lib/memchr.c

   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).
   7
   8 NOTE: The canonical source of this file is maintained with the GNU C Library.
   9 Bugs can be reported to bug-glibc@prep.ai.mit.edu.
  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 2, or (at your option) 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, write to the Free Software
  23 Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.  */
  24
  25 #ifdef HAVE_CONFIG_H
  26 #include <config.h>
  27 #endif
  28
  29 #undef __ptr_t
  30 #if defined (__cplusplus) || (defined (__STDC__) && __STDC__)
  31 # define __ptr_t void *
  32 #else /* Not C++ or ANSI C.  */
  33 # define __ptr_t char *
  34 #endif /* C++ or ANSI C.  */
  35
  36 #if defined (_LIBC)
  37 # include <string.h>
  38 #endif
  39
  40 #if defined (HAVE_LIMITS_H) || defined (_LIBC)
  41 # include <limits.h>
  42 #endif
  43
  44 #define LONG_MAX_32_BITS 2147483647
  45
  46 #ifndef LONG_MAX
  47 #define LONG_MAX LONG_MAX_32_BITS
  48 #endif
  49
  50 #include <sys/types.h>
  51
  52
  53 /* Search no more than N bytes of S for C.  */
  54
  55 __ptr_t
  56 memchr (s, c, n)
  57      const __ptr_t s;
  58      int c;
  59      size_t n;
  60 {
  61   const unsigned char *char_ptr;
  62   const unsigned long int *longword_ptr;
  63   unsigned long int longword, magic_bits, charmask;
  64
  65   c = (unsigned char) c;
  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 && ((unsigned long int) char_ptr
  71                  & (sizeof (longword) - 1)) != 0;
  72        --n, ++char_ptr)
  73     if (*char_ptr == c)
  74       return (__ptr_t) char_ptr;
  75
  76   /* All these elucidatory comments refer to 4-byte longwords,
  77      but the theory applies equally well to 8-byte longwords.  */
  78
  79   longword_ptr = (unsigned long int *) char_ptr;
  80
  81   /* Bits 31, 24, 16, and 8 of this number are zero.  Call these bits
  82      the "holes."  Note that there is a hole just to the left of
  83      each byte, with an extra at the end:
  84
  85      bits:  01111110 11111110 11111110 11111111
  86      bytes: AAAAAAAA BBBBBBBB CCCCCCCC DDDDDDDD
  87
  88      The 1-bits make sure that carries propagate to the next 0-bit.
  89      The 0-bits provide holes for carries to fall into.  */
  90
  91   if (sizeof (longword) != 4 && sizeof (longword) != 8)
  92     abort ();
  93
  94 #if LONG_MAX <= LONG_MAX_32_BITS
  95   magic_bits = 0x7efefeff;
  96 #else
  97   magic_bits = ((unsigned long int) 0x7efefefe << 32) | 0xfefefeff;
  98 #endif
  99
 100   /* Set up a longword, each of whose bytes is C.  */
 101   charmask = c | (c << 8);
 102   charmask |= charmask << 16;
 103 #if LONG_MAX > LONG_MAX_32_BITS
 104   charmask |= charmask << 32;
 105 #endif
 106
 107   /* Instead of the traditional loop which tests each character,
 108      we will test a longword at a time.  The tricky part is testing
 109      if *any of the four* bytes in the longword in question are zero.  */
 110   while (n >= sizeof (longword))
 111     {
 112       /* We tentatively exit the loop if adding MAGIC_BITS to
 113          LONGWORD fails to change any of the hole bits of LONGWORD.
 114
 115          1) Is this safe?  Will it catch all the zero bytes?
 116          Suppose there is a byte with all zeros.  Any carry bits
 117          propagating from its left will fall into the hole at its
 118          least significant bit and stop.  Since there will be no
 119          carry from its most significant bit, the LSB of the
 120          byte to the left will be unchanged, and the zero will be
 121          detected.
 122
 123          2) Is this worthwhile?  Will it ignore everything except
 124          zero bytes?  Suppose every byte of LONGWORD has a bit set
 125          somewhere.  There will be a carry into bit 8.  If bit 8
 126          is set, this will carry into bit 16.  If bit 8 is clear,
 127          one of bits 9-15 must be set, so there will be a carry
 128          into bit 16.  Similarly, there will be a carry into bit
 129          24.  If one of bits 24-30 is set, there will be a carry
 130          into bit 31, so all of the hole bits will be changed.
 131
 132          The one misfire occurs when bits 24-30 are clear and bit
 133          31 is set; in this case, the hole at bit 31 is not
 134          changed.  If we had access to the processor carry flag,
 135          we could close this loophole by putting the fourth hole
 136          at bit 32!
 137
 138          So it ignores everything except 128's, when they're aligned
 139          properly.
 140
 141          3) But wait!  Aren't we looking for C, not zero?
 142          Good point.  So what we do is XOR LONGWORD with a longword,
 143          each of whose bytes is C.  This turns each byte that is C
 144          into a zero.  */
 145
 146       longword = *longword_ptr++ ^ charmask;
 147
 148       /* Add MAGIC_BITS to LONGWORD.  */
 149       if ((((longword + magic_bits)
 150
 151             /* Set those bits that were unchanged by the addition.  */
 152             ^ ~longword)
 153
 154            /* Look at only the hole bits.  If any of the hole bits
 155               are unchanged, most likely one of the bytes was a
 156               zero.  */
 157            & ~magic_bits) != 0)
 158         {
 159           /* Which of the bytes was C?  If none of them were, it was
 160              a misfire; continue the search.  */
 161
 162           const unsigned char *cp = (const unsigned char *) (longword_ptr - 1);
 163
 164           if (cp[0] == c)
 165             return (__ptr_t) cp;
 166           if (cp[1] == c)
 167             return (__ptr_t) &cp[1];
 168           if (cp[2] == c)
 169             return (__ptr_t) &cp[2];
 170           if (cp[3] == c)
 171             return (__ptr_t) &cp[3];
 172 #if LONG_MAX > 2147483647
 173           if (cp[4] == c)
 174             return (__ptr_t) &cp[4];
 175           if (cp[5] == c)
 176             return (__ptr_t) &cp[5];
 177           if (cp[6] == c)
 178             return (__ptr_t) &cp[6];
 179           if (cp[7] == c)
 180             return (__ptr_t) &cp[7];
 181 #endif
 182         }
 183
 184       n -= sizeof (longword);
 185     }
 186
 187   char_ptr = (const unsigned char *) longword_ptr;
 188
 189   while (n-- > 0)
 190     {
 191       if (*char_ptr == c)
 192         return (__ptr_t) char_ptr;
 193       else
 194         ++char_ptr;
 195     }
 196
 197   return 0;
 198 }