1; RUN: opt < %s  -loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -dce -instcombine -S | FileCheck %s
2
3target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
4target triple = "x86_64-unknown-linux-gnu"
5
6; A tricky loop:
7;
8; void loop(int *a, int *b) {
9;    for (int i = 0; i < 512; ++i) {
10;        a[a[i]] = b[i];
11;        a[i] = b[i+1];
12;    }
13;}
14
15;CHECK-LABEL: @loop(
16;CHECK-NOT: <4 x i32>
17define void @loop(i32* nocapture %a, i32* nocapture %b) nounwind uwtable {
18entry:
19  br label %for.body
20
21for.body:                                         ; preds = %for.body, %entry
22  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
23  %arrayidx = getelementptr inbounds i32* %b, i64 %indvars.iv
24  %0 = load i32* %arrayidx, align 4
25  %arrayidx2 = getelementptr inbounds i32* %a, i64 %indvars.iv
26  %1 = load i32* %arrayidx2, align 4
27  %idxprom3 = sext i32 %1 to i64
28  %arrayidx4 = getelementptr inbounds i32* %a, i64 %idxprom3
29  store i32 %0, i32* %arrayidx4, align 4
30  %indvars.iv.next = add i64 %indvars.iv, 1
31  %arrayidx6 = getelementptr inbounds i32* %b, i64 %indvars.iv.next
32  %2 = load i32* %arrayidx6, align 4
33  store i32 %2, i32* %arrayidx2, align 4
34  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
35  %exitcond = icmp eq i32 %lftr.wideiv, 512
36  br i1 %exitcond, label %for.end, label %for.body
37
38for.end:                                          ; preds = %for.body
39  ret void
40}
41
42; The same loop with parallel loop metadata added to the loop branch
43; and the memory instructions.
44
45;CHECK-LABEL: @parallel_loop(
46;CHECK: <4 x i32>
47define void @parallel_loop(i32* nocapture %a, i32* nocapture %b) nounwind uwtable {
48entry:
49  br label %for.body
50
51for.body:                                         ; preds = %for.body, %entry
52  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
53  %arrayidx = getelementptr inbounds i32* %b, i64 %indvars.iv
54  %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !3
55  %arrayidx2 = getelementptr inbounds i32* %a, i64 %indvars.iv
56  %1 = load i32* %arrayidx2, align 4, !llvm.mem.parallel_loop_access !3
57  %idxprom3 = sext i32 %1 to i64
58  %arrayidx4 = getelementptr inbounds i32* %a, i64 %idxprom3
59  ; This store might have originated from inlining a function with a parallel
60  ; loop. Refers to a list with the "original loop reference" (!4) also included.
61  store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !5
62  %indvars.iv.next = add i64 %indvars.iv, 1
63  %arrayidx6 = getelementptr inbounds i32* %b, i64 %indvars.iv.next
64  %2 = load i32* %arrayidx6, align 4, !llvm.mem.parallel_loop_access !3
65  store i32 %2, i32* %arrayidx2, align 4, !llvm.mem.parallel_loop_access !3
66  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
67  %exitcond = icmp eq i32 %lftr.wideiv, 512
68  br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !3
69
70for.end:                                          ; preds = %for.body
71  ret void
72}
73
74; The same loop with an illegal parallel loop metadata: the memory
75; accesses refer to a different loop's identifier.
76
77;CHECK-LABEL: @mixed_metadata(
78;CHECK-NOT: <4 x i32>
79
80define void @mixed_metadata(i32* nocapture %a, i32* nocapture %b) nounwind uwtable {
81entry:
82  br label %for.body
83
84for.body:                                         ; preds = %for.body, %entry
85  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
86  %arrayidx = getelementptr inbounds i32* %b, i64 %indvars.iv
87  %0 = load i32* %arrayidx, align 4, !llvm.mem.parallel_loop_access !6
88  %arrayidx2 = getelementptr inbounds i32* %a, i64 %indvars.iv
89  %1 = load i32* %arrayidx2, align 4, !llvm.mem.parallel_loop_access !6
90  %idxprom3 = sext i32 %1 to i64
91  %arrayidx4 = getelementptr inbounds i32* %a, i64 %idxprom3
92  ; This refers to the loop marked with !7 which we are not in at the moment.
93  ; It should prevent detecting as a parallel loop.
94  store i32 %0, i32* %arrayidx4, align 4, !llvm.mem.parallel_loop_access !7
95  %indvars.iv.next = add i64 %indvars.iv, 1
96  %arrayidx6 = getelementptr inbounds i32* %b, i64 %indvars.iv.next
97  %2 = load i32* %arrayidx6, align 4, !llvm.mem.parallel_loop_access !6
98  store i32 %2, i32* %arrayidx2, align 4, !llvm.mem.parallel_loop_access !6
99  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
100  %exitcond = icmp eq i32 %lftr.wideiv, 512
101  br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !6
102
103for.end:                                          ; preds = %for.body
104  ret void
105}
106
107!3 = !{!3}
108!4 = !{!4}
109!5 = !{!3, !4}
110!6 = !{!6}
111!7 = !{!7}
112