1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * arch/alpha/lib/ev67-strcat.S
4 * 21264 version contributed by Rick Gorton <rick.gorton@alpha-processor.com>
5 *
6 * Append a null-terminated string from SRC to DST.
7 *
8 * Much of the information about 21264 scheduling/coding comes from:
9 *	Compiler Writer's Guide for the Alpha 21264
10 *	abbreviated as 'CWG' in other comments here
11 *	ftp.digital.com/pub/Digital/info/semiconductor/literature/dsc-library.html
12 * Scheduling notation:
13 *	E	- either cluster
14 *	U	- upper subcluster; U0 - subcluster U0; U1 - subcluster U1
15 *	L	- lower subcluster; L0 - subcluster L0; L1 - subcluster L1
16 * Try not to change the actual algorithm if possible for consistency.
17 * Commentary: It seems bogus to walk the input string twice - once
18 * to determine the length, and then again while doing the copy.
19 * A significant (future) enhancement would be to only read the input
20 * string once.
21 */
22
23#include <asm/export.h>
24	.text
25
26	.align 4
27	.globl strcat
28	.ent strcat
29strcat:
30	.frame $30, 0, $26
31	.prologue 0
32
33	mov	$16, $0		# E : set up return value
34	/* Find the end of the string.  */
35	ldq_u   $1, 0($16)	# L : load first quadword (a0 may be misaligned)
36	lda     $2, -1		# E :
37	insqh   $2, $16, $2	# U :
38
39	andnot  $16, 7, $16	# E :
40	or      $2, $1, $1	# E :
41	cmpbge  $31, $1, $2	# E : bits set iff byte == 0
42	bne     $2, $found	# U :
43
44$loop:	ldq     $1, 8($16)	# L :
45	addq    $16, 8, $16	# E :
46	cmpbge  $31, $1, $2	# E :
47	beq     $2, $loop	# U :
48
49$found:	cttz	$2, $3		# U0 :
50	addq	$16, $3, $16	# E :
51	/* Now do the append.  */
52	mov	$26, $23	# E :
53	br	__stxcpy	# L0 :
54
55	.end strcat
56	EXPORT_SYMBOL(strcat)
57