xref: /llvm-project/llvm/test/Transforms/LoopVectorize/AArch64/sve-tail-folding-forced.ll (revision 7f3428d3ed71d87a2088b77b6cab9f3d86544234)
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; REQUIRES: asserts
3; RUN: opt -S -passes=loop-vectorize -debug-only=loop-vectorize < %s 2>%t | FileCheck %s
4; RUN: cat %t | FileCheck %s --check-prefix=VPLANS
5
6; These tests ensure that tail-folding is enabled when the predicate.enable
7; loop attribute is set to true.
8
9target triple = "aarch64-unknown-linux-gnu"
10
11; VPLANS-LABEL: Checking a loop in 'simple_memset'
12; VPLANS:      VPlan 'Initial VPlan for VF={vscale x 1,vscale x 2,vscale x 4},UF>=1' {
13; VPLANS-NEXT: Live-in vp<[[VFxUF:%.+]]> = VF * UF
14; VPLANS:      vp<[[TC:%[0-9]+]]> = original trip-count
15; VPLANS-EMPTY:
16; VPLANS-NEXT: ir-bb<entry>:
17; VPLANS-NEXT:  EMIT vp<[[TC]]> = EXPAND SCEV (1 umax %n)
18; VPLANS-NEXT: Successor(s): vector.ph
19; VPLANS-EMPTY:
20; VPLANS-NEXT: vector.ph:
21; VPLANS-NEXT:   EMIT vp<[[NEWTC:%[0-9]+]]> = TC > VF ? TC - VF : 0 vp<[[TC]]>
22; VPLANS-NEXT:   EMIT vp<[[VF:%.+]]> = VF * Part + ir<0>
23; VPLANS-NEXT:   EMIT vp<[[LANEMASK_ENTRY:%.+]]> = active lane mask vp<[[VF]]>, vp<[[TC]]>
24; VPLANS-NEXT: Successor(s): vector loop
25; VPLANS-EMPTY:
26; VPLANS-NEXT: <x1> vector loop: {
27; VPLANS-NEXT:   vector.body:
28; VPLANS-NEXT:     EMIT vp<[[INDV:%[0-9]+]]> = CANONICAL-INDUCTION
29; VPLANS-NEXT:     ACTIVE-LANE-MASK-PHI vp<[[LANEMASK_PHI:%[0-9]+]]> = phi vp<[[LANEMASK_ENTRY]]>, vp<[[LANEMASK_LOOP:%.+]]>
30; VPLANS-NEXT:     vp<[[STEP:%[0-9]+]]>    = SCALAR-STEPS vp<[[INDV]]>, ir<1>
31; VPLANS-NEXT:     CLONE ir<%gep> = getelementptr ir<%ptr>, vp<[[STEP]]>
32; VPLANS-NEXT:     vp<[[VEC_PTR:%[0-9]+]]> = vector-pointer ir<%gep>
33; VPLANS-NEXT:     WIDEN store vp<[[VEC_PTR]]>, ir<%val>, vp<[[LANEMASK_PHI]]>
34; VPLANS-NEXT:     EMIT vp<[[INDV_UPDATE:%.+]]> = add vp<[[INDV]]>, vp<[[VFxUF]]>
35; VPLANS-NEXT:     EMIT vp<[[INC:%[0-9]+]]> = VF * Part + vp<[[INDV]]>
36; VPLANS-NEXT:     EMIT vp<[[LANEMASK_LOOP]]> = active lane mask vp<[[INC]]>, vp<[[NEWTC]]>
37; VPLANS-NEXT:     EMIT vp<[[NOT:%[0-9]+]]> = not vp<[[LANEMASK_LOOP]]>
38; VPLANS-NEXT:     EMIT branch-on-cond vp<[[NOT]]>
39; VPLANS-NEXT:   No successors
40; VPLANS-NEXT: }
41
42define void @simple_memset(i32 %val, ptr %ptr, i64 %n) #0 {
43; CHECK-LABEL: @simple_memset(
44; CHECK-NEXT:  entry:
45; CHECK-NEXT:    [[UMAX:%.*]] = call i64 @llvm.umax.i64(i64 [[N:%.*]], i64 1)
46; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
47; CHECK:       vector.ph:
48; CHECK-NEXT:    [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
49; CHECK-NEXT:    [[TMP1:%.*]] = mul i64 [[TMP0]], 4
50; CHECK-NEXT:    [[TMP4:%.*]] = sub i64 [[TMP1]], 1
51; CHECK-NEXT:    [[N_RND_UP:%.*]] = add i64 [[UMAX]], [[TMP4]]
52; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N_RND_UP]], [[TMP1]]
53; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]]
54; CHECK-NEXT:    [[TMP13:%.*]] = call i64 @llvm.vscale.i64()
55; CHECK-NEXT:    [[TMP14:%.*]] = mul i64 [[TMP13]], 4
56; CHECK-NEXT:    [[TMP5:%.*]] = call i64 @llvm.vscale.i64()
57; CHECK-NEXT:    [[TMP6:%.*]] = mul i64 [[TMP5]], 4
58; CHECK-NEXT:    [[TMP7:%.*]] = sub i64 [[UMAX]], [[TMP6]]
59; CHECK-NEXT:    [[TMP8:%.*]] = icmp ugt i64 [[UMAX]], [[TMP6]]
60; CHECK-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], i64 [[TMP7]], i64 0
61; CHECK-NEXT:    [[ACTIVE_LANE_MASK_ENTRY:%.*]] = call <vscale x 4 x i1> @llvm.get.active.lane.mask.nxv4i1.i64(i64 0, i64 [[UMAX]])
62; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[VAL:%.*]], i64 0
63; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[BROADCAST_SPLATINSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
64; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
65; CHECK:       vector.body:
66; CHECK-NEXT:    [[INDEX1:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT2:%.*]], [[VECTOR_BODY]] ]
67; CHECK-NEXT:    [[ACTIVE_LANE_MASK:%.*]] = phi <vscale x 4 x i1> [ [[ACTIVE_LANE_MASK_ENTRY]], [[VECTOR_PH]] ], [ [[ACTIVE_LANE_MASK_NEXT:%.*]], [[VECTOR_BODY]] ]
68; CHECK-NEXT:    [[TMP10:%.*]] = add i64 [[INDEX1]], 0
69; CHECK-NEXT:    [[TMP11:%.*]] = getelementptr i32, ptr [[PTR:%.*]], i64 [[TMP10]]
70; CHECK-NEXT:    [[TMP12:%.*]] = getelementptr i32, ptr [[TMP11]], i32 0
71; CHECK-NEXT:    call void @llvm.masked.store.nxv4i32.p0(<vscale x 4 x i32> [[BROADCAST_SPLAT]], ptr [[TMP12]], i32 4, <vscale x 4 x i1> [[ACTIVE_LANE_MASK]])
72; CHECK-NEXT:    [[INDEX_NEXT2]] = add i64 [[INDEX1]], [[TMP14]]
73; CHECK-NEXT:    [[ACTIVE_LANE_MASK_NEXT]] = call <vscale x 4 x i1> @llvm.get.active.lane.mask.nxv4i1.i64(i64 [[INDEX1]], i64 [[TMP9]])
74; CHECK-NEXT:    [[TMP15:%.*]] = xor <vscale x 4 x i1> [[ACTIVE_LANE_MASK_NEXT]], splat (i1 true)
75; CHECK-NEXT:    [[TMP16:%.*]] = extractelement <vscale x 4 x i1> [[TMP15]], i32 0
76; CHECK-NEXT:    br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
77; CHECK:       middle.block:
78; CHECK-NEXT:    br i1 true, label [[WHILE_END_LOOPEXIT:%.*]], label [[SCALAR_PH]]
79; CHECK:       scalar.ph:
80; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
81; CHECK-NEXT:    br label [[WHILE_BODY:%.*]]
82; CHECK:       while.body:
83; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ [[INDEX_NEXT:%.*]], [[WHILE_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
84; CHECK-NEXT:    [[GEP:%.*]] = getelementptr i32, ptr [[PTR]], i64 [[INDEX]]
85; CHECK-NEXT:    store i32 [[VAL]], ptr [[GEP]], align 4
86; CHECK-NEXT:    [[INDEX_NEXT]] = add nsw i64 [[INDEX]], 1
87; CHECK-NEXT:    [[CMP10:%.*]] = icmp ult i64 [[INDEX_NEXT]], [[N]]
88; CHECK-NEXT:    br i1 [[CMP10]], label [[WHILE_BODY]], label [[WHILE_END_LOOPEXIT]], !llvm.loop [[LOOP3:![0-9]+]]
89; CHECK:       while.end.loopexit:
90; CHECK-NEXT:    ret void
91;
92entry:
93  br label %while.body
94
95while.body:                                       ; preds = %while.body, %entry
96  %index = phi i64 [ %index.next, %while.body ], [ 0, %entry ]
97  %gep = getelementptr i32, ptr %ptr, i64 %index
98  store i32 %val, ptr %gep
99  %index.next = add nsw i64 %index, 1
100  %cmp10 = icmp ult i64 %index.next, %n
101  br i1 %cmp10, label %while.body, label %while.end.loopexit, !llvm.loop !0
102
103while.end.loopexit:                               ; preds = %while.body
104  ret void
105}
106
107
108attributes #0 = { "target-features"="+sve" }
109
110!0 = distinct !{!0, !1}
111!1 = !{!"llvm.loop.vectorize.predicate.enable", i1 true}
112