1 /*
2 * This string-include defines all string functions as inline
3 * functions. Use gcc. It also assumes ds=es=data space, this should be
4 * normal. Most of the string-functions are rather heavily hand-optimized,
5 * see especially strtok,strstr,str[c]spn. They should work, but are not
6 * very easy to understand. Everything is done entirely within the register
7 * set, making the functions fast and clean. String instructions have been
8 * used through-out, making for "slightly" unclear code :-)
9 *
10 * NO Copyright (C) 1991, 1992 Linus Torvalds,
11 * consider these trivial functions to be PD.
12 */
13
14 /*
15 * Copyright (C) 2000-2005 Erik Andersen <andersen@uclibc.org>
16 *
17 * Licensed under the LGPL v2.1, see the file COPYING.LIB in this tarball.
18 */
19
20 /*
21 * Modified for uClibc by Erik Andersen <andersen@codepoet.org>
22 * These make no attempt to use nifty things like mmx/3dnow/etc.
23 * These are not inline, and will therefore not be as fast as
24 * modifying the headers to use inlines (and cannot therefore
25 * do tricky things when dealing with const memory). But they
26 * should (I hope!) be faster than their generic equivalents....
27 *
28 * More importantly, these should provide a good example for
29 * others to follow when adding arch specific optimizations.
30 * -Erik
31 */
32
33 #include <string.h>
34
35 #undef memcpy
memcpy(void * to,const void * from,size_t n)36 void *memcpy(void * to, const void * from, size_t n)
37 {
38 int d0, d1, d2;
39 __asm__ __volatile__(
40 " rep; movsl\n"
41 " movl %4, %%ecx\n"
42 " andl $3, %%ecx\n"
43 /* jz is optional. avoids "rep; movsb" with ecx == 0,
44 * but adds a branch, which is currently (2008) faster */
45 " jz 1f\n"
46 " rep; movsb\n"
47 "1:\n"
48 : "=&c" (d0), "=&D" (d1), "=&S" (d2)
49 : "0" (n / 4), "g" (n), "1" ((long)to), "2" ((long)from)
50 : "memory"
51 );
52 return to;
53 }
54 libc_hidden_def(memcpy)
55