1 ; RUN: opt -S -basicaa -slp-vectorizer -dce < %s | FileCheck %s
2 target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128"
3 target triple = "arm64-apple-ios5.0.0"
6 define void @foo(float* noalias %a, float* noalias %b, float* noalias %c) {
8 ; Check that we don't lose !nontemporal hint when vectorizing loads.
9 ; CHECK: %{{[0-9]*}} = load <4 x float>, <4 x float>* %{{[0-9]+}}, align 4, !nontemporal !0
10 %b1 = load float, float* %b, align 4, !nontemporal !0
11 %arrayidx.1 = getelementptr inbounds float, float* %b, i64 1
12 %b2 = load float, float* %arrayidx.1, align 4, !nontemporal !0
13 %arrayidx.2 = getelementptr inbounds float, float* %b, i64 2
14 %b3 = load float, float* %arrayidx.2, align 4, !nontemporal !0
15 %arrayidx.3 = getelementptr inbounds float, float* %b, i64 3
16 %b4 = load float, float* %arrayidx.3, align 4, !nontemporal !0
18 ; Check that we don't introduce !nontemporal hint when the original scalar loads didn't have it.
19 ; CHECK: %{{[0-9]*}} = load <4 x float>, <4 x float>* %{{[0-9]+}}, align 4{{$}}
20 %c1 = load float, float* %c, align 4
21 %arrayidx2.1 = getelementptr inbounds float, float* %c, i64 1
22 %c2 = load float, float* %arrayidx2.1, align 4
23 %arrayidx2.2 = getelementptr inbounds float, float* %c, i64 2
24 %c3 = load float, float* %arrayidx2.2, align 4
25 %arrayidx2.3 = getelementptr inbounds float, float* %c, i64 3
26 %c4 = load float, float* %arrayidx2.3, align 4
28 %a1 = fadd float %b1, %c1
29 %a2 = fadd float %b2, %c2
30 %a3 = fadd float %b3, %c3
31 %a4 = fadd float %b4, %c4
33 ; Check that we don't lose !nontemporal hint when vectorizing stores.
34 ; CHECK: store <4 x float> %{{[0-9]+}}, <4 x float>* %{{[0-9]+}}, align 4, !nontemporal !0
35 store float %a1, float* %a, align 4, !nontemporal !0
36 %arrayidx3.1 = getelementptr inbounds float, float* %a, i64 1
37 store float %a2, float* %arrayidx3.1, align 4, !nontemporal !0
38 %arrayidx3.2 = getelementptr inbounds float, float* %a, i64 2
39 store float %a3, float* %arrayidx3.2, align 4, !nontemporal !0
40 %arrayidx3.3 = getelementptr inbounds float, float* %a, i64 3
41 store float %a4, float* %arrayidx3.3, align 4, !nontemporal !0
48 define void @foo2(float* noalias %a, float* noalias %b) {
50 ; Check that we don't mark vector load with !nontemporal attribute if some of
51 ; the original scalar loads don't have it.
52 ; CHECK: %{{[0-9]*}} = load <4 x float>, <4 x float>* %{{[0-9]+}}, align 4{{$}}
53 %b1 = load float, float* %b, align 4, !nontemporal !0
54 %arrayidx.1 = getelementptr inbounds float, float* %b, i64 1
55 %b2 = load float, float* %arrayidx.1, align 4
56 %arrayidx.2 = getelementptr inbounds float, float* %b, i64 2
57 %b3 = load float, float* %arrayidx.2, align 4
58 %arrayidx.3 = getelementptr inbounds float, float* %b, i64 3
59 %b4 = load float, float* %arrayidx.3, align 4, !nontemporal !0
61 ; Check that we don't mark vector store with !nontemporal attribute if some of
62 ; the original scalar stores don't have it.
63 ; CHECK: store <4 x float> %{{[0-9]+}}, <4 x float>* %{{[0-9]+}}, align 4{{$}}
64 store float %b1, float* %a, align 4, !nontemporal !0
65 %arrayidx3.1 = getelementptr inbounds float, float* %a, i64 1
66 store float %b2, float* %arrayidx3.1, align 4
67 %arrayidx3.2 = getelementptr inbounds float, float* %a, i64 2
68 store float %b3, float* %arrayidx3.2, align 4
69 %arrayidx3.3 = getelementptr inbounds float, float* %a, i64 3
70 store float %b4, float* %arrayidx3.3, align 4, !nontemporal !0