1 ; RUN: opt -passes=loop-vectorize -debug-only=loop-vectorize -disable-output -prefer-predicate-over-epilogue=scalar-epilogue 2>&1 < %s | FileCheck %s
4 target triple = "aarch64"
6 ; Test that shows how many registers the loop vectorizer thinks an illegal <VF x i1> will consume.
8 ; CHECK-LABEL: LV: Checking a loop in 'or_reduction_neon' from <stdin>
9 ; CHECK: LV(REG): VF = 32
10 ; CHECK-NEXT: LV(REG): Found max usage: 2 item
11 ; CHECK-NEXT: LV(REG): RegisterClass: Generic::VectorRC, 72 registers
12 ; CHECK-NEXT: LV(REG): RegisterClass: Generic::ScalarRC, 1 registers
14 define i1 @or_reduction_neon(i32 %arg, ptr %ptr) {
18 ret i1 %reduction_next
20 %induction = phi i32 [ 0, %entry ], [ %induction_next, %loop ]
21 %reduction = phi i1 [ 0, %entry ], [ %reduction_next, %loop ]
22 %gep = getelementptr inbounds i32, ptr %ptr, i32 %induction
23 %loaded = load i32, ptr %gep
24 %i1 = icmp eq i32 %loaded, %induction
25 %reduction_next = or i1 %i1, %reduction
26 %induction_next = add nuw i32 %induction, 1
27 %cond = icmp eq i32 %induction_next, %arg
28 br i1 %cond, label %exit, label %loop, !llvm.loop !32
31 ; CHECK-LABEL: LV: Checking a loop in 'or_reduction_sve'
32 ; CHECK: LV(REG): VF = 64
33 ; CHECK-NEXT: LV(REG): Found max usage: 2 item
34 ; CHECK-NEXT: LV(REG): RegisterClass: Generic::VectorRC, 136 registers
35 ; CHECK-NEXT: LV(REG): RegisterClass: Generic::ScalarRC, 1 registers
37 define i1 @or_reduction_sve(i32 %arg, ptr %ptr) vscale_range(2,2) "target-features"="+sve" {
41 ret i1 %reduction_next
43 %induction = phi i32 [ 0, %entry ], [ %induction_next, %loop ]
44 %reduction = phi i1 [ true, %entry ], [ %reduction_next, %loop ]
45 %gep = getelementptr inbounds i32, ptr %ptr, i32 %induction
46 %loaded = load i32, ptr %gep
47 %i1 = icmp eq i32 %loaded, %induction
48 %reduction_next = or i1 %i1, %reduction
49 %induction_next = add nuw i32 %induction, 1
50 %cond = icmp eq i32 %induction_next, %arg
51 br i1 %cond, label %exit, label %loop, !llvm.loop !64
54 !32 = distinct !{!32, !33}
55 !33 = !{!"llvm.loop.vectorize.width", i32 32}
56 !64 = distinct !{!64, !65}
57 !65 = !{!"llvm.loop.vectorize.width", i32 64}