1/* Copyright (C) 1991, 1993, 1996, 1997, 1999, 2000, 2003, 2004 Free
2   Software Foundation, Inc.
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 by Roland McGrath (roland@ai.mit.edu).
9
10NOTE: The canonical source of this file is maintained with the GNU C Library.
11Bugs can be reported to bug-glibc@prep.ai.mit.edu.
12
13This program is free software; you can redistribute it and/or modify it
14under the terms of the GNU General Public License as published by the
15Free Software Foundation; either version 2, or (at your option) any
16later version.
17
18This program is distributed in the hope that it will be useful,
19but WITHOUT ANY WARRANTY; without even the implied warranty of
20MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
21GNU General Public License for more details.
22
23You should have received a copy of the GNU General Public License
24along with this program; if not, write to the Free Software Foundation,
25Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.  */
26#include <sys/cdefs.h>
27__RCSID("$NetBSD: memchr.c,v 1.2 2016/05/17 14:00:09 christos Exp $");
28
29
30#ifdef HAVE_CONFIG_H
31# include <config.h>
32#endif
33
34#include <string.h>
35
36#include <stddef.h>
37
38#if defined _LIBC
39# include <memcopy.h>
40#else
41# define reg_char char
42#endif
43
44#include <limits.h>
45
46#if HAVE_BP_SYM_H || defined _LIBC
47# include <bp-sym.h>
48#else
49# define BP_SYM(sym) sym
50#endif
51
52#undef memchr
53#undef __memchr
54
55/* Search no more than N bytes of S for C.  */
56void *
57__memchr (void const *s, int c_in, size_t n)
58{
59  const unsigned char *char_ptr;
60  const unsigned long int *longword_ptr;
61  unsigned long int longword, magic_bits, charmask;
62  unsigned reg_char c;
63  int i;
64
65  c = (unsigned char) c_in;
66
67  /* Handle the first few characters by reading one character at a time.
68     Do this until CHAR_PTR is aligned on a longword boundary.  */
69  for (char_ptr = (const unsigned char *) s;
70       n > 0 && (size_t) char_ptr % sizeof longword != 0;
71       --n, ++char_ptr)
72    if (*char_ptr == c)
73      return (void *) char_ptr;
74
75  /* All these elucidatory comments refer to 4-byte longwords,
76     but the theory applies equally well to any size longwords.  */
77
78  longword_ptr = (const unsigned long int *) char_ptr;
79
80  /* Bits 31, 24, 16, and 8 of this number are zero.  Call these bits
81     the "holes."  Note that there is a hole just to the left of
82     each byte, with an extra at the end:
83
84     bits:  01111110 11111110 11111110 11111111
85     bytes: AAAAAAAA BBBBBBBB CCCCCCCC DDDDDDDD
86
87     The 1-bits make sure that carries propagate to the next 0-bit.
88     The 0-bits provide holes for carries to fall into.  */
89
90  /* Set MAGIC_BITS to be this pattern of 1 and 0 bits.
91     Set CHARMASK to be a longword, each of whose bytes is C.  */
92
93  magic_bits = 0xfefefefe;
94  charmask = c | (c << 8);
95  charmask |= charmask << 16;
96#if 0xffffffffU < ULONG_MAX
97  magic_bits |= magic_bits << 32;
98  charmask |= charmask << 32;
99  if (8 < sizeof longword)
100    for (i = 64; i < sizeof longword * 8; i *= 2)
101      {
102	magic_bits |= magic_bits << i;
103	charmask |= charmask << i;
104      }
105#endif
106  magic_bits = (ULONG_MAX >> 1) & (magic_bits | 1);
107
108  /* Instead of the traditional loop which tests each character,
109     we will test a longword at a time.  The tricky part is testing
110     if *any of the four* bytes in the longword in question are zero.  */
111  while (n >= sizeof longword)
112    {
113      /* We tentatively exit the loop if adding MAGIC_BITS to
114	 LONGWORD fails to change any of the hole bits of LONGWORD.
115
116	 1) Is this safe?  Will it catch all the zero bytes?
117	 Suppose there is a byte with all zeros.  Any carry bits
118	 propagating from its left will fall into the hole at its
119	 least significant bit and stop.  Since there will be no
120	 carry from its most significant bit, the LSB of the
121	 byte to the left will be unchanged, and the zero will be
122	 detected.
123
124	 2) Is this worthwhile?  Will it ignore everything except
125	 zero bytes?  Suppose every byte of LONGWORD has a bit set
126	 somewhere.  There will be a carry into bit 8.  If bit 8
127	 is set, this will carry into bit 16.  If bit 8 is clear,
128	 one of bits 9-15 must be set, so there will be a carry
129	 into bit 16.  Similarly, there will be a carry into bit
130	 24.  If one of bits 24-30 is set, there will be a carry
131	 into bit 31, so all of the hole bits will be changed.
132
133	 The one misfire occurs when bits 24-30 are clear and bit
134	 31 is set; in this case, the hole at bit 31 is not
135	 changed.  If we had access to the processor carry flag,
136	 we could close this loophole by putting the fourth hole
137	 at bit 32!
138
139	 So it ignores everything except 128's, when they're aligned
140	 properly.
141
142	 3) But wait!  Aren't we looking for C, not zero?
143	 Good point.  So what we do is XOR LONGWORD with a longword,
144	 each of whose bytes is C.  This turns each byte that is C
145	 into a zero.  */
146
147      longword = *longword_ptr++ ^ charmask;
148
149      /* Add MAGIC_BITS to LONGWORD.  */
150      if ((((longword + magic_bits)
151
152	    /* Set those bits that were unchanged by the addition.  */
153	    ^ ~longword)
154
155	   /* Look at only the hole bits.  If any of the hole bits
156	      are unchanged, most likely one of the bytes was a
157	      zero.  */
158	   & ~magic_bits) != 0)
159	{
160	  /* Which of the bytes was C?  If none of them were, it was
161	     a misfire; continue the search.  */
162
163	  const unsigned char *cp = (const unsigned char *) (longword_ptr - 1);
164
165	  if (cp[0] == c)
166	    return (void *) cp;
167	  if (cp[1] == c)
168	    return (void *) &cp[1];
169	  if (cp[2] == c)
170	    return (void *) &cp[2];
171	  if (cp[3] == c)
172	    return (void *) &cp[3];
173	  if (4 < sizeof longword && cp[4] == c)
174	    return (void *) &cp[4];
175	  if (5 < sizeof longword && cp[5] == c)
176	    return (void *) &cp[5];
177	  if (6 < sizeof longword && cp[6] == c)
178	    return (void *) &cp[6];
179	  if (7 < sizeof longword && cp[7] == c)
180	    return (void *) &cp[7];
181	  if (8 < sizeof longword)
182	    for (i = 8; i < sizeof longword; i++)
183	      if (cp[i] == c)
184		return (void *) &cp[i];
185	}
186
187      n -= sizeof longword;
188    }
189
190  char_ptr = (const unsigned char *) longword_ptr;
191
192  while (n-- > 0)
193    {
194      if (*char_ptr == c)
195	return (void *) char_ptr;
196      else
197	++char_ptr;
198    }
199
200  return 0;
201}
202#ifdef weak_alias
203weak_alias (__memchr, BP_SYM (memchr))
204#endif
205