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
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
; RUN: llc -mtriple=aarch64 -mattr=+sve < %s -o - | FileCheck --check-prefixes=CHECK,SVE %s
; RUN: llc -mtriple=aarch64 -mattr=+sve2 < %s -o - | FileCheck --check-prefixes=CHECK,SVE2 %s
; Wrong add/shift amount. Should be 32 for shift of 6.
define <vscale x 2 x i64> @neg_urshr_1(<vscale x 2 x i64> %x) {
; CHECK-LABEL: neg_urshr_1:
; CHECK: // %bb.0:
; CHECK-NEXT: add z0.d, z0.d, #16 // =0x10
; CHECK-NEXT: lsr z0.d, z0.d, #6
; CHECK-NEXT: ret
%add = add nuw nsw <vscale x 2 x i64> %x, splat (i64 16)
%sh = lshr <vscale x 2 x i64> %add, splat (i64 6)
ret <vscale x 2 x i64> %sh
}
; Vector Shift.
define <vscale x 2 x i64> @neg_urshr_2(<vscale x 2 x i64> %x, <vscale x 2 x i64> %y) {
; CHECK-LABEL: neg_urshr_2:
; CHECK: // %bb.0:
; CHECK-NEXT: add z0.d, z0.d, #32 // =0x20
; CHECK-NEXT: ptrue p0.d
; CHECK-NEXT: lsr z0.d, p0/m, z0.d, z1.d
; CHECK-NEXT: ret
%add = add nuw nsw <vscale x 2 x i64> %x, splat (i64 32)
%sh = lshr <vscale x 2 x i64> %add, %y
ret <vscale x 2 x i64> %sh
}
; Vector Add.
define <vscale x 2 x i64> @neg_urshr_3(<vscale x 2 x i64> %x, <vscale x 2 x i64> %y) {
; CHECK-LABEL: neg_urshr_3:
; CHECK: // %bb.0:
; CHECK-NEXT: add z0.d, z0.d, z1.d
; CHECK-NEXT: lsr z0.d, z0.d, #6
; CHECK-NEXT: ret
%add = add nuw nsw <vscale x 2 x i64> %x, %y
%sh = lshr <vscale x 2 x i64> %add, splat (i64 6)
ret <vscale x 2 x i64> %sh
}
; Add has two uses.
define <vscale x 2 x i64> @neg_urshr_4(<vscale x 2 x i64> %x, ptr %p) {
; CHECK-LABEL: neg_urshr_4:
; CHECK: // %bb.0:
; CHECK-NEXT: mov z1.d, z0.d
; CHECK-NEXT: ptrue p0.d
; CHECK-NEXT: add z1.d, z1.d, #32 // =0x20
; CHECK-NEXT: lsr z0.d, z1.d, #6
; CHECK-NEXT: st1d { z1.d }, p0, [x0]
; CHECK-NEXT: ret
%add = add nuw nsw <vscale x 2 x i64> %x, splat (i64 32)
%sh = lshr <vscale x 2 x i64> %add, splat (i64 6)
store <vscale x 2 x i64> %add, ptr %p
ret <vscale x 2 x i64> %sh
}
; Add can overflow.
define <vscale x 2 x i64> @neg_urshr_5(<vscale x 2 x i64> %x) {
; CHECK-LABEL: neg_urshr_5:
; CHECK: // %bb.0:
; CHECK-NEXT: add z0.d, z0.d, #32 // =0x20
; CHECK-NEXT: lsr z0.d, z0.d, #6
; CHECK-NEXT: ret
%add = add <vscale x 2 x i64> %x, splat (i64 32)
%sh = lshr <vscale x 2 x i64> %add, splat (i64 6)
ret <vscale x 2 x i64> %sh
}
define <vscale x 16 x i8> @urshr_i8(<vscale x 16 x i8> %x) {
; SVE-LABEL: urshr_i8:
; SVE: // %bb.0:
; SVE-NEXT: add z0.b, z0.b, #32 // =0x20
; SVE-NEXT: lsr z0.b, z0.b, #6
; SVE-NEXT: ret
;
; SVE2-LABEL: urshr_i8:
; SVE2: // %bb.0:
; SVE2-NEXT: ptrue p0.b
; SVE2-NEXT: urshr z0.b, p0/m, z0.b, #6
; SVE2-NEXT: ret
%add = add nuw nsw <vscale x 16 x i8> %x, splat (i8 32)
%sh = lshr <vscale x 16 x i8> %add, splat (i8 6)
ret <vscale x 16 x i8> %sh
}
define <vscale x 16 x i8> @urshr_8_wide_trunc(<vscale x 16 x i8> %x) {
; SVE-LABEL: urshr_8_wide_trunc:
; SVE: // %bb.0:
; SVE-NEXT: uunpkhi z1.h, z0.b
; SVE-NEXT: uunpklo z0.h, z0.b
; SVE-NEXT: add z0.h, z0.h, #32 // =0x20
; SVE-NEXT: add z1.h, z1.h, #32 // =0x20
; SVE-NEXT: lsr z1.h, z1.h, #6
; SVE-NEXT: lsr z0.h, z0.h, #6
; SVE-NEXT: uzp1 z0.b, z0.b, z1.b
; SVE-NEXT: ret
;
; SVE2-LABEL: urshr_8_wide_trunc:
; SVE2: // %bb.0:
; SVE2-NEXT: ptrue p0.b
; SVE2-NEXT: urshr z0.b, p0/m, z0.b, #6
; SVE2-NEXT: ret
%ext = zext <vscale x 16 x i8> %x to <vscale x 16 x i16>
%add = add nuw nsw <vscale x 16 x i16> %ext, splat (i16 32)
%sh = lshr <vscale x 16 x i16> %add, splat (i16 6)
%sht = trunc <vscale x 16 x i16> %sh to <vscale x 16 x i8>
ret <vscale x 16 x i8> %sht
}
define <vscale x 16 x i8> @urshr_8_wide_trunc_nomerge(<vscale x 16 x i16> %ext) {
; SVE-LABEL: urshr_8_wide_trunc_nomerge:
; SVE: // %bb.0:
; SVE-NEXT: add z0.h, z0.h, #256 // =0x100
; SVE-NEXT: add z1.h, z1.h, #256 // =0x100
; SVE-NEXT: lsr z1.h, z1.h, #9
; SVE-NEXT: lsr z0.h, z0.h, #9
; SVE-NEXT: uzp1 z0.b, z0.b, z1.b
; SVE-NEXT: ret
;
; SVE2-LABEL: urshr_8_wide_trunc_nomerge:
; SVE2: // %bb.0:
; SVE2-NEXT: ptrue p0.h
; SVE2-NEXT: urshr z1.h, p0/m, z1.h, #9
; SVE2-NEXT: urshr z0.h, p0/m, z0.h, #9
; SVE2-NEXT: uzp1 z0.b, z0.b, z1.b
; SVE2-NEXT: ret
%add = add nuw nsw <vscale x 16 x i16> %ext, splat (i16 256)
%sh = lshr <vscale x 16 x i16> %add, splat (i16 9)
%sht = trunc <vscale x 16 x i16> %sh to <vscale x 16 x i8>
ret <vscale x 16 x i8> %sht
}
define <vscale x 8 x i16> @urshr_i16(<vscale x 8 x i16> %x) {
; SVE-LABEL: urshr_i16:
; SVE: // %bb.0:
; SVE-NEXT: add z0.h, z0.h, #32 // =0x20
; SVE-NEXT: lsr z0.h, z0.h, #6
; SVE-NEXT: ret
;
; SVE2-LABEL: urshr_i16:
; SVE2: // %bb.0:
; SVE2-NEXT: ptrue p0.h
; SVE2-NEXT: urshr z0.h, p0/m, z0.h, #6
; SVE2-NEXT: ret
%add = add nuw nsw <vscale x 8 x i16> %x, splat (i16 32)
%sh = lshr <vscale x 8 x i16> %add, splat (i16 6)
ret <vscale x 8 x i16> %sh
}
define <vscale x 8 x i16> @urshr_16_wide_trunc(<vscale x 8 x i16> %x) {
; SVE-LABEL: urshr_16_wide_trunc:
; SVE: // %bb.0:
; SVE-NEXT: uunpkhi z1.s, z0.h
; SVE-NEXT: uunpklo z0.s, z0.h
; SVE-NEXT: add z0.s, z0.s, #32 // =0x20
; SVE-NEXT: add z1.s, z1.s, #32 // =0x20
; SVE-NEXT: lsr z1.s, z1.s, #6
; SVE-NEXT: lsr z0.s, z0.s, #6
; SVE-NEXT: uzp1 z0.h, z0.h, z1.h
; SVE-NEXT: ret
;
; SVE2-LABEL: urshr_16_wide_trunc:
; SVE2: // %bb.0:
; SVE2-NEXT: ptrue p0.h
; SVE2-NEXT: urshr z0.h, p0/m, z0.h, #6
; SVE2-NEXT: ret
%ext = zext <vscale x 8 x i16> %x to <vscale x 8 x i32>
%add = add nuw nsw <vscale x 8 x i32> %ext, splat (i32 32)
%sh = lshr <vscale x 8 x i32> %add, splat (i32 6)
%sht = trunc <vscale x 8 x i32> %sh to <vscale x 8 x i16>
ret <vscale x 8 x i16> %sht
}
define <vscale x 8 x i16> @urshr_16_wide_trunc_nomerge(<vscale x 8 x i32> %ext) {
; SVE-LABEL: urshr_16_wide_trunc_nomerge:
; SVE: // %bb.0:
; SVE-NEXT: mov z2.s, #0x10000
; SVE-NEXT: add z0.s, z0.s, z2.s
; SVE-NEXT: add z1.s, z1.s, z2.s
; SVE-NEXT: lsr z1.s, z1.s, #17
; SVE-NEXT: lsr z0.s, z0.s, #17
; SVE-NEXT: uzp1 z0.h, z0.h, z1.h
; SVE-NEXT: ret
;
; SVE2-LABEL: urshr_16_wide_trunc_nomerge:
; SVE2: // %bb.0:
; SVE2-NEXT: ptrue p0.s
; SVE2-NEXT: urshr z1.s, p0/m, z1.s, #17
; SVE2-NEXT: urshr z0.s, p0/m, z0.s, #17
; SVE2-NEXT: uzp1 z0.h, z0.h, z1.h
; SVE2-NEXT: ret
%add = add nuw nsw <vscale x 8 x i32> %ext, splat (i32 65536)
%sh = lshr <vscale x 8 x i32> %add, splat (i32 17)
%sht = trunc <vscale x 8 x i32> %sh to <vscale x 8 x i16>
ret <vscale x 8 x i16> %sht
}
define <vscale x 4 x i32> @urshr_i32(<vscale x 4 x i32> %x) {
; SVE-LABEL: urshr_i32:
; SVE: // %bb.0:
; SVE-NEXT: add z0.s, z0.s, #32 // =0x20
; SVE-NEXT: lsr z0.s, z0.s, #6
; SVE-NEXT: ret
;
; SVE2-LABEL: urshr_i32:
; SVE2: // %bb.0:
; SVE2-NEXT: ptrue p0.s
; SVE2-NEXT: urshr z0.s, p0/m, z0.s, #6
; SVE2-NEXT: ret
%add = add nuw nsw <vscale x 4 x i32> %x, splat (i32 32)
%sh = lshr <vscale x 4 x i32> %add, splat (i32 6)
ret <vscale x 4 x i32> %sh
}
define <vscale x 4 x i32> @urshr_32_wide_trunc(<vscale x 4 x i32> %x) {
; SVE-LABEL: urshr_32_wide_trunc:
; SVE: // %bb.0:
; SVE-NEXT: uunpkhi z1.d, z0.s
; SVE-NEXT: uunpklo z0.d, z0.s
; SVE-NEXT: add z0.d, z0.d, #32 // =0x20
; SVE-NEXT: add z1.d, z1.d, #32 // =0x20
; SVE-NEXT: lsr z1.d, z1.d, #6
; SVE-NEXT: lsr z0.d, z0.d, #6
; SVE-NEXT: uzp1 z0.s, z0.s, z1.s
; SVE-NEXT: ret
;
; SVE2-LABEL: urshr_32_wide_trunc:
; SVE2: // %bb.0:
; SVE2-NEXT: ptrue p0.s
; SVE2-NEXT: urshr z0.s, p0/m, z0.s, #6
; SVE2-NEXT: ret
%ext = zext <vscale x 4 x i32> %x to <vscale x 4 x i64>
%add = add nuw nsw <vscale x 4 x i64> %ext, splat (i64 32)
%sh = lshr <vscale x 4 x i64> %add, splat (i64 6)
%sht = trunc <vscale x 4 x i64> %sh to <vscale x 4 x i32>
ret <vscale x 4 x i32> %sht
}
define <vscale x 4 x i32> @urshr_32_wide_trunc_nomerge(<vscale x 4 x i64> %ext) {
; SVE-LABEL: urshr_32_wide_trunc_nomerge:
; SVE: // %bb.0:
; SVE-NEXT: mov z2.d, #0x100000000
; SVE-NEXT: add z0.d, z0.d, z2.d
; SVE-NEXT: add z1.d, z1.d, z2.d
; SVE-NEXT: lsr z1.d, z1.d, #33
; SVE-NEXT: lsr z0.d, z0.d, #33
; SVE-NEXT: uzp1 z0.s, z0.s, z1.s
; SVE-NEXT: ret
;
; SVE2-LABEL: urshr_32_wide_trunc_nomerge:
; SVE2: // %bb.0:
; SVE2-NEXT: ptrue p0.d
; SVE2-NEXT: urshr z1.d, p0/m, z1.d, #33
; SVE2-NEXT: urshr z0.d, p0/m, z0.d, #33
; SVE2-NEXT: uzp1 z0.s, z0.s, z1.s
; SVE2-NEXT: ret
%add = add nuw nsw <vscale x 4 x i64> %ext, splat (i64 4294967296)
%sh = lshr <vscale x 4 x i64> %add, splat (i64 33)
%sht = trunc <vscale x 4 x i64> %sh to <vscale x 4 x i32>
ret <vscale x 4 x i32> %sht
}
define <vscale x 2 x i64> @urshr_i64(<vscale x 2 x i64> %x) {
; SVE-LABEL: urshr_i64:
; SVE: // %bb.0:
; SVE-NEXT: add z0.d, z0.d, #32 // =0x20
; SVE-NEXT: lsr z0.d, z0.d, #6
; SVE-NEXT: ret
;
; SVE2-LABEL: urshr_i64:
; SVE2: // %bb.0:
; SVE2-NEXT: ptrue p0.d
; SVE2-NEXT: urshr z0.d, p0/m, z0.d, #6
; SVE2-NEXT: ret
%add = add nuw nsw <vscale x 2 x i64> %x, splat (i64 32)
%sh = lshr <vscale x 2 x i64> %add, splat (i64 6)
ret <vscale x 2 x i64> %sh
}
|
