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
|
/* memchr - find a character in a memory zone using base integer registers
Copyright (C) 2018 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
<http://www.gnu.org/licenses/>. */
#include <sysdep.h>
/* Assumptions:
*
* ARMv8-a, AArch64
* Use base integer registers.
*/
#ifndef MEMCHR
# define MEMCHR __memchr_nosimd
#endif
/* Arguments and results. */
#define srcin x0
#define chrin x1
#define cntin x2
#define result x0
#define repchr x1
#define tmp1 x2
#define tmp2 x3
#define tmp3 x4
#define tmp4 x5
#define src x6
#define srcend x7
#define srcend16 x8
#define anymore x9
#define zeroones x10
#define data1 x11
#define data2 x12
#define has_chr1 x13
#define has_chr2 x14
#define REP8_01 0x0101010101010101
#define REP8_7f 0x7f7f7f7f7f7f7f7f
ENTRY_ALIGN (MEMCHR, 6)
DELOUSE (0)
DELOUSE (2)
/* Do not dereference srcin if no bytes to compare. */
cbz cntin, L(none_chr)
/* Start address is 16-byte aligned or not? */
tst srcin, 15
bic src, srcin, 15
mov zeroones, REP8_01
and repchr, chrin, 255
/* Generate a qword integer as |c|c|c|c|c|c|c|c|. */
mul repchr, repchr, zeroones
add srcend, srcin, cntin
/*
* srcend16 is address of the block following the last block.
*
* [A block is 16-byte aligned and sized.]
*/
add srcend16, srcend, 15
bic srcend16, srcend16, 15
b.eq L(loop)
/* Load the first block containing start address. */
ldp data1, data2, [src], 16
lsl tmp1, srcin, 3
mov tmp2, ~0
#ifdef __AARCH64EB__
lsr tmp3, tmp2, tmp1
#else
lsl tmp3, tmp2, tmp1
#endif
/* Start address is in the first or the second qword? */
tst srcin, 8
/*
* Transform any byte in the block to zero using XOR operation,
* if that byte equals the char to search. In this way, searching
* the char becomes detecting zero in the resulting two qwords.
*/
eor data1, data1, repchr
eor data2, data2, repchr
/*
* Set those unused bytes(before start address) to 0xff, so
* that they will not hit any zero detection.
*/
orn tmp1, data1, tmp3
orn tmp2, data2, tmp3
csinv data1, tmp1, xzr, eq
csel data2, data2, tmp2, eq
/*
* When the first and last block are the same, there are two cases:
* o. Memory range to search is just in one block.
* ( start address - end address) < 0
*
* o. Memory range is so large that end address wrap-around.
* ( start address - end address) > 0
*/
cmp srcin, srcend
ccmp src, srcend16, 0, mi
csetm anymore, ne
b L(find_chr)
.p2align 4
L(loop):
ldp data1, data2, [src], 16
subs anymore, src, srcend16
/*
* Transform any byte in the block to zero using XOR operation,
* if that byte equals the char to search.
*/
eor data1, data1, repchr
eor data2, data2, repchr
L(find_chr):
/*
* Use the following integer test to find out if any byte in a
* qword is zero. If do not contain zero-valued byte, test result
* is zero.
*
* (qword - 0x0101010101010101) & ~(qword) & 0x8080808080808080
* =
* (qword - 0x0101010101010101) & ~(qword | 0x7f7f7f7f7f7f7f7f)
*
*/
sub tmp1, data1, zeroones
sub tmp2, data2, zeroones
orr tmp3, data1, REP8_7f
orr tmp4, data2, REP8_7f
bic has_chr1, tmp1, tmp3
bic has_chr2, tmp2, tmp4
orr tmp1, has_chr1, has_chr2
ccmp tmp1, 0, 0, ne
b.eq L(loop)
cbz has_chr1, 1f
sub result, src, 16
#ifdef __AARCH64EB__
rev data1, data1
#else
rev has_chr1, has_chr1
#endif
b L(done)
1: cbz has_chr2, L(none_chr)
sub result, src, 8
#ifdef __AARCH64EB__
rev data1, data2
#else
rev has_chr1, has_chr2
#endif
L(done):
#ifdef __AARCH64EB__
/*
* For big-endian, can not directly use has_chr1/has_chr2 because
* two qwords has been reversed after loading from memory.
* Thus, have to perform char detection on two qwords again, which
* should be byte-swapped this time.
*/
sub tmp1, data1, zeroones
orr tmp3, data1, REP8_7f
bic has_chr1, tmp1, tmp3
rev has_chr1, has_chr1
#endif
/*
* If the specified char is found in a qword, the corresponding
* byte of in has_chr has value of 1, while this is only true for
* the first occurrence, not other occurrences.
*/
cmp anymore, 0
clz tmp1, has_chr1
add result, result, tmp1, lsr 3
ccmp result, srcend, 8, eq /* NZCV = 8000 */
csel result, result, xzr, mi
ret
L(none_chr):
mov result, 0
ret
END (MEMCHR)
libc_hidden_builtin_def (MEMCHR)
|
