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
11; Function Attrs: nounwind uwtable
12define i32 @main() local_unnamed_addr #0 {
13; CHECK-LABEL: @main(
14; CHECK-NEXT:  entry:
15; CHECK-NEXT:    [[I:%.*]] = alloca i32, align 4
16; CHECK-NEXT:    [[S:%.*]] = alloca i16, align 2
17; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i32* [[I]] to i8*
18; CHECK-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull [[TMP0]])
19; CHECK-NEXT:    store i32 0, i32* [[I]], align 4
20; CHECK-NEXT:    [[TMP1:%.*]] = bitcast i16* [[S]] to i8*
21; CHECK-NEXT:    call void @llvm.lifetime.start.p0i8(i64 2, i8* nonnull [[TMP1]])
22; CHECK-NEXT:    [[CALL:%.*]] = call i32 (i32*, ...) bitcast (i32 (...)* @goo to i32 (i32*, ...)*)(i32* nonnull [[I]])
23; CHECK-NEXT:    [[TMP2:%.*]] = load i32, i32* [[I]], align 4
24; CHECK-NEXT:    [[STOREMERGE6:%.*]] = trunc i32 [[TMP2]] to i16
25; CHECK-NEXT:    store i16 [[STOREMERGE6]], i16* [[S]], align 2
26; CHECK-NEXT:    [[CONV17:%.*]] = and i32 [[TMP2]], 65472
27; CHECK-NEXT:    [[CMP8:%.*]] = icmp eq i32 [[CONV17]], 0
28; CHECK-NEXT:    br i1 [[CMP8]], label [[FOR_BODY_LR_PH:%.*]], label [[FOR_END12:%.*]]
29; CHECK:       for.body.lr.ph:
30; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
31; CHECK:       for.body:
32; CHECK-NEXT:    [[STOREMERGE_IN9:%.*]] = phi i32 [ [[TMP2]], [[FOR_BODY_LR_PH]] ], [ [[ADD:%.*]], [[FOR_INC9:%.*]] ]
33; CHECK-NEXT:    [[CONV52:%.*]] = and i32 [[STOREMERGE_IN9]], 255
34; CHECK-NEXT:    [[CMP63:%.*]] = icmp ult i32 [[TMP2]], [[CONV52]]
35; CHECK-NEXT:    br i1 [[CMP63]], label [[FOR_BODY8_LR_PH:%.*]], label [[FOR_INC9]]
36; CHECK:       for.body8.lr.ph:
37; CHECK-NEXT:    [[CONV3:%.*]] = trunc i32 [[STOREMERGE_IN9]] to i8
38; 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
39; CHECK-NEXT:    [[TMP3:%.*]] = add i8 [[CONV3]], -1
40; CHECK-NEXT:    [[TMP4:%.*]] = zext i8 [[TMP3]] to i32
41; CHECK-NEXT:    [[TMP5:%.*]] = add i32 [[TMP4]], 1
42; CHECK-NEXT:    [[UMIN1:%.*]] = call i32 @llvm.umin.i32(i32 [[TMP2]], i32 [[TMP4]])
43; CHECK-NEXT:    [[TMP6:%.*]] = sub i32 [[TMP5]], [[UMIN1]]
44; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP6]], 8
45; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
46; CHECK:       vector.scevcheck:
47; CHECK-NEXT:    [[TMP7:%.*]] = add i8 [[CONV3]], -1
48; CHECK-NEXT:    [[TMP8:%.*]] = zext i8 [[TMP7]] to i32
49; CHECK-NEXT:    [[UMIN:%.*]] = call i32 @llvm.umin.i32(i32 [[TMP2]], i32 [[TMP8]])
50; CHECK-NEXT:    [[TMP9:%.*]] = sub i32 [[TMP8]], [[UMIN]]
51; CHECK-NEXT:    [[TMP10:%.*]] = trunc i32 [[TMP9]] to i8
52; CHECK-NEXT:    [[MUL:%.*]] = call { i8, i1 } @llvm.umul.with.overflow.i8(i8 1, i8 [[TMP10]])
53; CHECK-NEXT:    [[MUL_RESULT:%.*]] = extractvalue { i8, i1 } [[MUL]], 0
54; CHECK-NEXT:    [[MUL_OVERFLOW:%.*]] = extractvalue { i8, i1 } [[MUL]], 1
55; CHECK-NEXT:    [[TMP12:%.*]] = sub i8 [[TMP7]], [[MUL_RESULT]]
56; CHECK-NEXT:    [[TMP14:%.*]] = icmp ugt i8 [[TMP12]], [[TMP7]]
57; CHECK-NEXT:    [[TMP17:%.*]] = or i1 [[TMP14]], [[MUL_OVERFLOW]]
58; CHECK-NEXT:    [[TMP16:%.*]] = icmp ugt i32 [[TMP9]], 255
59; CHECK-NEXT:    [[TMP18:%.*]] = or i1 [[TMP17]], [[TMP16]]
60; CHECK-NEXT:    br i1 [[TMP18]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
61; CHECK:       vector.ph:
62; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i32 [[TMP6]], 8
63; CHECK-NEXT:    [[N_VEC:%.*]] = sub i32 [[TMP6]], [[N_MOD_VF]]
64; CHECK-NEXT:    [[CAST_CRD:%.*]] = trunc i32 [[N_VEC]] to i8
65; CHECK-NEXT:    [[IND_END:%.*]] = sub i8 [[CONV3]], [[CAST_CRD]]
66; CHECK-NEXT:    [[TMP20:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[DOTPROMOTED]], i32 0
67; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
68; CHECK:       vector.body:
69; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
70; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <4 x i32> [ [[TMP20]], [[VECTOR_PH]] ], [ [[TMP24:%.*]], [[VECTOR_BODY]] ]
71; CHECK-NEXT:    [[VEC_PHI2:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP25:%.*]], [[VECTOR_BODY]] ]
72; CHECK-NEXT:    [[TMP21:%.*]] = trunc i32 [[INDEX]] to i8
73; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = sub i8 [[CONV3]], [[TMP21]]
74; CHECK-NEXT:    [[TMP22:%.*]] = add i8 [[OFFSET_IDX]], 0
75; CHECK-NEXT:    [[TMP23:%.*]] = add i8 [[OFFSET_IDX]], -4
76; CHECK-NEXT:    [[TMP24]] = add <4 x i32> [[VEC_PHI]], <i32 1, i32 1, i32 1, i32 1>
77; CHECK-NEXT:    [[TMP25]] = add <4 x i32> [[VEC_PHI2]], <i32 1, i32 1, i32 1, i32 1>
78; CHECK-NEXT:    [[TMP26:%.*]] = add i8 [[TMP22]], -1
79; CHECK-NEXT:    [[TMP27:%.*]] = add i8 [[TMP23]], -1
80; CHECK-NEXT:    [[TMP28:%.*]] = zext i8 [[TMP26]] to i32
81; CHECK-NEXT:    [[TMP29:%.*]] = zext i8 [[TMP27]] to i32
82; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 8
83; CHECK-NEXT:    [[TMP30:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
84; CHECK-NEXT:    br i1 [[TMP30]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
85; CHECK:       middle.block:
86; CHECK-NEXT:    [[BIN_RDX:%.*]] = add <4 x i32> [[TMP25]], [[TMP24]]
87; CHECK-NEXT:    [[TMP31:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[BIN_RDX]])
88; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[TMP6]], [[N_VEC]]
89; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_COND4_FOR_INC9_CRIT_EDGE:%.*]], label [[SCALAR_PH]]
90; CHECK:       scalar.ph:
91; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[CONV3]], [[FOR_BODY8_LR_PH]] ], [ [[CONV3]], [[VECTOR_SCEVCHECK]] ]
92; CHECK-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ [[DOTPROMOTED]], [[FOR_BODY8_LR_PH]] ], [ [[DOTPROMOTED]], [[VECTOR_SCEVCHECK]] ], [ [[TMP31]], [[MIDDLE_BLOCK]] ]
93; CHECK-NEXT:    br label [[FOR_BODY8:%.*]]
94; CHECK:       for.body8:
95; CHECK-NEXT:    [[INC5:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY8]] ]
96; CHECK-NEXT:    [[C_04:%.*]] = phi i8 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[DEC:%.*]], [[FOR_BODY8]] ]
97; CHECK-NEXT:    [[INC]] = add i32 [[INC5]], 1
98; CHECK-NEXT:    [[DEC]] = add i8 [[C_04]], -1
99; CHECK-NEXT:    [[CONV5:%.*]] = zext i8 [[DEC]] to i32
100; CHECK-NEXT:    [[CMP6:%.*]] = icmp ult i32 [[TMP2]], [[CONV5]]
101; CHECK-NEXT:    br i1 [[CMP6]], label [[FOR_BODY8]], label [[FOR_COND4_FOR_INC9_CRIT_EDGE]], !llvm.loop [[LOOP2:![0-9]+]]
102; CHECK:       for.cond4.for.inc9_crit_edge:
103; CHECK-NEXT:    [[INC_LCSSA:%.*]] = phi i32 [ [[INC]], [[FOR_BODY8]] ], [ [[TMP31]], [[MIDDLE_BLOCK]] ]
104; 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
105; CHECK-NEXT:    br label [[FOR_INC9]]
106; CHECK:       for.inc9:
107; CHECK-NEXT:    [[CONV10:%.*]] = and i32 [[STOREMERGE_IN9]], 65535
108; CHECK-NEXT:    [[ADD]] = add nuw nsw i32 [[CONV10]], 1
109; CHECK-NEXT:    [[CONV1:%.*]] = and i32 [[ADD]], 65472
110; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 [[CONV1]], 0
111; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_COND_FOR_END12_CRIT_EDGE:%.*]]
112; CHECK:       for.cond.for.end12_crit_edge:
113; CHECK-NEXT:    [[ADD_LCSSA:%.*]] = phi i32 [ [[ADD]], [[FOR_INC9]] ]
114; CHECK-NEXT:    [[STOREMERGE:%.*]] = trunc i32 [[ADD_LCSSA]] to i16
115; CHECK-NEXT:    store i16 [[STOREMERGE]], i16* [[S]], align 2
116; CHECK-NEXT:    br label [[FOR_END12]]
117; CHECK:       for.end12:
118; CHECK-NEXT:    [[CALL13:%.*]] = call i32 (i16*, ...) bitcast (i32 (...)* @foo to i32 (i16*, ...)*)(i16* nonnull [[S]])
119; CHECK-NEXT:    call void @llvm.lifetime.end.p0i8(i64 2, i8* nonnull [[TMP1]])
120; CHECK-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull [[TMP0]])
121; CHECK-NEXT:    ret i32 0
122;
123entry:
124  %i = alloca i32, align 4
125  %s = alloca i16, align 2
126  %0 = bitcast i32* %i to i8*
127  call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %0) #3
128  store i32 0, i32* %i, align 4
129  %1 = bitcast i16* %s to i8*
130  call void @llvm.lifetime.start.p0i8(i64 2, i8* nonnull %1) #3
131  %call = call i32 (i32*, ...) bitcast (i32 (...)* @goo to i32 (i32*, ...)*)(i32* nonnull %i) #3
132  %2 = load i32, i32* %i, align 4
133  %storemerge6 = trunc i32 %2 to i16
134  store i16 %storemerge6, i16* %s, align 2
135  %conv17 = and i32 %2, 65472
136  %cmp8 = icmp eq i32 %conv17, 0
137  br i1 %cmp8, label %for.body.lr.ph, label %for.end12
138
139for.body.lr.ph:                                   ; preds = %entry
140  br label %for.body
141
142for.body:                                         ; preds = %for.body.lr.ph, %for.inc9
143  %storemerge.in9 = phi i32 [ %2, %for.body.lr.ph ], [ %add, %for.inc9 ]
144  %conv52 = and i32 %storemerge.in9, 255
145  %cmp63 = icmp ult i32 %2, %conv52
146  br i1 %cmp63, label %for.body8.lr.ph, label %for.inc9
147
148for.body8.lr.ph:                                  ; preds = %for.body
149  %conv3 = trunc i32 %storemerge.in9 to i8
150  %.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
151  br label %for.body8
152
153for.body8:                                        ; preds = %for.body8.lr.ph, %for.body8
154  %inc5 = phi i32 [ %.promoted, %for.body8.lr.ph ], [ %inc, %for.body8 ]
155  %c.04 = phi i8 [ %conv3, %for.body8.lr.ph ], [ %dec, %for.body8 ]
156  %inc = add i32 %inc5, 1
157  %dec = add i8 %c.04, -1
158  %conv5 = zext i8 %dec to i32
159  %cmp6 = icmp ult i32 %2, %conv5
160  br i1 %cmp6, label %for.body8, label %for.cond4.for.inc9_crit_edge
161
162for.cond4.for.inc9_crit_edge:                     ; preds = %for.body8
163  %inc.lcssa = phi i32 [ %inc, %for.body8 ]
164  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
165  br label %for.inc9
166
167for.inc9:                                         ; preds = %for.cond4.for.inc9_crit_edge, %for.body
168  %conv10 = and i32 %storemerge.in9, 65535
169  %add = add nuw nsw i32 %conv10, 1
170  %conv1 = and i32 %add, 65472
171  %cmp = icmp eq i32 %conv1, 0
172  br i1 %cmp, label %for.body, label %for.cond.for.end12_crit_edge
173
174for.cond.for.end12_crit_edge:                     ; preds = %for.inc9
175  %add.lcssa = phi i32 [ %add, %for.inc9 ]
176  %storemerge = trunc i32 %add.lcssa to i16
177  store i16 %storemerge, i16* %s, align 2
178  br label %for.end12
179
180for.end12:                                        ; preds = %for.cond.for.end12_crit_edge, %entry
181  %call13 = call i32 (i16*, ...) bitcast (i32 (...)* @foo to i32 (i16*, ...)*)(i16* nonnull %s) #3
182  call void @llvm.lifetime.end.p0i8(i64 2, i8* nonnull %1) #3
183  call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %0) #3
184  ret i32 0
185}
186
187; Function Attrs: argmemonly nounwind
188declare void @llvm.lifetime.start.p0i8(i64, i8* nocapture) #1
189
190declare i32 @goo(...) local_unnamed_addr #2
191
192declare i32 @foo(...) local_unnamed_addr #2
193
194; Function Attrs: argmemonly nounwind
195declare void @llvm.lifetime.end.p0i8(i64, i8* nocapture) #1
196