sysdeps/generic/memchr.c

   1 /* Copyright (C) 1991, 1993, 1996 Free Software Foundation, Inc.
   2    Based on strlen implementation 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    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.
  12
  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.
  17
  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 not,
  20    write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  21    Boston, MA 02111-1307, USA.  */
  22
  23 #ifdef HAVE_CONFIG_H
  24 #include <config.h>
  25 #endif
  26
  27 #undef __ptr_t
  28 #if defined (__cplusplus) || (defined (__STDC__) && __STDC__)
  29 # define __ptr_t void *
  30 #else /* Not C++ or ANSI C.  */
  31 # define __ptr_t char *
  32 #endif /* C++ or ANSI C.  */
  33
  34 #if defined (_LIBC)
  35 # include <string.h>
  36 #endif
  37
  38 #if defined (HAVE_LIMITS_H) || defined (_LIBC)
  39 # include <limits.h>
  40 #endif
  41
  42 #define LONG_MAX_32_BITS 2147483647
  43
  44 #ifndef LONG_MAX
  45 #define LONG_MAX LONG_MAX_32_BITS
  46 #endif
  47
  48 #include <sys/types.h>
  49
  50
  51 /* Search no more than N bytes of S for C.  */
  52
  53 __ptr_t
  54 memchr (s, c, n)
  55      const __ptr_t s;
  56      int c;
  57      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
  63   c = (unsigned char) c;
  64
  65   /* Handle the first few characters by reading one character at a time.
  66      Do this until CHAR_PTR is aligned on a longword boundary.  */
  67   for (char_ptr = (const unsigned char *) s;
  68        n > 0 && ((unsigned long int) char_ptr
  69                  & (sizeof (longword) - 1)) != 0;
  70        --n, ++char_ptr)
  71     if (*char_ptr == c)
  72       return (__ptr_t) char_ptr;
  73
  74   /* All these elucidatory comments refer to 4-byte longwords,
  75      but the theory applies equally well to 8-byte longwords.  */
  76
  77   longword_ptr = (unsigned long int *) char_ptr;
  78
  79   /* Bits 31, 24, 16, and 8 of this number are zero.  Call these bits
  80      the "holes."  Note that there is a hole just to the left of
  81      each byte, with an extra at the end:
  82
  83      bits:  01111110 11111110 11111110 11111111
  84      bytes: AAAAAAAA BBBBBBBB CCCCCCCC DDDDDDDD
  85
  86      The 1-bits make sure that carries propagate to the next 0-bit.
  87      The 0-bits provide holes for carries to fall into.  */
  88
  89   if (sizeof (longword) != 4 && sizeof (longword) != 8)
  90     abort ();
  91
  92 #if LONG_MAX <= LONG_MAX_32_BITS
  93   magic_bits = 0x7efefeff;
  94 #else
  95   magic_bits = ((unsigned long int) 0x7efefefe << 32) | 0xfefefeff;
  96 #endif
  97
  98   /* Set up a longword, each of whose bytes is C.  */
  99   charmask = c | (c << 8);
 100   charmask |= charmask << 16;
 101 #if LONG_MAX > LONG_MAX_32_BITS
 102   charmask |= charmask << 32;
 103 #endif
 104
 105   /* Instead of the traditional loop which tests each character,
 106      we will test a longword at a time.  The tricky part is testing
 107      if *any of the four* bytes in the longword in question are zero.  */
 108   while (n >= sizeof (longword))
 109     {
 110       /* We tentatively exit the loop if adding MAGIC_BITS to
 111          LONGWORD fails to change any of the hole bits of LONGWORD.
 112
 113          1) Is this safe?  Will it catch all the zero bytes?
 114          Suppose there is a byte with all zeros.  Any carry bits
 115          propagating from its left will fall into the hole at its
 116          least significant bit and stop.  Since there will be no
 117          carry from its most significant bit, the LSB of the
 118          byte to the left will be unchanged, and the zero will be
 119          detected.
 120
 121          2) Is this worthwhile?  Will it ignore everything except
 122          zero bytes?  Suppose every byte of LONGWORD has a bit set
 123          somewhere.  There will be a carry into bit 8.  If bit 8
 124          is set, this will carry into bit 16.  If bit 8 is clear,
 125          one of bits 9-15 must be set, so there will be a carry
 126          into bit 16.  Similarly, there will be a carry into bit
 127          24.  If one of bits 24-30 is set, there will be a carry
 128          into bit 31, so all of the hole bits will be changed.
 129
 130          The one misfire occurs when bits 24-30 are clear and bit
 131          31 is set; in this case, the hole at bit 31 is not
 132          changed.  If we had access to the processor carry flag,
 133          we could close this loophole by putting the fourth hole
 134          at bit 32!
 135
 136          So it ignores everything except 128's, when they're aligned
 137          properly.
 138
 139          3) But wait!  Aren't we looking for C, not zero?
 140          Good point.  So what we do is XOR LONGWORD with a longword,
 141          each of whose bytes is C.  This turns each byte that is C
 142          into a zero.  */
 143
 144       longword = *longword_ptr++ ^ charmask;
 145
 146       /* Add MAGIC_BITS to LONGWORD.  */
 147       if ((((longword + magic_bits)
 148
 149             /* Set those bits that were unchanged by the addition.  */
 150             ^ ~longword)
 151
 152            /* Look at only the hole bits.  If any of the hole bits
 153               are unchanged, most likely one of the bytes was a
 154               zero.  */
 155            & ~magic_bits) != 0)
 156         {
 157           /* Which of the bytes was C?  If none of them were, it was
 158              a misfire; continue the search.  */
 159
 160           const unsigned char *cp = (const unsigned char *) (longword_ptr - 1);
 161
 162           if (cp[0] == c)
 163             return (__ptr_t) cp;
 164           if (cp[1] == c)
 165             return (__ptr_t) &cp[1];
 166           if (cp[2] == c)
 167             return (__ptr_t) &cp[2];
 168           if (cp[3] == c)
 169             return (__ptr_t) &cp[3];
 170 #if LONG_MAX > 2147483647
 171           if (cp[4] == c)
 172             return (__ptr_t) &cp[4];
 173           if (cp[5] == c)
 174             return (__ptr_t) &cp[5];
 175           if (cp[6] == c)
 176             return (__ptr_t) &cp[6];
 177           if (cp[7] == c)
 178             return (__ptr_t) &cp[7];
 179 #endif
 180         }
 181
 182       n -= sizeof (longword);
 183     }
 184
 185   char_ptr = (const unsigned char *) longword_ptr;
 186
 187   while (n-- > 0)
 188     {
 189       if (*char_ptr == c)
 190         return (__ptr_t) char_ptr;
 191       else
 192         ++char_ptr;
 193     }
 194
 195   return 0;
 196 }