Debian Codegraph @

lib/memrchr.c

raw
1 /* memrchr -- find the last occurrence of a byte in a memory block
2
3 Copyright (C) 1991, 1993, 1996-1997, 1999-2000, 2003-2026 Free Software
4 Foundation, Inc.
5
6 Based on strlen implementation by Torbjorn Granlund ([email protected]),
7 with help from Dan Sahlin ([email protected]) and
8 commentary by Jim Blandy ([email protected]);
9 adaptation to memchr suggested by Dick Karpinski ([email protected]),
10 and implemented by Roland McGrath ([email protected]).
11
12 This file is free software: you can redistribute it and/or modify
13 it under the terms of the GNU Lesser General Public License as
14 published by the Free Software Foundation, either version 3 of the
15 License, or (at your option) any later version.
16
17 This file is distributed in the hope that it will be useful,
18 but WITHOUT ANY WARRANTY; without even the implied warranty of
19 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
20 GNU Lesser General Public License for more details.
21
22 You should have received a copy of the GNU Lesser General Public License
23 along with this program. If not, see <https://www.gnu.org/licenses/>. */
24
25 #if defined _LIBC
26 # include <memcopy.h>
27 #else
28 # include <config.h>
29 # define reg_char char
30 #endif
31
32 #include <string.h>
33 #include <limits.h>
34
35 #undef __memrchr
36 #ifdef _LIBC
37 # undef memrchr
38 #endif
39
40 #ifndef weak_alias
41 # define __memrchr memrchr
42 #endif
43
44 /* Search no more than N bytes of S for C. */
45 void *
46 __memrchr (void const *s, int c_in, size_t n)
47 {
48 /* On 32-bit hardware, choosing longword to be a 32-bit unsigned
49 long instead of a 64-bit uintmax_t tends to give better
50 performance. On 64-bit hardware, unsigned long is generally 64
51 bits already. Change this typedef to experiment with
52 performance. */
53 typedef unsigned long int longword;
54
55 unsigned reg_char c = (unsigned char) c_in;
56
57 const longword *longword_ptr;
58
59 /* Handle the last few bytes by reading one byte at a time.
60 Do this until CHAR_PTR is aligned on a longword boundary. */
61 {
62 const unsigned char *char_ptr;
63 for (char_ptr = (const unsigned char *) s + n;
64 n > 0 && (size_t) char_ptr % sizeof (longword) != 0;
65 --n)
66 if (*--char_ptr == c)
67 return (void *) char_ptr;
68
69 longword_ptr = (const void *) char_ptr;
70 }
71
72 /* All these elucidatory comments refer to 4-byte longwords,
73 but the theory applies equally well to any size longwords. */
74 {
75 /* Compute auxiliary longword values:
76 repeated_one is a value which has a 1 in every byte.
77 repeated_c has c in every byte. */
78 longword repeated_one = 0x01010101;
79 longword repeated_c = c | (c << 8);
80 repeated_c |= repeated_c << 16;
81 if (0xffffffffU < (longword) -1)
82 {
83 repeated_one |= repeated_one << 31 << 1;
84 repeated_c |= repeated_c << 31 << 1;
85 if (8 < sizeof (longword))
86 for (size_t i = 64; i < sizeof (longword) * 8; i *= 2)
87 {
88 repeated_one |= repeated_one << i;
89 repeated_c |= repeated_c << i;
90 }
91 }
92
93 /* Instead of the traditional loop which tests each byte, we will test a
94 longword at a time. The tricky part is testing if *any of the four*
95 bytes in the longword in question are equal to c. We first use an xor
96 with repeated_c. This reduces the task to testing whether *any of the
97 four* bytes in longword1 is zero.
98
99 We compute tmp =
100 ((longword1 - repeated_one) & ~longword1) & (repeated_one << 7).
101 That is, we perform the following operations:
102 1. Subtract repeated_one.
103 2. & ~longword1.
104 3. & a mask consisting of 0x80 in every byte.
105 Consider what happens in each byte:
106 - If a byte of longword1 is zero, step 1 and 2 transform it into 0xff,
107 and step 3 transforms it into 0x80. A carry can also be propagated
108 to more significant bytes.
109 - If a byte of longword1 is nonzero, let its lowest 1 bit be at
110 position k (0 <= k <= 7); so the lowest k bits are 0. After step 1,
111 the byte ends in a single bit of value 0 and k bits of value 1.
112 After step 2, the result is just k bits of value 1: 2^k - 1. After
113 step 3, the result is 0. And no carry is produced.
114 So, if longword1 has only non-zero bytes, tmp is zero.
115 Whereas if longword1 has a zero byte, call j the position of the least
116 significant zero byte. Then the result has a zero at positions 0, ...,
117 j-1 and a 0x80 at position j. We cannot predict the result at the more
118 significant bytes (positions j+1..3), but it does not matter since we
119 already have a non-zero bit at position 8*j+7.
120
121 So, the test whether any byte in longword1 is zero is equivalent to
122 testing whether tmp is nonzero. */
123
124 while (n >= sizeof (longword))
125 {
126 longword longword1 = *--longword_ptr ^ repeated_c;
127
128 if ((((longword1 - repeated_one) & ~longword1)
129 & (repeated_one << 7)) != 0)
130 {
131 longword_ptr++;
132 break;
133 }
134 n -= sizeof (longword);
135 }
136 }
137
138 {
139 const unsigned char *char_ptr = (const unsigned char *) longword_ptr;
140
141 /* At this point, we know that either n < sizeof (longword), or one of the
142 sizeof (longword) bytes starting at char_ptr is == c. On little-endian
143 machines, we could determine the first such byte without any further
144 memory accesses, just by looking at the tmp result from the last loop
145 iteration. But this does not work on big-endian machines. Choose code
146 that works in both cases. */
147 while (n-- > 0)
148 {
149 if (*--char_ptr == c)
150 return (void *) char_ptr;
151 }
152 }
153
154 return NULL;
155 }
156 #ifdef weak_alias
157 weak_alias (__memrchr, memrchr)
158 #endif