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