1/*
2 * arch/alpha/lib/ev67-strlen.S
3 * 21264 version by Rick Gorton <rick.gorton@alpha-processor.com>
4 *
5 * Finds length of a 0-terminated string.  Optimized for the
6 * Alpha architecture:
7 *
8 *	- memory accessed as aligned quadwords only
9 *	- uses bcmpge to compare 8 bytes in parallel
10 *
11 * Much of the information about 21264 scheduling/coding comes from:
12 *	Compiler Writer's Guide for the Alpha 21264
13 *	abbreviated as 'CWG' in other comments here
14 *	ftp.digital.com/pub/Digital/info/semiconductor/literature/dsc-library.html
15 * Scheduling notation:
16 *	E	- either cluster
17 *	U	- upper subcluster; U0 - subcluster U0; U1 - subcluster U1
18 *	L	- lower subcluster; L0 - subcluster L0; L1 - subcluster L1
19 */
20
21	.set noreorder
22	.set noat
23
24	.globl	strlen
25	.ent	strlen
26	.align 4
27strlen:
28	ldq_u	$1, 0($16)	# L : load first quadword ($16  may be misaligned)
29	lda	$2, -1($31)	# E :
30	insqh	$2, $16, $2	# U :
31	andnot	$16, 7, $0	# E :
32
33	or	$2, $1, $1	# E :
34	cmpbge	$31, $1, $2	# E : $2  <- bitmask: bit i == 1 <==> i-th byte == 0
35	nop			# E :
36	bne	$2, $found	# U :
37
38$loop:	ldq	$1, 8($0)	# L :
39	addq	$0, 8, $0	# E : addr += 8
40	cmpbge	$31, $1, $2	# E :
41	beq	$2, $loop	# U :
42
43$found:
44	cttz	$2, $3		# U0 :
45	addq	$0, $3, $0	# E :
46	subq	$0, $16, $0	# E :
47	ret	$31, ($26)	# L0 :
48
49	.end	strlen
50