOAT_TY]], [[S_FLOAT_TY]]* %{{.+}}, i64 6
907 // CHECK: [[LD:%.+]] = load [[S_FLOAT_TY]]**, [[S_FLOAT_TY]]*** [[VAR2_ORIG]],
908 // CHECK: [[ORIG_START:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[LD]],
909 // CHECK: [[LAST:%.+]] = ptrtoint [[S_FLOAT_TY]]* %{{.+}} to i64
910 // CHECK: [[FIRST:%.+]] = ptrtoint [[S_FLOAT_TY]]* [[LOW]] to i64
911 // CHECK: [[BYTE_DIF:%.+]] = sub i64 [[LAST]], [[FIRST]]
912 // CHECK: [[DIF:%.+]] = sdiv exact i64 [[BYTE_DIF]], ptrtoint (float* getelementptr (float, float* null, i32 1) to i64)
913 // CHECK: [[SIZE:%.+]] = add nuw i64 [[DIF]], 1
914 // CHECK: call i8* @llvm.stacksave()
915 // CHECK: [[VAR2_PRIV:%.+]] = alloca [[S_FLOAT_TY]], i64 [[SIZE]],
916 // CHECK: [[START:%.+]] = ptrtoint [[S_FLOAT_TY]]* [[ORIG_START]] to i64
917 // CHECK: [[LOW_BOUND:%.+]] = ptrtoint [[S_FLOAT_TY]]* [[LOW]] to i64
918 // CHECK: [[OFFSET_BYTES:%.+]] = sub i64 [[START]], [[LOW_BOUND]]
919 // CHECK: [[OFFSET:%.+]] = sdiv exact i64 [[OFFSET_BYTES]], ptrtoint (float* getelementptr (float, float* null, i32 1) to i64)
920 // CHECK: [[PSEUDO_VAR2_PRIV:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[VAR2_PRIV]], i64 [[OFFSET]]
921 // CHECK: store [[S_FLOAT_TY]]** [[REF:.+]], [[S_FLOAT_TY]]*** %
922 // CHECK: store [[S_FLOAT_TY]]* [[PSEUDO_VAR2_PRIV]], [[S_FLOAT_TY]]** [[REF]]
923 // CHECK: ret void
924 
925 // CHECK: define internal void [[MAIN_MICROTASK4]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [2 x [[S_FLOAT_TY]]]* dereferenceable(8) %{{.+}})
926 
927 // CHECK: [[VVAR2_ORIG_ADDR:%.+]] = alloca [2 x [[S_FLOAT_TY]]]*,
928 
929 // Reduction list for runtime.
930 // CHECK: [[RED_LIST:%.+]] = alloca [2 x i8*],
931 
932 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]],
933 // CHECK: [[VVAR2_ORIG:%.+]] = load [2 x [[S_FLOAT_TY]]]*, [2 x [[S_FLOAT_TY]]]** [[VVAR2_ORIG_ADDR]],
934 
935 // CHECK: [[LOW:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[VVAR2_ORIG]], i64 0, i64 0
936 // CHECK: getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[VVAR2_ORIG]], i64 0, i64 4
937 // CHECK: [[ORIG_START:%.+]] = bitcast [2 x [[S_FLOAT_TY]]]* [[VVAR2_ORIG]] to [[S_FLOAT_TY]]*
938 // CHECK: [[LAST:%.+]] = ptrtoint [[S_FLOAT_TY]]* %{{.+}} to i64
939 // CHECK: [[FIRST:%.+]] = ptrtoint [[S_FLOAT_TY]]* [[LOW]] to i64
940 // CHECK: [[BYTE_DIF:%.+]] = sub i64 [[LAST]], [[FIRST]]
941 // CHECK: [[DIF:%.+]] = sdiv exact i64 [[BYTE_DIF]], ptrtoint (float* getelementptr (float, float* null, i32 1) to i64)
942 // CHECK: [[SIZE:%.+]] = add nuw i64 [[DIF]], 1
943 // CHECK: call i8* @llvm.stacksave()
944 // CHECK: [[VVAR2_PRIV:%.+]] = alloca [[S_FLOAT_TY]], i64 [[SIZE]],
945 // CHECK: [[START:%.+]] = ptrtoint [[S_FLOAT_TY]]* [[ORIG_START]] to i64
946 // CHECK: [[LOW_BOUND:%.+]] = ptrtoint [[S_FLOAT_TY]]* [[LOW]] to i64
947 // CHECK: [[OFFSET_BYTES:%.+]] = sub i64 [[START]], [[LOW_BOUND]]
948 // CHECK: [[OFFSET:%.+]] = sdiv exact i64 [[OFFSET_BYTES]], ptrtoint (float* getelementptr (float, float* null, i32 1) to i64)
949 // CHECK: [[PSEUDO_VVAR2_PRIV:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[VVAR2_PRIV]], i64 [[OFFSET]]
950 // CHECK: [[VVAR2_PRIV:%.+]] = bitcast [[S_FLOAT_TY]]* [[PSEUDO_VVAR2_PRIV]] to [2 x [[S_FLOAT_TY]]]*
951 // CHECK: ret void
952 
953 // CHECK: define internal void [[MAIN_MICROTASK5]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [2 x [[S_FLOAT_TY]]]* dereferenceable(8) %{{.+}})
954 
955 // CHECK: [[VAR3_ORIG_ADDR:%.+]] = alloca [2 x [[S_FLOAT_TY]]]*,
956 
957 // Reduction list for runtime.
958 // CHECK: [[RED_LIST:%.+]] = alloca [2 x i8*],
959 
960 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]],
961 
962 // CHECK: [[VAR3_ORIG:%.+]] = load [2 x [[S_FLOAT_TY]]]*, [2 x [[S_FLOAT_TY]]]** [[VAR3_ORIG_ADDR]],
963 // CHECK: store [2 x [[S_FLOAT_TY]]]* [[VAR3_ORIG]], [2 x [[S_FLOAT_TY]]]** [[VAR3_ORIG_ADDR:%.+]],
964 // CHECK: [[VAR3_ORIG:%.+]] = load [2 x [[S_FLOAT_TY]]]*, [2 x [[S_FLOAT_TY]]]** [[VAR3_ORIG_ADDR]],
965 // CHECK: [[LOW:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[VAR3_ORIG]], i64 0, i64 1
966 // CHECK: [[VAR3_ORIG:%.+]] = load [2 x [[S_FLOAT_TY]]]*, [2 x [[S_FLOAT_TY]]]** [[VAR3_ORIG_ADDR]],
967 // CHECK: getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[VAR3_ORIG]], i64 0, i64 2
968 // CHECK: [[VAR3_ORIG:%.+]] = load [2 x [[S_FLOAT_TY]]]*, [2 x [[S_FLOAT_TY]]]** [[VAR3_ORIG_ADDR]],
969 // CHECK: [[ORIG_START:%.+]] = bitcast [2 x [[S_FLOAT_TY]]]* [[VAR3_ORIG]] to [[S_FLOAT_TY]]*
970 // CHECK: [[LAST:%.+]] = ptrtoint [[S_FLOAT_TY]]* %{{.+}} to i64
971 // CHECK: [[FIRST:%.+]] = ptrtoint [[S_FLOAT_TY]]* [[LOW]] to i64
972 // CHECK: [[BYTE_DIF:%.+]] = sub i64 [[LAST]], [[FIRST]]
973 // CHECK: [[DIF:%.+]] = sdiv exact i64 [[BYTE_DIF]], ptrtoint (float* getelementptr (float, float* null, i32 1) to i64)
974 // CHECK: [[SIZE:%.+]] = add nuw i64 [[DIF]], 1
975 // CHECK: call i8* @llvm.stacksave()
976 // CHECK: [[VAR3_PRIV:%.+]] = alloca [[S_FLOAT_TY]], i64 [[SIZE]],
977 // CHECK: [[START:%.+]] = ptrtoint [[S_FLOAT_TY]]* [[ORIG_START]] to i64
978 // CHECK: [[LOW_BOUND:%.+]] = ptrtoint [[S_FLOAT_TY]]* [[LOW]] to i64
979 // CHECK: [[OFFSET_BYTES:%.+]] = sub i64 [[START]], [[LOW_BOUND]]
980 // CHECK: [[OFFSET:%.+]] = sdiv exact i64 [[OFFSET_BYTES]], ptrtoint (float* getelementptr (float, float* null, i32 1) to i64)
981 // CHECK: [[PSEUDO_VAR3_PRIV:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[VAR3_PRIV]], i64 [[OFFSET]]
982 // CHECK: [[VAR3_PRIV:%.+]] = bitcast [[S_FLOAT_TY]]* [[PSEUDO_VAR3_PRIV]] to [2 x [[S_FLOAT_TY]]]*
983 
984 // CHECK: store [2 x [[S_FLOAT_TY]]]* [[VAR3_PRIV]], [2 x [[S_FLOAT_TY]]]** %
985 
986 // CHECK: ret void
987 
988 // CHECK: define internal void [[MAIN_MICROTASK6]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [2 x [[S_FLOAT_TY]]]* dereferenceable(8) %{{.+}})
989 
990 // CHECK: [[VAR3_ORIG_ADDR:%.+]] = alloca [2 x [[S_FLOAT_TY]]]*,
991 // CHECK: [[VAR3_PRIV:%.+]] = alloca [2 x [[S_FLOAT_TY]]],
992 
993 // Reduction list for runtime.
994 // CHECK: [[RED_LIST:%.+]] = alloca [1 x i8*],
995 
996 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]],
997 
998 // CHECK: [[VAR3_ORIG:%.+]] = load [2 x [[S_FLOAT_TY]]]*, [2 x [[S_FLOAT_TY]]]** [[VAR3_ORIG_ADDR]],
999 // CHECK: store [2 x [[S_FLOAT_TY]]]* [[VAR3_ORIG]], [2 x [[S_FLOAT_TY]]]** [[VAR3_ORIG_ADDR:%.+]],
1000 // CHECK: [[VAR3_ORIG:%.+]] = load [2 x [[S_FLOAT_TY]]]*, [2 x [[S_FLOAT_TY]]]** [[VAR3_ORIG_ADDR]],
1001 // CHECK: bitcast [2 x [[S_FLOAT_TY]]]* [[VAR3_ORIG]] to [[S_FLOAT_TY]]*
1002 // CHECK: getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[VAR3_PRIV]], i32 0, i32 0
1003 // CHECK: getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* %{{.+}}, i64 2
1004 
1005 // CHECK: store [2 x [[S_FLOAT_TY]]]* [[VAR3_PRIV]], [2 x [[S_FLOAT_TY]]]** %
1006 
1007 // CHECK: ret void
1008 
1009 // CHECK: define {{.*}} i{{[0-9]+}} [[TMAIN_INT]]()
1010 // CHECK: [[TEST:%.+]] = alloca [[S_INT_TY]],
1011 // CHECK: call {{.*}} [[S_INT_TY_CONSTR:@.+]]([[S_INT_TY]]* [[TEST]])
1012 // 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
1013 // CHECK: call {{.*}} [[S_INT_TY_DESTR:@.+]]([[S_INT_TY]]*
1014 // CHECK: ret
1015 //
1016 // 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) %{{.+}})
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: alloca i{{[0-9]+}},
1022 // CHECK: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}},
1023 // CHECK: [[VAR_PRIV:%.+]] = alloca [[S_INT_TY]],
1024 // CHECK: [[VAR1_PRIV:%.+]] = alloca [[S_INT_TY]],
1025 // CHECK: [[T_VAR1_PRIV:%.+]] = alloca i{{[0-9]+}},
1026 
1027 // Reduction list for runtime.
1028 // CHECK: [[RED_LIST:%.+]] = alloca [4 x i8*],
1029 
1030 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]],
1031 
1032 // CHECK: [[T_VAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** %
1033 // CHECK: [[VAR1_REF:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** %
1034 // CHECK: [[T_VAR1_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** %
1035 
1036 // For + reduction operation initial value of private variable is 0.
1037 // CHECK: store i{{[0-9]+}} 0, i{{[0-9]+}}* [[T_VAR_PRIV]],
1038 
1039 // For & reduction operation initial value of private variable is ones in all bits.
1040 // CHECK: [[VAR_REF:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** %
1041 // CHECK: call {{.*}} [[S_INT_TY_CONSTR:@.+]]([[S_INT_TY]]* [[VAR_PRIV]])
1042 
1043 // For && reduction operation initial value of private variable is 1.0.
1044 // CHECK: call {{.*}} [[S_INT_TY_CONSTR:@.+]]([[S_INT_TY]]* [[VAR1_PRIV]])
1045 
1046 // For min reduction operation initial value of private variable is largest repesentable value.
1047 // CHECK: store i{{[0-9]+}} 2147483647, i{{[0-9]+}}* [[T_VAR1_PRIV]],
1048 
1049 // CHECK: [[GTID_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_ADDR]]
1050 // CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_REF]]
1051 // CHECK: call void @__kmpc_for_static_init_4(
1052 // Skip checks for internal operations.
1053 // CHECK: call void @__kmpc_for_static_fini(
1054 
1055 // void *RedList[<n>] = {<ReductionVars>[0], ..., <ReductionVars>[<n>-1]};
1056 
1057 // CHECK: [[T_VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 0
1058 // CHECK: [[BITCAST:%.+]] = bitcast i{{[0-9]+}}* [[T_VAR_PRIV]] to i8*
1059 // CHECK: store i8* [[BITCAST]], i8** [[T_VAR_PRIV_REF]],
1060 // CHECK: [[VAR_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 1
1061 // CHECK: [[BITCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR_PRIV]] to i8*
1062 // CHECK: store i8* [[BITCAST]], i8** [[VAR_PRIV_REF]],
1063 // CHECK: [[VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 2
1064 // CHECK: [[BITCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_PRIV]] to i8*
1065 // CHECK: store i8* [[BITCAST]], i8** [[VAR1_PRIV_REF]],
1066 // CHECK: [[T_VAR1_PRIV_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST]], i64 0, i64 3
1067 // CHECK: [[BITCAST:%.+]] = bitcast i{{[0-9]+}}* [[T_VAR1_PRIV]] to i8*
1068 // CHECK: store i8* [[BITCAST]], i8** [[T_VAR1_PRIV_REF]],
1069 
1070 // res = __kmpc_reduce_nowait(<loc>, <gtid>, <n>, sizeof(RedList), RedList, reduce_func, &<lock>);
1071 
1072 // CHECK: [[BITCAST:%.+]] = bitcast [4 x i8*]* [[RED_LIST]] to i8*
1073 // 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]])
1074 
1075 // switch(res)
1076 // CHECK: switch i32 [[RES]], label %[[RED_DONE:.+]] [
1077 // CHECK: i32 1, label %[[CASE1:.+]]
1078 // CHECK: i32 2, label %[[CASE2:.+]]
1079 // CHECK: ]
1080 
1081 // case 1:
1082 // t_var += t_var_reduction;
1083 // CHECK: [[T_VAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_REF]],
1084 // CHECK: [[T_VAR_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_PRIV]],
1085 // CHECK: [[UP:%.+]] = add nsw i{{[0-9]+}} [[T_VAR_VAL]], [[T_VAR_PRIV_VAL]]
1086 // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR_REF]],
1087 
1088 // var = var.operator &(var_reduction);
1089 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_REF]], [[S_INT_TY]]* dereferenceable(4) [[VAR_PRIV]])
1090 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_REF]] to i8*
1091 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8*
1092 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
1093 
1094 // var1 = var1.operator &&(var1_reduction);
1095 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_REF]])
1096 // CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
1097 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]]
1098 // CHECK: [[TRUE]]
1099 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_PRIV]])
1100 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
1101 // CHECK: br label %[[END2]]
1102 // CHECK: [[END2]]
1103 // CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ]
1104 // CHECK: [[CONV:%.+]] = zext i1 [[COND_LVALUE]] to i32
1105 // CHECK:  call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]])
1106 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8*
1107 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8*
1108 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
1109 
1110 // t_var1 = min(t_var1, t_var1_reduction);
1111 // CHECK: [[T_VAR1_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_REF]],
1112 // CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_PRIV]],
1113 // CHECK: [[CMP:%.+]] = icmp slt i{{[0-9]+}} [[T_VAR1_VAL]], [[T_VAR1_PRIV_VAL]]
1114 // CHECK: br i1 [[CMP]]
1115 // CHECK: [[UP:%.+]] = phi i32
1116 // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR1_REF]],
1117 
1118 // __kmpc_end_reduce_nowait(<loc>, <gtid>, &<lock>);
1119 // CHECK: call void @__kmpc_end_reduce_nowait(%{{.+}}* [[REDUCTION_LOC]], i32 [[GTID]], [8 x i32]* [[REDUCTION_LOCK]])
1120 
1121 // break;
1122 // CHECK: br label %[[RED_DONE]]
1123 
1124 // case 2:
1125 // t_var += t_var_reduction;
1126 // CHECK: [[T_VAR_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_PRIV]]
1127 // CHECK: atomicrmw add i32* [[T_VAR_REF]], i32 [[T_VAR_PRIV_VAL]] monotonic
1128 
1129 // var = var.operator &(var_reduction);
1130 // CHECK: call void @__kmpc_critical(
1131 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_REF]], [[S_INT_TY]]* dereferenceable(4) [[VAR_PRIV]])
1132 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_REF]] to i8*
1133 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8*
1134 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
1135 // CHECK: call void @__kmpc_end_critical(
1136 
1137 // var1 = var1.operator &&(var1_reduction);
1138 // CHECK: call void @__kmpc_critical(
1139 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_REF]])
1140 // CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
1141 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]]
1142 // CHECK: [[TRUE]]
1143 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_PRIV]])
1144 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
1145 // CHECK: br label %[[END2]]
1146 // CHECK: [[END2]]
1147 // CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ]
1148 // CHECK: [[CONV:%.+]] = zext i1 [[COND_LVALUE]] to i32
1149 // CHECK:  call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]])
1150 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8*
1151 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8*
1152 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
1153 // CHECK: call void @__kmpc_end_critical(
1154 
1155 // t_var1 = min(t_var1, t_var1_reduction);
1156 // CHECK: [[T_VAR1_PRIV_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_PRIV]]
1157 // CHECK: atomicrmw min i32* [[T_VAR1_REF]], i32 [[T_VAR1_PRIV_VAL]] monotonic
1158 
1159 // break;
1160 // CHECK: br label %[[RED_DONE]]
1161 // CHECK: [[RED_DONE]]
1162 // CHECK-DAG: call {{.*}} [[S_INT_TY_DESTR]]([[S_INT_TY]]* [[VAR_PRIV]])
1163 // CHECK-DAG: call {{.*}} [[S_INT_TY_DESTR]]([[S_INT_TY]]*
1164 // CHECK: ret void
1165 
1166 // void reduce_func(void *lhs[<n>], void *rhs[<n>]) {
1167 //  *(Type0*)lhs[0] = ReductionOperation0(*(Type0*)lhs[0], *(Type0*)rhs[0]);
1168 //  ...
1169 //  *(Type<n>-1*)lhs[<n>-1] = ReductionOperation<n>-1(*(Type<n>-1*)lhs[<n>-1],
1170 //  *(Type<n>-1*)rhs[<n>-1]);
1171 // }
1172 // CHECK: define internal void [[REDUCTION_FUNC]](i8*, i8*)
1173 // t_var_lhs = (i{{[0-9]+}}*)lhs[0];
1174 // CHECK: [[T_VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS:%.+]], i64 0, i64 0
1175 // CHECK: [[T_VAR_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR_RHS_REF]],
1176 // CHECK: [[T_VAR_RHS:%.+]] = bitcast i8* [[T_VAR_RHS_VOID]] to i{{[0-9]+}}*
1177 // t_var_rhs = (i{{[0-9]+}}*)rhs[0];
1178 // CHECK: [[T_VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS:%.+]], i64 0, i64 0
1179 // CHECK: [[T_VAR_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR_LHS_REF]],
1180 // CHECK: [[T_VAR_LHS:%.+]] = bitcast i8* [[T_VAR_LHS_VOID]] to i{{[0-9]+}}*
1181 
1182 // var_lhs = (S<i{{[0-9]+}}>*)lhs[1];
1183 // CHECK: [[VAR_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 1
1184 // CHECK: [[VAR_RHS_VOID:%.+]] = load i8*, i8** [[VAR_RHS_REF]],
1185 // CHECK: [[VAR_RHS:%.+]] = bitcast i8* [[VAR_RHS_VOID]] to [[S_INT_TY]]*
1186 // var_rhs = (S<i{{[0-9]+}}>*)rhs[1];
1187 // CHECK: [[VAR_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 1
1188 // CHECK: [[VAR_LHS_VOID:%.+]] = load i8*, i8** [[VAR_LHS_REF]],
1189 // CHECK: [[VAR_LHS:%.+]] = bitcast i8* [[VAR_LHS_VOID]] to [[S_INT_TY]]*
1190 
1191 // var1_lhs = (S<i{{[0-9]+}}>*)lhs[2];
1192 // CHECK: [[VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 2
1193 // CHECK: [[VAR1_RHS_VOID:%.+]] = load i8*, i8** [[VAR1_RHS_REF]],
1194 // CHECK: [[VAR1_RHS:%.+]] = bitcast i8* [[VAR1_RHS_VOID]] to [[S_INT_TY]]*
1195 // var1_rhs = (S<i{{[0-9]+}}>*)rhs[2];
1196 // CHECK: [[VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 2
1197 // CHECK: [[VAR1_LHS_VOID:%.+]] = load i8*, i8** [[VAR1_LHS_REF]],
1198 // CHECK: [[VAR1_LHS:%.+]] = bitcast i8* [[VAR1_LHS_VOID]] to [[S_INT_TY]]*
1199 
1200 // t_var1_lhs = (i{{[0-9]+}}*)lhs[3];
1201 // CHECK: [[T_VAR1_RHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_RHS]], i64 0, i64 3
1202 // CHECK: [[T_VAR1_RHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_RHS_REF]],
1203 // CHECK: [[T_VAR1_RHS:%.+]] = bitcast i8* [[T_VAR1_RHS_VOID]] to i{{[0-9]+}}*
1204 // t_var1_rhs = (i{{[0-9]+}}*)rhs[3];
1205 // CHECK: [[T_VAR1_LHS_REF:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[RED_LIST_LHS]], i64 0, i64 3
1206 // CHECK: [[T_VAR1_LHS_VOID:%.+]] = load i8*, i8** [[T_VAR1_LHS_REF]],
1207 // CHECK: [[T_VAR1_LHS:%.+]] = bitcast i8* [[T_VAR1_LHS_VOID]] to i{{[0-9]+}}*
1208 
1209 // t_var_lhs += t_var_rhs;
1210 // CHECK: [[T_VAR_LHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_LHS]],
1211 // CHECK: [[T_VAR_RHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_RHS]],
1212 // CHECK: [[UP:%.+]] = add nsw i{{[0-9]+}} [[T_VAR_LHS_VAL]], [[T_VAR_RHS_VAL]]
1213 // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR_LHS]],
1214 
1215 // var_lhs = var_lhs.operator &(var_rhs);
1216 // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_LHS]], [[S_INT_TY]]* dereferenceable(4) [[VAR_RHS]])
1217 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_LHS]] to i8*
1218 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8*
1219 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
1220 
1221 // var1_lhs = var1_lhs.operator &&(var1_rhs);
1222 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_LHS]])
1223 // CHECK: [[VAR1_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
1224 // CHECK: br i1 [[VAR1_BOOL]], label %[[TRUE:.+]], label %[[END2:.+]]
1225 // CHECK: [[TRUE]]
1226 // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_RHS]])
1227 // CHECK: [[VAR1_REDUCTION_BOOL:%.+]] = icmp ne i{{[0-9]+}} [[TO_INT]], 0
1228 // CHECK: br label %[[END2]]
1229 // CHECK: [[END2]]
1230 // CHECK: [[COND_LVALUE:%.+]] = phi i1 [ false, %{{.+}} ], [ [[VAR1_REDUCTION_BOOL]], %[[TRUE]] ]
1231 // CHECK: [[CONV:%.+]] = zext i1 [[COND_LVALUE]] to i32
1232 // CHECK:  call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]])
1233 // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_LHS]] to i8*
1234 // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8*
1235 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false)
1236 
1237 // t_var1_lhs = min(t_var1_lhs, t_var1_rhs);
1238 // CHECK: [[T_VAR1_LHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_LHS]],
1239 // CHECK: [[T_VAR1_RHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_RHS]],
1240 // CHECK: [[CMP:%.+]] = icmp slt i{{[0-9]+}} [[T_VAR1_LHS_VAL]], [[T_VAR1_RHS_VAL]]
1241 // CHECK: br i1 [[CMP]]
1242 // CHECK: [[UP:%.+]] = phi i32
1243 // CHECK: store i{{[0-9]+}} [[UP]], i{{[0-9]+}}* [[T_VAR1_LHS]],
1244 // CHECK: ret void
1245 
1246 #endif
1247 
1248