1 ; RUN: opt -passes=loop-vectorize,dce,instcombine -mtriple aarch64-linux-gnu -S < %s | FileCheck %s
3 target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
4 target triple = "aarch64-unknown-linux-gnu"
6 define void @inv_store_last_lane(ptr noalias nocapture %a, ptr noalias nocapture %inv, ptr noalias nocapture readonly %b, i64 %n) #0 {
7 ; CHECK-LABEL: @inv_store_last_lane
9 ; CHECK: store <vscale x 4 x i32> %[[VEC_VAL:.*]], ptr
10 ; CHECK: middle.block:
11 ; CHECK: %[[VSCALE:.*]] = call i32 @llvm.vscale.i32()
12 ; CHECK-NEXT: %[[VSCALE2:.*]] = shl i32 %[[VSCALE]], 2
13 ; CHECK-NEXT: %[[LAST_LANE:.*]] = add i32 %[[VSCALE2]], -1
14 ; CHECK-NEXT: %{{.*}} = extractelement <vscale x 4 x i32> %[[VEC_VAL]], i32 %[[LAST_LANE]]
19 for.body: ; preds = %for.body.lr.ph, %for.body
20 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
21 %arrayidx = getelementptr inbounds i32, ptr %b, i64 %indvars.iv
22 %0 = load i32, ptr %arrayidx, align 4
23 %mul = shl nsw i32 %0, 1
24 %arrayidx2 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv
25 store i32 %mul, ptr %arrayidx2, align 4
26 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
27 %exitcond.not = icmp eq i64 %indvars.iv.next, %n
28 br i1 %exitcond.not, label %exit, label %for.body, !llvm.loop !0
30 exit: ; preds = %for.body
31 %arrayidx5 = getelementptr inbounds i32, ptr %inv, i64 42
32 store i32 %mul, ptr %arrayidx5, align 4
36 define float @ret_last_lane(ptr noalias nocapture %a, ptr noalias nocapture readonly %b, i64 %n) #0 {
37 ; CHECK-LABEL: @ret_last_lane
39 ; CHECK: store <vscale x 4 x float> %[[VEC_VAL:.*]], ptr
40 ; CHECK: middle.block:
41 ; CHECK: %[[VSCALE:.*]] = call i32 @llvm.vscale.i32()
42 ; CHECK-NEXT: %[[VSCALE2:.*]] = shl i32 %[[VSCALE]], 2
43 ; CHECK-NEXT: %[[LAST_LANE:.*]] = add i32 %[[VSCALE2]], -1
44 ; CHECK-NEXT: %{{.*}} = extractelement <vscale x 4 x float> %[[VEC_VAL]], i32 %[[LAST_LANE]]
49 for.body: ; preds = %for.body.preheader, %for.body
50 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
51 %arrayidx = getelementptr inbounds float, ptr %b, i64 %indvars.iv
52 %0 = load float, ptr %arrayidx, align 4
53 %mul = fmul float %0, 2.000000e+00
54 %arrayidx2 = getelementptr inbounds float, ptr %a, i64 %indvars.iv
55 store float %mul, ptr %arrayidx2, align 4
56 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
57 %exitcond.not = icmp eq i64 %indvars.iv.next, %n
58 br i1 %exitcond.not, label %exit, label %for.body, !llvm.loop !6
60 exit: ; preds = %for.body, %entry
64 attributes #0 = { "target-cpu"="generic" "target-features"="+neon,+sve" }
66 !0 = distinct !{!0, !1, !2, !3, !4, !5}
67 !1 = !{!"llvm.loop.mustprogress"}
68 !2 = !{!"llvm.loop.vectorize.width", i32 4}
69 !3 = !{!"llvm.loop.vectorize.scalable.enable", i1 true}
70 !4 = !{!"llvm.loop.interleave.count", i32 1}
71 !5 = !{!"llvm.loop.vectorize.enable", i1 true}
72 !6 = distinct !{!6, !1, !2, !3, !4, !5}