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
|
/* Optimized memcpy for SVE.
Copyright (C) 2021-2022 Free Software Foundation, Inc.
This file is part of the GNU C Library.
The GNU C Library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
The GNU C Library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with the GNU C Library. If not, see
<https://www.gnu.org/licenses/>. */
#include <sysdep.h>
/* Assumptions:
*
* ARMv8-a, AArch64, Advanced SIMD, SVE, unaligned accesses.
*
*/
#define dstin x0
#define src x1
#define count x2
#define dst x3
#define srcend x4
#define dstend x5
#define tmp1 x6
#define vlen x6
#define A_q q0
#define B_q q1
#define C_q q2
#define D_q q3
#define E_q q4
#define F_q q5
#define G_q q6
#define H_q q7
/* This implementation supports both memcpy and memmove and shares most code.
It uses unaligned accesses and branchless sequences to keep the code small,
simple and improve performance.
Copies are split into 3 main cases: small copies of up to 32 bytes, medium
copies of up to 128 bytes, and large copies. The overhead of the overlap
check in memmove is negligible since it is only required for large copies.
Large copies use a software pipelined loop processing 64 bytes per iteration.
The source pointer is 16-byte aligned to minimize unaligned accesses.
The loop tail is handled by always copying 64 bytes from the end.
*/
#if HAVE_AARCH64_SVE_ASM
.arch armv8.2-a+sve
ENTRY (__memcpy_sve)
PTR_ARG (0)
PTR_ARG (1)
SIZE_ARG (2)
cmp count, 128
b.hi L(copy_long)
cmp count, 32
b.hi L(copy32_128)
whilelo p0.b, xzr, count
cntb vlen
tbnz vlen, 4, L(vlen128)
ld1b z0.b, p0/z, [src]
st1b z0.b, p0, [dstin]
ret
/* Medium copies: 33..128 bytes. */
L(copy32_128):
add srcend, src, count
add dstend, dstin, count
ldp A_q, B_q, [src]
ldp C_q, D_q, [srcend, -32]
cmp count, 64
b.hi L(copy128)
stp A_q, B_q, [dstin]
stp C_q, D_q, [dstend, -32]
ret
/* Copy 65..128 bytes. */
L(copy128):
ldp E_q, F_q, [src, 32]
cmp count, 96
b.ls L(copy96)
ldp G_q, H_q, [srcend, -64]
stp G_q, H_q, [dstend, -64]
L(copy96):
stp A_q, B_q, [dstin]
stp E_q, F_q, [dstin, 32]
stp C_q, D_q, [dstend, -32]
ret
L(vlen128):
whilelo p1.b, vlen, count
ld1b z0.b, p0/z, [src, 0, mul vl]
ld1b z1.b, p1/z, [src, 1, mul vl]
st1b z0.b, p0, [dstin, 0, mul vl]
st1b z1.b, p1, [dstin, 1, mul vl]
ret
.p2align 4
/* Copy more than 128 bytes. */
L(copy_long):
add srcend, src, count
add dstend, dstin, count
/* Copy 16 bytes and then align src to 16-byte alignment. */
ldr D_q, [src]
and tmp1, src, 15
bic src, src, 15
sub dst, dstin, tmp1
add count, count, tmp1 /* Count is now 16 too large. */
ldp A_q, B_q, [src, 16]
str D_q, [dstin]
ldp C_q, D_q, [src, 48]
subs count, count, 128 + 16 /* Test and readjust count. */
b.ls L(copy64_from_end)
L(loop64):
stp A_q, B_q, [dst, 16]
ldp A_q, B_q, [src, 80]
stp C_q, D_q, [dst, 48]
ldp C_q, D_q, [src, 112]
add src, src, 64
add dst, dst, 64
subs count, count, 64
b.hi L(loop64)
/* Write the last iteration and copy 64 bytes from the end. */
L(copy64_from_end):
ldp E_q, F_q, [srcend, -64]
stp A_q, B_q, [dst, 16]
ldp A_q, B_q, [srcend, -32]
stp C_q, D_q, [dst, 48]
stp E_q, F_q, [dstend, -64]
stp A_q, B_q, [dstend, -32]
ret
END (__memcpy_sve)
libc_hidden_builtin_def (__memcpy_sve)
ENTRY (__memmove_sve)
PTR_ARG (0)
PTR_ARG (1)
SIZE_ARG (2)
cmp count, 128
b.hi L(move_long)
cmp count, 32
b.hi L(copy32_128)
whilelo p0.b, xzr, count
cntb vlen
tbnz vlen, 4, L(vlen128)
ld1b z0.b, p0/z, [src]
st1b z0.b, p0, [dstin]
ret
.p2align 4
L(move_long):
add srcend, src, count
add dstend, dstin, count
/* Only use backward copy if there is an overlap. */
sub tmp1, dstin, src
cbz tmp1, L(return)
cmp tmp1, count
b.hs L(copy_long)
/* Large backwards copy for overlapping copies.
Copy 16 bytes and then align srcend to 16-byte alignment. */
ldr D_q, [srcend, -16]
and tmp1, srcend, 15
bic srcend, srcend, 15
sub count, count, tmp1
ldp A_q, B_q, [srcend, -32]
str D_q, [dstend, -16]
ldp C_q, D_q, [srcend, -64]
sub dstend, dstend, tmp1
subs count, count, 128
b.ls L(copy64_from_start)
L(loop64_backwards):
str B_q, [dstend, -16]
str A_q, [dstend, -32]
ldp A_q, B_q, [srcend, -96]
str D_q, [dstend, -48]
str C_q, [dstend, -64]!
ldp C_q, D_q, [srcend, -128]
sub srcend, srcend, 64
subs count, count, 64
b.hi L(loop64_backwards)
/* Write the last iteration and copy 64 bytes from the start. */
L(copy64_from_start):
ldp E_q, F_q, [src, 32]
stp A_q, B_q, [dstend, -32]
ldp A_q, B_q, [src]
stp C_q, D_q, [dstend, -64]
stp E_q, F_q, [dstin, 32]
stp A_q, B_q, [dstin]
L(return):
ret
END (__memmove_sve)
libc_hidden_builtin_def (__memmove_sve)
#endif
|
