1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --function-signature
2; RUN: opt -S -attributor -openmpopt  < %s | FileCheck %s
3; RUN: opt -S -passes='attributor,cgscc(openmpopt)'  < %s | FileCheck %s
4;
5target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
6
7%struct.ident_t = type { i32, i32, i32, i32, i8* }
8
9@.str = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00", align 1
10@0 = private unnamed_addr global %struct.ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* @.str, i32 0, i32 0) }, align 8
11@1 = private unnamed_addr global %struct.ident_t { i32 0, i32 322, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* @.str, i32 0, i32 0) }, align 8
12@.gomp_critical_user_.reduction.var = common global [8 x i32] zeroinitializer
13@2 = private unnamed_addr global %struct.ident_t { i32 0, i32 18, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* @.str, i32 0, i32 0) }, align 8
14
15;    void delete_parallel_0(void) {
16;    #pragma omp parallel
17;      { unknown_willreturn(); }
18;    #pragma omp parallel
19;      { readonly_willreturn(); }
20;    #pragma omp parallel
21;      { readnone_willreturn(); }
22;    #pragma omp parallel
23;      {}
24;    }
25;
26; We delete all but the first of the parallel regions in this test.
27define void @delete_parallel_0() {
28; CHECK-LABEL: define {{[^@]+}}@delete_parallel_0()
29; CHECK-NEXT:  entry:
30; CHECK-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* noundef nonnull align 8 dereferenceable(24) @0, i32 noundef 0, void (i32*, i32*, ...)* noundef bitcast (void (i32*, i32*)* @.omp_outlined.willreturn to void (i32*, i32*, ...)*))
31; CHECK-NEXT:    ret void
32;
33entry:
34  call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* nonnull @0, i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* @.omp_outlined.willreturn to void (i32*, i32*, ...)*))
35  call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* nonnull @0, i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* @.omp_outlined.willreturn.0 to void (i32*, i32*, ...)*))
36  call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* nonnull @0, i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* @.omp_outlined.willreturn.1 to void (i32*, i32*, ...)*))
37  call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* nonnull @0, i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* @.omp_outlined.willreturn.2 to void (i32*, i32*, ...)*))
38  ret void
39}
40
41define internal void @.omp_outlined.willreturn(i32* noalias %.global_tid., i32* noalias %.bound_tid.) {
42; CHECK-LABEL: define {{[^@]+}}@.omp_outlined.willreturn
43; CHECK-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #0
44; CHECK-NEXT:  entry:
45; CHECK-NEXT:    call void @unknown() #0
46; CHECK-NEXT:    ret void
47;
48entry:
49  call void @unknown() willreturn
50  ret void
51}
52
53define internal void @.omp_outlined.willreturn.0(i32* noalias %.global_tid., i32* noalias %.bound_tid.) willreturn {
54; CHECK-LABEL: define {{[^@]+}}@.omp_outlined.willreturn.0
55; CHECK-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #1
56; CHECK-NEXT:  entry:
57; CHECK-NEXT:    call void @readonly() #4
58; CHECK-NEXT:    ret void
59;
60entry:
61  call void @readonly()
62  ret void
63}
64
65define internal void @.omp_outlined.willreturn.1(i32* noalias %.global_tid., i32* noalias %.bound_tid.) {
66; CHECK-LABEL: define {{[^@]+}}@.omp_outlined.willreturn.1
67; CHECK-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #2
68; CHECK-NEXT:  entry:
69; CHECK-NEXT:    call void @readnone() #0
70; CHECK-NEXT:    ret void
71;
72entry:
73  call void @readnone() willreturn
74  ret void
75}
76
77define internal void @.omp_outlined.willreturn.2(i32* noalias %.global_tid., i32* noalias %.bound_tid.) {
78; CHECK-LABEL: define {{[^@]+}}@.omp_outlined.willreturn.2
79; CHECK-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #3
80; CHECK-NEXT:  entry:
81; CHECK-NEXT:    ret void
82;
83entry:
84  ret void
85}
86
87;    void delete_parallel_1(void) {
88;    #pragma omp parallel
89;      { unknown(); }
90;    #pragma omp parallel
91;      { readonly(); }
92;    #pragma omp parallel
93;      { readnone(); }
94;    #pragma omp parallel
95;      {}
96;    }
97;
98; We delete only the last parallel regions in this test because the others might not return.
99define void @delete_parallel_1() {
100; CHECK-LABEL: define {{[^@]+}}@delete_parallel_1()
101; CHECK-NEXT:  entry:
102; CHECK-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* noundef nonnull align 8 dereferenceable(24) @0, i32 noundef 0, void (i32*, i32*, ...)* noundef bitcast (void (i32*, i32*)* @.omp_outlined. to void (i32*, i32*, ...)*))
103; CHECK-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* noundef nonnull align 8 dereferenceable(24) @0, i32 noundef 0, void (i32*, i32*, ...)* noundef bitcast (void (i32*, i32*)* @.omp_outlined..0 to void (i32*, i32*, ...)*))
104; CHECK-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* noundef nonnull align 8 dereferenceable(24) @0, i32 noundef 0, void (i32*, i32*, ...)* noundef bitcast (void (i32*, i32*)* @.omp_outlined..1 to void (i32*, i32*, ...)*))
105; CHECK-NEXT:    ret void
106;
107entry:
108  call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* nonnull @0, i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* @.omp_outlined. to void (i32*, i32*, ...)*))
109  call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* nonnull @0, i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* @.omp_outlined..0 to void (i32*, i32*, ...)*))
110  call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* nonnull @0, i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* @.omp_outlined..1 to void (i32*, i32*, ...)*))
111  call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* nonnull @0, i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* @.omp_outlined..2 to void (i32*, i32*, ...)*))
112  ret void
113}
114
115define internal void @.omp_outlined.(i32* noalias %.global_tid., i32* noalias %.bound_tid.) {
116; CHECK-LABEL: define {{[^@]+}}@.omp_outlined.
117; CHECK-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]])
118; CHECK-NEXT:  entry:
119; CHECK-NEXT:    call void @unknown()
120; CHECK-NEXT:    ret void
121;
122entry:
123  call void @unknown()
124  ret void
125}
126
127define internal void @.omp_outlined..0(i32* noalias %.global_tid., i32* noalias %.bound_tid.) {
128; CHECK-LABEL: define {{[^@]+}}@.omp_outlined..0
129; CHECK-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #4
130; CHECK-NEXT:  entry:
131; CHECK-NEXT:    call void @readonly() #4
132; CHECK-NEXT:    ret void
133;
134entry:
135  call void @readonly()
136  ret void
137}
138
139define internal void @.omp_outlined..1(i32* noalias %.global_tid., i32* noalias %.bound_tid.) {
140; CHECK-LABEL: define {{[^@]+}}@.omp_outlined..1
141; CHECK-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #5
142; CHECK-NEXT:  entry:
143; CHECK-NEXT:    call void @readnone()
144; CHECK-NEXT:    ret void
145;
146entry:
147  call void @readnone()
148  ret void
149}
150
151define internal void @.omp_outlined..2(i32* noalias %.global_tid., i32* noalias %.bound_tid.) {
152; CHECK-LABEL: define {{[^@]+}}@.omp_outlined..2
153; CHECK-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #3
154; CHECK-NEXT:  entry:
155; CHECK-NEXT:    ret void
156;
157entry:
158  ret void
159}
160
161;    void delete_parallel_2(void) {
162;      int a = 0;
163;    #pragma omp parallel
164;      {
165;        if (omp_get_thread_num() == 0)
166;          ++a;
167;      }
168;    #pragma omp parallel
169;      {
170;    #pragma omp master
171;        ++a;
172;      }
173;    #pragma omp parallel
174;      {
175;    #pragma omp single
176;        ++a;
177;      }
178;    #pragma omp parallel reduction(+: a)
179;      {
180;        ++a;
181;      }
182;    }
183;
184; FIXME: We do not realize that `a` is dead and all accesses to it can be removed
185;        making the parallel regions readonly and deletable.
186define void @delete_parallel_2() {
187; CHECK-LABEL: define {{[^@]+}}@delete_parallel_2()
188; CHECK-NEXT:  entry:
189; CHECK-NEXT:    [[A:%.*]] = alloca i32, align 4
190; CHECK-NEXT:    [[TMP:%.*]] = bitcast i32* [[A]] to i8*
191; CHECK-NEXT:    call void @llvm.lifetime.start.p0i8(i64 noundef 4, i8* noundef nonnull align 4 dereferenceable(4) [[TMP]]) #0
192; CHECK-NEXT:    store i32 0, i32* [[A]], align 4
193; CHECK-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* noundef nonnull align 8 dereferenceable(24) @0, i32 noundef 1, void (i32*, i32*, ...)* noundef bitcast (void (i32*, i32*, i32*)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i32* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[A]])
194; CHECK-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* noundef nonnull align 8 dereferenceable(24) @0, i32 noundef 1, void (i32*, i32*, ...)* noundef bitcast (void (i32*, i32*, i32*)* @.omp_outlined..4 to void (i32*, i32*, ...)*), i32* nocapture noundef nonnull align 4 dereferenceable(4) [[A]])
195; CHECK-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* noundef nonnull align 8 dereferenceable(24) @0, i32 noundef 1, void (i32*, i32*, ...)* noundef bitcast (void (i32*, i32*, i32*)* @.omp_outlined..5 to void (i32*, i32*, ...)*), i32* nocapture noundef nonnull align 4 dereferenceable(4) [[A]])
196; CHECK-NEXT:    call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* noundef nonnull align 8 dereferenceable(24) @0, i32 noundef 1, void (i32*, i32*, ...)* noundef bitcast (void (i32*, i32*, i32*)* @.omp_outlined..6 to void (i32*, i32*, ...)*), i32* nocapture noundef nonnull align 4 dereferenceable(4) [[A]])
197; CHECK-NEXT:    [[TMP1:%.*]] = bitcast i32* [[A]] to i8*
198; CHECK-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 4, i8* noundef nonnull [[TMP1]])
199; CHECK-NEXT:    ret void
200;
201entry:
202  %a = alloca i32, align 4
203  %tmp = bitcast i32* %a to i8*
204  call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %tmp)
205  store i32 0, i32* %a, align 4
206  call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* nonnull @0, i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*)* @.omp_outlined..3 to void (i32*, i32*, ...)*), i32* nonnull %a)
207  call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* nonnull @0, i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*)* @.omp_outlined..4 to void (i32*, i32*, ...)*), i32* nonnull %a)
208  call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* nonnull @0, i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*)* @.omp_outlined..5 to void (i32*, i32*, ...)*), i32* nonnull %a)
209  call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* nonnull @0, i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*)* @.omp_outlined..6 to void (i32*, i32*, ...)*), i32* nonnull %a)
210  %tmp1 = bitcast i32* %a to i8*
211  call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %tmp1)
212  ret void
213}
214
215define internal void @.omp_outlined..3(i32* noalias %.global_tid., i32* noalias %.bound_tid., i32* dereferenceable(4) %a) {
216; CHECK-LABEL: define {{[^@]+}}@.omp_outlined..3
217; CHECK-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]], i32* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #6
218; CHECK-NEXT:  entry:
219; CHECK-NEXT:    [[CALL:%.*]] = call i32 @omp_get_thread_num() #4
220; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 [[CALL]], 0
221; CHECK-NEXT:    br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_END:%.*]]
222; CHECK:       if.then:
223; CHECK-NEXT:    [[TMP:%.*]] = load i32, i32* [[A]], align 4
224; CHECK-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP]], 1
225; CHECK-NEXT:    store i32 [[INC]], i32* [[A]], align 4
226; CHECK-NEXT:    br label [[IF_END]]
227; CHECK:       if.end:
228; CHECK-NEXT:    ret void
229;
230entry:
231  %call = call i32 @omp_get_thread_num()
232  %cmp = icmp eq i32 %call, 0
233  br i1 %cmp, label %if.then, label %if.end
234
235if.then:                                          ; preds = %entry
236  %tmp = load i32, i32* %a, align 4
237  %inc = add nsw i32 %tmp, 1
238  store i32 %inc, i32* %a, align 4
239  br label %if.end
240
241if.end:                                           ; preds = %if.then, %entry
242  ret void
243}
244
245define internal void @.omp_outlined..4(i32* noalias %.global_tid., i32* noalias %.bound_tid., i32* dereferenceable(4) %a) {
246; CHECK-LABEL: define {{[^@]+}}@.omp_outlined..4
247; CHECK-SAME: (i32* noalias nocapture nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]], i32* nocapture noundef nonnull align 4 dereferenceable(4) [[A:%.*]])
248; CHECK-NEXT:  entry:
249; CHECK-NEXT:    [[TMP:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
250; CHECK-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_master(%struct.ident_t* noundef nonnull @0, i32 [[TMP]])
251; CHECK-NEXT:    [[TMP2:%.*]] = icmp eq i32 [[TMP1]], 0
252; CHECK-NEXT:    br i1 [[TMP2]], label [[OMP_IF_END:%.*]], label [[OMP_IF_THEN:%.*]]
253; CHECK:       omp_if.then:
254; CHECK-NEXT:    [[TMP3:%.*]] = load i32, i32* [[A]], align 4
255; CHECK-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP3]], 1
256; CHECK-NEXT:    store i32 [[INC]], i32* [[A]], align 4
257; CHECK-NEXT:    call void @__kmpc_end_master(%struct.ident_t* noundef nonnull @0, i32 [[TMP]])
258; CHECK-NEXT:    br label [[OMP_IF_END]]
259; CHECK:       omp_if.end:
260; CHECK-NEXT:    ret void
261;
262entry:
263  %tmp = load i32, i32* %.global_tid., align 4
264  %tmp1 = call i32 @__kmpc_master(%struct.ident_t* nonnull @0, i32 %tmp)
265  %tmp2 = icmp eq i32 %tmp1, 0
266  br i1 %tmp2, label %omp_if.end, label %omp_if.then
267
268omp_if.then:                                      ; preds = %entry
269  %tmp3 = load i32, i32* %a, align 4
270  %inc = add nsw i32 %tmp3, 1
271  store i32 %inc, i32* %a, align 4
272  call void @__kmpc_end_master(%struct.ident_t* nonnull @0, i32 %tmp)
273  br label %omp_if.end
274
275omp_if.end:                                       ; preds = %entry, %omp_if.then
276  ret void
277}
278
279declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture)
280
281declare i32 @omp_get_thread_num() inaccessiblememonly nofree nosync nounwind readonly
282
283declare i32 @__kmpc_master(%struct.ident_t*, i32)
284
285declare void @__kmpc_end_master(%struct.ident_t*, i32)
286
287define internal void @.omp_outlined..5(i32* noalias %.global_tid., i32* noalias %.bound_tid., i32* dereferenceable(4) %a) {
288; CHECK-LABEL: define {{[^@]+}}@.omp_outlined..5
289; CHECK-SAME: (i32* noalias nocapture nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]], i32* nocapture noundef nonnull align 4 dereferenceable(4) [[A:%.*]])
290; CHECK-NEXT:  entry:
291; CHECK-NEXT:    [[OMP_GLOBAL_THREAD_NUM:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* noundef nonnull @0)
292; CHECK-NEXT:    [[TMP:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
293; CHECK-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_single(%struct.ident_t* noundef nonnull @0, i32 [[TMP]])
294; CHECK-NEXT:    [[TMP2:%.*]] = icmp eq i32 [[TMP1]], 0
295; CHECK-NEXT:    br i1 [[TMP2]], label [[OMP_IF_END:%.*]], label [[OMP_IF_THEN:%.*]]
296; CHECK:       omp_if.then:
297; CHECK-NEXT:    [[TMP3:%.*]] = load i32, i32* [[A]], align 4
298; CHECK-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP3]], 1
299; CHECK-NEXT:    store i32 [[INC]], i32* [[A]], align 4
300; CHECK-NEXT:    call void @__kmpc_end_single(%struct.ident_t* noundef nonnull @0, i32 [[TMP]])
301; CHECK-NEXT:    br label [[OMP_IF_END]]
302; CHECK:       omp_if.end:
303; CHECK-NEXT:    call void @__kmpc_barrier(%struct.ident_t* noundef nonnull @1, i32 [[OMP_GLOBAL_THREAD_NUM]])
304; CHECK-NEXT:    ret void
305;
306entry:
307  %omp_global_thread_num = call i32 @__kmpc_global_thread_num(%struct.ident_t* nonnull @0)
308  %tmp = load i32, i32* %.global_tid., align 4
309  %tmp1 = call i32 @__kmpc_single(%struct.ident_t* nonnull @0, i32 %tmp)
310  %tmp2 = icmp eq i32 %tmp1, 0
311  br i1 %tmp2, label %omp_if.end, label %omp_if.then
312
313omp_if.then:                                      ; preds = %entry
314  %tmp3 = load i32, i32* %a, align 4
315  %inc = add nsw i32 %tmp3, 1
316  store i32 %inc, i32* %a, align 4
317  call void @__kmpc_end_single(%struct.ident_t* nonnull @0, i32 %tmp)
318  br label %omp_if.end
319
320omp_if.end:                                       ; preds = %entry, %omp_if.then
321  call void @__kmpc_barrier(%struct.ident_t* nonnull @1, i32 %omp_global_thread_num) #6
322  ret void
323}
324
325define internal void @.omp_outlined..6(i32* noalias %.global_tid., i32* noalias %.bound_tid., i32* dereferenceable(4) %a) {
326; CHECK-LABEL: define {{[^@]+}}@.omp_outlined..6
327; CHECK-SAME: (i32* noalias nocapture nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]], i32* nocapture noundef nonnull align 4 dereferenceable(4) [[A:%.*]])
328; CHECK-NEXT:  entry:
329; CHECK-NEXT:    [[A1:%.*]] = alloca i32, align 4
330; CHECK-NEXT:    [[DOTOMP_REDUCTION_RED_LIST:%.*]] = alloca [1 x i8*], align 8
331; CHECK-NEXT:    [[TMP:%.*]] = bitcast i32* [[A1]] to i8*
332; CHECK-NEXT:    call void @llvm.lifetime.start.p0i8(i64 noundef 4, i8* noundef nonnull align 4 [[TMP]]) #0
333; CHECK-NEXT:    store i32 1, i32* [[A1]], align 4
334; CHECK-NEXT:    [[TMP1:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i32**
335; CHECK-NEXT:    store i32* [[A1]], i32** [[TMP1]], align 8
336; CHECK-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
337; CHECK-NEXT:    [[TMP3:%.*]] = bitcast [1 x i8*]* [[DOTOMP_REDUCTION_RED_LIST]] to i8*
338; CHECK-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_reduce_nowait(%struct.ident_t* noundef nonnull @2, i32 [[TMP2]], i32 noundef 1, i64 noundef 8, i8* noundef nonnull align 8 [[TMP3]], void (i8*, i8*)* noundef nonnull @.omp.reduction.reduction_func, [8 x i32]* noundef nonnull @.gomp_critical_user_.reduction.var)
339; CHECK-NEXT:    switch i32 [[TMP4]], label [[DOTOMP_REDUCTION_DEFAULT:%.*]] [
340; CHECK-NEXT:    i32 1, label [[DOTOMP_REDUCTION_CASE1:%.*]]
341; CHECK-NEXT:    i32 2, label [[DOTOMP_REDUCTION_CASE2:%.*]]
342; CHECK-NEXT:    ]
343; CHECK:       .omp.reduction.case1:
344; CHECK-NEXT:    [[TMP5:%.*]] = load i32, i32* [[A]], align 4
345; CHECK-NEXT:    [[TMP6:%.*]] = load i32, i32* [[A1]], align 4
346; CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP5]], [[TMP6]]
347; CHECK-NEXT:    store i32 [[ADD]], i32* [[A]], align 4
348; CHECK-NEXT:    call void @__kmpc_end_reduce_nowait(%struct.ident_t* noundef nonnull @2, i32 [[TMP2]], [8 x i32]* noundef nonnull @.gomp_critical_user_.reduction.var)
349; CHECK-NEXT:    br label [[DOTOMP_REDUCTION_DEFAULT]]
350; CHECK:       .omp.reduction.case2:
351; CHECK-NEXT:    [[TMP7:%.*]] = load i32, i32* [[A1]], align 4
352; CHECK-NEXT:    [[TMP8:%.*]] = atomicrmw add i32* [[A]], i32 [[TMP7]] monotonic
353; CHECK-NEXT:    br label [[DOTOMP_REDUCTION_DEFAULT]]
354; CHECK:       .omp.reduction.default:
355; CHECK-NEXT:    [[TMP9:%.*]] = bitcast i32* [[A1]] to i8*
356; CHECK-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 4, i8* noundef nonnull [[TMP9]])
357; CHECK-NEXT:    ret void
358;
359entry:
360  %a1 = alloca i32, align 4
361  %.omp.reduction.red_list = alloca [1 x i8*], align 8
362  %tmp = bitcast i32* %a1 to i8*
363  call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %tmp)
364  store i32 1, i32* %a1, align 4
365  %tmp1 = bitcast [1 x i8*]* %.omp.reduction.red_list to i32**
366  store i32* %a1, i32** %tmp1, align 8
367  %tmp2 = load i32, i32* %.global_tid., align 4
368  %tmp3 = bitcast [1 x i8*]* %.omp.reduction.red_list to i8*
369  %tmp4 = call i32 @__kmpc_reduce_nowait(%struct.ident_t* nonnull @2, i32 %tmp2, i32 1, i64 8, i8* nonnull %tmp3, void (i8*, i8*)* nonnull @.omp.reduction.reduction_func, [8 x i32]* nonnull @.gomp_critical_user_.reduction.var)
370  switch i32 %tmp4, label %.omp.reduction.default [
371  i32 1, label %.omp.reduction.case1
372  i32 2, label %.omp.reduction.case2
373  ]
374
375.omp.reduction.case1:                             ; preds = %entry
376  %tmp5 = load i32, i32* %a, align 4
377  %tmp6 = load i32, i32* %a1, align 4
378  %add = add nsw i32 %tmp5, %tmp6
379  store i32 %add, i32* %a, align 4
380  call void @__kmpc_end_reduce_nowait(%struct.ident_t* nonnull @2, i32 %tmp2, [8 x i32]* nonnull @.gomp_critical_user_.reduction.var)
381  br label %.omp.reduction.default
382
383.omp.reduction.case2:                             ; preds = %entry
384  %tmp7 = load i32, i32* %a1, align 4
385  %tmp8 = atomicrmw add i32* %a, i32 %tmp7 monotonic
386  br label %.omp.reduction.default
387
388.omp.reduction.default:                           ; preds = %.omp.reduction.case2, %.omp.reduction.case1, %entry
389  %tmp9 = bitcast i32* %a1 to i8*
390  call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %tmp9)
391  ret void
392}
393
394define internal void @.omp.reduction.reduction_func(i8* %arg, i8* %arg1) {
395; CHECK-LABEL: define {{[^@]+}}@.omp.reduction.reduction_func
396; CHECK-SAME: (i8* nocapture nofree nonnull readonly align 8 dereferenceable(8) [[ARG:%.*]], i8* nocapture nofree nonnull readonly align 8 dereferenceable(8) [[ARG1:%.*]]) #{{[0-9]+}}
397; CHECK-NEXT:  entry:
398; CHECK-NEXT:    [[TMP:%.*]] = bitcast i8* [[ARG1]] to i32**
399; CHECK-NEXT:    [[TMP2:%.*]] = load i32*, i32** [[TMP]], align 8
400; CHECK-NEXT:    [[TMP3:%.*]] = bitcast i8* [[ARG]] to i32**
401; CHECK-NEXT:    [[TMP4:%.*]] = load i32*, i32** [[TMP3]], align 8
402; CHECK-NEXT:    [[TMP5:%.*]] = load i32, i32* [[TMP4]], align 4
403; CHECK-NEXT:    [[TMP6:%.*]] = load i32, i32* [[TMP2]], align 4
404; CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP5]], [[TMP6]]
405; CHECK-NEXT:    store i32 [[ADD]], i32* [[TMP4]], align 4
406; CHECK-NEXT:    ret void
407;
408entry:
409  %tmp = bitcast i8* %arg1 to i32**
410  %tmp2 = load i32*, i32** %tmp, align 8
411  %tmp3 = bitcast i8* %arg to i32**
412  %tmp4 = load i32*, i32** %tmp3, align 8
413  %tmp5 = load i32, i32* %tmp4, align 4
414  %tmp6 = load i32, i32* %tmp2, align 4
415  %add = add nsw i32 %tmp5, %tmp6
416  store i32 %add, i32* %tmp4, align 4
417  ret void
418}
419
420declare i32 @__kmpc_single(%struct.ident_t*, i32)
421
422declare void @__kmpc_end_single(%struct.ident_t*, i32)
423
424declare void @__kmpc_barrier(%struct.ident_t*, i32)
425
426declare i32 @__kmpc_global_thread_num(%struct.ident_t*) nofree nosync nounwind readonly
427
428declare i32 @__kmpc_reduce_nowait(%struct.ident_t*, i32, i32, i64, i8*, void (i8*, i8*)*, [8 x i32]*)
429
430declare void @__kmpc_end_reduce_nowait(%struct.ident_t*, i32, [8 x i32]*)
431
432declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture)
433
434declare !callback !2 void @__kmpc_fork_call(%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...)
435
436declare void @unknown()
437
438declare void @readonly() readonly
439
440declare void @readnone() readnone
441
442!0 = !{i32 1, !"wchar_size", i32 4}
443!1 = !{!"clang"}
444!2 = !{!3}
445!3 = !{i64 2, i64 -1, i64 -1, i1 true}
446!4 = !{!5, !5, i64 0}
447!5 = !{!"int", !6, i64 0}
448!6 = !{!"omnipotent char", !7, i64 0}
449!7 = !{!"Simple C/C++ TBAA"}
450