1 ; RUN: opt < %s -loop-vectorize -force-vector-width=4 -force-vector-interleave=1 -instcombine -S | FileCheck %s
3 target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128"
4 target triple = "arm64-apple-ios5.0.0"
7 define void @foo(float* noalias %a, float* noalias %b, float* noalias %c, i32 %N) {
9 %cmp.4 = icmp sgt i32 %N, 0
10 br i1 %cmp.4, label %for.body.preheader, label %for.end
12 for.body.preheader: ; preds = %entry
15 for.body: ; preds = %for.body.preheader, %for.body
16 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ]
18 ; Check that we don't lose !nontemporal hint when vectorizing loads.
19 ; CHECK: %wide.load{{[0-9]*}} = load <4 x float>, <4 x float>* %{{[0-9]+}}, align 4, !nontemporal !0
20 %arrayidx = getelementptr inbounds float, float* %b, i64 %indvars.iv
21 %0 = load float, float* %arrayidx, align 4, !nontemporal !0
23 ; Check that we don't introduce !nontemporal hint when the original scalar loads didn't have it.
24 ; CHECK: %wide.load{{[0-9]+}} = load <4 x float>, <4 x float>* %{{[0-9]+}}, align 4{{$}}
25 %arrayidx2 = getelementptr inbounds float, float* %c, i64 %indvars.iv
26 %1 = load float, float* %arrayidx2, align 4
27 %add = fadd float %0, %1
29 ; Check that we don't lose !nontemporal hint when vectorizing stores.
30 ; CHECK: store <4 x float> %{{[0-9]+}}, <4 x float>* %{{[0-9]+}}, align 4, !nontemporal !0
31 %arrayidx4 = getelementptr inbounds float, float* %a, i64 %indvars.iv
32 store float %add, float* %arrayidx4, align 4, !nontemporal !0
34 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
35 %lftr.wideiv = trunc i64 %indvars.iv.next to i32
36 %exitcond = icmp eq i32 %lftr.wideiv, %N
37 br i1 %exitcond, label %for.end.loopexit, label %for.body
39 for.end.loopexit: ; preds = %for.body
42 for.end: ; preds = %for.end.loopexit, %entry