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.fp -verify-machineinstrs -tail-predication=enabled -o - %s | FileCheck %s
3%struct.SpeexPreprocessState_ = type { i32, i32, half*, half* }
4
5define void @foo(%struct.SpeexPreprocessState_* nocapture readonly %st, i16* %x) {
6; CHECK-LABEL: foo:
7; CHECK:       @ %bb.0: @ %entry
8; CHECK-NEXT:    .save {r4, lr}
9; CHECK-NEXT:    push {r4, lr}
10; CHECK-NEXT:    ldrd r12, r2, [r0]
11; CHECK-NEXT:    ldrd r4, r3, [r0, #8]
12; CHECK-NEXT:    rsb r12, r12, r2, lsl #1
13; CHECK-NEXT:    dlstp.16 lr, r12
14; CHECK-NEXT:  .LBB0_1: @ %do.body
15; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
16; CHECK-NEXT:    vldrh.u16 q0, [r3], #16
17; CHECK-NEXT:    vstrh.16 q0, [r4], #16
18; CHECK-NEXT:    letp lr, .LBB0_1
19; CHECK-NEXT:  @ %bb.2: @ %do.end
20; CHECK-NEXT:    ldr r2, [r0]
21; CHECK-NEXT:    ldr r0, [r0, #8]
22; CHECK-NEXT:    vmov.i16 q0, #0x1800
23; CHECK-NEXT:    add.w r0, r0, r12, lsl #1
24; CHECK-NEXT:    dlstp.16 lr, r2
25; CHECK-NEXT:  .LBB0_3: @ %do.body6
26; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
27; CHECK-NEXT:    vldrh.u16 q1, [r1], #16
28; CHECK-NEXT:    vcvt.f16.s16 q1, q1
29; CHECK-NEXT:    vmul.f16 q1, q1, q0
30; CHECK-NEXT:    vstrh.16 q1, [r0], #16
31; CHECK-NEXT:    letp lr, .LBB0_3
32; CHECK-NEXT:  @ %bb.4: @ %do.end13
33; CHECK-NEXT:    pop {r4, pc}
34entry:
35  %ps_size = getelementptr inbounds %struct.SpeexPreprocessState_, %struct.SpeexPreprocessState_* %st, i32 0, i32 1
36  %0 = load i32, i32* %ps_size, align 4
37  %mul = shl nsw i32 %0, 1
38  %frame_size = getelementptr inbounds %struct.SpeexPreprocessState_, %struct.SpeexPreprocessState_* %st, i32 0, i32 0
39  %1 = load i32, i32* %frame_size, align 4
40  %sub = sub nsw i32 %mul, %1
41  %inbuf = getelementptr inbounds %struct.SpeexPreprocessState_, %struct.SpeexPreprocessState_* %st, i32 0, i32 3
42  %2 = load half*, half** %inbuf, align 4
43  %frame = getelementptr inbounds %struct.SpeexPreprocessState_, %struct.SpeexPreprocessState_* %st, i32 0, i32 2
44  %3 = load half*, half** %frame, align 4
45  br label %do.body
46
47do.body:                                          ; preds = %do.body, %entry
48  %pinbuff16.0 = phi half* [ %2, %entry ], [ %add.ptr, %do.body ]
49  %blkCnt.0 = phi i32 [ %sub, %entry ], [ %sub2, %do.body ]
50  %pframef16.0 = phi half* [ %3, %entry ], [ %add.ptr1, %do.body ]
51  %4 = tail call <8 x i1> @llvm.arm.mve.vctp16(i32 %blkCnt.0)
52  %5 = bitcast half* %pinbuff16.0 to <8 x half>*
53  %6 = tail call fast <8 x half> @llvm.masked.load.v8f16.p0v8f16(<8 x half>* %5, i32 2, <8 x i1> %4, <8 x half> zeroinitializer)
54  %7 = bitcast half* %pframef16.0 to <8 x half>*
55  tail call void @llvm.masked.store.v8f16.p0v8f16(<8 x half> %6, <8 x half>* %7, i32 2, <8 x i1> %4)
56  %add.ptr = getelementptr inbounds half, half* %pinbuff16.0, i32 8
57  %add.ptr1 = getelementptr inbounds half, half* %pframef16.0, i32 8
58  %sub2 = add nsw i32 %blkCnt.0, -8
59  %cmp = icmp sgt i32 %blkCnt.0, 8
60  br i1 %cmp, label %do.body, label %do.end
61
62do.end:                                           ; preds = %do.body
63  %8 = load half*, half** %frame, align 4
64  %add.ptr4 = getelementptr inbounds half, half* %8, i32 %sub
65  %9 = load i32, i32* %frame_size, align 4
66  br label %do.body6
67
68do.body6:                                         ; preds = %do.body6, %do.end
69  %px.0 = phi i16* [ %x, %do.end ], [ %add.ptr8, %do.body6 ]
70  %blkCnt.1 = phi i32 [ %9, %do.end ], [ %sub10, %do.body6 ]
71  %pframef16.1 = phi half* [ %add.ptr4, %do.end ], [ %add.ptr9, %do.body6 ]
72  %10 = tail call <8 x i1> @llvm.arm.mve.vctp16(i32 %blkCnt.1)
73  %11 = bitcast i16* %px.0 to <8 x i16>*
74  %12 = tail call <8 x i16> @llvm.masked.load.v8i16.p0v8i16(<8 x i16>* %11, i32 2, <8 x i1> %10, <8 x i16> zeroinitializer)
75  %13 = tail call fast <8 x half> @llvm.arm.mve.vcvt.fp.int.predicated.v8f16.v8i16.v8i1(<8 x i16> %12, i32 0, <8 x i1> %10, <8 x half> undef)
76  %14 = tail call fast <8 x half> @llvm.arm.mve.mul.predicated.v8f16.v8i1(<8 x half> %13, <8 x half> <half 0xH1800, half 0xH1800, half 0xH1800, half 0xH1800, half 0xH1800, half 0xH1800, half 0xH1800, half 0xH1800>, <8 x i1> %10, <8 x half> undef)
77  %15 = bitcast half* %pframef16.1 to <8 x half>*
78  tail call void @llvm.masked.store.v8f16.p0v8f16(<8 x half> %14, <8 x half>* %15, i32 2, <8 x i1> %10)
79  %add.ptr8 = getelementptr inbounds i16, i16* %px.0, i32 8
80  %add.ptr9 = getelementptr inbounds half, half* %pframef16.1, i32 8
81  %sub10 = add nsw i32 %blkCnt.1, -8
82  %cmp12 = icmp sgt i32 %blkCnt.1, 8
83  br i1 %cmp12, label %do.body6, label %do.end13
84
85do.end13:                                         ; preds = %do.body6
86  ret void
87}
88
89declare <8 x i1> @llvm.arm.mve.vctp16(i32)
90
91declare <8 x half> @llvm.masked.load.v8f16.p0v8f16(<8 x half>*, i32 immarg, <8 x i1>, <8 x half>)
92
93declare void @llvm.masked.store.v8f16.p0v8f16(<8 x half>, <8 x half>*, i32 immarg, <8 x i1>)
94
95declare <8 x i16> @llvm.masked.load.v8i16.p0v8i16(<8 x i16>*, i32 immarg, <8 x i1>, <8 x i16>)
96
97declare <8 x half> @llvm.arm.mve.vcvt.fp.int.predicated.v8f16.v8i16.v8i1(<8 x i16>, i32, <8 x i1>, <8 x half>)
98
99declare <8 x half> @llvm.arm.mve.mul.predicated.v8f16.v8i1(<8 x half>, <8 x half>, <8 x i1>, <8 x half>)
100