]> git.ipfire.org Git - thirdparty/glibc.git/blob - sysdeps/aarch64/strnlen.S
Update copyright dates with scripts/update-copyrights.
[thirdparty/glibc.git] / sysdeps / aarch64 / strnlen.S
1 /* strnlen - calculate the length of a string with limit.
2
3 Copyright (C) 2013-2016 Free Software Foundation, Inc.
4
5 This file is part of the GNU C Library.
6
7 The GNU C Library is free software; you can redistribute it and/or
8 modify it under the terms of the GNU Lesser General Public
9 License as published by the Free Software Foundation; either
10 version 2.1 of the License, or (at your option) any later version.
11
12 The GNU C Library is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 Lesser General Public License for more details.
16
17 You should have received a copy of the GNU Lesser General Public
18 License along with the GNU C Library. If not, see
19 <http://www.gnu.org/licenses/>. */
20
21 #include <sysdep.h>
22
23 /* Assumptions:
24 *
25 * ARMv8-a, AArch64
26 */
27
28 /* Arguments and results. */
29 #define srcin x0
30 #define len x0
31 #define limit x1
32
33 /* Locals and temporaries. */
34 #define src x2
35 #define data1 x3
36 #define data2 x4
37 #define data2a x5
38 #define has_nul1 x6
39 #define has_nul2 x7
40 #define tmp1 x8
41 #define tmp2 x9
42 #define tmp3 x10
43 #define tmp4 x11
44 #define zeroones x12
45 #define pos x13
46 #define limit_wd x14
47
48 #define REP8_01 0x0101010101010101
49 #define REP8_7f 0x7f7f7f7f7f7f7f7f
50 #define REP8_80 0x8080808080808080
51
52 ENTRY_ALIGN_AND_PAD (__strnlen, 6, 9)
53 cbz limit, L(hit_limit)
54 mov zeroones, #REP8_01
55 bic src, srcin, #15
56 ands tmp1, srcin, #15
57 b.ne L(misaligned)
58 /* Calculate the number of full and partial words -1. */
59 sub limit_wd, limit, #1 /* Limit != 0, so no underflow. */
60 lsr limit_wd, limit_wd, #4 /* Convert to Qwords. */
61
62 /* NUL detection works on the principle that (X - 1) & (~X) & 0x80
63 (=> (X - 1) & ~(X | 0x7f)) is non-zero iff a byte is zero, and
64 can be done in parallel across the entire word. */
65 /* The inner loop deals with two Dwords at a time. This has a
66 slightly higher start-up cost, but we should win quite quickly,
67 especially on cores with a high number of issue slots per
68 cycle, as we get much better parallelism out of the operations. */
69
70 /* Start of critial section -- keep to one 64Byte cache line. */
71 L(loop):
72 ldp data1, data2, [src], #16
73 L(realigned):
74 sub tmp1, data1, zeroones
75 orr tmp2, data1, #REP8_7f
76 sub tmp3, data2, zeroones
77 orr tmp4, data2, #REP8_7f
78 bic has_nul1, tmp1, tmp2
79 bic has_nul2, tmp3, tmp4
80 subs limit_wd, limit_wd, #1
81 orr tmp1, has_nul1, has_nul2
82 ccmp tmp1, #0, #0, pl /* NZCV = 0000 */
83 b.eq L(loop)
84 /* End of critical section -- keep to one 64Byte cache line. */
85
86 orr tmp1, has_nul1, has_nul2
87 cbz tmp1, L(hit_limit) /* No null in final Qword. */
88
89 /* We know there's a null in the final Qword. The easiest thing
90 to do now is work out the length of the string and return
91 MIN (len, limit). */
92
93 sub len, src, srcin
94 cbz has_nul1, L(nul_in_data2)
95 #ifdef __AARCH64EB__
96 mov data2, data1
97 #endif
98 sub len, len, #8
99 mov has_nul2, has_nul1
100 L(nul_in_data2):
101 #ifdef __AARCH64EB__
102 /* For big-endian, carry propagation (if the final byte in the
103 string is 0x01) means we cannot use has_nul directly. The
104 easiest way to get the correct byte is to byte-swap the data
105 and calculate the syndrome a second time. */
106 rev data2, data2
107 sub tmp1, data2, zeroones
108 orr tmp2, data2, #REP8_7f
109 bic has_nul2, tmp1, tmp2
110 #endif
111 sub len, len, #8
112 rev has_nul2, has_nul2
113 clz pos, has_nul2
114 add len, len, pos, lsr #3 /* Bits to bytes. */
115 cmp len, limit
116 csel len, len, limit, ls /* Return the lower value. */
117 RET
118
119 L(misaligned):
120 /* Deal with a partial first word.
121 We're doing two things in parallel here;
122 1) Calculate the number of words (but avoiding overflow if
123 limit is near ULONG_MAX) - to do this we need to work out
124 limit + tmp1 - 1 as a 65-bit value before shifting it;
125 2) Load and mask the initial data words - we force the bytes
126 before the ones we are interested in to 0xff - this ensures
127 early bytes will not hit any zero detection. */
128 sub limit_wd, limit, #1
129 neg tmp4, tmp1
130 cmp tmp1, #8
131
132 and tmp3, limit_wd, #15
133 lsr limit_wd, limit_wd, #4
134 mov tmp2, #~0
135
136 ldp data1, data2, [src], #16
137 lsl tmp4, tmp4, #3 /* Bytes beyond alignment -> bits. */
138 add tmp3, tmp3, tmp1
139
140 #ifdef __AARCH64EB__
141 /* Big-endian. Early bytes are at MSB. */
142 lsl tmp2, tmp2, tmp4 /* Shift (tmp1 & 63). */
143 #else
144 /* Little-endian. Early bytes are at LSB. */
145 lsr tmp2, tmp2, tmp4 /* Shift (tmp1 & 63). */
146 #endif
147 add limit_wd, limit_wd, tmp3, lsr #4
148
149 orr data1, data1, tmp2
150 orr data2a, data2, tmp2
151
152 csinv data1, data1, xzr, le
153 csel data2, data2, data2a, le
154 b L(realigned)
155
156 L(hit_limit):
157 mov len, limit
158 RET
159 END (__strnlen)
160 libc_hidden_def (__strnlen)
161 weak_alias (__strnlen, strnlen)
162 libc_hidden_def (strnlen)