1 | /* memrchr -- find the last occurrence of a byte in a memory block |
2 | Copyright (C) 1991-2022 Free Software Foundation, Inc. |
3 | This file is part of the GNU C Library. |
4 | Based on strlen implementation by Torbjorn Granlund (tege@sics.se), |
5 | with help from Dan Sahlin (dan@sics.se) and |
6 | commentary by Jim Blandy (jimb@ai.mit.edu); |
7 | adaptation to memchr suggested by Dick Karpinski (dick@cca.ucsf.edu), |
8 | and implemented by Roland McGrath (roland@ai.mit.edu). |
9 | |
10 | The GNU C Library is free software; you can redistribute it and/or |
11 | modify it under the terms of the GNU Lesser General Public |
12 | License as published by the Free Software Foundation; either |
13 | version 2.1 of the License, or (at your option) any later version. |
14 | |
15 | The GNU C Library is distributed in the hope that it will be useful, |
16 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
17 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
18 | Lesser General Public License for more details. |
19 | |
20 | You should have received a copy of the GNU Lesser General Public |
21 | License along with the GNU C Library; if not, see |
22 | <https://www.gnu.org/licenses/>. */ |
23 | |
24 | #include <stdlib.h> |
25 | |
26 | #ifdef HAVE_CONFIG_H |
27 | # include <config.h> |
28 | #endif |
29 | |
30 | #if defined _LIBC |
31 | # include <string.h> |
32 | # include <memcopy.h> |
33 | #endif |
34 | |
35 | #if defined HAVE_LIMITS_H || defined _LIBC |
36 | # include <limits.h> |
37 | #endif |
38 | |
39 | #define LONG_MAX_32_BITS 2147483647 |
40 | |
41 | #ifndef LONG_MAX |
42 | # define LONG_MAX LONG_MAX_32_BITS |
43 | #endif |
44 | |
45 | #include <sys/types.h> |
46 | |
47 | #undef __memrchr |
48 | #undef memrchr |
49 | |
50 | #ifndef weak_alias |
51 | # define __memrchr memrchr |
52 | #endif |
53 | |
54 | /* Search no more than N bytes of S for C. */ |
55 | void * |
56 | #ifndef MEMRCHR |
57 | __memrchr |
58 | #else |
59 | MEMRCHR |
60 | #endif |
61 | (const void *s, int c_in, size_t n) |
62 | { |
63 | const unsigned char *char_ptr; |
64 | const unsigned long int *longword_ptr; |
65 | unsigned long int longword, magic_bits, charmask; |
66 | unsigned char c; |
67 | |
68 | c = (unsigned char) c_in; |
69 | |
70 | /* Handle the last few characters by reading one character at a time. |
71 | Do this until CHAR_PTR is aligned on a longword boundary. */ |
72 | for (char_ptr = (const unsigned char *) s + n; |
73 | n > 0 && ((unsigned long int) char_ptr |
74 | & (sizeof (longword) - 1)) != 0; |
75 | --n) |
76 | if (*--char_ptr == c) |
77 | return (void *) char_ptr; |
78 | |
79 | /* All these elucidatory comments refer to 4-byte longwords, |
80 | but the theory applies equally well to 8-byte longwords. */ |
81 | |
82 | longword_ptr = (const unsigned long int *) char_ptr; |
83 | |
84 | /* Bits 31, 24, 16, and 8 of this number are zero. Call these bits |
85 | the "holes." Note that there is a hole just to the left of |
86 | each byte, with an extra at the end: |
87 | |
88 | bits: 01111110 11111110 11111110 11111111 |
89 | bytes: AAAAAAAA BBBBBBBB CCCCCCCC DDDDDDDD |
90 | |
91 | The 1-bits make sure that carries propagate to the next 0-bit. |
92 | The 0-bits provide holes for carries to fall into. */ |
93 | magic_bits = -1; |
94 | magic_bits = magic_bits / 0xff * 0xfe << 1 >> 1 | 1; |
95 | |
96 | /* Set up a longword, each of whose bytes is C. */ |
97 | charmask = c | (c << 8); |
98 | charmask |= charmask << 16; |
99 | #if LONG_MAX > LONG_MAX_32_BITS |
100 | charmask |= charmask << 32; |
101 | #endif |
102 | |
103 | /* Instead of the traditional loop which tests each character, |
104 | we will test a longword at a time. The tricky part is testing |
105 | if *any of the four* bytes in the longword in question are zero. */ |
106 | while (n >= sizeof (longword)) |
107 | { |
108 | /* We tentatively exit the loop if adding MAGIC_BITS to |
109 | LONGWORD fails to change any of the hole bits of LONGWORD. |
110 | |
111 | 1) Is this safe? Will it catch all the zero bytes? |
112 | Suppose there is a byte with all zeros. Any carry bits |
113 | propagating from its left will fall into the hole at its |
114 | least significant bit and stop. Since there will be no |
115 | carry from its most significant bit, the LSB of the |
116 | byte to the left will be unchanged, and the zero will be |
117 | detected. |
118 | |
119 | 2) Is this worthwhile? Will it ignore everything except |
120 | zero bytes? Suppose every byte of LONGWORD has a bit set |
121 | somewhere. There will be a carry into bit 8. If bit 8 |
122 | is set, this will carry into bit 16. If bit 8 is clear, |
123 | one of bits 9-15 must be set, so there will be a carry |
124 | into bit 16. Similarly, there will be a carry into bit |
125 | 24. If one of bits 24-30 is set, there will be a carry |
126 | into bit 31, so all of the hole bits will be changed. |
127 | |
128 | The one misfire occurs when bits 24-30 are clear and bit |
129 | 31 is set; in this case, the hole at bit 31 is not |
130 | changed. If we had access to the processor carry flag, |
131 | we could close this loophole by putting the fourth hole |
132 | at bit 32! |
133 | |
134 | So it ignores everything except 128's, when they're aligned |
135 | properly. |
136 | |
137 | 3) But wait! Aren't we looking for C, not zero? |
138 | Good point. So what we do is XOR LONGWORD with a longword, |
139 | each of whose bytes is C. This turns each byte that is C |
140 | into a zero. */ |
141 | |
142 | longword = *--longword_ptr ^ charmask; |
143 | |
144 | /* Add MAGIC_BITS to LONGWORD. */ |
145 | if ((((longword + magic_bits) |
146 | |
147 | /* Set those bits that were unchanged by the addition. */ |
148 | ^ ~longword) |
149 | |
150 | /* Look at only the hole bits. If any of the hole bits |
151 | are unchanged, most likely one of the bytes was a |
152 | zero. */ |
153 | & ~magic_bits) != 0) |
154 | { |
155 | /* Which of the bytes was C? If none of them were, it was |
156 | a misfire; continue the search. */ |
157 | |
158 | const unsigned char *cp = (const unsigned char *) longword_ptr; |
159 | |
160 | #if LONG_MAX > 2147483647 |
161 | if (cp[7] == c) |
162 | return (void *) &cp[7]; |
163 | if (cp[6] == c) |
164 | return (void *) &cp[6]; |
165 | if (cp[5] == c) |
166 | return (void *) &cp[5]; |
167 | if (cp[4] == c) |
168 | return (void *) &cp[4]; |
169 | #endif |
170 | if (cp[3] == c) |
171 | return (void *) &cp[3]; |
172 | if (cp[2] == c) |
173 | return (void *) &cp[2]; |
174 | if (cp[1] == c) |
175 | return (void *) &cp[1]; |
176 | if (cp[0] == c) |
177 | return (void *) cp; |
178 | } |
179 | |
180 | n -= sizeof (longword); |
181 | } |
182 | |
183 | char_ptr = (const unsigned char *) longword_ptr; |
184 | |
185 | while (n-- > 0) |
186 | { |
187 | if (*--char_ptr == c) |
188 | return (void *) char_ptr; |
189 | } |
190 | |
191 | return 0; |
192 | } |
193 | #ifndef MEMRCHR |
194 | # ifdef weak_alias |
195 | weak_alias (__memrchr, memrchr) |
196 | # endif |
197 | #endif |
198 | |