sysdeps/generic/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 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
  20 not, write to the Free Software Foundation, Inc., 675 Mass Ave,
  21 Cambridge, MA 02139, 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 (HAVE_LIMITS_H) || defined (_LIBC)
  35 # include <limits.h>
  36 #endif
  37
  38 #define LONG_MAX_32_BITS 2147483647
  39
  40 #ifndef LONG_MAX
  41 #define LONG_MAX LONG_MAX_32_BITS
  42 #endif
  43
  44 #include <sys/types.h>
  45
  46
  47 /* Search no more than N bytes of S for C.  */
  48
  49 __ptr_t
  50 memchr (s, c, n)
  51      const __ptr_t s;
  52      int c;
  53      size_t n;
  54 {
  55   const unsigned char *char_ptr;
  56   const unsigned long int *longword_ptr;
  57   unsigned long int longword, magic_bits, charmask;
  58
  59   c = (unsigned char) c;
  60
  61   /* Handle the first few characters by reading one character at a time.
  62      Do this until CHAR_PTR is aligned on a longword boundary.  */
  63   for (char_ptr = s; n > 0 && ((unsigned long int) char_ptr
  64                                & (sizeof (longword) - 1)) != 0;
  65        --n, ++char_ptr)
  66     if (*char_ptr == c)
  67       return (__ptr_t) char_ptr;
  68
  69   /* All these elucidatory comments refer to 4-byte longwords,
  70      but the theory applies equally well to 8-byte longwords.  */
  71
  72   longword_ptr = (unsigned long int *) char_ptr;
  73
  74   /* Bits 31, 24, 16, and 8 of this number are zero.  Call these bits
  75      the "holes."  Note that there is a hole just to the left of
  76      each byte, with an extra at the end:
  77
  78      bits:  01111110 11111110 11111110 11111111
  79      bytes: AAAAAAAA BBBBBBBB CCCCCCCC DDDDDDDD
  80
  81      The 1-bits make sure that carries propagate to the next 0-bit.
  82      The 0-bits provide holes for carries to fall into.  */
  83
  84   if (sizeof (longword) != 4 && sizeof (longword) != 8)
  85     abort ();
  86
  87 #if LONG_MAX <= LONG_MAX_32_BITS
  88   magic_bits = 0x7efefeff;
  89 #else
  90   magic_bits = ((unsigned long int) 0x7efefefe << 32) | 0xfefefeff;
  91 #endif
  92
  93   /* Set up a longword, each of whose bytes is C.  */
  94   charmask = c | (c << 8);
  95   charmask |= charmask << 16;
  96 #if LONG_MAX > LONG_MAX_32_BITS
  97   charmask |= charmask << 32;
  98 #endif
  99
 100   /* Instead of the traditional loop which tests each character,
 101      we will test a longword at a time.  The tricky part is testing
 102      if *any of the four* bytes in the longword in question are zero.  */
 103   while (n >= sizeof (longword))
 104     {
 105       /* We tentatively exit the loop if adding MAGIC_BITS to
 106          LONGWORD fails to change any of the hole bits of LONGWORD.
 107
 108          1) Is this safe?  Will it catch all the zero bytes?
 109          Suppose there is a byte with all zeros.  Any carry bits
 110          propagating from its left will fall into the hole at its
 111          least significant bit and stop.  Since there will be no
 112          carry from its most significant bit, the LSB of the
 113          byte to the left will be unchanged, and the zero will be
 114          detected.
 115
 116          2) Is this worthwhile?  Will it ignore everything except
 117          zero bytes?  Suppose every byte of LONGWORD has a bit set
 118          somewhere.  There will be a carry into bit 8.  If bit 8
 119          is set, this will carry into bit 16.  If bit 8 is clear,
 120          one of bits 9-15 must be set, so there will be a carry
 121          into bit 16.  Similarly, there will be a carry into bit
 122          24.  If one of bits 24-30 is set, there will be a carry
 123          into bit 31, so all of the hole bits will be changed.
 124
 125          The one misfire occurs when bits 24-30 are clear and bit
 126          31 is set; in this case, the hole at bit 31 is not
 127          changed.  If we had access to the processor carry flag,
 128          we could close this loophole by putting the fourth hole
 129          at bit 32!
 130
 131          So it ignores everything except 128's, when they're aligned
 132          properly.
 133
 134          3) But wait!  Aren't we looking for C, not zero?
 135          Good point.  So what we do is XOR LONGWORD with a longword,
 136          each of whose bytes is C.  This turns each byte that is C
 137          into a zero.  */
 138
 139       longword = *longword_ptr++ ^ charmask;
 140
 141       /* Add MAGIC_BITS to LONGWORD.  */
 142       if ((((longword + magic_bits)
 143
 144             /* Set those bits that were unchanged by the addition.  */
 145             ^ ~longword)
 146
 147            /* Look at only the hole bits.  If any of the hole bits
 148               are unchanged, most likely one of the bytes was a
 149               zero.  */
 150            & ~magic_bits) != 0)
 151         {
 152           /* Which of the bytes was C?  If none of them were, it was
 153              a misfire; continue the search.  */
 154
 155           const unsigned char *cp = (const unsigned char *) (longword_ptr - 1);
 156
 157           if (cp[0] == c)
 158             return (__ptr_t) cp;
 159           if (cp[1] == c)
 160             return (__ptr_t) &cp[1];
 161           if (cp[2] == c)
 162             return (__ptr_t) &cp[2];
 163           if (cp[3] == c)
 164             return (__ptr_t) &cp[3];
 165 #if LONG_MAX > 2147483647
 166           if (cp[4] == c)
 167             return (__ptr_t) &cp[4];
 168           if (cp[5] == c)
 169             return (__ptr_t) &cp[5];
 170           if (cp[6] == c)
 171             return (__ptr_t) &cp[6];
 172           if (cp[7] == c)
 173             return (__ptr_t) &cp[7];
 174 #endif
 175         }
 176
 177       n -= sizeof (longword);
 178     }
 179
 180   char_ptr = (const unsigned char *) longword_ptr;
 181
 182   while (n-- > 0)
 183     {
 184       if (*char_ptr == c)
 185         return (__ptr_t) char_ptr;
 186       else
 187         ++char_ptr;
 188     }
 189
 190   return 0;
 191 }