1 /* memchr (str, ch, n) -- Return pointer to first occurrence of CH in STR less
4 Copyright (C) 1998, 1999, 2000, 2003 Free Software Foundation, Inc.
5 This file is part of the GNU C Library.
6 Contributed by Jan Vondrak <jvon4518@ss1000.ms.mff.cuni.cz> and
7 Jakub Jelinek <jj@ultra.linux.cz>.
8 This version is developed using the same algorithm as the fast C
9 version which carries the following introduction:
10 Based on strlen implementation by Torbjorn Granlund (tege@sics.se),
11 with help from Dan Sahlin (dan@sics.se) and
12 commentary by Jim Blandy (jimb@ai.mit.edu);
13 adaptation to memchr suggested by Dick Karpinski (dick@cca.ucsf.edu),
14 and implemented by Roland McGrath (roland@ai.mit.edu).
16 The GNU C Library is free software; you can redistribute it and/or
17 modify it under the terms of the GNU Lesser General Public
18 License as published by the Free Software Foundation; either
19 version 2.1 of the License, or (at your option) any later version.
21 The GNU C Library is distributed in the hope that it will be useful,
22 but WITHOUT ANY WARRANTY; without even the implied warranty of
23 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
24 Lesser General Public License for more details.
26 You should have received a copy of the GNU Lesser General Public
27 License along with the GNU C Library; if not, see
28 <http://www.gnu.org/licenses/>. */
34 .register %g2, #scratch
35 .register %g3, #scratch
38 /* Normally, this uses
39 ((xword - 0x0101010101010101) & 0x8080808080808080) test
40 to find out if any byte in xword could be zero. This is fast, but
41 also gives false alarm for any byte in range 0x81-0xff. It does
42 not matter for correctness, as if this test tells us there could
43 be some zero byte, we check it byte by byte, but if bytes with
44 high bits set are common in the strings, then this will give poor
45 performance. You can #define EIGHTBIT_NOT_RARE and the algorithm
46 will use one tick slower, but more precise test
47 ((xword - 0x0101010101010101) & (~xword) & 0x8080808080808080),
48 which does not give any false alarms (but if some bits are set,
49 one cannot assume from it which bytes are zero and which are not).
50 It is yet to be measured, what is the correct default for glibc
51 in these days for an average user.
57 and %o1, 0xff, %o1 /* IEU0 Group */
59 brz,pn %o2, 12f /* CTI+IEU1 */
62 be,pn %XCC, 12f /* CTI */
64 sll %o1, 8, %g3 /* IEU0 Group */
65 add %o0, %o2, %o2 /* IEU1 */
67 sethi %hi(0x01010101), %g1 /* IEU0 Group */
68 or %g3, %o1, %g3 /* IEU1 */
69 ldub [%o0], %o3 /* Load */
70 sllx %g3, 16, %g5 /* IEU0 Group */
72 or %g1, %lo(0x01010101), %g1 /* IEU1 */
73 sllx %g1, 32, %g2 /* IEU0 Group */
74 or %g3, %g5, %g3 /* IEU1 */
75 sllx %g3, 32, %g5 /* IEU0 Group */
77 cmp %o3, %o1 /* IEU1 */
78 be,pn %xcc, 13f /* CTI */
79 or %g1, %g2, %g1 /* IEU0 Group */
80 andcc %o0, 7, %g0 /* IEU1 */
82 bne,a,pn %icc, 21f /* CTI */
83 add %o0, 1, %o0 /* IEU0 Group */
84 ldx [%o0], %o3 /* Load Group */
85 sllx %g1, 7, %g2 /* IEU0 */
87 or %g3, %g5, %g3 /* IEU1 */
88 1: add %o0, 8, %o0 /* IEU0 Group */
89 xor %o3, %g3, %o4 /* IEU1 */
90 /* %g1 = 0101010101010101 *
91 * %g2 = 8080088080808080 *
92 * %g3 = c c c c c c c c *
94 * %o4 = value XOR c */
95 2: cmp %o0, %o2 /* IEU1 Group */
97 bg,pn %XCC, 11f /* CTI */
98 ldxa [%o0] ASI_PNF, %o3 /* Load */
99 sub %o4, %g1, %o5 /* IEU0 Group */
100 add %o0, 8, %o0 /* IEU1 */
101 #ifdef EIGHTBIT_NOT_RARE
102 andn %o5, %o4, %o5 /* IEU0 Group */
105 andcc %o5, %g2, %g0 /* IEU1 Group */
106 be,a,pt %xcc, 2b /* CTI */
107 xor %o3, %g3, %o4 /* IEU0 */
108 srlx %o4, 56, %g5 /* IEU0 */
110 andcc %g5, 0xff, %g0 /* IEU1 Group */
111 be,pn %icc, 3f /* CTI */
112 srlx %o4, 48, %g5 /* IEU0 */
113 andcc %g5, 0xff, %g0 /* IEU1 Group */
115 be,pn %icc, 4f /* CTI */
116 srlx %o4, 40, %g5 /* IEU0 */
117 andcc %g5, 0xff, %g0 /* IEU1 Group */
118 be,pn %icc, 5f /* CTI */
120 srlx %o4, 32, %g5 /* IEU0 */
121 andcc %g5, 0xff, %g0 /* IEU1 Group */
122 be,pn %icc, 6f /* CTI */
123 srlx %o4, 24, %g5 /* IEU0 */
125 andcc %g5, 0xff, %g0 /* IEU1 Group */
126 be,pn %icc, 7f /* CTI */
127 srlx %o4, 16, %g5 /* IEU0 */
128 andcc %g5, 0xff, %g0 /* IEU1 Group */
130 be,pn %icc, 8f /* CTI */
131 srlx %o4, 8, %g5 /* IEU0 */
132 andcc %g5, 0xff, %g0 /* IEU1 Group */
133 be,pn %icc, 9f /* CTI */
135 andcc %o4, 0xff, %g0 /* IEU1 Group */
136 bne,pt %icc, 2b /* CTI */
137 xor %o3, %g3, %o4 /* IEU0 */
138 retl /* CTI+IEU1 Group */
140 add %o0, -9, %o0 /* IEU0 */
143 3: retl /* CTI+IEU1 Group */
144 add %o0, -16, %o0 /* IEU0 */
145 4: retl /* CTI+IEU1 Group */
146 add %o0, -15, %o0 /* IEU0 */
148 5: retl /* CTI+IEU1 Group */
149 add %o0, -14, %o0 /* IEU0 */
150 6: retl /* CTI+IEU1 Group */
151 add %o0, -13, %o0 /* IEU0 */
153 7: retl /* CTI+IEU1 Group */
154 add %o0, -12, %o0 /* IEU0 */
155 8: retl /* CTI+IEU1 Group */
156 add %o0, -11, %o0 /* IEU0 */
158 9: retl /* CTI+IEU1 Group */
159 add %o0, -10, %o0 /* IEU0 */
160 11: sub %o4, %g1, %o5 /* IEU0 Group */
161 sub %o0, 8, %o0 /* IEU1 */
163 andcc %o5, %g2, %g0 /* IEU1 Group */
164 be,pt %xcc, 12f /* CTI */
165 sub %o2, %o0, %o2 /* IEU0 */
166 tst %o2 /* IEU1 Group */
168 be,pn %XCC, 12f /* CTI */
169 srlx %o4, 56, %g5 /* IEU0 */
170 andcc %g5, 0xff, %g0 /* IEU1 Group */
171 be,pn %icc, 13f /* CTI */
173 cmp %o2, 1 /* IEU0 */
174 be,pn %XCC, 12f /* CTI Group */
175 srlx %o4, 48, %g5 /* IEU0 */
176 andcc %g5, 0xff, %g0 /* IEU1 Group */
178 be,pn %icc, 14f /* CTI */
179 cmp %o2, 2 /* IEU1 Group */
180 be,pn %XCC, 12f /* CTI */
181 srlx %o4, 40, %g5 /* IEU0 */
183 andcc %g5, 0xff, %g0 /* IEU1 Group */
184 be,pn %icc, 15f /* CTI */
185 cmp %o2, 3 /* IEU1 Group */
186 be,pn %XCC, 12f /* CTI */
188 srlx %o4, 32, %g5 /* IEU0 */
189 andcc %g5, 0xff, %g0 /* IEU1 Group */
190 be,pn %icc, 16f /* CTI */
191 cmp %o2, 4 /* IEU1 Group */
193 be,pn %XCC, 12f /* CTI */
194 srlx %o4, 24, %g5 /* IEU0 */
195 andcc %g5, 0xff, %g0 /* IEU1 Group */
196 be,pn %icc, 17f /* CTI */
198 cmp %o2, 5 /* IEU1 Group */
199 be,pn %XCC, 12f /* CTI */
200 srlx %o4, 16, %g5 /* IEU0 */
201 andcc %g5, 0xff, %g0 /* IEU1 Group */
203 be,pn %icc, 18f /* CTI */
204 cmp %o2, 6 /* IEU1 Group */
205 be,pn %XCC, 12f /* CTI */
206 srlx %o4, 8, %g5 /* IEU0 */
208 andcc %g5, 0xff, %g0 /* IEU1 Group */
209 be,pn %icc, 19f /* CTI */
211 12: retl /* CTI+IEU1 Group */
215 13: retl /* CTI+IEU1 Group */
218 14: retl /* CTI+IEU1 Group */
219 add %o0, 1, %o0 /* IEU0 */
220 15: retl /* CTI+IEU1 Group */
221 add %o0, 2, %o0 /* IEU0 */
223 16: retl /* CTI+IEU1 Group */
224 add %o0, 3, %o0 /* IEU0 */
225 17: retl /* CTI+IEU1 Group */
226 add %o0, 4, %o0 /* IEU0 */
228 18: retl /* CTI+IEU1 Group */
229 add %o0, 5, %o0 /* IEU0 */
230 19: retl /* CTI+IEU1 Group */
231 add %o0, 6, %o0 /* IEU0 */
233 21: cmp %o0, %o2 /* IEU1 */
234 be,pn %XCC, 12b /* CTI */
235 sllx %g1, 7, %g2 /* IEU0 Group */
236 ldub [%o0], %o3 /* Load */
238 or %g3, %g5, %g3 /* IEU1 */
239 22: andcc %o0, 7, %g0 /* IEU1 Group */
240 be,a,pn %icc, 1b /* CTI */
241 ldx [%o0], %o3 /* Load */
243 cmp %o3, %o1 /* IEU1 Group */
244 be,pn %xcc, 23f /* CTI */
245 add %o0, 1, %o0 /* IEU0 */
246 cmp %o0, %o2 /* IEU1 Group */
248 bne,a,pt %XCC, 22b /* CTI */
249 ldub [%o0], %o3 /* Load */
250 retl /* CTI+IEU1 Group */
253 23: retl /* CTI+IEU1 Group */
254 add %o0, -1, %o0 /* IEU0 */
257 libc_hidden_def(memchr)
258 weak_alias(memchr,__ubp_memchr)