1; RUN: opt < %s -force-vector-width=4 -force-vector-interleave=2 -loop-vectorize -instcombine -S | FileCheck %s
2; RUN: opt < %s -force-vector-width=4 -force-vector-interleave=2 -loop-vectorize -S | FileCheck %s --check-prefix=NO-IC
3
4target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"
5
6; CHECK-LABEL: @scalar_after_vectorization_0
7;
8; CHECK: vector.body:
9; CHECK:   %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]
10; CHECK:   %offset.idx = or i64 %index, 1
11; CHECK:   %[[T2:.+]] = add nuw nsw i64 %offset.idx, %tmp0
12; CHECK:   %[[T3:.+]] = sub nsw i64 %[[T2]], %x
13; CHECK:   %[[T4:.+]] = getelementptr inbounds i32, i32* %a, i64 %[[T3]]
14; CHECK:   %[[T5:.+]] = bitcast i32* %[[T4]] to <4 x i32>*
15; CHECK:   load <4 x i32>, <4 x i32>* %[[T5]], align 4
16; CHECK:   %[[T6:.+]] = getelementptr inbounds i32, i32* %[[T4]], i64 4
17; CHECK:   %[[T7:.+]] = bitcast i32* %[[T6]] to <4 x i32>*
18; CHECK:   load <4 x i32>, <4 x i32>* %[[T7]], align 4
19; CHECK:   br {{.*}}, label %middle.block, label %vector.body
20;
21; NO-IC-LABEL: @scalar_after_vectorization_0
22;
23; NO-IC: vector.body:
24; NO-IC:   %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]
25; NO-IC:   %offset.idx = add i64 1, %index
26; NO-IC:   %[[T2:.+]] = add i64 %offset.idx, 0
27; NO-IC:   %[[T3:.+]] = add i64 %offset.idx, 4
28; NO-IC:   %[[T4:.+]] = add nuw nsw i64 %[[T2]], %tmp0
29; NO-IC:   %[[T5:.+]] = add nuw nsw i64 %[[T3]], %tmp0
30; NO-IC:   %[[T6:.+]] = sub nsw i64 %[[T4]], %x
31; NO-IC:   %[[T7:.+]] = sub nsw i64 %[[T5]], %x
32; NO-IC:   %[[T8:.+]] = getelementptr inbounds i32, i32* %a, i64 %[[T6]]
33; NO-IC:   %[[T9:.+]] = getelementptr inbounds i32, i32* %a, i64 %[[T7]]
34; NO-IC:   %[[T10:.+]] = getelementptr inbounds i32, i32* %[[T8]], i32 0
35; NO-IC:   %[[T11:.+]] = bitcast i32* %[[T10]] to <4 x i32>*
36; NO-IC:   load <4 x i32>, <4 x i32>* %[[T11]], align 4
37; NO-IC:   %[[T12:.+]] = getelementptr inbounds i32, i32* %[[T8]], i32 4
38; NO-IC:   %[[T13:.+]] = bitcast i32* %[[T12]] to <4 x i32>*
39; NO-IC:   load <4 x i32>, <4 x i32>* %[[T13]], align 4
40; NO-IC:   br {{.*}}, label %middle.block, label %vector.body
41;
42define void @scalar_after_vectorization_0(i32* noalias %a, i32* noalias %b, i64 %x, i64 %y) {
43
44outer.ph:
45  br label %outer.body
46
47outer.body:
48  %i = phi i64 [ 1, %outer.ph ], [ %i.next, %inner.end ]
49  %tmp0 = mul nuw nsw i64 %i, %x
50  br label %inner.ph
51
52inner.ph:
53  br label %inner.body
54
55inner.body:
56  %j = phi i64 [ 1, %inner.ph ], [ %j.next, %inner.body ]
57  %tmp1 = add nuw nsw i64 %j, %tmp0
58  %tmp2 = sub nsw i64 %tmp1, %x
59  %tmp3 = getelementptr inbounds i32, i32* %a, i64 %tmp2
60  %tmp4 = load i32, i32* %tmp3, align 4
61  %tmp5 = getelementptr inbounds i32, i32* %b, i64 %tmp1
62  store i32 %tmp4, i32* %tmp5, align 4
63  %j.next = add i64 %j, 1
64  %cond.j = icmp slt i64 %j.next, %y
65  br i1 %cond.j, label %inner.body, label %inner.end
66
67inner.end:
68  %i.next = add i64 %i, 1
69  %cond.i = icmp slt i64 %i.next, %y
70  br i1 %cond.i, label %outer.body, label %outer.end
71
72outer.end:
73  ret void
74}
75