1; RUN: opt -load-store-vectorizer %s -S | FileCheck %s
2; RUN: opt -aa-pipeline=basic-aa -passes='function(load-store-vectorizer)' %s -S | FileCheck %s
3
4; Check that setting wrapping flags after a SCEV node is created
5; does not invalidate "sorted by complexity" invariant for
6; operands of commutative and associative SCEV operators.
7
8target triple = "x86_64--"
9
10@global_value0 = external constant i32
11@global_value1 = external constant i32
12@other_value = external global float
13@a = external global float
14@b = external global float
15@c = external global float
16@d = external global float
17@plus1 = external global i32
18@cnd = external global i8
19
20; Function Attrs: nounwind
21define void @main() local_unnamed_addr #0 {
22; CHECK-LABEL: @main()
23; CHECK: [[PTR:%[0-9]+]] = bitcast float* %preheader.load0.address to <2 x float>*
24; CHECK:  = load <2 x float>, <2 x float>* [[PTR]]
25; CHECK-LABEL: for.body23:
26entry:
27  %tmp = load i32, i32* @global_value0, !range !0
28  %tmp2 = load i32, i32* @global_value1
29  %and.i.i = and i32 %tmp2, 2
30  %add.nuw.nsw.i.i = add nuw nsw i32 %and.i.i, 0
31  %mul.i.i = shl nuw nsw i32 %add.nuw.nsw.i.i, 1
32  %and6.i.i = and i32 %tmp2, 3
33  %and9.i.i = and i32 %tmp2, 4
34  %add.nuw.nsw10.i.i = add nuw nsw i32 %and6.i.i, %and9.i.i
35  %conv3.i42.i = add nuw nsw i32 %mul.i.i, 1
36  %reass.add346.7 = add nuw nsw i32 %add.nuw.nsw10.i.i, 56
37  %reass.mul347.7 = mul nuw nsw i32 %tmp, %reass.add346.7
38  %add7.i.7 = add nuw nsw i32 %reass.mul347.7, 0
39  %preheader.address0.idx = add nuw nsw i32 %add7.i.7, %mul.i.i
40  %preheader.address0.idx.zext = zext i32 %preheader.address0.idx to i64
41  %preheader.load0.address = getelementptr inbounds float, float* @other_value, i64 %preheader.address0.idx.zext
42  %preheader.load0. = load float, float* %preheader.load0.address, align 4, !tbaa !1
43  %common.address.idx = add nuw nsw i32 %add7.i.7, %conv3.i42.i
44  %preheader.header.common.address.idx.zext = zext i32 %common.address.idx to i64
45  %preheader.load1.address = getelementptr inbounds float, float* @other_value, i64 %preheader.header.common.address.idx.zext
46  %preheader.load1. = load float, float* %preheader.load1.address, align 4, !tbaa !1
47  br label %for.body23
48
49for.body23:                                       ; preds = %for.body23, %entry
50  %loop.header.load0.address = getelementptr inbounds float, float* @other_value, i64 %preheader.header.common.address.idx.zext
51  %loop.header.load0. = load float, float* %loop.header.load0.address, align 4, !tbaa !1
52  %reass.mul343.7 = mul nuw nsw i32 %reass.add346.7, 72
53  %add7.i286.7.7 = add nuw nsw i32 %reass.mul343.7, 56
54  %add9.i288.7.7 = add nuw nsw i32 %add7.i286.7.7, %mul.i.i
55  %loop.header.address1.idx = add nuw nsw i32 %add9.i288.7.7, 1
56  %loop.header.address1.idx.zext = zext i32 %loop.header.address1.idx to i64
57  %loop.header.load1.address = getelementptr inbounds float, float* @other_value, i64 %loop.header.address1.idx.zext
58  %loop.header.load1. = load float, float* %loop.header.load1.address, align 4, !tbaa !1
59  store float %preheader.load0., float* @a, align 4, !tbaa !1
60  store float %preheader.load1., float* @b, align 4, !tbaa !1
61  store float %loop.header.load0., float* @c, align 4, !tbaa !1
62  store float %loop.header.load1., float* @d, align 4, !tbaa !1
63  %loaded.cnd = load i8, i8* @cnd
64  %condition = trunc i8 %loaded.cnd to i1
65  br i1 %condition, label %for.body23, label %exit
66
67exit:
68  ret void
69}
70
71attributes #0 = { nounwind }
72
73!0 = !{i32 0, i32 65536}
74!1 = !{!2, !2, i64 0}
75!2 = !{!"float", !3, i64 0}
76!3 = !{!"omnipotent char", !4, i64 0}
77!4 = !{!"Simple C++ TBAA"}
78