43 lines · plain
1; RUN: opt < %s -passes=loop-vectorize -mtriple=aarch64-none-linux-gnu -mattr=+neon -S | FileCheck %s2target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"3 4; Function Attrs: nounwind5define ptr @array_add(ptr noalias nocapture readonly %a, ptr noalias nocapture readonly %b, ptr %c, i32 %size) {6;CHECK-LABEL: array_add7;CHECK: load <4 x i32>8;CHECK: load <4 x i32>9;CHECK: load <4 x i32>10;CHECK: load <4 x i32>11;CHECK: add nsw <4 x i32>12;CHECK: add nsw <4 x i32>13;CHECK: store <4 x i32>14;CHECK: store <4 x i32>15;CHECK: ret16entry:17 %cmp10 = icmp sgt i32 %size, 018 br i1 %cmp10, label %for.body.preheader, label %for.end19 20for.body.preheader: ; preds = %entry21 br label %for.body22 23for.body: ; preds = %for.body.preheader, %for.body24 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ]25 %arrayidx = getelementptr inbounds i32, ptr %a, i64 %indvars.iv26 %0 = load i32, ptr %arrayidx, align 427 %arrayidx2 = getelementptr inbounds i32, ptr %b, i64 %indvars.iv28 %1 = load i32, ptr %arrayidx2, align 429 %add = add nsw i32 %1, %030 %arrayidx4 = getelementptr inbounds i32, ptr %c, i64 %indvars.iv31 store i32 %add, ptr %arrayidx4, align 432 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 133 %lftr.wideiv = trunc i64 %indvars.iv.next to i3234 %exitcond = icmp eq i32 %lftr.wideiv, %size35 br i1 %exitcond, label %for.end.loopexit, label %for.body36 37for.end.loopexit: ; preds = %for.body38 br label %for.end39 40for.end: ; preds = %for.end.loopexit, %entry41 ret ptr %c42}43