Mercurial > hg > octave-nkf > gnulib-hg
annotate lib/memchr2.c @ 14610:b427a1938336
use _GL_ATTRIBUTE_CONST and _GL_ATTRIBUTE_PURE
author | Jim Meyering <meyering@redhat.com> |
---|---|
date | Sun, 24 Apr 2011 19:02:10 +0200 |
parents | 97fc9a21a8fb |
children | 6ef4f1f39105 |
rev | line source |
---|---|
14079
97fc9a21a8fb
maint: update almost all copyright ranges to include 2011
Jim Meyering <meyering@redhat.com>
parents:
12559
diff
changeset
|
1 /* Copyright (C) 1991, 1993, 1996-1997, 1999-2000, 2003-2004, 2006, 2008-2011 |
97fc9a21a8fb
maint: update almost all copyright ranges to include 2011
Jim Meyering <meyering@redhat.com>
parents:
12559
diff
changeset
|
2 Free Software Foundation, Inc. |
9742 | 3 |
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 in glibc by Roland McGrath (roland@ai.mit.edu). | |
9 Extension to memchr2 implemented by Eric Blake (ebb9@byu.net). | |
10 | |
11 This program is free software: you can redistribute it and/or modify it | |
12 under the terms of the GNU General Public License as published by the | |
13 Free Software Foundation; either version 3 of the License, or any | |
14 later version. | |
15 | |
16 This program is distributed in the hope that it will be useful, | |
17 but WITHOUT ANY WARRANTY; without even the implied warranty of | |
18 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
19 GNU General Public License for more details. | |
20 | |
21 You should have received a copy of the GNU General Public License | |
22 along with this program. If not, see <http://www.gnu.org/licenses/>. */ | |
23 | |
24 #include <config.h> | |
25 | |
26 #include "memchr2.h" | |
27 | |
28 #include <limits.h> | |
29 #include <stdint.h> | |
30 #include <string.h> | |
31 | |
14610
b427a1938336
use _GL_ATTRIBUTE_CONST and _GL_ATTRIBUTE_PURE
Jim Meyering <meyering@redhat.com>
parents:
14079
diff
changeset
|
32 /* The attribute __pure__ was added in gcc 2.96. */ |
b427a1938336
use _GL_ATTRIBUTE_CONST and _GL_ATTRIBUTE_PURE
Jim Meyering <meyering@redhat.com>
parents:
14079
diff
changeset
|
33 #undef _GL_ATTRIBUTE_PURE |
b427a1938336
use _GL_ATTRIBUTE_CONST and _GL_ATTRIBUTE_PURE
Jim Meyering <meyering@redhat.com>
parents:
14079
diff
changeset
|
34 #if __GNUC__ > 2 || (__GNUC__ == 2 && __GNUC_MINOR__ >= 96) |
b427a1938336
use _GL_ATTRIBUTE_CONST and _GL_ATTRIBUTE_PURE
Jim Meyering <meyering@redhat.com>
parents:
14079
diff
changeset
|
35 # define _GL_ATTRIBUTE_PURE __attribute__ ((__pure__)) |
b427a1938336
use _GL_ATTRIBUTE_CONST and _GL_ATTRIBUTE_PURE
Jim Meyering <meyering@redhat.com>
parents:
14079
diff
changeset
|
36 #else |
b427a1938336
use _GL_ATTRIBUTE_CONST and _GL_ATTRIBUTE_PURE
Jim Meyering <meyering@redhat.com>
parents:
14079
diff
changeset
|
37 # define _GL_ATTRIBUTE_PURE /* empty */ |
b427a1938336
use _GL_ATTRIBUTE_CONST and _GL_ATTRIBUTE_PURE
Jim Meyering <meyering@redhat.com>
parents:
14079
diff
changeset
|
38 #endif |
b427a1938336
use _GL_ATTRIBUTE_CONST and _GL_ATTRIBUTE_PURE
Jim Meyering <meyering@redhat.com>
parents:
14079
diff
changeset
|
39 |
9742 | 40 /* Return the first address of either C1 or C2 (treated as unsigned |
41 char) that occurs within N bytes of the memory region S. If | |
42 neither byte appears, return NULL. */ | |
14610
b427a1938336
use _GL_ATTRIBUTE_CONST and _GL_ATTRIBUTE_PURE
Jim Meyering <meyering@redhat.com>
parents:
14079
diff
changeset
|
43 void * _GL_ATTRIBUTE_PURE |
9742 | 44 memchr2 (void const *s, int c1_in, int c2_in, size_t n) |
45 { | |
9967 | 46 /* On 32-bit hardware, choosing longword to be a 32-bit unsigned |
47 long instead of a 64-bit uintmax_t tends to give better | |
48 performance. On 64-bit hardware, unsigned long is generally 64 | |
49 bits already. Change this typedef to experiment with | |
50 performance. */ | |
9995 | 51 typedef unsigned long int longword; |
9967 | 52 |
9742 | 53 const unsigned char *char_ptr; |
9967 | 54 const longword *longword_ptr; |
9971
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
55 longword repeated_one; |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
56 longword repeated_c1; |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
57 longword repeated_c2; |
9742 | 58 unsigned char c1; |
59 unsigned char c2; | |
60 | |
61 c1 = (unsigned char) c1_in; | |
62 c2 = (unsigned char) c2_in; | |
63 | |
64 if (c1 == c2) | |
65 return memchr (s, c1, n); | |
66 | |
9971
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
67 /* Handle the first few bytes by reading one byte at a time. |
9742 | 68 Do this until CHAR_PTR is aligned on a longword boundary. */ |
69 for (char_ptr = (const unsigned char *) s; | |
9971
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
70 n > 0 && (size_t) char_ptr % sizeof (longword) != 0; |
9742 | 71 --n, ++char_ptr) |
72 if (*char_ptr == c1 || *char_ptr == c2) | |
73 return (void *) char_ptr; | |
74 | |
9971
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
75 longword_ptr = (const longword *) char_ptr; |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
76 |
9742 | 77 /* All these elucidatory comments refer to 4-byte longwords, |
78 but the theory applies equally well to any size longwords. */ | |
79 | |
9971
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
80 /* Compute auxiliary longword values: |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
81 repeated_one is a value which has a 1 in every byte. |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
82 repeated_c1 has c1 in every byte. |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
83 repeated_c2 has c2 in every byte. */ |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
84 repeated_one = 0x01010101; |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
85 repeated_c1 = c1 | (c1 << 8); |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
86 repeated_c2 = c2 | (c2 << 8); |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
87 repeated_c1 |= repeated_c1 << 16; |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
88 repeated_c2 |= repeated_c2 << 16; |
9995 | 89 if (0xffffffffU < (longword) -1) |
9951
03b212920228
Work around preprocessors that don't handle UINTMAX_MAX.
Eric Blake <ebb9@byu.net>
parents:
9745
diff
changeset
|
90 { |
9971
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
91 repeated_one |= repeated_one << 31 << 1; |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
92 repeated_c1 |= repeated_c1 << 31 << 1; |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
93 repeated_c2 |= repeated_c2 << 31 << 1; |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
94 if (8 < sizeof (longword)) |
12421
e8d2c6fc33ad
Use spaces for indentation, not tabs.
Bruno Haible <bruno@clisp.org>
parents:
9995
diff
changeset
|
95 { |
e8d2c6fc33ad
Use spaces for indentation, not tabs.
Bruno Haible <bruno@clisp.org>
parents:
9995
diff
changeset
|
96 size_t i; |
9971
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
97 |
12421
e8d2c6fc33ad
Use spaces for indentation, not tabs.
Bruno Haible <bruno@clisp.org>
parents:
9995
diff
changeset
|
98 for (i = 64; i < sizeof (longword) * 8; i *= 2) |
e8d2c6fc33ad
Use spaces for indentation, not tabs.
Bruno Haible <bruno@clisp.org>
parents:
9995
diff
changeset
|
99 { |
e8d2c6fc33ad
Use spaces for indentation, not tabs.
Bruno Haible <bruno@clisp.org>
parents:
9995
diff
changeset
|
100 repeated_one |= repeated_one << i; |
e8d2c6fc33ad
Use spaces for indentation, not tabs.
Bruno Haible <bruno@clisp.org>
parents:
9995
diff
changeset
|
101 repeated_c1 |= repeated_c1 << i; |
e8d2c6fc33ad
Use spaces for indentation, not tabs.
Bruno Haible <bruno@clisp.org>
parents:
9995
diff
changeset
|
102 repeated_c2 |= repeated_c2 << i; |
e8d2c6fc33ad
Use spaces for indentation, not tabs.
Bruno Haible <bruno@clisp.org>
parents:
9995
diff
changeset
|
103 } |
e8d2c6fc33ad
Use spaces for indentation, not tabs.
Bruno Haible <bruno@clisp.org>
parents:
9995
diff
changeset
|
104 } |
9951
03b212920228
Work around preprocessors that don't handle UINTMAX_MAX.
Eric Blake <ebb9@byu.net>
parents:
9745
diff
changeset
|
105 } |
9742 | 106 |
9971
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
107 /* Instead of the traditional loop which tests each byte, we will test a |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
108 longword at a time. The tricky part is testing if *any of the four* |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
109 bytes in the longword in question are equal to c1 or c2. We first use |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
110 an xor with repeated_c1 and repeated_c2, respectively. This reduces |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
111 the task to testing whether *any of the four* bytes in longword1 or |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
112 longword2 is zero. |
9967 | 113 |
9971
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
114 Let's consider longword1. We compute tmp1 = |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
115 ((longword1 - repeated_one) & ~longword1) & (repeated_one << 7). |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
116 That is, we perform the following operations: |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
117 1. Subtract repeated_one. |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
118 2. & ~longword1. |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
119 3. & a mask consisting of 0x80 in every byte. |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
120 Consider what happens in each byte: |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
121 - If a byte of longword1 is zero, step 1 and 2 transform it into 0xff, |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
122 and step 3 transforms it into 0x80. A carry can also be propagated |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
123 to more significant bytes. |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
124 - If a byte of longword1 is nonzero, let its lowest 1 bit be at |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
125 position k (0 <= k <= 7); so the lowest k bits are 0. After step 1, |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
126 the byte ends in a single bit of value 0 and k bits of value 1. |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
127 After step 2, the result is just k bits of value 1: 2^k - 1. After |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
128 step 3, the result is 0. And no carry is produced. |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
129 So, if longword1 has only non-zero bytes, tmp1 is zero. |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
130 Whereas if longword1 has a zero byte, call j the position of the least |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
131 significant zero byte. Then the result has a zero at positions 0, ..., |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
132 j-1 and a 0x80 at position j. We cannot predict the result at the more |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
133 significant bytes (positions j+1..3), but it does not matter since we |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
134 already have a non-zero bit at position 8*j+7. |
9967 | 135 |
9971
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
136 Similary, we compute tmp2 = |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
137 ((longword2 - repeated_one) & ~longword2) & (repeated_one << 7). |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
138 |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
139 The test whether any byte in longword1 or longword2 is zero is equivalent |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
140 to testing whether tmp1 is nonzero or tmp2 is nonzero. We can combine |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
141 this into a single test, whether (tmp1 | tmp2) is nonzero. */ |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
142 |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
143 while (n >= sizeof (longword)) |
9742 | 144 { |
9971
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
145 longword longword1 = *longword_ptr ^ repeated_c1; |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
146 longword longword2 = *longword_ptr ^ repeated_c2; |
9742 | 147 |
9971
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
148 if (((((longword1 - repeated_one) & ~longword1) |
12421
e8d2c6fc33ad
Use spaces for indentation, not tabs.
Bruno Haible <bruno@clisp.org>
parents:
9995
diff
changeset
|
149 | ((longword2 - repeated_one) & ~longword2)) |
e8d2c6fc33ad
Use spaces for indentation, not tabs.
Bruno Haible <bruno@clisp.org>
parents:
9995
diff
changeset
|
150 & (repeated_one << 7)) != 0) |
e8d2c6fc33ad
Use spaces for indentation, not tabs.
Bruno Haible <bruno@clisp.org>
parents:
9995
diff
changeset
|
151 break; |
9967 | 152 longword_ptr++; |
9971
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
153 n -= sizeof (longword); |
9742 | 154 } |
155 | |
156 char_ptr = (const unsigned char *) longword_ptr; | |
157 | |
9971
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
158 /* At this point, we know that either n < sizeof (longword), or one of the |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
159 sizeof (longword) bytes starting at char_ptr is == c1 or == c2. On |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
160 little-endian machines, we could determine the first such byte without |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
161 any further memory accesses, just by looking at the (tmp1 | tmp2) result |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
162 from the last loop iteration. But this does not work on big-endian |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
163 machines. Choose code that works in both cases. */ |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
164 |
e70f61b955b9
Superficial improvements and comments.
Bruno Haible <bruno@clisp.org>
parents:
9967
diff
changeset
|
165 for (; n > 0; --n, ++char_ptr) |
9742 | 166 { |
167 if (*char_ptr == c1 || *char_ptr == c2) | |
12421
e8d2c6fc33ad
Use spaces for indentation, not tabs.
Bruno Haible <bruno@clisp.org>
parents:
9995
diff
changeset
|
168 return (void *) char_ptr; |
9742 | 169 } |
170 | |
9967 | 171 return NULL; |
9742 | 172 } |