1*43a90889SApple OSS Distributions /*-
2*43a90889SApple OSS Distributions * Copyright (c) 2009 Xin LI <[email protected]>
3*43a90889SApple OSS Distributions * All rights reserved.
4*43a90889SApple OSS Distributions *
5*43a90889SApple OSS Distributions * Redistribution and use in source and binary forms, with or without
6*43a90889SApple OSS Distributions * modification, are permitted provided that the following conditions
7*43a90889SApple OSS Distributions * are met:
8*43a90889SApple OSS Distributions * 1. Redistributions of source code must retain the above copyright
9*43a90889SApple OSS Distributions * notice, this list of conditions and the following disclaimer.
10*43a90889SApple OSS Distributions * 2. Redistributions in binary form must reproduce the above copyright
11*43a90889SApple OSS Distributions * notice, this list of conditions and the following disclaimer in the
12*43a90889SApple OSS Distributions * documentation and/or other materials provided with the distribution.
13*43a90889SApple OSS Distributions *
14*43a90889SApple OSS Distributions * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15*43a90889SApple OSS Distributions * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16*43a90889SApple OSS Distributions * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17*43a90889SApple OSS Distributions * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18*43a90889SApple OSS Distributions * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19*43a90889SApple OSS Distributions * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20*43a90889SApple OSS Distributions * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21*43a90889SApple OSS Distributions * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22*43a90889SApple OSS Distributions * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23*43a90889SApple OSS Distributions * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24*43a90889SApple OSS Distributions * SUCH DAMAGE.
25*43a90889SApple OSS Distributions */
26*43a90889SApple OSS Distributions
27*43a90889SApple OSS Distributions #include "string.h"
28*43a90889SApple OSS Distributions #include <limits.h>
29*43a90889SApple OSS Distributions
30*43a90889SApple OSS Distributions /*
31*43a90889SApple OSS Distributions * Portable strlen() for 32-bit and 64-bit systems.
32*43a90889SApple OSS Distributions *
33*43a90889SApple OSS Distributions * Rationale: it is generally much more efficient to do word length
34*43a90889SApple OSS Distributions * operations and avoid branches on modern computer systems, as
35*43a90889SApple OSS Distributions * compared to byte-length operations with a lot of branches.
36*43a90889SApple OSS Distributions *
37*43a90889SApple OSS Distributions * The expression:
38*43a90889SApple OSS Distributions *
39*43a90889SApple OSS Distributions * ((x - 0x01....01) & ~x & 0x80....80)
40*43a90889SApple OSS Distributions *
41*43a90889SApple OSS Distributions * would evaluate to a non-zero value iff any of the bytes in the
42*43a90889SApple OSS Distributions * original word is zero. However, we can further reduce ~1/3 of
43*43a90889SApple OSS Distributions * time if we consider that strlen() usually operate on 7-bit ASCII
44*43a90889SApple OSS Distributions * by employing the following expression, which allows false positive
45*43a90889SApple OSS Distributions * when high bit of 1 and use the tail case to catch these case:
46*43a90889SApple OSS Distributions *
47*43a90889SApple OSS Distributions * ((x - 0x01....01) & 0x80....80)
48*43a90889SApple OSS Distributions *
49*43a90889SApple OSS Distributions * This is more than 5.2 times as fast as the raw implementation on
50*43a90889SApple OSS Distributions * Intel T7300 under long mode for strings longer than word length.
51*43a90889SApple OSS Distributions */
52*43a90889SApple OSS Distributions
53*43a90889SApple OSS Distributions /* Magic numbers for the algorithm */
54*43a90889SApple OSS Distributions #if LONG_BIT == 32
55*43a90889SApple OSS Distributions static const unsigned long mask01 = 0x01010101;
56*43a90889SApple OSS Distributions static const unsigned long mask80 = 0x80808080;
57*43a90889SApple OSS Distributions #elif LONG_BIT == 64
58*43a90889SApple OSS Distributions static const unsigned long mask01 = 0x0101010101010101;
59*43a90889SApple OSS Distributions static const unsigned long mask80 = 0x8080808080808080;
60*43a90889SApple OSS Distributions #else
61*43a90889SApple OSS Distributions #error Unsupported word size
62*43a90889SApple OSS Distributions #endif
63*43a90889SApple OSS Distributions
64*43a90889SApple OSS Distributions #define LONGPTR_MASK (sizeof(long) - 1)
65*43a90889SApple OSS Distributions
66*43a90889SApple OSS Distributions /*
67*43a90889SApple OSS Distributions * Helper macro to return string length if we caught the zero
68*43a90889SApple OSS Distributions * byte.
69*43a90889SApple OSS Distributions */
70*43a90889SApple OSS Distributions #define testbyte(x) \
71*43a90889SApple OSS Distributions do { \
72*43a90889SApple OSS Distributions if (p[x] == '\0') \
73*43a90889SApple OSS Distributions return (p - str + x); \
74*43a90889SApple OSS Distributions } while (0)
75*43a90889SApple OSS Distributions
76*43a90889SApple OSS Distributions __attribute__((visibility("hidden")))
77*43a90889SApple OSS Distributions size_t
_libkernel_strlen(const char * str)78*43a90889SApple OSS Distributions _libkernel_strlen(const char *str)
79*43a90889SApple OSS Distributions {
80*43a90889SApple OSS Distributions const char *p;
81*43a90889SApple OSS Distributions const unsigned long *lp;
82*43a90889SApple OSS Distributions
83*43a90889SApple OSS Distributions /* Skip the first few bytes until we have an aligned p */
84*43a90889SApple OSS Distributions for (p = str; (uintptr_t)p & LONGPTR_MASK; p++) {
85*43a90889SApple OSS Distributions if (*p == '\0') {
86*43a90889SApple OSS Distributions return p - str;
87*43a90889SApple OSS Distributions }
88*43a90889SApple OSS Distributions }
89*43a90889SApple OSS Distributions
90*43a90889SApple OSS Distributions /* Scan the rest of the string using word sized operation */
91*43a90889SApple OSS Distributions for (lp = (const unsigned long *)p;; lp++) {
92*43a90889SApple OSS Distributions if ((*lp - mask01) & mask80) {
93*43a90889SApple OSS Distributions p = (const char *)(lp);
94*43a90889SApple OSS Distributions testbyte(0);
95*43a90889SApple OSS Distributions testbyte(1);
96*43a90889SApple OSS Distributions testbyte(2);
97*43a90889SApple OSS Distributions testbyte(3);
98*43a90889SApple OSS Distributions #if (LONG_BIT >= 64)
99*43a90889SApple OSS Distributions testbyte(4);
100*43a90889SApple OSS Distributions testbyte(5);
101*43a90889SApple OSS Distributions testbyte(6);
102*43a90889SApple OSS Distributions testbyte(7);
103*43a90889SApple OSS Distributions #endif
104*43a90889SApple OSS Distributions }
105*43a90889SApple OSS Distributions }
106*43a90889SApple OSS Distributions
107*43a90889SApple OSS Distributions /* NOTREACHED */
108*43a90889SApple OSS Distributions return 0;
109*43a90889SApple OSS Distributions }
110