xref: /titanic_50/usr/src/lib/libc/sparcv9/gen/strchr.s (revision ff3124eff995e6cd8ebd8c6543648e0670920034)
1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22/*
23 * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
24 * Use is subject to license terms.
25 */
26
27#pragma ident	"%Z%%M%	%I%	%E% SMI"
28
29	.file	"%M%"
30
31/*
32 * The strchr() function returns a pointer to the first occurrence of c
33 * (converted to a char) in string s, or a null pointer if c does not occur
34 * in the string.
35 */
36
37#include <sys/asm_linkage.h>
38
39	! Here, we start by checking to see if we're searching the dest
40	! string for a null byte.  We have fast code for this, so it's
41	! an important special case.  Otherwise, if the string is not
42	! word aligned, we check a for the search char a byte at a time
43	! until we've reached a word boundary.  Once this has happened
44	! some zero-byte finding values are initialized and the string
45	! is checked a word at a time
46
47	ENTRY(strchr)
48
49	.align 32
50
51	andcc	%o1, 0xff, %o1		! search only for this one byte
52	bz,pn	%ncc, .searchnullbyte	! faster code for searching null
53	andcc	%o0, 3, %o4		! str word aligned ?
54	bz,a,pn	%ncc, .prepword2	! yup, prepare for word-wise search
55	sll	%o1, 8, %g1		! start spreading findchar across word
56
57	ldub	[%o0], %o2		! str[0]
58	cmp	%o2, %o1		! str[0] == findchar ?
59	be,pn	%ncc, .done		! yup, done
60	tst	%o2			! str[0] == 0 ?
61	bz,pn	%ncc, .notfound		! yup, return null pointer
62	cmp	%o4, 3			! only one byte needed to align?
63	bz,pn	%ncc, .prepword		! yup, prepare for word-wise search
64	inc	%o0			! str++
65	ldub	[%o0], %o2		! str[1]
66	cmp	%o2, %o1		! str[1] == findchar ?
67	be,pn	%ncc, .done		! yup, done
68	tst	%o2			! str[1] == 0 ?
69	bz,pn	%ncc, .notfound		! yup, return null pointer
70	cmp	%o4, 2			! only two bytes needed to align?
71	bz,pn	%ncc, .prepword		! yup, prepare for word-wise search
72	inc	%o0			! str++
73	ldub	[%o0], %o2		! str[2]
74	cmp	%o2, %o1		! str[2] == findchar ?
75	be,pn	%ncc, .done		! yup, done
76	tst	%o2			! str[2] == 0 ?
77	bz,pn	%ncc, .notfound		! yup, return null pointer
78	inc	%o0			! str++
79
80.prepword:
81	sll	%o1, 8, %g1		! spread findchar ------+
82.prepword2:							!
83	sethi	%hi(0x01010101), %o4	! Alan Mycroft's magic1 !
84	or	%o1, %g1, %o1		!  across all <---------+
85	sethi	%hi(0x80808080), %o5	! Alan Mycroft's magic2	!
86	sll	%o1, 16, %g1		!   four bytes <--------+
87	or	%o4, %lo(0x01010101), %o4			!
88	or	%o1, %g1, %o1		!    of a word <--------+
89	or	%o5, %lo(0x80808080), %o5
90
91.searchchar:
92	lduw	[%o0], %o2		! src word
93	andn	%o5, %o2, %o3		! ~word & 0x80808080
94	sub	%o2, %o4, %g1		! word = (word - 0x01010101)
95	andcc	%o3, %g1, %g0		! ((word - 0x01010101) & ~word & 0x80808080)
96	bnz,pn	%ncc, .haszerobyte	! zero byte if magic expression != 0
97	xor	%o2, %o1, %g1		! tword = word ^ findchar
98	andn	%o5, %g1, %o3		! ~tword & 0x80808080
99	sub	%g1, %o4, %o2		! (tword - 0x01010101)
100	andcc	%o3, %o2, %g0		! ((tword - 0x01010101) & ~tword & 0x80808080)
101	bz,a,pt	%ncc, .searchchar	! no findchar if magic expression == 0
102	add	%o0, 4, %o0		! str += 4
103
104	! here we know "word" contains the searched character, but no null
105	! byte. if there was a null byte, we would have gone to .haszerobyte
106	! "tword" has null bytes where "word" had findchar. Examine "tword"
107
108.foundchar:
109	set	0xff000000, %o4		! mask for 1st byte
110	andcc	%g1, %o4, %g0		! first byte zero (= found search char) ?
111	bz,pn	%ncc, .done		! yup, done
112	set	0x00ff0000, %o5		! mask for 2nd byte
113	inc	%o0			! str++
114	andcc	%g1, %o5, %g0		! second byte zero (= found search char) ?
115	bz,pn	%ncc, .done		! yup, done
116	srl	%o4, 16, %o4		! 0x0000ff00 = mask for 3rd byte
117	inc	%o0			! str++
118	andcc	%g1, %o4, %g0		! third byte zero (= found search char) ?
119	bnz,a	%ncc, .done		! nope, increment in delay slot
120	inc	%o0			! str++
121
122.done:
123	retl				! done with leaf function
124	nop				! padding
125
126	! Here we know that "word" contains a null byte indicating the
127	! end of the string. However, "word" might also contain findchar
128	! "tword" (in %g1) has null bytes where "word" had findchar. So
129	! check both "tword" and "word"
130
131.haszerobyte:
132	set	0xff000000, %o4		! mask for 1st byte
133	andcc	%g1, %o4, %g0		! first byte == findchar ?
134	bz,pn	%ncc, .done		! yup, done
135	andcc	%o2, %o4, %g0		! first byte == 0 ?
136	bz,pn	%ncc, .notfound		! yup, return null pointer
137	set	0x00ff0000, %o4		! mask for 2nd byte
138	inc	%o0			! str++
139	andcc	%g1, %o4, %g0		! second byte == findchar ?
140	bz,pn	%ncc, .done		! yup, done
141	andcc	%o2, %o4, %g0		! second byte == 0 ?
142	bz,pn	%ncc, .notfound		! yup, return null pointer
143	srl	%o4, 8, %o4		! mask for 3rd byte = 0x0000ff00
144	inc	%o0			! str++
145	andcc	%g1, %o4, %g0		! third byte == findchar ?
146	bz,pn	%ncc, .done		! yup, done
147	andcc	%o2, %o4, %g0		! third byte == 0 ?
148	bz,pn	%ncc, .notfound		! yup, return null pointer
149	andcc	%g1, 0xff, %g0		! fourth byte == findchar ?
150	bz,pn	%ncc, .done		! yup, done
151	inc	%o0			! str++
152
153.notfound:
154	retl				! done with leaf function
155	xor	%o0, %o0, %o0		! return null pointer
156
157	! since findchar == 0, we only have to do one test per item
158	! instead of two. This makes the search much faster.
159
160.searchnullbyte:
161	bz,pn	%ncc, .straligned	! str is word aligned
162	nop				! padding
163
164	cmp	%o4, 2			! str halfword aligned ?
165	be,pn	%ncc, .s2aligned	! yup
166	ldub	[%o0], %o1		! str[0]
167	tst	%o1			! byte zero?
168	bz,pn	%ncc, .done		! yup, done
169	cmp	%o4, 3			! only one byte needed to align?
170	bz,pn	%ncc, .straligned	! yup
171	inc	%o0			! str++
172
173	! check to see if we're half word aligned, which it better than
174	! not being aligned at all.  Search the first half of the word
175	! if we are, and then search by whole word.
176
177.s2aligned:
178	lduh	[%o0], %o1		! str[]
179	srl	%o1, 8, %o4		! %o4<7:0> = first byte
180	tst	%o4			! first byte zero ?
181	bz,pn	%ncc, .done2		! yup, done
182	andcc	%o1, 0xff, %g0		! second byte zero ?
183	bz,a,pn	%ncc, .done2		! yup, done
184	inc	%o0			! str++
185	add	%o0, 2, %o0		! str+=2
186
187.straligned:
188	sethi	%hi(0x01010101), %o4	! Alan Mycroft's magic1
189	sethi	%hi(0x80808080), %o5	! Alan Mycroft's magic2
190	or	%o4, %lo(0x01010101), %o4
191	or	%o5, %lo(0x80808080), %o5
192
193.searchword:
194	lduw	[%o0], %o1		! src word
195	andn	%o5, %o1, %o3		! ~word & 0x80808080
196	sub	%o1, %o4, %g1		! word = (word - 0x01010101)
197	andcc	%o3, %g1, %g0		! ((word - 0x01010101) & ~word & 0x80808080)
198	bz,a,pt	%ncc, .searchword	! no zero byte if magic expression == 0
199	add	%o0, 4, %o0		! str += 4
200
201.zerobyte:
202	set	0xff000000, %o4		! mask for 1st byte
203	andcc	%o1, %o4, %g0		! first byte zero?
204	bz,pn	%ncc, .done2		! yup, done
205	set	0x00ff0000, %o5		! mask for 2nd byte
206	inc	%o0			! str++
207	andcc	%o1, %o5, %g0		! second byte zero?
208	bz,pn	%ncc, .done2		! yup, done
209	srl	%o4, 16, %o4		! 0x0000ff00 = mask for 3rd byte
210	inc	%o0			! str++
211	andcc	%o1, %o4, %g0		! third byte zero?
212	bnz,a	%ncc, .done2		! nope, increment in delay slot
213	inc	%o0			! str++
214.done2:
215    	retl				! return from leaf function
216	nop				! padding
217
218	SET_SIZE(strchr)
219