1; RUN: opt < %s  -tbaa -basicaa -loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -dce -instcombine -simplifycfg -S | FileCheck %s
2; RUN: opt < %s  -basicaa -loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -dce -instcombine -simplifycfg -S | FileCheck %s --check-prefix=CHECK-NOTBAA
3target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
4target triple = "x86_64-unknown-linux-gnu"
5
6; Function Attrs: nounwind uwtable
7define i32 @test1(i32* nocapture %a, float* nocapture readonly %b) #0 {
8entry:
9  br label %for.body
10
11for.body:                                         ; preds = %for.body, %entry
12  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
13  %arrayidx = getelementptr inbounds float, float* %b, i64 %indvars.iv
14  %0 = load float, float* %arrayidx, align 4, !tbaa !0
15  %conv = fptosi float %0 to i32
16  %arrayidx2 = getelementptr inbounds i32, i32* %a, i64 %indvars.iv
17  store i32 %conv, i32* %arrayidx2, align 4, !tbaa !4
18  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
19  %exitcond = icmp eq i64 %indvars.iv.next, 1600
20  br i1 %exitcond, label %for.end, label %for.body
21
22for.end:                                          ; preds = %for.body
23  ret i32 0
24
25; TBAA partitions the accesses in this loop, so it can be vectorized without
26; runtime checks.
27
28; CHECK-LABEL: @test1
29; CHECK: entry:
30; CHECK-NEXT: br label %vector.body
31; CHECK: vector.body:
32
33; CHECK: load <4 x float>, <4 x float>* %{{.*}}, align 4, !tbaa
34; CHECK: store <4 x i32> %{{.*}}, <4 x i32>* %{{.*}}, align 4, !tbaa
35
36; CHECK: ret i32 0
37
38; CHECK-NOTBAA-LABEL: @test1
39; CHECK-NOTBAA: icmp uge i32*
40
41; CHECK-NOTBAA: load <4 x float>, <4 x float>* %{{.*}}, align 4, !tbaa
42; CHECK-NOTBAA: store <4 x i32> %{{.*}}, <4 x i32>* %{{.*}}, align 4, !tbaa
43
44; CHECK-NOTBAA: ret i32 0
45}
46
47; Function Attrs: nounwind uwtable
48define i32 @test2(i32* nocapture readonly %a, float* nocapture readonly %b, float* nocapture %c) #0 {
49entry:
50  br label %for.body
51
52for.body:                                         ; preds = %for.body, %entry
53  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
54  %arrayidx = getelementptr inbounds float, float* %b, i64 %indvars.iv
55  %0 = load float, float* %arrayidx, align 4, !tbaa !0
56  %arrayidx2 = getelementptr inbounds i32, i32* %a, i64 %indvars.iv
57  %1 = load i32, i32* %arrayidx2, align 4, !tbaa !4
58  %conv = sitofp i32 %1 to float
59  %mul = fmul float %0, %conv
60  %arrayidx4 = getelementptr inbounds float, float* %c, i64 %indvars.iv
61  store float %mul, float* %arrayidx4, align 4, !tbaa !0
62  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
63  %exitcond = icmp eq i64 %indvars.iv.next, 1600
64  br i1 %exitcond, label %for.end, label %for.body
65
66for.end:                                          ; preds = %for.body
67  ret i32 0
68
69; This test is like the first, except here there is still one runtime check
70; required. Without TBAA, however, two checks are required.
71
72; CHECK-LABEL: @test2
73; CHECK: icmp uge float*
74; CHECK: icmp uge float*
75; CHECK-NOT: icmp uge i32*
76
77; CHECK: load <4 x float>, <4 x float>* %{{.*}}, align 4, !tbaa
78; CHECK: store <4 x float> %{{.*}}, <4 x float>* %{{.*}}, align 4, !tbaa
79
80; CHECK: ret i32 0
81
82; CHECK-NOTBAA-LABEL: @test2
83; CHECK-NOTBAA: icmp uge float*
84; CHECK-NOTBAA: icmp uge float*
85; CHECK-NOTBAA-DAG: icmp uge float*
86; CHECK-NOTBAA-DAG: icmp uge i32*
87
88; CHECK-NOTBAA: load <4 x float>, <4 x float>* %{{.*}}, align 4, !tbaa
89; CHECK-NOTBAA: store <4 x float> %{{.*}}, <4 x float>* %{{.*}}, align 4, !tbaa
90
91; CHECK-NOTBAA: ret i32 0
92}
93
94attributes #0 = { nounwind uwtable }
95
96!0 = !{!1, !1, i64 0}
97!1 = !{!"float", !2, i64 0}
98!2 = !{!"omnipotent char", !3, i64 0}
99!3 = !{!"Simple C/C++ TBAA"}
100!4 = !{!5, !5, i64 0}
101!5 = !{!"int", !2, i64 0}
102
103