]>
Commit | Line | Data |
---|---|---|
cc3fa755 | 1 | /* Highly optimized version for i586. |
b168057a | 2 | Copyright (C) 1997-2015 Free Software Foundation, Inc. |
cc3fa755 UD |
3 | This file is part of the GNU C Library. |
4 | Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997. | |
5 | ||
6 | The GNU C Library is free software; you can redistribute it and/or | |
41bdb6e2 AJ |
7 | modify it under the terms of the GNU Lesser General Public |
8 | License as published by the Free Software Foundation; either | |
9 | version 2.1 of the License, or (at your option) any later version. | |
cc3fa755 UD |
10 | |
11 | The GNU C Library is distributed in the hope that it will be useful, | |
12 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
41bdb6e2 | 14 | Lesser General Public License for more details. |
cc3fa755 | 15 | |
41bdb6e2 | 16 | You should have received a copy of the GNU Lesser General Public |
59ba27a6 PE |
17 | License along with the GNU C Library; if not, see |
18 | <http://www.gnu.org/licenses/>. */ | |
cc3fa755 UD |
19 | |
20 | #include <sysdep.h> | |
21 | #include "asm-syntax.h" | |
22 | ||
7475d016 UD |
23 | /* BEWARE: `#ifdef memcpy' means that memcpy is redefined as `mempcpy', |
24 | and the return value is the byte after the last one copied in | |
25 | the destination. */ | |
26 | #define MEMPCPY_P (defined memcpy) | |
cc3fa755 | 27 | |
2366713d | 28 | #define PARMS 4+8 /* space for 2 saved regs */ |
7475d016 | 29 | #define RTN PARMS |
2366713d JM |
30 | #define DEST RTN |
31 | #define SRC DEST+4 | |
32 | #define LEN SRC+4 | |
cc3fa755 UD |
33 | |
34 | .text | |
4f41c682 | 35 | #if defined PIC && IS_IN (libc) |
9bdfff60 UD |
36 | ENTRY (__memcpy_chk) |
37 | movl 12(%esp), %eax | |
38 | cmpl %eax, 16(%esp) | |
39 | jb HIDDEN_JUMPTARGET (__chk_fail) | |
40 | END (__memcpy_chk) | |
41 | #endif | |
2366713d | 42 | ENTRY (memcpy) |
3f02f778 | 43 | |
cc3fa755 | 44 | pushl %edi |
1ad9da69 | 45 | cfi_adjust_cfa_offset (4) |
cc3fa755 | 46 | pushl %esi |
1ad9da69 | 47 | cfi_adjust_cfa_offset (4) |
cc3fa755 | 48 | |
3f02f778 | 49 | movl DEST(%esp), %edi |
1ad9da69 | 50 | cfi_rel_offset (edi, 4) |
3f02f778 | 51 | movl SRC(%esp), %esi |
fee732e5 | 52 | cfi_rel_offset (esi, 0) |
3f02f778 | 53 | movl LEN(%esp), %ecx |
cc3fa755 UD |
54 | movl %edi, %eax |
55 | ||
56 | /* We need this in any case. */ | |
57 | cld | |
58 | ||
59 | /* Cutoff for the big loop is a size of 32 bytes since otherwise | |
60 | the loop will never be entered. */ | |
61 | cmpl $32, %ecx | |
62 | jbe L(1) | |
63 | ||
64 | negl %eax | |
65 | andl $3, %eax | |
66 | subl %eax, %ecx | |
67 | xchgl %eax, %ecx | |
68 | ||
69 | rep; movsb | |
70 | ||
71 | movl %eax, %ecx | |
72 | subl $32, %ecx | |
73 | js L(2) | |
74 | ||
75 | /* Read ahead to make sure we write in the cache since the stupid | |
76 | i586 designers haven't implemented read-on-write-miss. */ | |
77 | movl (%edi), %eax | |
78 | L(3): movl 28(%edi), %edx | |
79 | ||
80 | /* Now correct the loop counter. Please note that in the following | |
81 | code the flags are not changed anymore. */ | |
82 | subl $32, %ecx | |
83 | ||
84 | movl (%esi), %eax | |
85 | movl 4(%esi), %edx | |
86 | movl %eax, (%edi) | |
87 | movl %edx, 4(%edi) | |
88 | movl 8(%esi), %eax | |
89 | movl 12(%esi), %edx | |
90 | movl %eax, 8(%edi) | |
91 | movl %edx, 12(%edi) | |
92 | movl 16(%esi), %eax | |
93 | movl 20(%esi), %edx | |
94 | movl %eax, 16(%edi) | |
95 | movl %edx, 20(%edi) | |
96 | movl 24(%esi), %eax | |
97 | movl 28(%esi), %edx | |
98 | movl %eax, 24(%edi) | |
99 | movl %edx, 28(%edi) | |
100 | ||
101 | leal 32(%esi), %esi | |
102 | leal 32(%edi), %edi | |
103 | ||
104 | jns L(3) | |
105 | ||
106 | /* Correct extra loop counter modification. */ | |
107 | L(2): addl $32, %ecx | |
7475d016 | 108 | #if !MEMPCPY_P |
3f02f778 | 109 | movl DEST(%esp), %eax |
cc3fa755 UD |
110 | #endif |
111 | ||
112 | L(1): rep; movsb | |
113 | ||
7475d016 | 114 | #if MEMPCPY_P |
cc3fa755 UD |
115 | movl %edi, %eax |
116 | #endif | |
117 | ||
118 | popl %esi | |
1ad9da69 UD |
119 | cfi_adjust_cfa_offset (-4) |
120 | cfi_restore (esi) | |
cc3fa755 | 121 | popl %edi |
1ad9da69 UD |
122 | cfi_adjust_cfa_offset (-4) |
123 | cfi_restore (edi) | |
cc3fa755 | 124 | |
2366713d JM |
125 | ret |
126 | END (memcpy) | |
44809672 | 127 | #if !MEMPCPY_P |
85dd1003 | 128 | libc_hidden_builtin_def (memcpy) |
44809672 | 129 | #endif |