xref: /linux/arch/alpha/lib/ev67-strlen.S (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * arch/alpha/lib/ev67-strlen.S
4 * 21264 version by Rick Gorton <rick.gorton@alpha-processor.com>
5 *
6 * Finds length of a 0-terminated string.  Optimized for the
7 * Alpha architecture:
8 *
9 *	- memory accessed as aligned quadwords only
10 *	- uses bcmpge to compare 8 bytes in parallel
11 *
12 * Much of the information about 21264 scheduling/coding comes from:
13 *	Compiler Writer's Guide for the Alpha 21264
14 *	abbreviated as 'CWG' in other comments here
15 *	ftp.digital.com/pub/Digital/info/semiconductor/literature/dsc-library.html
16 * Scheduling notation:
17 *	E	- either cluster
18 *	U	- upper subcluster; U0 - subcluster U0; U1 - subcluster U1
19 *	L	- lower subcluster; L0 - subcluster L0; L1 - subcluster L1
20 */
21#include <linux/export.h>
22	.set noreorder
23	.set noat
24
25	.globl	strlen
26	.ent	strlen
27	.align 4
28strlen:
29	ldq_u	$1, 0($16)	# L : load first quadword ($16  may be misaligned)
30	lda	$2, -1($31)	# E :
31	insqh	$2, $16, $2	# U :
32	andnot	$16, 7, $0	# E :
33
34	or	$2, $1, $1	# E :
35	cmpbge	$31, $1, $2	# E : $2  <- bitmask: bit i == 1 <==> i-th byte == 0
36	nop			# E :
37	bne	$2, $found	# U :
38
39$loop:	ldq	$1, 8($0)	# L :
40	addq	$0, 8, $0	# E : addr += 8
41	cmpbge	$31, $1, $2	# E :
42	beq	$2, $loop	# U :
43
44$found:
45	cttz	$2, $3		# U0 :
46	addq	$0, $3, $0	# E :
47	subq	$0, $16, $0	# E :
48	ret	$31, ($26)	# L0 :
49
50	.end	strlen
51	EXPORT_SYMBOL(strlen)
52