66 lines
1.2 KiB
ArmAsm
66 lines
1.2 KiB
ArmAsm
|
/*
|
||
|
* Written by J.T. Conklin <jtc@netbsd.org>.
|
||
|
* Public domain.
|
||
|
* Adapted for NetBSD/x86_64 by Frank van der Linden <fvdl@wasabisystems.com>
|
||
|
*/
|
||
|
|
||
|
#include <machine/asm.h>
|
||
|
|
||
|
#if defined(LIBC_SCCS)
|
||
|
RCSID("$NetBSD: strcat.S,v 1.1 2001/06/19 00:25:05 fvdl Exp $")
|
||
|
#endif
|
||
|
|
||
|
/*
|
||
|
* NOTE: I've unrolled the loop eight times: large enough to make a
|
||
|
* significant difference, and small enough not to totally trash the
|
||
|
* cache.
|
||
|
*/
|
||
|
|
||
|
ENTRY(strcat)
|
||
|
movq %rdi,%r11
|
||
|
|
||
|
cld /* set search forward */
|
||
|
xorl %eax,%eax /* set search for null terminator */
|
||
|
movq $-1,%rcx /* set search for lots of characters */
|
||
|
repne /* search! */
|
||
|
scasb
|
||
|
|
||
|
decq %rdi
|
||
|
|
||
|
L1: movb (%rsi),%al /* unroll loop, but not too much */
|
||
|
movb %al,(%rdi)
|
||
|
testb %al,%al
|
||
|
jz L2
|
||
|
movb 1(%rsi),%al
|
||
|
movb %al,1(%rdi)
|
||
|
testb %al,%al
|
||
|
jz L2
|
||
|
movb 2(%rsi),%al
|
||
|
movb %al,2(%rdi)
|
||
|
testb %al,%al
|
||
|
jz L2
|
||
|
movb 3(%rsi),%al
|
||
|
movb %al,3(%rdi)
|
||
|
testb %al,%al
|
||
|
jz L2
|
||
|
movb 4(%rsi),%al
|
||
|
movb %al,4(%rdi)
|
||
|
testb %al,%al
|
||
|
jz L2
|
||
|
movb 5(%rsi),%al
|
||
|
movb %al,5(%rdi)
|
||
|
testb %al,%al
|
||
|
jz L2
|
||
|
movb 6(%rsi),%al
|
||
|
movb %al,6(%rdi)
|
||
|
testb %al,%al
|
||
|
jz L2
|
||
|
movb 7(%rsi),%al
|
||
|
movb %al,7(%rdi)
|
||
|
addq $8,%rsi
|
||
|
addq $8,%rdi
|
||
|
testb %al,%al
|
||
|
jnz L1
|
||
|
L2: movq %r11,%rax
|
||
|
ret
|