1; RUN: opt -S  -mtriple=s390x-unknown-linux -mcpu=z13 -loop-vectorize -dce \
2; RUN:   -instcombine -force-vector-width=2  < %s | FileCheck %s
3;
4; Test that loop vectorizer does not generate vector addresses that must then
5; always be extracted.
6
7; Check that the addresses for a scalarized memory access is not extracted
8; from a vector register.
9define i32 @foo(i32* nocapture %A) {
10;CHECK-LABEL: @foo(
11;CHECK:  %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]
12;CHECK:  %0 = shl nsw i64 %index, 2
13;CHECK:  %1 = shl i64 %index, 2
14;CHECK:  %2 = or i64 %1, 4
15;CHECK:  %3 = getelementptr inbounds i32, i32* %A, i64 %0
16;CHECK:  %4 = getelementptr inbounds i32, i32* %A, i64 %2
17;CHECK:  store i32 4, i32* %3, align 4
18;CHECK:  store i32 4, i32* %4, align 4
19
20entry:
21  br label %for.body
22
23for.body:
24  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
25  %0 = shl nsw i64 %indvars.iv, 2
26  %arrayidx = getelementptr inbounds i32, i32* %A, i64 %0
27  store i32 4, i32* %arrayidx, align 4
28  %indvars.iv.next = add i64 %indvars.iv, 1
29  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
30  %exitcond = icmp eq i32 %lftr.wideiv, 10000
31  br i1 %exitcond, label %for.end, label %for.body
32
33for.end:
34  ret i32 undef
35}
36
37
38; Check that a load of address is scalarized.
39define i32 @foo1(i32* nocapture noalias %A, i32** nocapture %PtrPtr) {
40;CHECK-LABEL: @foo1(
41;CHECK:  %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]
42;CHECK:  %0 = or i64 %index, 1
43;CHECK:  %1 = getelementptr inbounds i32*, i32** %PtrPtr, i64 %index
44;CHECK:  %2 = getelementptr inbounds i32*, i32** %PtrPtr, i64 %0
45;CHECK:  %3 = load i32*, i32** %1, align 8
46;CHECK:  %4 = load i32*, i32** %2, align 8
47;CHECK:  %5 = load i32, i32* %3, align 4
48;CHECK:  %6 = load i32, i32* %4, align 4
49;CHECK:  %7 = insertelement <2 x i32> poison, i32 %5, i32 0
50;CHECK:  %8 = insertelement <2 x i32> %7, i32 %6, i32 1
51;CHECK:  %9 = getelementptr inbounds i32, i32* %A, i64 %index
52;CHECK:  %10 = bitcast i32* %9 to <2 x i32>*
53;CHECK:  store <2 x i32> %8, <2 x i32>* %10, align 4
54
55entry:
56  br label %for.body
57
58for.body:
59  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
60  %ptr = getelementptr inbounds i32*, i32** %PtrPtr, i64 %indvars.iv
61  %el = load i32*, i32** %ptr
62  %v = load i32, i32* %el
63  %arrayidx = getelementptr inbounds i32, i32* %A, i64 %indvars.iv
64  store i32 %v, i32* %arrayidx, align 4
65  %indvars.iv.next = add i64 %indvars.iv, 1
66  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
67  %exitcond = icmp eq i32 %lftr.wideiv, 10000
68  br i1 %exitcond, label %for.end, label %for.body
69
70for.end:
71  ret i32 undef
72}
73