]>
Commit | Line | Data |
---|---|---|
22312b71 AD |
1 | /* memrchr -- find the last occurrence of a byte in a memory block |
2 | Copyright (C) 1991, 93, 96, 97, 99, 2000 Free Software Foundation, Inc. | |
3 | Based on strlen implementation by Torbjorn Granlund (tege@sics.se), | |
4 | with help from Dan Sahlin (dan@sics.se) and | |
5 | commentary by Jim Blandy (jimb@ai.mit.edu); | |
6 | adaptation to memchr suggested by Dick Karpinski (dick@cca.ucsf.edu), | |
7 | and implemented by Roland McGrath (roland@ai.mit.edu). | |
8 | ||
9 | The GNU C Library is free software; you can redistribute it and/or | |
10 | modify it under the terms of the GNU Library General Public License as | |
11 | published by the Free Software Foundation; either version 2 of the | |
12 | License, or (at your option) any later version. | |
13 | ||
14 | The GNU C Library is distributed in the hope that it will be useful, | |
15 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
17 | Library General Public License for more details. | |
18 | ||
19 | You should have received a copy of the GNU Library General Public | |
20 | License along with the GNU C Library; see the file COPYING.LIB. If not, | |
21 | write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
22 | Boston, MA 02111-1307, USA. */ | |
23 | ||
24 | #ifdef HAVE_CONFIG_H | |
25 | # include <config.h> | |
26 | #endif | |
27 | ||
28 | #include <stdlib.h> | |
29 | ||
30 | #undef __ptr_t | |
31 | #if defined (__cplusplus) || (defined (__STDC__) && __STDC__) | |
32 | # define __ptr_t void * | |
33 | #else /* Not C++ or ANSI C. */ | |
34 | # define __ptr_t char * | |
35 | #endif /* C++ or ANSI C. */ | |
36 | ||
37 | #if defined (_LIBC) | |
38 | # include <string.h> | |
39 | # include <memcopy.h> | |
40 | #else | |
41 | # define reg_char char | |
42 | #endif | |
43 | ||
44 | #if defined (HAVE_LIMITS_H) || defined (_LIBC) | |
45 | # include <limits.h> | |
46 | #endif | |
47 | ||
48 | #define LONG_MAX_32_BITS 2147483647 | |
49 | ||
50 | #ifndef LONG_MAX | |
51 | # define LONG_MAX LONG_MAX_32_BITS | |
52 | #endif | |
53 | ||
54 | #include <sys/types.h> | |
55 | ||
56 | #undef __memrchr | |
57 | #undef memrchr | |
58 | ||
59 | #ifndef weak_alias | |
60 | # define __memrchr memrchr | |
61 | #endif | |
62 | ||
63 | /* Search no more than N bytes of S for C. */ | |
64 | __ptr_t | |
65 | __memrchr (s, c_in, n) | |
66 | const __ptr_t s; | |
67 | int c_in; | |
68 | size_t n; | |
69 | { | |
70 | const unsigned char *char_ptr; | |
71 | const unsigned long int *longword_ptr; | |
72 | unsigned long int longword, magic_bits, charmask; | |
73 | unsigned reg_char c; | |
74 | ||
75 | c = (unsigned char) c_in; | |
76 | ||
77 | /* Handle the last few characters by reading one character at a time. | |
78 | Do this until CHAR_PTR is aligned on a longword boundary. */ | |
79 | for (char_ptr = (const unsigned char *) s + n; | |
80 | n > 0 && ((unsigned long int) char_ptr | |
81 | & (sizeof (longword) - 1)) != 0; | |
82 | --n) | |
83 | if (*--char_ptr == c) | |
84 | return (__ptr_t) char_ptr; | |
85 | ||
86 | /* All these elucidatory comments refer to 4-byte longwords, | |
87 | but the theory applies equally well to 8-byte longwords. */ | |
88 | ||
89 | longword_ptr = (unsigned long int *) char_ptr; | |
90 | ||
91 | /* Bits 31, 24, 16, and 8 of this number are zero. Call these bits | |
92 | the "holes." Note that there is a hole just to the left of | |
93 | each byte, with an extra at the end: | |
94 | ||
95 | bits: 01111110 11111110 11111110 11111111 | |
96 | bytes: AAAAAAAA BBBBBBBB CCCCCCCC DDDDDDDD | |
97 | ||
98 | The 1-bits make sure that carries propagate to the next 0-bit. | |
99 | The 0-bits provide holes for carries to fall into. */ | |
100 | ||
101 | if (sizeof (longword) != 4 && sizeof (longword) != 8) | |
102 | abort (); | |
103 | ||
104 | #if LONG_MAX <= LONG_MAX_32_BITS | |
105 | magic_bits = 0x7efefeff; | |
106 | #else | |
107 | magic_bits = ((unsigned long int) 0x7efefefe << 32) | 0xfefefeff; | |
108 | #endif | |
109 | ||
110 | /* Set up a longword, each of whose bytes is C. */ | |
111 | charmask = c | (c << 8); | |
112 | charmask |= charmask << 16; | |
113 | #if LONG_MAX > LONG_MAX_32_BITS | |
114 | charmask |= charmask << 32; | |
115 | #endif | |
116 | ||
117 | /* Instead of the traditional loop which tests each character, | |
118 | we will test a longword at a time. The tricky part is testing | |
119 | if *any of the four* bytes in the longword in question are zero. */ | |
120 | while (n >= sizeof (longword)) | |
121 | { | |
122 | /* We tentatively exit the loop if adding MAGIC_BITS to | |
123 | LONGWORD fails to change any of the hole bits of LONGWORD. | |
124 | ||
125 | 1) Is this safe? Will it catch all the zero bytes? | |
126 | Suppose there is a byte with all zeros. Any carry bits | |
127 | propagating from its left will fall into the hole at its | |
128 | least significant bit and stop. Since there will be no | |
129 | carry from its most significant bit, the LSB of the | |
130 | byte to the left will be unchanged, and the zero will be | |
131 | detected. | |
132 | ||
133 | 2) Is this worthwhile? Will it ignore everything except | |
134 | zero bytes? Suppose every byte of LONGWORD has a bit set | |
135 | somewhere. There will be a carry into bit 8. If bit 8 | |
136 | is set, this will carry into bit 16. If bit 8 is clear, | |
137 | one of bits 9-15 must be set, so there will be a carry | |
138 | into bit 16. Similarly, there will be a carry into bit | |
139 | 24. If one of bits 24-30 is set, there will be a carry | |
140 | into bit 31, so all of the hole bits will be changed. | |
141 | ||
142 | The one misfire occurs when bits 24-30 are clear and bit | |
143 | 31 is set; in this case, the hole at bit 31 is not | |
144 | changed. If we had access to the processor carry flag, | |
145 | we could close this loophole by putting the fourth hole | |
146 | at bit 32! | |
147 | ||
148 | So it ignores everything except 128's, when they're aligned | |
149 | properly. | |
150 | ||
151 | 3) But wait! Aren't we looking for C, not zero? | |
152 | Good point. So what we do is XOR LONGWORD with a longword, | |
153 | each of whose bytes is C. This turns each byte that is C | |
154 | into a zero. */ | |
155 | ||
156 | longword = *--longword_ptr ^ charmask; | |
157 | ||
158 | /* Add MAGIC_BITS to LONGWORD. */ | |
159 | if ((((longword + magic_bits) | |
160 | ||
161 | /* Set those bits that were unchanged by the addition. */ | |
162 | ^ ~longword) | |
163 | ||
164 | /* Look at only the hole bits. If any of the hole bits | |
165 | are unchanged, most likely one of the bytes was a | |
166 | zero. */ | |
167 | & ~magic_bits) != 0) | |
168 | { | |
169 | /* Which of the bytes was C? If none of them were, it was | |
170 | a misfire; continue the search. */ | |
171 | ||
172 | const unsigned char *cp = (const unsigned char *) longword_ptr; | |
173 | ||
174 | #if LONG_MAX > 2147483647 | |
175 | if (cp[7] == c) | |
176 | return (__ptr_t) &cp[7]; | |
177 | if (cp[6] == c) | |
178 | return (__ptr_t) &cp[6]; | |
179 | if (cp[5] == c) | |
180 | return (__ptr_t) &cp[5]; | |
181 | if (cp[4] == c) | |
182 | return (__ptr_t) &cp[4]; | |
183 | #endif | |
184 | if (cp[3] == c) | |
185 | return (__ptr_t) &cp[3]; | |
186 | if (cp[2] == c) | |
187 | return (__ptr_t) &cp[2]; | |
188 | if (cp[1] == c) | |
189 | return (__ptr_t) &cp[1]; | |
190 | if (cp[0] == c) | |
191 | return (__ptr_t) cp; | |
192 | } | |
193 | ||
194 | n -= sizeof (longword); | |
195 | } | |
196 | ||
197 | char_ptr = (const unsigned char *) longword_ptr; | |
198 | ||
199 | while (n-- > 0) | |
200 | { | |
201 | if (*--char_ptr == c) | |
202 | return (__ptr_t) char_ptr; | |
203 | } | |
204 | ||
205 | return 0; | |
206 | } | |
207 | #ifdef weak_alias | |
208 | weak_alias (__memrchr, memrchr) | |
209 | #endif |