| xf.li | bdd93d5 | 2023-05-12 07:10:14 -0700 | [diff] [blame] | 1 | /* strlen(str) -- determine the length of the string STR. |
| 2 | Optimized for Intel 80x86, x>=4. |
| 3 | Copyright (C) 1991-2016 Free Software Foundation, Inc. |
| 4 | Contributed by Ulrich Drepper <drepper@gnu.ai.mit.edu>. |
| 5 | This file is part of the GNU C Library. |
| 6 | |
| 7 | The GNU C Library is free software; you can redistribute it and/or |
| 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. |
| 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 |
| 15 | Lesser General Public License for more details. |
| 16 | |
| 17 | You should have received a copy of the GNU Lesser General Public |
| 18 | License along with the GNU C Library; if not, see |
| 19 | <http://www.gnu.org/licenses/>. */ |
| 20 | |
| 21 | #include <sysdep.h> |
| 22 | #include "asm-syntax.h" |
| 23 | |
| 24 | #define PARMS 4 /* no space for saved regs */ |
| 25 | #define STR PARMS |
| 26 | |
| 27 | .text |
| 28 | ENTRY (strlen) |
| 29 | |
| 30 | movl STR(%esp), %ecx |
| 31 | movl %ecx, %eax /* duplicate it */ |
| 32 | |
| 33 | andl $3, %ecx /* mask alignment bits */ |
| 34 | jz L(1) /* aligned => start loop */ |
| 35 | cmpb %ch, (%eax) /* is byte NUL? */ |
| 36 | je L(2) /* yes => return */ |
| 37 | incl %eax /* increment pointer */ |
| 38 | |
| 39 | xorl $3, %ecx /* was alignment = 3? */ |
| 40 | jz L(1) /* yes => now it is aligned and start loop */ |
| 41 | cmpb %ch, (%eax) /* is byte NUL? */ |
| 42 | je L(2) /* yes => return */ |
| 43 | addl $1, %eax /* increment pointer */ |
| 44 | |
| 45 | subl $1, %ecx /* was alignment = 2? */ |
| 46 | jz L(1) /* yes => now it is aligned and start loop */ |
| 47 | cmpb %ch, (%eax) /* is byte NUL? */ |
| 48 | je L(2) /* yes => return */ |
| 49 | |
| 50 | /* Don't change the above `addl $1,%eax' and `subl $1, %ecx' into `incl %eax' |
| 51 | and `decl %ecx' resp. The additional two byte per instruction make the |
| 52 | label 4 to be aligned on a 16 byte boundary with nops. |
| 53 | |
| 54 | The following `sub $15, %eax' is part of this trick, too. Together with |
| 55 | the next instruction (`addl $16, %eax') it is in fact a `incl %eax', just |
| 56 | as expected from the algorithm. But doing so has the advantage that |
| 57 | no jump to label 1 is necessary and so the pipeline is not flushed. */ |
| 58 | |
| 59 | subl $15, %eax /* effectively +1 */ |
| 60 | |
| 61 | |
| 62 | L(4): addl $16, %eax /* adjust pointer for full loop */ |
| 63 | |
| 64 | L(1): movl (%eax), %ecx /* get word (= 4 bytes) in question */ |
| 65 | movl $0xfefefeff, %edx /* magic value */ |
| 66 | addl %ecx, %edx /* add the magic value to the word. We get |
| 67 | carry bits reported for each byte which |
| 68 | is *not* 0 */ |
| 69 | jnc L(3) /* highest byte is NUL => return pointer */ |
| 70 | xorl %ecx, %edx /* (word+magic)^word */ |
| 71 | orl $0xfefefeff, %edx /* set all non-carry bits */ |
| 72 | incl %edx /* add 1: if one carry bit was *not* set |
| 73 | the addition will not result in 0. */ |
| 74 | jnz L(3) /* found NUL => return pointer */ |
| 75 | |
| 76 | movl 4(%eax), %ecx /* get word (= 4 bytes) in question */ |
| 77 | movl $0xfefefeff, %edx /* magic value */ |
| 78 | addl %ecx, %edx /* add the magic value to the word. We get |
| 79 | carry bits reported for each byte which |
| 80 | is *not* 0 */ |
| 81 | jnc L(5) /* highest byte is NUL => return pointer */ |
| 82 | xorl %ecx, %edx /* (word+magic)^word */ |
| 83 | orl $0xfefefeff, %edx /* set all non-carry bits */ |
| 84 | incl %edx /* add 1: if one carry bit was *not* set |
| 85 | the addition will not result in 0. */ |
| 86 | jnz L(5) /* found NUL => return pointer */ |
| 87 | |
| 88 | movl 8(%eax), %ecx /* get word (= 4 bytes) in question */ |
| 89 | movl $0xfefefeff, %edx /* magic value */ |
| 90 | addl %ecx, %edx /* add the magic value to the word. We get |
| 91 | carry bits reported for each byte which |
| 92 | is *not* 0 */ |
| 93 | jnc L(6) /* highest byte is NUL => return pointer */ |
| 94 | xorl %ecx, %edx /* (word+magic)^word */ |
| 95 | orl $0xfefefeff, %edx /* set all non-carry bits */ |
| 96 | incl %edx /* add 1: if one carry bit was *not* set |
| 97 | the addition will not result in 0. */ |
| 98 | jnz L(6) /* found NUL => return pointer */ |
| 99 | |
| 100 | movl 12(%eax), %ecx /* get word (= 4 bytes) in question */ |
| 101 | movl $0xfefefeff, %edx /* magic value */ |
| 102 | addl %ecx, %edx /* add the magic value to the word. We get |
| 103 | carry bits reported for each byte which |
| 104 | is *not* 0 */ |
| 105 | jnc L(7) /* highest byte is NUL => return pointer */ |
| 106 | xorl %ecx, %edx /* (word+magic)^word */ |
| 107 | orl $0xfefefeff, %edx /* set all non-carry bits */ |
| 108 | incl %edx /* add 1: if one carry bit was *not* set |
| 109 | the addition will not result in 0. */ |
| 110 | jz L(4) /* no NUL found => continue loop */ |
| 111 | |
| 112 | L(7): addl $4, %eax /* adjust pointer */ |
| 113 | L(6): addl $4, %eax |
| 114 | L(5): addl $4, %eax |
| 115 | |
| 116 | L(3): testb %cl, %cl /* is first byte NUL? */ |
| 117 | jz L(2) /* yes => return */ |
| 118 | incl %eax /* increment pointer */ |
| 119 | |
| 120 | testb %ch, %ch /* is second byte NUL? */ |
| 121 | jz L(2) /* yes => return */ |
| 122 | incl %eax /* increment pointer */ |
| 123 | |
| 124 | testl $0xff0000, %ecx /* is third byte NUL? */ |
| 125 | jz L(2) /* yes => return pointer */ |
| 126 | incl %eax /* increment pointer */ |
| 127 | |
| 128 | L(2): subl STR(%esp), %eax /* compute difference to string start */ |
| 129 | |
| 130 | ret |
| 131 | END (strlen) |
| 132 | libc_hidden_builtin_def (strlen) |