xref: /onnv-gate/usr/src/lib/libc/sparc/gen/strchr.s (revision 7298:b69e27387f74)
10Sstevel@tonic-gate/*
20Sstevel@tonic-gate * CDDL HEADER START
30Sstevel@tonic-gate *
40Sstevel@tonic-gate * The contents of this file are subject to the terms of the
56812Sraf * Common Development and Distribution License (the "License").
66812Sraf * You may not use this file except in compliance with the License.
70Sstevel@tonic-gate *
80Sstevel@tonic-gate * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
90Sstevel@tonic-gate * or http://www.opensolaris.org/os/licensing.
100Sstevel@tonic-gate * See the License for the specific language governing permissions
110Sstevel@tonic-gate * and limitations under the License.
120Sstevel@tonic-gate *
130Sstevel@tonic-gate * When distributing Covered Code, include this CDDL HEADER in each
140Sstevel@tonic-gate * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
150Sstevel@tonic-gate * If applicable, add the following below this CDDL HEADER, with the
160Sstevel@tonic-gate * fields enclosed by brackets "[]" replaced with your own identifying
170Sstevel@tonic-gate * information: Portions Copyright [yyyy] [name of copyright owner]
180Sstevel@tonic-gate *
190Sstevel@tonic-gate * CDDL HEADER END
200Sstevel@tonic-gate */
216812Sraf
220Sstevel@tonic-gate/*
236812Sraf * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
240Sstevel@tonic-gate * Use is subject to license terms.
250Sstevel@tonic-gate */
260Sstevel@tonic-gate
27*7298SMark.J.Nelson@Sun.COM	.file	"strchr.s"
280Sstevel@tonic-gate
290Sstevel@tonic-gate/*
300Sstevel@tonic-gate * The strchr() function returns a pointer to the first occurrence of c
310Sstevel@tonic-gate * (converted to a char) in string s, or a null pointer if c does not occur
320Sstevel@tonic-gate * in the string.
330Sstevel@tonic-gate */
340Sstevel@tonic-gate
350Sstevel@tonic-gate#include <sys/asm_linkage.h>
360Sstevel@tonic-gate
370Sstevel@tonic-gate	! Here, we start by checking to see if we're searching the dest
380Sstevel@tonic-gate	! string for a null byte.  We have fast code for this, so it's
390Sstevel@tonic-gate	! an important special case.  Otherwise, if the string is not
400Sstevel@tonic-gate	! word aligned, we check a for the search char a byte at a time
410Sstevel@tonic-gate	! until we've reached a word boundary.  Once this has happened
420Sstevel@tonic-gate	! some zero-byte finding values are initialized and the string
430Sstevel@tonic-gate	! is checked a word at a time
440Sstevel@tonic-gate
450Sstevel@tonic-gate	ENTRY(strchr)
460Sstevel@tonic-gate
470Sstevel@tonic-gate	.align 32
480Sstevel@tonic-gate
490Sstevel@tonic-gate	andcc	%o1, 0xff, %o1		! search only for this one byte
500Sstevel@tonic-gate	bz	.searchnullbyte		! faster code for searching null
510Sstevel@tonic-gate	andcc	%o0, 3, %o4		! str word aligned ?
520Sstevel@tonic-gate	bz,a	.prepword2		! yup, prepare for word-wise search
530Sstevel@tonic-gate	sll	%o1, 8, %g1		! start spreading findchar across word
540Sstevel@tonic-gate
550Sstevel@tonic-gate	ldub	[%o0], %o2		! str[0]
560Sstevel@tonic-gate	cmp	%o2, %o1		! str[0] == findchar ?
570Sstevel@tonic-gate	be	.done			! yup, done
580Sstevel@tonic-gate	tst	%o2			! str[0] == 0 ?
590Sstevel@tonic-gate	bz	.notfound		! yup, return null pointer
600Sstevel@tonic-gate	cmp	%o4, 3			! only one byte needed to align?
610Sstevel@tonic-gate	bz	.prepword		! yup, prepare for word-wise search
620Sstevel@tonic-gate	inc	%o0			! str++
630Sstevel@tonic-gate	ldub	[%o0], %o2		! str[1]
640Sstevel@tonic-gate	cmp	%o2, %o1		! str[1] == findchar ?
650Sstevel@tonic-gate	be	.done			! yup, done
660Sstevel@tonic-gate	tst	%o2			! str[1] == 0 ?
670Sstevel@tonic-gate	bz	.notfound		! yup, return null pointer
680Sstevel@tonic-gate	cmp	%o4, 2			! only two bytes needed to align?
690Sstevel@tonic-gate	bz	.prepword		! yup, prepare for word-wise search
700Sstevel@tonic-gate	inc	%o0			! str++
710Sstevel@tonic-gate	ldub	[%o0], %o2		! str[2]
720Sstevel@tonic-gate	cmp	%o2, %o1		! str[2] == findchar ?
730Sstevel@tonic-gate	be	.done			! yup, done
740Sstevel@tonic-gate	tst	%o2			! str[2] == 0 ?
750Sstevel@tonic-gate	bz	.notfound		! yup, return null pointer
760Sstevel@tonic-gate	inc	%o0			! str++
770Sstevel@tonic-gate
780Sstevel@tonic-gate.prepword:
790Sstevel@tonic-gate	sll	%o1, 8, %g1		! spread findchar ------+
800Sstevel@tonic-gate.prepword2:							!
810Sstevel@tonic-gate	sethi	%hi(0x01010101), %o4	! Alan Mycroft's magic1 !
820Sstevel@tonic-gate	or	%o1, %g1, %o1		!  across all <---------+
830Sstevel@tonic-gate	sethi	%hi(0x80808080), %o5	! Alan Mycroft's magic2	!
840Sstevel@tonic-gate	sll	%o1, 16, %g1		!   four bytes <--------+
850Sstevel@tonic-gate	or	%o4, %lo(0x01010101), %o4			!
860Sstevel@tonic-gate	or	%o1, %g1, %o1		!    of a word <--------+
870Sstevel@tonic-gate	or	%o5, %lo(0x80808080), %o5
880Sstevel@tonic-gate
890Sstevel@tonic-gate.searchchar:
900Sstevel@tonic-gate	lduw	[%o0], %o2		! src word
910Sstevel@tonic-gate	andn	%o5, %o2, %o3		! ~word & 0x80808080
920Sstevel@tonic-gate	sub	%o2, %o4, %g1		! word = (word - 0x01010101)
930Sstevel@tonic-gate	andcc	%o3, %g1, %g0		! ((word - 0x01010101) & ~word & 0x80808080)
940Sstevel@tonic-gate	bnz	.haszerobyte		! zero byte if magic expression != 0
950Sstevel@tonic-gate	xor	%o2, %o1, %g1		! tword = word ^ findchar
960Sstevel@tonic-gate	andn	%o5, %g1, %o3		! ~tword & 0x80808080
970Sstevel@tonic-gate	sub	%g1, %o4, %o2		! (tword - 0x01010101)
980Sstevel@tonic-gate	andcc	%o3, %o2, %g0		! ((tword - 0x01010101) & ~tword & 0x80808080)
990Sstevel@tonic-gate	bz,a	.searchchar		! no findchar if magic expression == 0
1000Sstevel@tonic-gate	add	%o0, 4, %o0		! str += 4
1010Sstevel@tonic-gate
1020Sstevel@tonic-gate	! here we know "word" contains the searched character, but no null
1030Sstevel@tonic-gate	! byte. if there was a null byte, we would have gone to .haszerobyte
1040Sstevel@tonic-gate	! "tword" has null bytes where "word" had findchar. Examine "tword"
1050Sstevel@tonic-gate
1060Sstevel@tonic-gate.foundchar:
1070Sstevel@tonic-gate	set	0xff000000, %o4		! mask for 1st byte
1080Sstevel@tonic-gate	andcc	%g1, %o4, %g0		! first byte zero (= found search char) ?
1090Sstevel@tonic-gate	bz	.done			! yup, done
1100Sstevel@tonic-gate	set	0x00ff0000, %o5		! mask for 2nd byte
1110Sstevel@tonic-gate	inc	%o0			! str++
1120Sstevel@tonic-gate	andcc	%g1, %o5, %g0		! second byte zero (= found search char) ?
1130Sstevel@tonic-gate	bz	.done			! yup, done
1140Sstevel@tonic-gate	srl	%o4, 16, %o4		! 0x0000ff00 = mask for 3rd byte
1150Sstevel@tonic-gate	inc	%o0			! str++
1160Sstevel@tonic-gate	andcc	%g1, %o4, %g0		! third byte zero (= found search char) ?
1170Sstevel@tonic-gate	bnz,a	.done			! nope, increment in delay slot
1180Sstevel@tonic-gate	inc	%o0			! str++
1190Sstevel@tonic-gate
1200Sstevel@tonic-gate.done:
1210Sstevel@tonic-gate	retl				! done with leaf function
1220Sstevel@tonic-gate	nop				! padding
1230Sstevel@tonic-gate
1240Sstevel@tonic-gate	! Here we know that "word" contains a null byte indicating the
1250Sstevel@tonic-gate	! end of the string. However, "word" might also contain findchar
1260Sstevel@tonic-gate	! "tword" (in %g1) has null bytes where "word" had findchar. So
1270Sstevel@tonic-gate	! check both "tword" and "word"
1280Sstevel@tonic-gate
1290Sstevel@tonic-gate.haszerobyte:
1300Sstevel@tonic-gate	set	0xff000000, %o4		! mask for 1st byte
1310Sstevel@tonic-gate	andcc	%g1, %o4, %g0		! first byte == findchar ?
1320Sstevel@tonic-gate	bz	.done			! yup, done
1330Sstevel@tonic-gate	andcc	%o2, %o4, %g0		! first byte == 0 ?
1340Sstevel@tonic-gate	bz	.notfound		! yup, return null pointer
1350Sstevel@tonic-gate	set	0x00ff0000, %o4		! mask for 2nd byte
1360Sstevel@tonic-gate	inc	%o0			! str++
1370Sstevel@tonic-gate	andcc	%g1, %o4, %g0		! second byte == findchar ?
1380Sstevel@tonic-gate	bz	.done			! yup, done
1390Sstevel@tonic-gate	andcc	%o2, %o4, %g0		! second byte == 0 ?
1400Sstevel@tonic-gate	bz	.notfound		! yup, return null pointer
1410Sstevel@tonic-gate	srl	%o4, 8, %o4		! mask for 3rd byte = 0x0000ff00
1420Sstevel@tonic-gate	inc	%o0			! str++
1430Sstevel@tonic-gate	andcc	%g1, %o4, %g0		! third byte == findchar ?
1440Sstevel@tonic-gate	bz	.done			! yup, done
1450Sstevel@tonic-gate	andcc	%o2, %o4, %g0		! third byte == 0 ?
1460Sstevel@tonic-gate	bz	.notfound		! yup, return null pointer
1470Sstevel@tonic-gate	andcc	%g1, 0xff, %g0		! fourth byte == findchar ?
1480Sstevel@tonic-gate	bz	.done			! yup, done
1490Sstevel@tonic-gate	inc	%o0			! str++
1500Sstevel@tonic-gate
1510Sstevel@tonic-gate.notfound:
1520Sstevel@tonic-gate	retl				! done with leaf function
1530Sstevel@tonic-gate	xor	%o0, %o0, %o0		! return null pointer
1540Sstevel@tonic-gate
1550Sstevel@tonic-gate	! since findchar == 0, we only have to do one test per item
1560Sstevel@tonic-gate	! instead of two. This makes the search much faster.
1570Sstevel@tonic-gate
1580Sstevel@tonic-gate.searchnullbyte:
1590Sstevel@tonic-gate	bz	.straligned		! str is word aligned
1600Sstevel@tonic-gate	nop				! padding
1610Sstevel@tonic-gate
1620Sstevel@tonic-gate	cmp	%o4, 2			! str halfword aligned ?
1630Sstevel@tonic-gate	be	.s2aligned		! yup
1640Sstevel@tonic-gate	ldub	[%o0], %o1		! str[0]
1650Sstevel@tonic-gate	tst	%o1			! byte zero?
1660Sstevel@tonic-gate	bz	.done2			! yup, done
1670Sstevel@tonic-gate	cmp	%o4, 3			! only one byte needed to align?
1680Sstevel@tonic-gate	bz	.straligned		! yup
1690Sstevel@tonic-gate	inc	%o0			! str++
1700Sstevel@tonic-gate
1710Sstevel@tonic-gate	! check to see if we're half word aligned, which it better than
1720Sstevel@tonic-gate	! not being aligned at all.  Search the first half of the word
1730Sstevel@tonic-gate	! if we are, and then search by whole word.
1740Sstevel@tonic-gate
1750Sstevel@tonic-gate.s2aligned:
1760Sstevel@tonic-gate	lduh	[%o0], %o1		! str[]
1770Sstevel@tonic-gate	srl	%o1, 8, %o4		! %o4<7:0> = first byte
1780Sstevel@tonic-gate	tst	%o4			! first byte zero ?
1790Sstevel@tonic-gate	bz	.done2			! yup, done
1800Sstevel@tonic-gate	andcc	%o1, 0xff, %g0		! second byte zero ?
1810Sstevel@tonic-gate	bz,a	.done2			! yup, done
1820Sstevel@tonic-gate	inc	%o0			! str++
1830Sstevel@tonic-gate	add	%o0, 2, %o0		! str+=2
1840Sstevel@tonic-gate
1850Sstevel@tonic-gate.straligned:
1860Sstevel@tonic-gate	sethi	%hi(0x01010101), %o4	! Alan Mycroft's magic1
1870Sstevel@tonic-gate	sethi	%hi(0x80808080), %o5	! Alan Mycroft's magic2
1880Sstevel@tonic-gate	or	%o4, %lo(0x01010101), %o4
1890Sstevel@tonic-gate	or	%o5, %lo(0x80808080), %o5
1900Sstevel@tonic-gate
1910Sstevel@tonic-gate.searchword:
1920Sstevel@tonic-gate	lduw	[%o0], %o1		! src word
1930Sstevel@tonic-gate	andn	%o5, %o1, %o3		! ~word & 0x80808080
1940Sstevel@tonic-gate	sub	%o1, %o4, %g1		! word = (word - 0x01010101)
1950Sstevel@tonic-gate	andcc	%o3, %g1, %g0		! ((word - 0x01010101) & ~word & 0x80808080)
1960Sstevel@tonic-gate	bz,a	.searchword		! no zero byte if magic expression == 0
1970Sstevel@tonic-gate	add	%o0, 4, %o0		! str += 4
1980Sstevel@tonic-gate
1990Sstevel@tonic-gate.zerobyte:
2000Sstevel@tonic-gate	set	0xff000000, %o4		! mask for 1st byte
2010Sstevel@tonic-gate	andcc	%o1, %o4, %g0		! first byte zero?
2020Sstevel@tonic-gate	bz	.done2			! yup, done
2030Sstevel@tonic-gate	set	0x00ff0000, %o5		! mask for 2nd byte
2040Sstevel@tonic-gate	inc	%o0			! str++
2050Sstevel@tonic-gate	andcc	%o1, %o5, %g0		! second byte zero?
2060Sstevel@tonic-gate	bz	.done2			! yup, done
2070Sstevel@tonic-gate	srl	%o4, 16, %o4		! 0x0000ff00 = mask for 3rd byte
2080Sstevel@tonic-gate	inc	%o0			! str++
2090Sstevel@tonic-gate	andcc	%o1, %o4, %g0		! third byte zero?
2100Sstevel@tonic-gate	bnz,a	.done2			! nope, increment in delay slot
2110Sstevel@tonic-gate	inc	%o0			! str++
2120Sstevel@tonic-gate.done2:
2130Sstevel@tonic-gate    	retl				! return from leaf function
2140Sstevel@tonic-gate	nop				! padding
2150Sstevel@tonic-gate
2160Sstevel@tonic-gate	SET_SIZE(strchr)
217