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
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
|
; RUN: llc -march=arm64 -arm64-neon-syntax=apple < %s -mcpu=cyclone | FileCheck %s
define signext i8 @test_vaddv_s8(<8 x i8> %a1) {
; CHECK-LABEL: test_vaddv_s8:
; CHECK: addv.8b b[[REGNUM:[0-9]+]], v0
; CHECK-NEXT: smov.b w0, v[[REGNUM]][0]
; CHECK-NEXT: ret
entry:
%vaddv.i = tail call i32 @llvm.arm64.neon.saddv.i32.v8i8(<8 x i8> %a1)
%0 = trunc i32 %vaddv.i to i8
ret i8 %0
}
define signext i16 @test_vaddv_s16(<4 x i16> %a1) {
; CHECK-LABEL: test_vaddv_s16:
; CHECK: addv.4h h[[REGNUM:[0-9]+]], v0
; CHECK-NEXT: smov.h w0, v[[REGNUM]][0]
; CHECK-NEXT: ret
entry:
%vaddv.i = tail call i32 @llvm.arm64.neon.saddv.i32.v4i16(<4 x i16> %a1)
%0 = trunc i32 %vaddv.i to i16
ret i16 %0
}
define i32 @test_vaddv_s32(<2 x i32> %a1) {
; CHECK-LABEL: test_vaddv_s32:
; 2 x i32 is not supported by the ISA, thus, this is a special case
; CHECK: addp.2s v[[REGNUM:[0-9]+]], v0, v0
; CHECK-NEXT: fmov w0, s[[REGNUM]]
; CHECK-NEXT: ret
entry:
%vaddv.i = tail call i32 @llvm.arm64.neon.saddv.i32.v2i32(<2 x i32> %a1)
ret i32 %vaddv.i
}
define i64 @test_vaddv_s64(<2 x i64> %a1) {
; CHECK-LABEL: test_vaddv_s64:
; CHECK: addp.2d [[REGNUM:d[0-9]+]], v0
; CHECK-NEXT: fmov x0, [[REGNUM]]
; CHECK-NEXT: ret
entry:
%vaddv.i = tail call i64 @llvm.arm64.neon.saddv.i64.v2i64(<2 x i64> %a1)
ret i64 %vaddv.i
}
define zeroext i8 @test_vaddv_u8(<8 x i8> %a1) {
; CHECK-LABEL: test_vaddv_u8:
; CHECK: addv.8b b[[REGNUM:[0-9]+]], v0
; CHECK-NEXT: fmov w0, s[[REGNUM]]
; CHECK-NEXT: ret
entry:
%vaddv.i = tail call i32 @llvm.arm64.neon.uaddv.i32.v8i8(<8 x i8> %a1)
%0 = trunc i32 %vaddv.i to i8
ret i8 %0
}
define i32 @test_vaddv_u8_masked(<8 x i8> %a1) {
; CHECK-LABEL: test_vaddv_u8_masked:
; CHECK: addv.8b b[[REGNUM:[0-9]+]], v0
; CHECK-NEXT: fmov w0, s[[REGNUM]]
; CHECK-NEXT: ret
entry:
%vaddv.i = tail call i32 @llvm.arm64.neon.uaddv.i32.v8i8(<8 x i8> %a1)
%0 = and i32 %vaddv.i, 511 ; 0x1ff
ret i32 %0
}
define zeroext i16 @test_vaddv_u16(<4 x i16> %a1) {
; CHECK-LABEL: test_vaddv_u16:
; CHECK: addv.4h h[[REGNUM:[0-9]+]], v0
; CHECK-NEXT: fmov w0, s[[REGNUM]]
; CHECK-NEXT: ret
entry:
%vaddv.i = tail call i32 @llvm.arm64.neon.uaddv.i32.v4i16(<4 x i16> %a1)
%0 = trunc i32 %vaddv.i to i16
ret i16 %0
}
define i32 @test_vaddv_u16_masked(<4 x i16> %a1) {
; CHECK-LABEL: test_vaddv_u16_masked:
; CHECK: addv.4h h[[REGNUM:[0-9]+]], v0
; CHECK-NEXT: fmov w0, s[[REGNUM]]
; CHECK-NEXT: ret
entry:
%vaddv.i = tail call i32 @llvm.arm64.neon.uaddv.i32.v4i16(<4 x i16> %a1)
%0 = and i32 %vaddv.i, 3276799 ; 0x31ffff
ret i32 %0
}
define i32 @test_vaddv_u32(<2 x i32> %a1) {
; CHECK-LABEL: test_vaddv_u32:
; 2 x i32 is not supported by the ISA, thus, this is a special case
; CHECK: addp.2s v[[REGNUM:[0-9]+]], v0, v0
; CHECK-NEXT: fmov w0, s[[REGNUM]]
; CHECK-NEXT: ret
entry:
%vaddv.i = tail call i32 @llvm.arm64.neon.uaddv.i32.v2i32(<2 x i32> %a1)
ret i32 %vaddv.i
}
define float @test_vaddv_f32(<2 x float> %a1) {
; CHECK-LABEL: test_vaddv_f32:
; CHECK: faddp.2s s0, v0
; CHECK-NEXT: ret
entry:
%vaddv.i = tail call float @llvm.arm64.neon.faddv.f32.v2f32(<2 x float> %a1)
ret float %vaddv.i
}
define float @test_vaddv_v4f32(<4 x float> %a1) {
; CHECK-LABEL: test_vaddv_v4f32:
; CHECK: faddp.4s [[REGNUM:v[0-9]+]], v0, v0
; CHECK: faddp.2s s0, [[REGNUM]]
; CHECK-NEXT: ret
entry:
%vaddv.i = tail call float @llvm.arm64.neon.faddv.f32.v4f32(<4 x float> %a1)
ret float %vaddv.i
}
define double @test_vaddv_f64(<2 x double> %a1) {
; CHECK-LABEL: test_vaddv_f64:
; CHECK: faddp.2d d0, v0
; CHECK-NEXT: ret
entry:
%vaddv.i = tail call double @llvm.arm64.neon.faddv.f64.v2f64(<2 x double> %a1)
ret double %vaddv.i
}
define i64 @test_vaddv_u64(<2 x i64> %a1) {
; CHECK-LABEL: test_vaddv_u64:
; CHECK: addp.2d [[REGNUM:d[0-9]+]], v0
; CHECK-NEXT: fmov x0, [[REGNUM]]
; CHECK-NEXT: ret
entry:
%vaddv.i = tail call i64 @llvm.arm64.neon.uaddv.i64.v2i64(<2 x i64> %a1)
ret i64 %vaddv.i
}
define <1 x i64> @test_vaddv_u64_to_vec(<2 x i64> %a1) {
; CHECK-LABEL: test_vaddv_u64_to_vec:
; CHECK: addp.2d d0, v0
; CHECK-NOT: fmov
; CHECK-NOT: ins
; CHECK: ret
entry:
%vaddv.i = tail call i64 @llvm.arm64.neon.uaddv.i64.v2i64(<2 x i64> %a1)
%vec = insertelement <1 x i64> undef, i64 %vaddv.i, i32 0
ret <1 x i64> %vec
}
define signext i8 @test_vaddvq_s8(<16 x i8> %a1) {
; CHECK-LABEL: test_vaddvq_s8:
; CHECK: addv.16b b[[REGNUM:[0-9]+]], v0
; CHECK-NEXT: smov.b w0, v[[REGNUM]][0]
; CHECK-NEXT: ret
entry:
%vaddv.i = tail call i32 @llvm.arm64.neon.saddv.i32.v16i8(<16 x i8> %a1)
%0 = trunc i32 %vaddv.i to i8
ret i8 %0
}
define signext i16 @test_vaddvq_s16(<8 x i16> %a1) {
; CHECK-LABEL: test_vaddvq_s16:
; CHECK: addv.8h h[[REGNUM:[0-9]+]], v0
; CHECK-NEXT: smov.h w0, v[[REGNUM]][0]
; CHECK-NEXT: ret
entry:
%vaddv.i = tail call i32 @llvm.arm64.neon.saddv.i32.v8i16(<8 x i16> %a1)
%0 = trunc i32 %vaddv.i to i16
ret i16 %0
}
define i32 @test_vaddvq_s32(<4 x i32> %a1) {
; CHECK-LABEL: test_vaddvq_s32:
; CHECK: addv.4s [[REGNUM:s[0-9]+]], v0
; CHECK-NEXT: fmov w0, [[REGNUM]]
; CHECK-NEXT: ret
entry:
%vaddv.i = tail call i32 @llvm.arm64.neon.saddv.i32.v4i32(<4 x i32> %a1)
ret i32 %vaddv.i
}
define zeroext i8 @test_vaddvq_u8(<16 x i8> %a1) {
; CHECK-LABEL: test_vaddvq_u8:
; CHECK: addv.16b b[[REGNUM:[0-9]+]], v0
; CHECK-NEXT: fmov w0, s[[REGNUM]]
; CHECK-NEXT: ret
entry:
%vaddv.i = tail call i32 @llvm.arm64.neon.uaddv.i32.v16i8(<16 x i8> %a1)
%0 = trunc i32 %vaddv.i to i8
ret i8 %0
}
define zeroext i16 @test_vaddvq_u16(<8 x i16> %a1) {
; CHECK-LABEL: test_vaddvq_u16:
; CHECK: addv.8h h[[REGNUM:[0-9]+]], v0
; CHECK-NEXT: fmov w0, s[[REGNUM]]
; CHECK-NEXT: ret
entry:
%vaddv.i = tail call i32 @llvm.arm64.neon.uaddv.i32.v8i16(<8 x i16> %a1)
%0 = trunc i32 %vaddv.i to i16
ret i16 %0
}
define i32 @test_vaddvq_u32(<4 x i32> %a1) {
; CHECK-LABEL: test_vaddvq_u32:
; CHECK: addv.4s [[REGNUM:s[0-9]+]], v0
; CHECK-NEXT: fmov [[FMOVRES:w[0-9]+]], [[REGNUM]]
; CHECK-NEXT: ret
entry:
%vaddv.i = tail call i32 @llvm.arm64.neon.uaddv.i32.v4i32(<4 x i32> %a1)
ret i32 %vaddv.i
}
declare i32 @llvm.arm64.neon.uaddv.i32.v4i32(<4 x i32>)
declare i32 @llvm.arm64.neon.uaddv.i32.v8i16(<8 x i16>)
declare i32 @llvm.arm64.neon.uaddv.i32.v16i8(<16 x i8>)
declare i32 @llvm.arm64.neon.saddv.i32.v4i32(<4 x i32>)
declare i32 @llvm.arm64.neon.saddv.i32.v8i16(<8 x i16>)
declare i32 @llvm.arm64.neon.saddv.i32.v16i8(<16 x i8>)
declare i64 @llvm.arm64.neon.uaddv.i64.v2i64(<2 x i64>)
declare i32 @llvm.arm64.neon.uaddv.i32.v2i32(<2 x i32>)
declare i32 @llvm.arm64.neon.uaddv.i32.v4i16(<4 x i16>)
declare i32 @llvm.arm64.neon.uaddv.i32.v8i8(<8 x i8>)
declare i32 @llvm.arm64.neon.saddv.i32.v2i32(<2 x i32>)
declare i64 @llvm.arm64.neon.saddv.i64.v2i64(<2 x i64>)
declare i32 @llvm.arm64.neon.saddv.i32.v4i16(<4 x i16>)
declare i32 @llvm.arm64.neon.saddv.i32.v8i8(<8 x i8>)
declare float @llvm.arm64.neon.faddv.f32.v2f32(<2 x float> %a1)
declare float @llvm.arm64.neon.faddv.f32.v4f32(<4 x float> %a1)
declare double @llvm.arm64.neon.faddv.f64.v2f64(<2 x double> %a1)
|
