1; RUN: opt < %s -indvars -S | FileCheck %s
2target triple = "aarch64--linux-gnu"
3
4; Check the loop exit i32 compare instruction and operand are widened to i64
5; instead of truncating IV before its use in the i32 compare instruction.
6
7@idx = common global i32 0, align 4
8@e = common global i32 0, align 4
9@ptr = common global i32* null, align 8
10
11; CHECK-LABEL: @test1
12; CHECK: for.body.lr.ph:
13; CHECK: sext i32
14; CHECK: for.cond:
15; CHECK: icmp slt i64
16; CHECK: for.body:
17; CHECK: phi i64
18
19define i32 @test1() {
20entry:
21  store i32 -1, i32* @idx, align 4
22  %0 = load i32* @e, align 4
23  %cmp4 = icmp slt i32 %0, 0
24  br i1 %cmp4, label %for.end.loopexit, label %for.body.lr.ph
25
26for.body.lr.ph:
27  %1 = load i32** @ptr, align 8
28  %2 = load i32* @e, align 4
29  br label %for.body
30
31for.cond:
32  %inc = add nsw i32 %i.05, 1
33  %cmp = icmp slt i32 %i.05, %2
34  br i1 %cmp, label %for.body, label %for.cond.for.end.loopexit_crit_edge
35
36for.body:
37  %i.05 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.cond ]
38  %idxprom = sext i32 %i.05 to i64
39  %arrayidx = getelementptr inbounds i32* %1, i64 %idxprom
40  %3 = load i32* %arrayidx, align 4
41  %tobool = icmp eq i32 %3, 0
42  br i1 %tobool, label %if.then, label %for.cond
43
44if.then:
45  %i.05.lcssa = phi i32 [ %i.05, %for.body ]
46  store i32 %i.05.lcssa, i32* @idx, align 4
47  br label %for.end
48
49for.cond.for.end.loopexit_crit_edge:
50  br label %for.end.loopexit
51
52for.end.loopexit:
53  br label %for.end
54
55for.end:
56  %4 = load i32* @idx, align 4
57  ret i32 %4
58}
59
60; CHECK-LABEL: @test2
61; CHECK: for.body4.us
62; CHECK: %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
63; CHECK: %cmp2.us = icmp slt i64
64; CHECK-NOT: %2 = trunc i64 %indvars.iv.next to i32
65; CHECK-NOT: %cmp2.us = icmp slt i32
66
67define void @test2([8 x i8]* %a, i8* %b, i8 %limit) {
68entry:
69  %conv = zext i8 %limit to i32
70  br i1 undef, label %for.cond1.preheader, label %for.cond1.preheader.us
71
72for.cond1.preheader.us:
73  %storemerge5.us = phi i32 [ 0, %entry ], [ %inc14.us, %for.inc13.us ]
74  br i1 true, label %for.body4.lr.ph.us, label %for.inc13.us
75
76for.inc13.us:
77  %inc14.us = add nsw i32 %storemerge5.us, 1
78  %cmp.us = icmp slt i32 %inc14.us, 4
79  br i1 %cmp.us, label %for.cond1.preheader.us, label %for.end
80
81for.body4.us:
82  %storemerge14.us = phi i32 [ 0, %for.body4.lr.ph.us ], [ %inc.us, %for.body4.us ]
83  %idxprom.us = sext i32 %storemerge14.us to i64
84  %arrayidx6.us = getelementptr inbounds [8 x i8]* %a, i64 %idxprom5.us, i64 %idxprom.us
85  %0 = load i8* %arrayidx6.us, align 1
86  %idxprom7.us = zext i8 %0 to i64
87  %arrayidx8.us = getelementptr inbounds i8* %b, i64 %idxprom7.us
88  %1 = load i8* %arrayidx8.us, align 1
89  store i8 %1, i8* %arrayidx6.us, align 1
90  %inc.us = add nsw i32 %storemerge14.us, 1
91  %cmp2.us = icmp slt i32 %inc.us, %conv
92  br i1 %cmp2.us, label %for.body4.us, label %for.inc13.us
93
94for.body4.lr.ph.us:
95  %idxprom5.us = sext i32 %storemerge5.us to i64
96  br label %for.body4.us
97
98for.cond1.preheader:
99  %storemerge5 = phi i32 [ 0, %entry ], [ %inc14, %for.inc13 ]
100  br i1 false, label %for.inc13, label %for.inc13
101
102for.inc13:
103  %inc14 = add nsw i32 %storemerge5, 1
104  %cmp = icmp slt i32 %inc14, 4
105  br i1 %cmp, label %for.cond1.preheader, label %for.end
106
107for.end:
108  ret void
109}
110
111; CHECK-LABEL: @test3
112; CHECK: sext i32 %b
113; CHECK: for.cond:
114; CHECK: phi i64
115; CHECK: icmp slt i64
116
117define i32 @test3(i32* %a, i32 %b) {
118entry:
119  br label %for.cond
120
121for.cond:
122  %sum.0 = phi i32 [ 0, %entry ], [ %add, %for.body ]
123  %i.0 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
124  %cmp = icmp slt i32 %i.0, %b
125  br i1 %cmp, label %for.body, label %for.end
126
127for.body:
128  %idxprom = sext i32 %i.0 to i64
129  %arrayidx = getelementptr inbounds i32* %a, i64 %idxprom
130  %0 = load i32* %arrayidx, align 4
131  %add = add nsw i32 %sum.0, %0
132  %inc = add nsw i32 %i.0, 1
133  br label %for.cond
134
135for.end:
136  ret i32 %sum.0
137}
138
139declare i32 @fn1(i8 signext)
140
141; PR21030
142; CHECK-LABEL: @test4
143; CHECK: for.body:
144; CHECK: phi i32
145; CHECK: icmp sgt i8
146
147define i32 @test4(i32 %a) {
148entry:
149  br label %for.body
150
151for.body:
152  %c.07 = phi i8 [ -3, %entry ], [ %dec, %for.body ]
153  %conv6 = zext i8 %c.07 to i32
154  %or = or i32 %a, %conv6
155  %conv3 = trunc i32 %or to i8
156  %call = call i32 @fn1(i8 signext %conv3)
157  %dec = add i8 %c.07, -1
158  %cmp = icmp sgt i8 %dec, -14
159  br i1 %cmp, label %for.body, label %for.end
160
161for.end:
162  ret i32 0
163}
164
165; CHECK-LABEL: @test5
166; CHECK: zext i32 %b
167; CHECK: for.cond:
168; CHECK: phi i64
169; CHECK: icmp ule i64
170
171define i32 @test5(i32* %a, i32 %b) {
172entry:
173  br label %for.cond
174
175for.cond:
176  %sum.0 = phi i32 [ 0, %entry ], [ %add, %for.body ]
177  %i.0 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
178  %cmp = icmp ule i32 %i.0, %b
179  br i1 %cmp, label %for.body, label %for.end
180
181for.body:
182  %idxprom = zext i32 %i.0 to i64
183  %arrayidx = getelementptr inbounds i32* %a, i64 %idxprom
184  %0 = load i32* %arrayidx, align 4
185  %add = add nsw i32 %sum.0, %0
186  %inc = add nsw i32 %i.0, 1
187  br label %for.cond
188
189for.end:
190  ret i32 %sum.0
191}
192