blob: 987b68b9ce4474bb9cfd44e77ef61d3a871d40f8 [file] [log] [blame]
zhichang.yuan0a42cb02014-04-28 13:11:34 +08001/*
2 * Copyright (C) 2013 ARM Ltd.
3 * Copyright (C) 2013 Linaro.
4 *
5 * This code is based on glibc cortex strings work originally authored by Linaro
6 * and re-licensed under GPLv2 for the Linux kernel. The original code can
7 * be found @
8 *
9 * http://bazaar.launchpad.net/~linaro-toolchain-dev/cortex-strings/trunk/
10 * files/head:/src/aarch64/
11 *
12 * This program is free software; you can redistribute it and/or modify
13 * it under the terms of the GNU General Public License version 2 as
14 * published by the Free Software Foundation.
15 *
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 * GNU General Public License for more details.
20 *
21 * You should have received a copy of the GNU General Public License
22 * along with this program. If not, see <http://www.gnu.org/licenses/>.
23 */
24
25#include <linux/linkage.h>
26#include <asm/assembler.h>
27
28/*
29 * calculate the length of a string
30 *
31 * Parameters:
32 * x0 - const string pointer
33 * Returns:
34 * x0 - the return length of specific string
35 */
36
37/* Arguments and results. */
38srcin .req x0
39len .req x0
40
41/* Locals and temporaries. */
42src .req x1
43data1 .req x2
44data2 .req x3
45data2a .req x4
46has_nul1 .req x5
47has_nul2 .req x6
48tmp1 .req x7
49tmp2 .req x8
50tmp3 .req x9
51tmp4 .req x10
52zeroones .req x11
53pos .req x12
54
55#define REP8_01 0x0101010101010101
56#define REP8_7f 0x7f7f7f7f7f7f7f7f
57#define REP8_80 0x8080808080808080
58
59ENTRY(strlen)
60 mov zeroones, #REP8_01
61 bic src, srcin, #15
62 ands tmp1, srcin, #15
63 b.ne .Lmisaligned
64 /*
65 * NUL detection works on the principle that (X - 1) & (~X) & 0x80
66 * (=> (X - 1) & ~(X | 0x7f)) is non-zero iff a byte is zero, and
67 * can be done in parallel across the entire word.
68 */
69 /*
70 * The inner loop deals with two Dwords at a time. This has a
71 * slightly higher start-up cost, but we should win quite quickly,
72 * especially on cores with a high number of issue slots per
73 * cycle, as we get much better parallelism out of the operations.
74 */
75.Lloop:
76 ldp data1, data2, [src], #16
77.Lrealigned:
78 sub tmp1, data1, zeroones
79 orr tmp2, data1, #REP8_7f
80 sub tmp3, data2, zeroones
81 orr tmp4, data2, #REP8_7f
82 bic has_nul1, tmp1, tmp2
83 bics has_nul2, tmp3, tmp4
84 ccmp has_nul1, #0, #0, eq /* NZCV = 0000 */
85 b.eq .Lloop
86
87 sub len, src, srcin
88 cbz has_nul1, .Lnul_in_data2
89CPU_BE( mov data2, data1 ) /*prepare data to re-calculate the syndrome*/
90 sub len, len, #8
91 mov has_nul2, has_nul1
92.Lnul_in_data2:
93 /*
94 * For big-endian, carry propagation (if the final byte in the
95 * string is 0x01) means we cannot use has_nul directly. The
96 * easiest way to get the correct byte is to byte-swap the data
97 * and calculate the syndrome a second time.
98 */
99CPU_BE( rev data2, data2 )
100CPU_BE( sub tmp1, data2, zeroones )
101CPU_BE( orr tmp2, data2, #REP8_7f )
102CPU_BE( bic has_nul2, tmp1, tmp2 )
103
104 sub len, len, #8
105 rev has_nul2, has_nul2
106 clz pos, has_nul2
107 add len, len, pos, lsr #3 /* Bits to bytes. */
108 ret
109
110.Lmisaligned:
111 cmp tmp1, #8
112 neg tmp1, tmp1
113 ldp data1, data2, [src], #16
114 lsl tmp1, tmp1, #3 /* Bytes beyond alignment -> bits. */
115 mov tmp2, #~0
116 /* Big-endian. Early bytes are at MSB. */
117CPU_BE( lsl tmp2, tmp2, tmp1 ) /* Shift (tmp1 & 63). */
118 /* Little-endian. Early bytes are at LSB. */
119CPU_LE( lsr tmp2, tmp2, tmp1 ) /* Shift (tmp1 & 63). */
120
121 orr data1, data1, tmp2
122 orr data2a, data2, tmp2
123 csinv data1, data1, xzr, le
124 csel data2, data2, data2a, le
125 b .Lrealigned
126ENDPROC(strlen)