1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
| ; RUN: llc < %s -mtriple=arm64-eabi -aarch64-neon-syntax=apple -asm-verbose=false | FileCheck %s
define signext i8 @test_vmaxv_s8(<8 x i8> %a1) {
; CHECK: test_vmaxv_s8
; CHECK: smaxv.8b b[[REGNUM:[0-9]+]], v0
; CHECK-NEXT: smov.b w0, v[[REGNUM]][0]
; CHECK-NEXT: ret
entry:
%vmaxv.i = tail call i32 @llvm.aarch64.neon.smaxv.i32.v8i8(<8 x i8> %a1)
%0 = trunc i32 %vmaxv.i to i8
ret i8 %0
}
define signext i16 @test_vmaxv_s16(<4 x i16> %a1) {
; CHECK: test_vmaxv_s16
; CHECK: smaxv.4h h[[REGNUM:[0-9]+]], v0
; CHECK-NEXT: smov.h w0, v[[REGNUM]][0]
; CHECK-NEXT: ret
entry:
%vmaxv.i = tail call i32 @llvm.aarch64.neon.smaxv.i32.v4i16(<4 x i16> %a1)
%0 = trunc i32 %vmaxv.i to i16
ret i16 %0
}
define i32 @test_vmaxv_s32(<2 x i32> %a1) {
; CHECK: test_vmaxv_s32
; 2 x i32 is not supported by the ISA, thus, this is a special case
; CHECK: smaxp.2s v[[REGNUM:[0-9]+]], v0, v0
; CHECK-NEXT: fmov w0, s[[REGNUM]]
; CHECK-NEXT: ret
entry:
%vmaxv.i = tail call i32 @llvm.aarch64.neon.smaxv.i32.v2i32(<2 x i32> %a1)
ret i32 %vmaxv.i
}
define signext i8 @test_vmaxvq_s8(<16 x i8> %a1) {
; CHECK: test_vmaxvq_s8
; CHECK: smaxv.16b b[[REGNUM:[0-9]+]], v0
; CHECK-NEXT: smov.b w0, v[[REGNUM]][0]
; CHECK-NEXT: ret
entry:
%vmaxv.i = tail call i32 @llvm.aarch64.neon.smaxv.i32.v16i8(<16 x i8> %a1)
%0 = trunc i32 %vmaxv.i to i8
ret i8 %0
}
define signext i16 @test_vmaxvq_s16(<8 x i16> %a1) {
; CHECK: test_vmaxvq_s16
; CHECK: smaxv.8h h[[REGNUM:[0-9]+]], v0
; CHECK-NEXT: smov.h w0, v[[REGNUM]][0]
; CHECK-NEXT: ret
entry:
%vmaxv.i = tail call i32 @llvm.aarch64.neon.smaxv.i32.v8i16(<8 x i16> %a1)
%0 = trunc i32 %vmaxv.i to i16
ret i16 %0
}
define i32 @test_vmaxvq_s32(<4 x i32> %a1) {
; CHECK: test_vmaxvq_s32
; CHECK: smaxv.4s [[REGNUM:s[0-9]+]], v0
; CHECK-NEXT: fmov w0, [[REGNUM]]
; CHECK-NEXT: ret
entry:
%vmaxv.i = tail call i32 @llvm.aarch64.neon.smaxv.i32.v4i32(<4 x i32> %a1)
ret i32 %vmaxv.i
}
define <8 x i8> @test_vmaxv_s8_used_by_laneop(<8 x i8> %a1, <8 x i8> %a2) {
; CHECK-LABEL: test_vmaxv_s8_used_by_laneop:
; CHECK: smaxv.8b b[[REGNUM:[0-9]+]], v1
; CHECK-NEXT: mov.b v0[3], v[[REGNUM]][0]
; CHECK-NEXT: ret
entry:
%0 = tail call i32 @llvm.aarch64.neon.smaxv.i32.v8i8(<8 x i8> %a2)
%1 = trunc i32 %0 to i8
%2 = insertelement <8 x i8> %a1, i8 %1, i32 3
ret <8 x i8> %2
}
define <4 x i16> @test_vmaxv_s16_used_by_laneop(<4 x i16> %a1, <4 x i16> %a2) {
; CHECK-LABEL: test_vmaxv_s16_used_by_laneop:
; CHECK: smaxv.4h h[[REGNUM:[0-9]+]], v1
; CHECK-NEXT: mov.h v0[3], v[[REGNUM]][0]
; CHECK-NEXT: ret
entry:
%0 = tail call i32 @llvm.aarch64.neon.smaxv.i32.v4i16(<4 x i16> %a2)
%1 = trunc i32 %0 to i16
%2 = insertelement <4 x i16> %a1, i16 %1, i32 3
ret <4 x i16> %2
}
define <2 x i32> @test_vmaxv_s32_used_by_laneop(<2 x i32> %a1, <2 x i32> %a2) {
; CHECK-LABEL: test_vmaxv_s32_used_by_laneop:
; CHECK: smaxp.2s v[[REGNUM:[0-9]+]], v1, v1
; CHECK-NEXT: mov.s v0[1], v[[REGNUM]][0]
; CHECK-NEXT: ret
entry:
%0 = tail call i32 @llvm.aarch64.neon.smaxv.i32.v2i32(<2 x i32> %a2)
%1 = insertelement <2 x i32> %a1, i32 %0, i32 1
ret <2 x i32> %1
}
define <16 x i8> @test_vmaxvq_s8_used_by_laneop(<16 x i8> %a1, <16 x i8> %a2) {
; CHECK-LABEL: test_vmaxvq_s8_used_by_laneop:
; CHECK: smaxv.16b b[[REGNUM:[0-9]+]], v1
; CHECK-NEXT: mov.b v0[3], v[[REGNUM]][0]
; CHECK-NEXT: ret
entry:
%0 = tail call i32 @llvm.aarch64.neon.smaxv.i32.v16i8(<16 x i8> %a2)
%1 = trunc i32 %0 to i8
%2 = insertelement <16 x i8> %a1, i8 %1, i32 3
ret <16 x i8> %2
}
define <8 x i16> @test_vmaxvq_s16_used_by_laneop(<8 x i16> %a1, <8 x i16> %a2) {
; CHECK-LABEL: test_vmaxvq_s16_used_by_laneop:
; CHECK: smaxv.8h h[[REGNUM:[0-9]+]], v1
; CHECK-NEXT: mov.h v0[3], v[[REGNUM]][0]
; CHECK-NEXT: ret
entry:
%0 = tail call i32 @llvm.aarch64.neon.smaxv.i32.v8i16(<8 x i16> %a2)
%1 = trunc i32 %0 to i16
%2 = insertelement <8 x i16> %a1, i16 %1, i32 3
ret <8 x i16> %2
}
define <4 x i32> @test_vmaxvq_s32_used_by_laneop(<4 x i32> %a1, <4 x i32> %a2) {
; CHECK-LABEL: test_vmaxvq_s32_used_by_laneop:
; CHECK: smaxv.4s s[[REGNUM:[0-9]+]], v1
; CHECK-NEXT: mov.s v0[3], v[[REGNUM]][0]
; CHECK-NEXT: ret
entry:
%0 = tail call i32 @llvm.aarch64.neon.smaxv.i32.v4i32(<4 x i32> %a2)
%1 = insertelement <4 x i32> %a1, i32 %0, i32 3
ret <4 x i32> %1
}
declare i32 @llvm.aarch64.neon.smaxv.i32.v4i32(<4 x i32>)
declare i32 @llvm.aarch64.neon.smaxv.i32.v8i16(<8 x i16>)
declare i32 @llvm.aarch64.neon.smaxv.i32.v16i8(<16 x i8>)
declare i32 @llvm.aarch64.neon.smaxv.i32.v2i32(<2 x i32>)
declare i32 @llvm.aarch64.neon.smaxv.i32.v4i16(<4 x i16>)
declare i32 @llvm.aarch64.neon.smaxv.i32.v8i8(<8 x i8>)
|