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
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mtriple=aarch64 -mattr=+sve %s -o - | FileCheck %s --check-prefix=CHECK-SVE
; RUN: llc -mtriple=aarch64 -mattr=+sve2 %s -o - | FileCheck %s --check-prefix=CHECK-SVE2
define <vscale x 2 x i64> @signed_wide_add_nxv4i32(<vscale x 2 x i64> %acc, <vscale x 4 x i32> %input){
; CHECK-SVE-LABEL: signed_wide_add_nxv4i32:
; CHECK-SVE: // %bb.0: // %entry
; CHECK-SVE-NEXT: sunpklo z2.d, z1.s
; CHECK-SVE-NEXT: sunpkhi z1.d, z1.s
; CHECK-SVE-NEXT: add z0.d, z0.d, z2.d
; CHECK-SVE-NEXT: add z0.d, z0.d, z1.d
; CHECK-SVE-NEXT: ret
;
; CHECK-SVE2-LABEL: signed_wide_add_nxv4i32:
; CHECK-SVE2: // %bb.0: // %entry
; CHECK-SVE2-NEXT: saddwb z0.d, z0.d, z1.s
; CHECK-SVE2-NEXT: saddwt z0.d, z0.d, z1.s
; CHECK-SVE2-NEXT: ret
entry:
%input.wide = sext <vscale x 4 x i32> %input to <vscale x 4 x i64>
%partial.reduce = tail call <vscale x 2 x i64> @llvm.experimental.vector.partial.reduce.add.nxv2i64.nxv4i64(<vscale x 2 x i64> %acc, <vscale x 4 x i64> %input.wide)
ret <vscale x 2 x i64> %partial.reduce
}
define <vscale x 2 x i64> @unsigned_wide_add_nxv4i32(<vscale x 2 x i64> %acc, <vscale x 4 x i32> %input){
; CHECK-SVE-LABEL: unsigned_wide_add_nxv4i32:
; CHECK-SVE: // %bb.0: // %entry
; CHECK-SVE-NEXT: uunpklo z2.d, z1.s
; CHECK-SVE-NEXT: uunpkhi z1.d, z1.s
; CHECK-SVE-NEXT: add z0.d, z0.d, z2.d
; CHECK-SVE-NEXT: add z0.d, z0.d, z1.d
; CHECK-SVE-NEXT: ret
;
; CHECK-SVE2-LABEL: unsigned_wide_add_nxv4i32:
; CHECK-SVE2: // %bb.0: // %entry
; CHECK-SVE2-NEXT: uaddwb z0.d, z0.d, z1.s
; CHECK-SVE2-NEXT: uaddwt z0.d, z0.d, z1.s
; CHECK-SVE2-NEXT: ret
entry:
%input.wide = zext <vscale x 4 x i32> %input to <vscale x 4 x i64>
%partial.reduce = tail call <vscale x 2 x i64> @llvm.experimental.vector.partial.reduce.add.nxv2i64.nxv4i64(<vscale x 2 x i64> %acc, <vscale x 4 x i64> %input.wide)
ret <vscale x 2 x i64> %partial.reduce
}
define <vscale x 4 x i32> @signed_wide_add_nxv8i16(<vscale x 4 x i32> %acc, <vscale x 8 x i16> %input){
; CHECK-SVE-LABEL: signed_wide_add_nxv8i16:
; CHECK-SVE: // %bb.0: // %entry
; CHECK-SVE-NEXT: sunpklo z2.s, z1.h
; CHECK-SVE-NEXT: sunpkhi z1.s, z1.h
; CHECK-SVE-NEXT: add z0.s, z0.s, z2.s
; CHECK-SVE-NEXT: add z0.s, z0.s, z1.s
; CHECK-SVE-NEXT: ret
;
; CHECK-SVE2-LABEL: signed_wide_add_nxv8i16:
; CHECK-SVE2: // %bb.0: // %entry
; CHECK-SVE2-NEXT: saddwb z0.s, z0.s, z1.h
; CHECK-SVE2-NEXT: saddwt z0.s, z0.s, z1.h
; CHECK-SVE2-NEXT: ret
entry:
%input.wide = sext <vscale x 8 x i16> %input to <vscale x 8 x i32>
%partial.reduce = tail call <vscale x 4 x i32> @llvm.experimental.vector.partial.reduce.add.nxv4i32.nxv8i32(<vscale x 4 x i32> %acc, <vscale x 8 x i32> %input.wide)
ret <vscale x 4 x i32> %partial.reduce
}
define <vscale x 4 x i32> @unsigned_wide_add_nxv8i16(<vscale x 4 x i32> %acc, <vscale x 8 x i16> %input){
; CHECK-SVE-LABEL: unsigned_wide_add_nxv8i16:
; CHECK-SVE: // %bb.0: // %entry
; CHECK-SVE-NEXT: uunpklo z2.s, z1.h
; CHECK-SVE-NEXT: uunpkhi z1.s, z1.h
; CHECK-SVE-NEXT: add z0.s, z0.s, z2.s
; CHECK-SVE-NEXT: add z0.s, z0.s, z1.s
; CHECK-SVE-NEXT: ret
;
; CHECK-SVE2-LABEL: unsigned_wide_add_nxv8i16:
; CHECK-SVE2: // %bb.0: // %entry
; CHECK-SVE2-NEXT: uaddwb z0.s, z0.s, z1.h
; CHECK-SVE2-NEXT: uaddwt z0.s, z0.s, z1.h
; CHECK-SVE2-NEXT: ret
entry:
%input.wide = zext <vscale x 8 x i16> %input to <vscale x 8 x i32>
%partial.reduce = tail call <vscale x 4 x i32> @llvm.experimental.vector.partial.reduce.add.nxv4i32.nxv8i32(<vscale x 4 x i32> %acc, <vscale x 8 x i32> %input.wide)
ret <vscale x 4 x i32> %partial.reduce
}
define <vscale x 8 x i16> @signed_wide_add_nxv16i8(<vscale x 8 x i16> %acc, <vscale x 16 x i8> %input){
; CHECK-SVE-LABEL: signed_wide_add_nxv16i8:
; CHECK-SVE: // %bb.0: // %entry
; CHECK-SVE-NEXT: sunpklo z2.h, z1.b
; CHECK-SVE-NEXT: sunpkhi z1.h, z1.b
; CHECK-SVE-NEXT: add z0.h, z0.h, z2.h
; CHECK-SVE-NEXT: add z0.h, z0.h, z1.h
; CHECK-SVE-NEXT: ret
;
; CHECK-SVE2-LABEL: signed_wide_add_nxv16i8:
; CHECK-SVE2: // %bb.0: // %entry
; CHECK-SVE2-NEXT: saddwb z0.h, z0.h, z1.b
; CHECK-SVE2-NEXT: saddwt z0.h, z0.h, z1.b
; CHECK-SVE2-NEXT: ret
entry:
%input.wide = sext <vscale x 16 x i8> %input to <vscale x 16 x i16>
%partial.reduce = tail call <vscale x 8 x i16> @llvm.experimental.vector.partial.reduce.add.nxv8i16.nxv16i16(<vscale x 8 x i16> %acc, <vscale x 16 x i16> %input.wide)
ret <vscale x 8 x i16> %partial.reduce
}
define <vscale x 8 x i16> @unsigned_wide_add_nxv16i8(<vscale x 8 x i16> %acc, <vscale x 16 x i8> %input){
; CHECK-SVE-LABEL: unsigned_wide_add_nxv16i8:
; CHECK-SVE: // %bb.0: // %entry
; CHECK-SVE-NEXT: uunpklo z2.h, z1.b
; CHECK-SVE-NEXT: uunpkhi z1.h, z1.b
; CHECK-SVE-NEXT: add z0.h, z0.h, z2.h
; CHECK-SVE-NEXT: add z0.h, z0.h, z1.h
; CHECK-SVE-NEXT: ret
;
; CHECK-SVE2-LABEL: unsigned_wide_add_nxv16i8:
; CHECK-SVE2: // %bb.0: // %entry
; CHECK-SVE2-NEXT: uaddwb z0.h, z0.h, z1.b
; CHECK-SVE2-NEXT: uaddwt z0.h, z0.h, z1.b
; CHECK-SVE2-NEXT: ret
entry:
%input.wide = zext <vscale x 16 x i8> %input to <vscale x 16 x i16>
%partial.reduce = tail call <vscale x 8 x i16> @llvm.experimental.vector.partial.reduce.add.nxv8i16.nxv16i16(<vscale x 8 x i16> %acc, <vscale x 16 x i16> %input.wide)
ret <vscale x 8 x i16> %partial.reduce
}
define <vscale x 2 x i32> @signed_wide_add_nxv4i16(<vscale x 2 x i32> %acc, <vscale x 4 x i16> %input){
; CHECK-SVE-LABEL: signed_wide_add_nxv4i16:
; CHECK-SVE: // %bb.0: // %entry
; CHECK-SVE-NEXT: ptrue p0.s
; CHECK-SVE-NEXT: sxth z1.s, p0/m, z1.s
; CHECK-SVE-NEXT: uunpklo z2.d, z1.s
; CHECK-SVE-NEXT: uunpkhi z1.d, z1.s
; CHECK-SVE-NEXT: add z0.d, z0.d, z2.d
; CHECK-SVE-NEXT: add z0.d, z1.d, z0.d
; CHECK-SVE-NEXT: ret
;
; CHECK-SVE2-LABEL: signed_wide_add_nxv4i16:
; CHECK-SVE2: // %bb.0: // %entry
; CHECK-SVE2-NEXT: ptrue p0.s
; CHECK-SVE2-NEXT: sxth z1.s, p0/m, z1.s
; CHECK-SVE2-NEXT: saddwb z0.d, z0.d, z1.s
; CHECK-SVE2-NEXT: saddwt z0.d, z0.d, z1.s
; CHECK-SVE2-NEXT: ret
entry:
%input.wide = sext <vscale x 4 x i16> %input to <vscale x 4 x i32>
%partial.reduce = tail call <vscale x 2 x i32> @llvm.experimental.vector.partial.reduce.add.nxv2i32.nxv4i32(<vscale x 2 x i32> %acc, <vscale x 4 x i32> %input.wide)
ret <vscale x 2 x i32> %partial.reduce
}
define <vscale x 2 x i32> @unsigned_wide_add_nxv4i16(<vscale x 2 x i32> %acc, <vscale x 4 x i16> %input){
; CHECK-SVE-LABEL: unsigned_wide_add_nxv4i16:
; CHECK-SVE: // %bb.0: // %entry
; CHECK-SVE-NEXT: and z1.s, z1.s, #0xffff
; CHECK-SVE-NEXT: uunpklo z2.d, z1.s
; CHECK-SVE-NEXT: uunpkhi z1.d, z1.s
; CHECK-SVE-NEXT: add z0.d, z0.d, z2.d
; CHECK-SVE-NEXT: add z0.d, z1.d, z0.d
; CHECK-SVE-NEXT: ret
;
; CHECK-SVE2-LABEL: unsigned_wide_add_nxv4i16:
; CHECK-SVE2: // %bb.0: // %entry
; CHECK-SVE2-NEXT: and z1.s, z1.s, #0xffff
; CHECK-SVE2-NEXT: uaddwb z0.d, z0.d, z1.s
; CHECK-SVE2-NEXT: uaddwt z0.d, z0.d, z1.s
; CHECK-SVE2-NEXT: ret
entry:
%input.wide = zext <vscale x 4 x i16> %input to <vscale x 4 x i32>
%partial.reduce = tail call <vscale x 2 x i32> @llvm.experimental.vector.partial.reduce.add.nxv2i32.nxv4i32(<vscale x 2 x i32> %acc, <vscale x 4 x i32> %input.wide)
ret <vscale x 2 x i32> %partial.reduce
}
define <vscale x 4 x i64> @signed_wide_add_nxv8i32(<vscale x 4 x i64> %acc, <vscale x 8 x i32> %input){
; CHECK-SVE-LABEL: signed_wide_add_nxv8i32:
; CHECK-SVE: // %bb.0: // %entry
; CHECK-SVE-NEXT: sunpklo z4.d, z3.s
; CHECK-SVE-NEXT: sunpklo z5.d, z2.s
; CHECK-SVE-NEXT: sunpkhi z3.d, z3.s
; CHECK-SVE-NEXT: sunpkhi z2.d, z2.s
; CHECK-SVE-NEXT: add z0.d, z0.d, z5.d
; CHECK-SVE-NEXT: add z1.d, z1.d, z4.d
; CHECK-SVE-NEXT: add z0.d, z0.d, z2.d
; CHECK-SVE-NEXT: add z1.d, z1.d, z3.d
; CHECK-SVE-NEXT: ret
;
; CHECK-SVE2-LABEL: signed_wide_add_nxv8i32:
; CHECK-SVE2: // %bb.0: // %entry
; CHECK-SVE2-NEXT: saddwb z1.d, z1.d, z3.s
; CHECK-SVE2-NEXT: saddwb z0.d, z0.d, z2.s
; CHECK-SVE2-NEXT: saddwt z1.d, z1.d, z3.s
; CHECK-SVE2-NEXT: saddwt z0.d, z0.d, z2.s
; CHECK-SVE2-NEXT: ret
entry:
%input.wide = sext <vscale x 8 x i32> %input to <vscale x 8 x i64>
%partial.reduce = tail call <vscale x 4 x i64> @llvm.experimental.vector.partial.reduce.add.nxv4i64.nxv8i64(<vscale x 4 x i64> %acc, <vscale x 8 x i64> %input.wide)
ret <vscale x 4 x i64> %partial.reduce
}
define <vscale x 4 x i64> @unsigned_wide_add_nxv8i32(<vscale x 4 x i64> %acc, <vscale x 8 x i32> %input){
; CHECK-SVE-LABEL: unsigned_wide_add_nxv8i32:
; CHECK-SVE: // %bb.0: // %entry
; CHECK-SVE-NEXT: uunpklo z4.d, z3.s
; CHECK-SVE-NEXT: uunpklo z5.d, z2.s
; CHECK-SVE-NEXT: uunpkhi z3.d, z3.s
; CHECK-SVE-NEXT: uunpkhi z2.d, z2.s
; CHECK-SVE-NEXT: add z0.d, z0.d, z5.d
; CHECK-SVE-NEXT: add z1.d, z1.d, z4.d
; CHECK-SVE-NEXT: add z0.d, z0.d, z2.d
; CHECK-SVE-NEXT: add z1.d, z1.d, z3.d
; CHECK-SVE-NEXT: ret
;
; CHECK-SVE2-LABEL: unsigned_wide_add_nxv8i32:
; CHECK-SVE2: // %bb.0: // %entry
; CHECK-SVE2-NEXT: uaddwb z1.d, z1.d, z3.s
; CHECK-SVE2-NEXT: uaddwb z0.d, z0.d, z2.s
; CHECK-SVE2-NEXT: uaddwt z1.d, z1.d, z3.s
; CHECK-SVE2-NEXT: uaddwt z0.d, z0.d, z2.s
; CHECK-SVE2-NEXT: ret
entry:
%input.wide = zext <vscale x 8 x i32> %input to <vscale x 8 x i64>
%partial.reduce = tail call <vscale x 4 x i64> @llvm.experimental.vector.partial.reduce.add.nxv4i64.nxv8i64(<vscale x 4 x i64> %acc, <vscale x 8 x i64> %input.wide)
ret <vscale x 4 x i64> %partial.reduce
}
|
