2 ; RUN: opt < %s -passes=loop-vectorize -debug -disable-output -force-ordered-reductions=true -hints-allow-reordering=false \
3 ; RUN: -prefer-predicate-over-epilogue=scalar-epilogue -force-vector-interleave=1 -S 2>&1 | FileCheck %s
4 ; RUN: opt < %s -passes=loop-vectorize -debug -disable-output -force-ordered-reductions=true -hints-allow-reordering=false \
5 ; RUN: -prefer-predicate-over-epilogue=scalar-epilogue -force-vector-interleave=1 \
6 ; RUN: -mcpu=neoverse-n2 -S 2>&1 | FileCheck %s --check-prefix=CHECK-CPU-NEOVERSE-N2
8 target triple="aarch64-unknown-linux-gnu"
10 ; CHECK: Found an estimated cost of 4 for VF vscale x 2 For instruction: %add = fadd float %0, %sum.07
11 ; CHECK: Found an estimated cost of 8 for VF vscale x 4 For instruction: %add = fadd float %0, %sum.07
12 ; CHECK-CPU-NEOVERSE-N2: Found an estimated cost of 2 for VF vscale x 2 For instruction: %add = fadd float %0, %sum.07
13 ; CHECK-CPU-NEOVERSE-N2: Found an estimated cost of 4 for VF vscale x 4 For instruction: %add = fadd float %0, %sum.07
15 define float @fadd_strict32(ptr noalias nocapture readonly %a, i64 %n) #0 {
20 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
21 %sum.07 = phi float [ 0.000000e+00, %entry ], [ %add, %for.body ]
22 %arrayidx = getelementptr inbounds float, ptr %a, i64 %iv
23 %0 = load float, ptr %arrayidx, align 4
24 %add = fadd float %0, %sum.07
25 %iv.next = add nuw nsw i64 %iv, 1
26 %exitcond.not = icmp eq i64 %iv.next, %n
27 br i1 %exitcond.not, label %for.end, label %for.body, !llvm.loop !0
34 ; CHECK: Found an estimated cost of 4 for VF vscale x 2 For instruction: %add = fadd double %0, %sum.07
35 ; CHECK-CPU-NEOVERSE-N2: Found an estimated cost of 2 for VF vscale x 2 For instruction: %add = fadd double %0, %sum.07
37 define double @fadd_strict64(ptr noalias nocapture readonly %a, i64 %n) #0 {
42 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
43 %sum.07 = phi double [ 0.000000e+00, %entry ], [ %add, %for.body ]
44 %arrayidx = getelementptr inbounds double, ptr %a, i64 %iv
45 %0 = load double, ptr %arrayidx, align 4
46 %add = fadd double %0, %sum.07
47 %iv.next = add nuw nsw i64 %iv, 1
48 %exitcond.not = icmp eq i64 %iv.next, %n
49 br i1 %exitcond.not, label %for.end, label %for.body, !llvm.loop !0
55 attributes #0 = { "target-features"="+sve" vscale_range(1, 16) }
57 !0 = distinct !{!0, !1}
58 !1 = !{!"llvm.loop.vectorize.scalable.enable", i1 true}