]>
Commit | Line | Data |
---|---|---|
53f770e0 | 1 | /* memset/bzero -- set memory area to CH/0 |
cc3fa755 | 2 | Highly optimized version for ix86, x>=5. |
b168057a | 3 | Copyright (C) 1996-2015 Free Software Foundation, Inc. |
5929563f UD |
4 | This file is part of the GNU C Library. |
5 | Contributed by Torbjorn Granlund, <tege@matematik.su.se> | |
6 | ||
7 | The GNU C Library is free software; you can redistribute it and/or | |
41bdb6e2 AJ |
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. | |
5929563f UD |
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 | |
41bdb6e2 | 15 | Lesser General Public License for more details. |
5929563f | 16 | |
41bdb6e2 | 17 | You should have received a copy of the GNU Lesser General Public |
59ba27a6 PE |
18 | License along with the GNU C Library; if not, see |
19 | <http://www.gnu.org/licenses/>. */ | |
53f770e0 RM |
20 | |
21 | #include <sysdep.h> | |
5929563f | 22 | #include "asm-syntax.h" |
53f770e0 | 23 | |
3f02f778 GM |
24 | /* BEWARE: `#ifdef memset' means that memset is redefined as `bzero' */ |
25 | #define BZERO_P (defined memset) | |
53f770e0 | 26 | |
2366713d | 27 | #define PARMS 4+4 /* space for 1 saved reg */ |
3f02f778 | 28 | #define RTN PARMS |
2366713d | 29 | #define DEST RTN |
3f02f778 | 30 | #if BZERO_P |
2366713d | 31 | # define LEN DEST+4 |
3f02f778 | 32 | #else |
2366713d | 33 | # define CHR DEST+4 |
3f02f778 GM |
34 | # define LEN CHR+4 |
35 | #endif | |
53f770e0 RM |
36 | |
37 | .text | |
4f41c682 | 38 | #if defined PIC && IS_IN (libc) && !BZERO_P |
9bdfff60 UD |
39 | ENTRY (__memset_chk) |
40 | movl 12(%esp), %eax | |
41 | cmpl %eax, 16(%esp) | |
42 | jb HIDDEN_JUMPTARGET (__chk_fail) | |
43 | END (__memset_chk) | |
44 | #endif | |
2366713d | 45 | ENTRY (memset) |
3f02f778 | 46 | |
53f770e0 | 47 | pushl %edi |
1ad9da69 | 48 | cfi_adjust_cfa_offset (4) |
53f770e0 | 49 | |
3f02f778 | 50 | movl DEST(%esp), %edi |
1ad9da69 | 51 | cfi_rel_offset (edi, 0) |
3f02f778 GM |
52 | movl LEN(%esp), %edx |
53 | #if BZERO_P | |
53f770e0 | 54 | xorl %eax, %eax /* we fill with 0 */ |
53f770e0 | 55 | #else |
3f02f778 | 56 | movb CHR(%esp), %al |
53f770e0 RM |
57 | movb %al, %ah |
58 | movl %eax, %ecx | |
59 | shll $16, %eax | |
60 | movw %cx, %ax | |
61 | #endif | |
62 | cld | |
63 | ||
64 | /* If less than 36 bytes to write, skip tricky code (it wouldn't work). */ | |
65 | cmpl $36, %edx | |
66 | movl %edx, %ecx /* needed when branch is taken! */ | |
5929563f | 67 | jl L(2) |
53f770e0 RM |
68 | |
69 | /* First write 0-3 bytes to make the pointer 32-bit aligned. */ | |
70 | movl %edi, %ecx /* Copy ptr to ecx... */ | |
71 | negl %ecx /* ...and negate that and... */ | |
72 | andl $3, %ecx /* ...mask to get byte count. */ | |
73 | subl %ecx, %edx /* adjust global byte count */ | |
74 | rep | |
75 | stosb | |
76 | ||
77 | subl $32, %edx /* offset count for unrolled loop */ | |
78 | movl (%edi), %ecx /* Fetch destination cache line */ | |
79 | ||
80 | .align 2, 0x90 /* supply 0x90 for broken assemblers */ | |
5929563f | 81 | L(1): movl 28(%edi), %ecx /* allocate cache line for destination */ |
53f770e0 RM |
82 | subl $32, %edx /* decr loop count */ |
83 | movl %eax, 0(%edi) /* store words pairwise */ | |
84 | movl %eax, 4(%edi) | |
85 | movl %eax, 8(%edi) | |
86 | movl %eax, 12(%edi) | |
87 | movl %eax, 16(%edi) | |
88 | movl %eax, 20(%edi) | |
89 | movl %eax, 24(%edi) | |
90 | movl %eax, 28(%edi) | |
91 | leal 32(%edi), %edi /* update destination pointer */ | |
5929563f | 92 | jge L(1) |
53f770e0 RM |
93 | |
94 | leal 32(%edx), %ecx /* reset offset count */ | |
95 | ||
96 | /* Write last 0-7 full 32-bit words (up to 8 words if loop was skipped). */ | |
5929563f | 97 | L(2): shrl $2, %ecx /* convert byte count to longword count */ |
53f770e0 RM |
98 | rep |
99 | stosl | |
100 | ||
101 | /* Finally write the last 0-3 bytes. */ | |
102 | movl %edx, %ecx | |
103 | andl $3, %ecx | |
104 | rep | |
105 | stosb | |
106 | ||
3f02f778 | 107 | #if !BZERO_P |
53f770e0 | 108 | /* Load result (only if used as memset). */ |
2fc08826 | 109 | movl DEST(%esp), %eax /* start address of destination is result */ |
53f770e0 RM |
110 | #endif |
111 | popl %edi | |
1ad9da69 UD |
112 | cfi_adjust_cfa_offset (-4) |
113 | cfi_restore (edi) | |
53f770e0 | 114 | |
3f02f778 | 115 | #if BZERO_P |
53f770e0 | 116 | ret |
3f02f778 | 117 | #else |
2366713d | 118 | ret |
3f02f778 | 119 | #endif |
2366713d | 120 | END (memset) |
85dd1003 | 121 | libc_hidden_builtin_def (memset) |