1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc -mtriple=thumbv8.1m.main-none-none-eabi -mattr=+mve -verify-machineinstrs %s -o - | FileCheck %s
4 ; This test has an instruction that gets sunk into the loop, that is a
5 ; active.lane.mask operand. (%exitcount.ptrcnt.to.int = ptrtoint). We
6 ; need to make sure it is loop invariant.
8 define i32 @a(i32* readnone %b, i8* %c) {
10 ; CHECK: @ %bb.0: @ %entry
11 ; CHECK-NEXT: .save {r4, lr}
12 ; CHECK-NEXT: push {r4, lr}
13 ; CHECK-NEXT: cmp r0, r1
15 ; CHECK-NEXT: popls {r4, pc}
16 ; CHECK-NEXT: .LBB0_1: @ %while.body.preheader
17 ; CHECK-NEXT: subs r4, r0, r1
18 ; CHECK-NEXT: movs r2, #0
19 ; CHECK-NEXT: mov r3, r1
20 ; CHECK-NEXT: dlstp.8 lr, r4
21 ; CHECK-NEXT: .LBB0_2: @ %vector.body
22 ; CHECK-NEXT: @ =>This Inner Loop Header: Depth=1
23 ; CHECK-NEXT: adds r0, r1, r2
24 ; CHECK-NEXT: adds r2, #16
25 ; CHECK-NEXT: vidup.u8 q0, r0, #1
26 ; CHECK-NEXT: vstrb.8 q0, [r3], #16
27 ; CHECK-NEXT: letp lr, .LBB0_2
28 ; CHECK-NEXT: @ %bb.3: @ %while.end
29 ; CHECK-NEXT: pop {r4, pc}
31 %0 = bitcast i32* %b to i8*
32 %cmp3 = icmp ugt i8* %0, %c
33 br i1 %cmp3, label %while.body.preheader, label %while.end
35 while.body.preheader: ; preds = %entry
36 %c5 = ptrtoint i8* %c to i32
38 %uglygep = getelementptr i8, i8* %0, i32 %1
39 %exitcount.ptrcnt.to.int = ptrtoint i8* %uglygep to i32
40 %n.rnd.up = add i32 %exitcount.ptrcnt.to.int, 15
41 %n.vec = and i32 %n.rnd.up, -16
44 vector.body: ; preds = %vector.body, %while.body.preheader
45 %index = phi i32 [ 0, %while.body.preheader ], [ %index.next, %vector.body ]
46 %next.gep = getelementptr i8, i8* %c, i32 %index
48 %next.gep7 = getelementptr i8, i8* %c, i32 %2
50 %next.gep8 = getelementptr i8, i8* %c, i32 %3
52 %next.gep9 = getelementptr i8, i8* %c, i32 %4
54 %next.gep10 = getelementptr i8, i8* %c, i32 %5
56 %next.gep11 = getelementptr i8, i8* %c, i32 %6
58 %next.gep12 = getelementptr i8, i8* %c, i32 %7
60 %next.gep13 = getelementptr i8, i8* %c, i32 %8
62 %next.gep14 = getelementptr i8, i8* %c, i32 %9
63 %10 = or i32 %index, 9
64 %next.gep15 = getelementptr i8, i8* %c, i32 %10
65 %11 = or i32 %index, 10
66 %next.gep16 = getelementptr i8, i8* %c, i32 %11
67 %12 = or i32 %index, 11
68 %next.gep17 = getelementptr i8, i8* %c, i32 %12
69 %13 = or i32 %index, 12
70 %next.gep18 = getelementptr i8, i8* %c, i32 %13
71 %14 = or i32 %index, 13
72 %next.gep19 = getelementptr i8, i8* %c, i32 %14
73 %15 = or i32 %index, 14
74 %next.gep20 = getelementptr i8, i8* %c, i32 %15
75 %16 = or i32 %index, 15
76 %next.gep21 = getelementptr i8, i8* %c, i32 %16
77 %17 = insertelement <16 x i8*> poison, i8* %next.gep, i32 0
78 %18 = insertelement <16 x i8*> %17, i8* %next.gep7, i32 1
79 %19 = insertelement <16 x i8*> %18, i8* %next.gep8, i32 2
80 %20 = insertelement <16 x i8*> %19, i8* %next.gep9, i32 3
81 %21 = insertelement <16 x i8*> %20, i8* %next.gep10, i32 4
82 %22 = insertelement <16 x i8*> %21, i8* %next.gep11, i32 5
83 %23 = insertelement <16 x i8*> %22, i8* %next.gep12, i32 6
84 %24 = insertelement <16 x i8*> %23, i8* %next.gep13, i32 7
85 %25 = insertelement <16 x i8*> %24, i8* %next.gep14, i32 8
86 %26 = insertelement <16 x i8*> %25, i8* %next.gep15, i32 9
87 %27 = insertelement <16 x i8*> %26, i8* %next.gep16, i32 10
88 %28 = insertelement <16 x i8*> %27, i8* %next.gep17, i32 11
89 %29 = insertelement <16 x i8*> %28, i8* %next.gep18, i32 12
90 %30 = insertelement <16 x i8*> %29, i8* %next.gep19, i32 13
91 %31 = insertelement <16 x i8*> %30, i8* %next.gep20, i32 14
92 %32 = insertelement <16 x i8*> %31, i8* %next.gep21, i32 15
93 %active.lane.mask = call <16 x i1> @llvm.get.active.lane.mask.v16i1.i32(i32 %index, i32 %exitcount.ptrcnt.to.int)
94 %33 = ptrtoint <16 x i8*> %32 to <16 x i32>
95 %34 = trunc <16 x i32> %33 to <16 x i8>
96 %35 = bitcast i8* %next.gep to <16 x i8>*
97 call void @llvm.masked.store.v16i8.p0v16i8(<16 x i8> %34, <16 x i8>* %35, i32 1, <16 x i1> %active.lane.mask)
98 %index.next = add i32 %index, 16
99 %36 = icmp eq i32 %index.next, %n.vec
100 br i1 %36, label %while.end, label %vector.body
102 while.end: ; preds = %vector.body, %entry
106 declare <16 x i1> @llvm.get.active.lane.mask.v16i1.i32(i32, i32)
107 declare void @llvm.masked.store.v16i8.p0v16i8(<16 x i8>, <16 x i8>*, i32 immarg, <16 x i1>)