1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -loop-vectorize -mtriple=x86_64-unknown-linux-gnu -S < %s | FileCheck %s
3
4; The test checks that there is no assert caused by issue described in PR35432
5
6target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
7target triple = "x86_64-unknown-linux-gnu"
8
9@a = common local_unnamed_addr global [192 x [192 x i32]] zeroinitializer, align 16
10
11define i32 @main(i32* %ptr) {
12; CHECK-LABEL: @main(
13; CHECK-NEXT:  entry:
14; CHECK-NEXT:    [[I:%.*]] = alloca i32, align 4
15; CHECK-NEXT:    [[S:%.*]] = alloca i16, align 2
16; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i32* [[I]] to i8*
17; CHECK-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull [[TMP0]])
18; CHECK-NEXT:    store i32 0, i32* [[I]], align 4
19; CHECK-NEXT:    [[TMP1:%.*]] = bitcast i16* [[S]] to i8*
20; CHECK-NEXT:    call void @llvm.lifetime.start.p0i8(i64 2, i8* nonnull [[TMP1]])
21; CHECK-NEXT:    [[CALL:%.*]] = call i32 (i32*, ...) bitcast (i32 (...)* @goo to i32 (i32*, ...)*)(i32* nonnull [[I]])
22; CHECK-NEXT:    [[TMP2:%.*]] = load i32, i32* [[I]], align 4
23; CHECK-NEXT:    [[STOREMERGE6:%.*]] = trunc i32 [[TMP2]] to i16
24; CHECK-NEXT:    store i16 [[STOREMERGE6]], i16* [[S]], align 2
25; CHECK-NEXT:    [[CONV17:%.*]] = and i32 [[TMP2]], 65472
26; CHECK-NEXT:    [[CMP8:%.*]] = icmp eq i32 [[CONV17]], 0
27; CHECK-NEXT:    br i1 [[CMP8]], label [[FOR_BODY_LR_PH:%.*]], label [[FOR_END12:%.*]]
28; CHECK:       for.body.lr.ph:
29; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
30; CHECK:       for.body:
31; CHECK-NEXT:    [[STOREMERGE_IN9:%.*]] = phi i32 [ [[TMP2]], [[FOR_BODY_LR_PH]] ], [ [[ADD:%.*]], [[FOR_INC9:%.*]] ]
32; CHECK-NEXT:    [[CONV52:%.*]] = and i32 [[STOREMERGE_IN9]], 255
33; CHECK-NEXT:    [[CMP63:%.*]] = icmp ult i32 [[TMP2]], [[CONV52]]
34; CHECK-NEXT:    br i1 [[CMP63]], label [[FOR_BODY8_LR_PH:%.*]], label [[FOR_INC9]]
35; CHECK:       for.body8.lr.ph:
36; CHECK-NEXT:    [[CONV3:%.*]] = trunc i32 [[STOREMERGE_IN9]] to i8
37; CHECK-NEXT:    [[DOTPROMOTED:%.*]] = load i32, i32* getelementptr inbounds ([192 x [192 x i32]], [192 x [192 x i32]]* @a, i64 0, i64 0, i64 0), align 16
38; CHECK-NEXT:    [[TMP3:%.*]] = add i8 [[CONV3]], -1
39; CHECK-NEXT:    [[TMP4:%.*]] = zext i8 [[TMP3]] to i32
40; CHECK-NEXT:    [[TMP5:%.*]] = add i32 [[TMP4]], 1
41; CHECK-NEXT:    [[UMIN1:%.*]] = call i32 @llvm.umin.i32(i32 [[TMP2]], i32 [[TMP4]])
42; CHECK-NEXT:    [[TMP6:%.*]] = sub i32 [[TMP5]], [[UMIN1]]
43; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP6]], 32
44; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
45; CHECK:       vector.scevcheck:
46; CHECK-NEXT:    [[TMP7:%.*]] = add i8 [[CONV3]], -1
47; CHECK-NEXT:    [[TMP8:%.*]] = zext i8 [[TMP7]] to i32
48; CHECK-NEXT:    [[UMIN:%.*]] = call i32 @llvm.umin.i32(i32 [[TMP2]], i32 [[TMP8]])
49; CHECK-NEXT:    [[TMP9:%.*]] = sub i32 [[TMP8]], [[UMIN]]
50; CHECK-NEXT:    [[TMP10:%.*]] = trunc i32 [[TMP9]] to i8
51; CHECK-NEXT:    [[MUL:%.*]] = call { i8, i1 } @llvm.umul.with.overflow.i8(i8 1, i8 [[TMP10]])
52; CHECK-NEXT:    [[MUL_RESULT:%.*]] = extractvalue { i8, i1 } [[MUL]], 0
53; CHECK-NEXT:    [[MUL_OVERFLOW:%.*]] = extractvalue { i8, i1 } [[MUL]], 1
54; CHECK-NEXT:    [[TMP11:%.*]] = sub i8 [[TMP7]], [[MUL_RESULT]]
55; CHECK-NEXT:    [[TMP12:%.*]] = icmp ugt i8 [[TMP11]], [[TMP7]]
56; CHECK-NEXT:    [[TMP13:%.*]] = or i1 [[TMP12]], [[MUL_OVERFLOW]]
57; CHECK-NEXT:    [[TMP14:%.*]] = icmp ugt i32 [[TMP9]], 255
58; CHECK-NEXT:    [[TMP15:%.*]] = or i1 [[TMP13]], [[TMP14]]
59; CHECK-NEXT:    [[TMP16:%.*]] = add i32 [[DOTPROMOTED]], 1
60; CHECK-NEXT:    [[TMP17:%.*]] = add i32 [[TMP16]], [[TMP9]]
61; CHECK-NEXT:    [[TMP18:%.*]] = icmp slt i32 [[TMP17]], [[TMP16]]
62; CHECK-NEXT:    [[TMP19:%.*]] = or i1 [[TMP15]], [[TMP18]]
63; CHECK-NEXT:    br i1 [[TMP19]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
64; CHECK:       vector.ph:
65; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i32 [[TMP6]], 8
66; CHECK-NEXT:    [[N_VEC:%.*]] = sub i32 [[TMP6]], [[N_MOD_VF]]
67; CHECK-NEXT:    [[IND_END:%.*]] = add i32 [[DOTPROMOTED]], [[N_VEC]]
68; CHECK-NEXT:    [[CAST_CRD:%.*]] = trunc i32 [[N_VEC]] to i8
69; CHECK-NEXT:    [[IND_END3:%.*]] = sub i8 [[CONV3]], [[CAST_CRD]]
70; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
71; CHECK:       vector.body:
72; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
73; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = add i32 [[DOTPROMOTED]], [[INDEX]]
74; CHECK-NEXT:    [[TMP20:%.*]] = add i32 [[OFFSET_IDX]], 0
75; CHECK-NEXT:    [[TMP21:%.*]] = add i32 [[OFFSET_IDX]], 4
76; CHECK-NEXT:    [[TMP25:%.*]] = add i32 [[TMP20]], 1
77; CHECK-NEXT:    [[TMP26:%.*]] = add i32 [[TMP21]], 1
78; CHECK-NEXT:    [[TMP29:%.*]] = getelementptr inbounds i32, i32* [[PTR:%.*]], i32 [[TMP25]]
79; CHECK-NEXT:    [[TMP30:%.*]] = getelementptr inbounds i32, i32* [[PTR]], i32 [[TMP26]]
80; CHECK-NEXT:    [[TMP31:%.*]] = getelementptr inbounds i32, i32* [[TMP29]], i32 0
81; CHECK-NEXT:    [[TMP32:%.*]] = bitcast i32* [[TMP31]] to <4 x i32>*
82; CHECK-NEXT:    store <4 x i32> zeroinitializer, <4 x i32>* [[TMP32]], align 4
83; CHECK-NEXT:    [[TMP33:%.*]] = getelementptr inbounds i32, i32* [[TMP29]], i32 4
84; CHECK-NEXT:    [[TMP34:%.*]] = bitcast i32* [[TMP33]] to <4 x i32>*
85; CHECK-NEXT:    store <4 x i32> zeroinitializer, <4 x i32>* [[TMP34]], align 4
86; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 8
87; CHECK-NEXT:    [[TMP37:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
88; CHECK-NEXT:    br i1 [[TMP37]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
89; CHECK:       middle.block:
90; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[TMP6]], [[N_VEC]]
91; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_COND4_FOR_INC9_CRIT_EDGE:%.*]], label [[SCALAR_PH]]
92; CHECK:       scalar.ph:
93; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[DOTPROMOTED]], [[FOR_BODY8_LR_PH]] ], [ [[DOTPROMOTED]], [[VECTOR_SCEVCHECK]] ]
94; CHECK-NEXT:    [[BC_RESUME_VAL2:%.*]] = phi i8 [ [[IND_END3]], [[MIDDLE_BLOCK]] ], [ [[CONV3]], [[FOR_BODY8_LR_PH]] ], [ [[CONV3]], [[VECTOR_SCEVCHECK]] ]
95; CHECK-NEXT:    br label [[FOR_BODY8:%.*]]
96; CHECK:       for.body8:
97; CHECK-NEXT:    [[INC5:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY8]] ]
98; CHECK-NEXT:    [[C_04:%.*]] = phi i8 [ [[BC_RESUME_VAL2]], [[SCALAR_PH]] ], [ [[DEC:%.*]], [[FOR_BODY8]] ]
99; CHECK-NEXT:    [[INC]] = add i32 [[INC5]], 1
100; CHECK-NEXT:    [[DEC]] = add i8 [[C_04]], -1
101; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds i32, i32* [[PTR]], i32 [[INC]]
102; CHECK-NEXT:    store i32 0, i32* [[GEP]], align 4
103; CHECK-NEXT:    [[CONV5:%.*]] = zext i8 [[DEC]] to i32
104; CHECK-NEXT:    [[CMP6:%.*]] = icmp ult i32 [[TMP2]], [[CONV5]]
105; CHECK-NEXT:    br i1 [[CMP6]], label [[FOR_BODY8]], label [[FOR_COND4_FOR_INC9_CRIT_EDGE]], !llvm.loop [[LOOP2:![0-9]+]]
106; CHECK:       for.cond4.for.inc9_crit_edge:
107; CHECK-NEXT:    [[INC_LCSSA:%.*]] = phi i32 [ [[INC]], [[FOR_BODY8]] ], [ [[IND_END]], [[MIDDLE_BLOCK]] ]
108; CHECK-NEXT:    store i32 [[INC_LCSSA]], i32* getelementptr inbounds ([192 x [192 x i32]], [192 x [192 x i32]]* @a, i64 0, i64 0, i64 0), align 16
109; CHECK-NEXT:    br label [[FOR_INC9]]
110; CHECK:       for.inc9:
111; CHECK-NEXT:    [[CONV10:%.*]] = and i32 [[STOREMERGE_IN9]], 65535
112; CHECK-NEXT:    [[ADD]] = add nuw nsw i32 [[CONV10]], 1
113; CHECK-NEXT:    [[CONV1:%.*]] = and i32 [[ADD]], 65472
114; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 [[CONV1]], 0
115; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_COND_FOR_END12_CRIT_EDGE:%.*]]
116; CHECK:       for.cond.for.end12_crit_edge:
117; CHECK-NEXT:    [[ADD_LCSSA:%.*]] = phi i32 [ [[ADD]], [[FOR_INC9]] ]
118; CHECK-NEXT:    [[STOREMERGE:%.*]] = trunc i32 [[ADD_LCSSA]] to i16
119; CHECK-NEXT:    store i16 [[STOREMERGE]], i16* [[S]], align 2
120; CHECK-NEXT:    br label [[FOR_END12]]
121; CHECK:       for.end12:
122; CHECK-NEXT:    [[CALL13:%.*]] = call i32 (i16*, ...) bitcast (i32 (...)* @foo to i32 (i16*, ...)*)(i16* nonnull [[S]])
123; CHECK-NEXT:    call void @llvm.lifetime.end.p0i8(i64 2, i8* nonnull [[TMP1]])
124; CHECK-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull [[TMP0]])
125; CHECK-NEXT:    ret i32 0
126;
127entry:
128  %i = alloca i32, align 4
129  %s = alloca i16, align 2
130  %0 = bitcast i32* %i to i8*
131  call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %0) #3
132  store i32 0, i32* %i, align 4
133  %1 = bitcast i16* %s to i8*
134  call void @llvm.lifetime.start.p0i8(i64 2, i8* nonnull %1) #3
135  %call = call i32 (i32*, ...) bitcast (i32 (...)* @goo to i32 (i32*, ...)*)(i32* nonnull %i) #3
136  %2 = load i32, i32* %i, align 4
137  %storemerge6 = trunc i32 %2 to i16
138  store i16 %storemerge6, i16* %s, align 2
139  %conv17 = and i32 %2, 65472
140  %cmp8 = icmp eq i32 %conv17, 0
141  br i1 %cmp8, label %for.body.lr.ph, label %for.end12
142
143for.body.lr.ph:                                   ; preds = %entry
144  br label %for.body
145
146for.body:                                         ; preds = %for.body.lr.ph, %for.inc9
147  %storemerge.in9 = phi i32 [ %2, %for.body.lr.ph ], [ %add, %for.inc9 ]
148  %conv52 = and i32 %storemerge.in9, 255
149  %cmp63 = icmp ult i32 %2, %conv52
150  br i1 %cmp63, label %for.body8.lr.ph, label %for.inc9
151
152for.body8.lr.ph:                                  ; preds = %for.body
153  %conv3 = trunc i32 %storemerge.in9 to i8
154  %.promoted = load i32, i32* getelementptr inbounds ([192 x [192 x i32]], [192 x [192 x i32]]* @a, i64 0, i64 0, i64 0), align 16
155  br label %for.body8
156
157for.body8:                                        ; preds = %for.body8.lr.ph, %for.body8
158  %inc5 = phi i32 [ %.promoted, %for.body8.lr.ph ], [ %inc, %for.body8 ]
159  %c.04 = phi i8 [ %conv3, %for.body8.lr.ph ], [ %dec, %for.body8 ]
160  %inc = add i32 %inc5, 1
161  %dec = add i8 %c.04, -1
162  %gep = getelementptr inbounds i32, i32* %ptr, i32 %inc
163  store i32 0, i32* %gep
164  %conv5 = zext i8 %dec to i32
165  %cmp6 = icmp ult i32 %2, %conv5
166  br i1 %cmp6, label %for.body8, label %for.cond4.for.inc9_crit_edge
167
168for.cond4.for.inc9_crit_edge:                     ; preds = %for.body8
169  %inc.lcssa = phi i32 [ %inc, %for.body8 ]
170  store i32 %inc.lcssa, i32* getelementptr inbounds ([192 x [192 x i32]], [192 x [192 x i32]]* @a, i64 0, i64 0, i64 0), align 16
171  br label %for.inc9
172
173for.inc9:                                         ; preds = %for.cond4.for.inc9_crit_edge, %for.body
174  %conv10 = and i32 %storemerge.in9, 65535
175  %add = add nuw nsw i32 %conv10, 1
176  %conv1 = and i32 %add, 65472
177  %cmp = icmp eq i32 %conv1, 0
178  br i1 %cmp, label %for.body, label %for.cond.for.end12_crit_edge
179
180for.cond.for.end12_crit_edge:                     ; preds = %for.inc9
181  %add.lcssa = phi i32 [ %add, %for.inc9 ]
182  %storemerge = trunc i32 %add.lcssa to i16
183  store i16 %storemerge, i16* %s, align 2
184  br label %for.end12
185
186for.end12:                                        ; preds = %for.cond.for.end12_crit_edge, %entry
187  %call13 = call i32 (i16*, ...) bitcast (i32 (...)* @foo to i32 (i16*, ...)*)(i16* nonnull %s) #3
188  call void @llvm.lifetime.end.p0i8(i64 2, i8* nonnull %1) #3
189  call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %0) #3
190  ret i32 0
191}
192
193; Function Attrs: argmemonly nounwind
194declare void @llvm.lifetime.start.p0i8(i64, i8* nocapture) #1
195
196declare i32 @goo(...) local_unnamed_addr #2
197
198declare i32 @foo(...) local_unnamed_addr #2
199
200; Function Attrs: argmemonly nounwind
201declare void @llvm.lifetime.end.p0i8(i64, i8* nocapture) #1
202