initial import
[glibc.git] / sysdeps / generic / memchr.c
blobd17f9c649ecff96cf5fc4681512cb5c918c72421
1 /* Copyright (C) 1991, 1993 Free Software Foundation, Inc.
2 Based on strlen implemention by Torbjorn Granlund (tege@sics.se),
3 with help from Dan Sahlin (dan@sics.se) and
4 commentary by Jim Blandy (jimb@ai.mit.edu);
5 adaptation to memchr suggested by Dick Karpinski (dick@cca.ucsf.edu),
6 and implemented by Roland McGrath (roland@ai.mit.edu).
8 The GNU C Library is free software; you can redistribute it and/or
9 modify it under the terms of the GNU Library General Public License as
10 published by the Free Software Foundation; either version 2 of the
11 License, or (at your option) any later version.
13 The GNU C Library is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 Library General Public License for more details.
18 You should have received a copy of the GNU Library General Public
19 License along with the GNU C Library; see the file COPYING.LIB. If
20 not, write to the Free Software Foundation, Inc., 675 Mass Ave,
21 Cambridge, MA 02139, USA. */
23 #include <ansidecl.h>
24 #include <string.h>
27 /* Search no more than N bytes of S for C. */
29 PTR
30 DEFUN(memchr, (s, c, n), CONST PTR s AND int c AND size_t n)
32 CONST unsigned char *char_ptr;
33 CONST unsigned long int *longword_ptr;
34 unsigned long int longword, magic_bits, charmask;
36 c = (unsigned char) c;
38 /* Handle the first few characters by reading one character at a time.
39 Do this until CHAR_PTR is aligned on a longword boundary. */
40 for (char_ptr = s; n > 0 && ((unsigned long int) char_ptr
41 & (sizeof (longword) - 1)) != 0;
42 --n, ++char_ptr)
43 if (*char_ptr == c)
44 return (PTR) char_ptr;
46 /* All these elucidatory comments refer to 4-byte longwords,
47 but the theory applies equally well to 8-byte longwords. */
49 longword_ptr = (unsigned long int *) char_ptr;
51 /* Bits 31, 24, 16, and 8 of this number are zero. Call these bits
52 the "holes." Note that there is a hole just to the left of
53 each byte, with an extra at the end:
55 bits: 01111110 11111110 11111110 11111111
56 bytes: AAAAAAAA BBBBBBBB CCCCCCCC DDDDDDDD
58 The 1-bits make sure that carries propagate to the next 0-bit.
59 The 0-bits provide holes for carries to fall into. */
60 switch (sizeof (longword))
62 case 4: magic_bits = 0x7efefeffL; break;
63 case 8: magic_bits = (0x7efefefeL << 32) | 0xfefefeffL; break;
64 default:
65 abort ();
68 /* Set up a longword, each of whose bytes is C. */
69 charmask = c | (c << 8);
70 charmask |= charmask << 16;
71 if (sizeof (longword) > 4)
72 charmask |= charmask << 32;
73 if (sizeof (longword) > 8)
74 abort ();
76 /* Instead of the traditional loop which tests each character,
77 we will test a longword at a time. The tricky part is testing
78 if *any of the four* bytes in the longword in question are zero. */
79 while (n >= sizeof (longword))
81 /* We tentatively exit the loop if adding MAGIC_BITS to
82 LONGWORD fails to change any of the hole bits of LONGWORD.
84 1) Is this safe? Will it catch all the zero bytes?
85 Suppose there is a byte with all zeros. Any carry bits
86 propagating from its left will fall into the hole at its
87 least significant bit and stop. Since there will be no
88 carry from its most significant bit, the LSB of the
89 byte to the left will be unchanged, and the zero will be
90 detected.
92 2) Is this worthwhile? Will it ignore everything except
93 zero bytes? Suppose every byte of LONGWORD has a bit set
94 somewhere. There will be a carry into bit 8. If bit 8
95 is set, this will carry into bit 16. If bit 8 is clear,
96 one of bits 9-15 must be set, so there will be a carry
97 into bit 16. Similarly, there will be a carry into bit
98 24. If one of bits 24-30 is set, there will be a carry
99 into bit 31, so all of the hole bits will be changed.
101 The one misfire occurs when bits 24-30 are clear and bit
102 31 is set; in this case, the hole at bit 31 is not
103 changed. If we had access to the processor carry flag,
104 we could close this loophole by putting the fourth hole
105 at bit 32!
107 So it ignores everything except 128's, when they're aligned
108 properly.
110 3) But wait! Aren't we looking for C, not zero?
111 Good point. So what we do is XOR LONGWORD with a longword,
112 each of whose bytes is C. This turns each byte that is C
113 into a zero. */
115 longword = *longword_ptr++ ^ charmask;
117 /* Add MAGIC_BITS to LONGWORD. */
118 if ((((longword + magic_bits)
120 /* Set those bits that were unchanged by the addition. */
121 ^ ~longword)
123 /* Look at only the hole bits. If any of the hole bits
124 are unchanged, most likely one of the bytes was a
125 zero. */
126 & ~magic_bits) != 0)
128 /* Which of the bytes was C? If none of them were, it was
129 a misfire; continue the search. */
131 CONST unsigned char *cp = (CONST unsigned char *) (longword_ptr - 1);
133 if (cp[0] == c)
134 return (PTR) cp;
135 if (cp[1] == c)
136 return (PTR) &cp[1];
137 if (cp[2] == c)
138 return (PTR) &cp[2];
139 if (cp[3] == c)
140 return (PTR) &cp[3];
141 if (sizeof (longword) > 4)
143 if (cp[4] == c)
144 return (PTR) &cp[4];
145 if (cp[5] == c)
146 return (PTR) &cp[5];
147 if (cp[6] == c)
148 return (PTR) &cp[6];
149 if (cp[7] == c)
150 return (PTR) &cp[7];
154 n -= sizeof (longword);
157 char_ptr = (CONST unsigned char *) longword_ptr;
159 while (n-- > 0)
161 if (*char_ptr == c)
162 return (PTR) char_ptr;
163 else
164 ++char_ptr;
167 return NULL;