string/rawmemchr.c

   1 /* Copyright (C) 1991,93,96,97,99,2000,2002,2012 Free Software Foundation, Inc.
   2    This file is part of the GNU C Library.
   3    Based on strlen implementation by Torbjorn Granlund (tege@sics.se),
   4    with help from Dan Sahlin (dan@sics.se) and
   5    commentary by Jim Blandy (jimb@ai.mit.edu);
   6    adaptation to memchr suggested by Dick Karpinski (dick@cca.ucsf.edu),
   7    and implemented by Roland McGrath (roland@ai.mit.edu).
   8
   9    The GNU C Library is free software; you can redistribute it and/or
  10    modify it under the terms of the GNU Lesser General Public
  11    License as published by the Free Software Foundation; either
  12    version 2.1 of the License, or (at your option) any later version.
  13
  14    The GNU C Library is distributed in the hope that it will be useful,
  15    but WITHOUT ANY WARRANTY; without even the implied warranty of
  16    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17    Lesser General Public License for more details.
  18
  19    You should have received a copy of the GNU Lesser General Public
  20    License along with the GNU C Library; if not, write to the Free
  21    Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
  22    02111-1307 USA.  */
  23
  24 #ifdef HAVE_CONFIG_H
  25 #include <config.h>
  26 #endif
  27
  28 #undef __ptr_t
  29 #define __ptr_t void *
  30
  31 #if defined (_LIBC)
  32 # include <string.h>
  33 # include <memcopy.h>
  34 # include <stdlib.h>
  35 #else
  36 # define reg_char char
  37 #endif
  38
  39 #if defined (HAVE_LIMITS_H) || defined (_LIBC)
  40 # include <limits.h>
  41 #endif
  42
  43 #define LONG_MAX_32_BITS 2147483647
  44
  45 #ifndef LONG_MAX
  46 #define LONG_MAX LONG_MAX_32_BITS
  47 #endif
  48
  49 #include <sys/types.h>
  50
  51 #undef memchr
  52
  53
  54 /* Find the first occurrence of C in S.  */
  55 __ptr_t
  56 __rawmemchr (s, c_in)
  57      const __ptr_t s;
  58      int c_in;
  59 {
  60   const unsigned char *char_ptr;
  61   const unsigned long int *longword_ptr;
  62   unsigned long int longword, magic_bits, charmask;
  63   unsigned reg_char c;
  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        ((unsigned long int) char_ptr & (sizeof (longword) - 1)) != 0;
  71        ++char_ptr)
  72     if (*char_ptr == c)
  73       return (__ptr_t) char_ptr;
  74
  75   /* All these elucidatory comments refer to 4-byte longwords,
  76      but the theory applies equally well to 8-byte longwords.  */
  77
  78   longword_ptr = (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   if (sizeof (longword) != 4 && sizeof (longword) != 8)
  91     abort ();
  92
  93 #if LONG_MAX <= LONG_MAX_32_BITS
  94   magic_bits = 0x7efefeff;
  95 #else
  96   magic_bits = ((unsigned long int) 0x7efefefe << 32) | 0xfefefeff;
  97 #endif
  98
  99   /* Set up a longword, each of whose bytes is C.  */
 100   charmask = c | (c << 8);
 101   charmask |= charmask << 16;
 102 #if LONG_MAX > LONG_MAX_32_BITS
 103   charmask |= charmask << 32;
 104 #endif
 105
 106   /* Instead of the traditional loop which tests each character,
 107      we will test a longword at a time.  The tricky part is testing
 108      if *any of the four* bytes in the longword in question are zero.  */
 109   while (1)
 110     {
 111       /* We tentatively exit the loop if adding MAGIC_BITS to
 112          LONGWORD fails to change any of the hole bits of LONGWORD.
 113
 114          1) Is this safe?  Will it catch all the zero bytes?
 115          Suppose there is a byte with all zeros.  Any carry bits
 116          propagating from its left will fall into the hole at its
 117          least significant bit and stop.  Since there will be no
 118          carry from its most significant bit, the LSB of the
 119          byte to the left will be unchanged, and the zero will be
 120          detected.
 121
 122          2) Is this worthwhile?  Will it ignore everything except
 123          zero bytes?  Suppose every byte of LONGWORD has a bit set
 124          somewhere.  There will be a carry into bit 8.  If bit 8
 125          is set, this will carry into bit 16.  If bit 8 is clear,
 126          one of bits 9-15 must be set, so there will be a carry
 127          into bit 16.  Similarly, there will be a carry into bit
 128          24.  If one of bits 24-30 is set, there will be a carry
 129          into bit 31, so all of the hole bits will be changed.
 130
 131          The one misfire occurs when bits 24-30 are clear and bit
 132          31 is set; in this case, the hole at bit 31 is not
 133          changed.  If we had access to the processor carry flag,
 134          we could close this loophole by putting the fourth hole
 135          at bit 32!
 136
 137          So it ignores everything except 128's, when they're aligned
 138          properly.
 139
 140          3) But wait!  Aren't we looking for C, not zero?
 141          Good point.  So what we do is XOR LONGWORD with a longword,
 142          each of whose bytes is C.  This turns each byte that is C
 143          into a zero.  */
 144
 145       longword = *longword_ptr++ ^ charmask;
 146
 147       /* Add MAGIC_BITS to LONGWORD.  */
 148       if ((((longword + magic_bits)
 149
 150             /* Set those bits that were unchanged by the addition.  */
 151             ^ ~longword)
 152
 153            /* Look at only the hole bits.  If any of the hole bits
 154               are unchanged, most likely one of the bytes was a
 155               zero.  */
 156            & ~magic_bits) != 0)
 157         {
 158           /* Which of the bytes was C?  If none of them were, it was
 159              a misfire; continue the search.  */
 160
 161           const unsigned char *cp = (const unsigned char *) (longword_ptr - 1);
 162
 163           if (cp[0] == c)
 164             return (__ptr_t) cp;
 165           if (cp[1] == c)
 166             return (__ptr_t) &cp[1];
 167           if (cp[2] == c)
 168             return (__ptr_t) &cp[2];
 169           if (cp[3] == c)
 170             return (__ptr_t) &cp[3];
 171 #if LONG_MAX > 2147483647
 172           if (cp[4] == c)
 173             return (__ptr_t) &cp[4];
 174           if (cp[5] == c)
 175             return (__ptr_t) &cp[5];
 176           if (cp[6] == c)
 177             return (__ptr_t) &cp[6];
 178           if (cp[7] == c)
 179             return (__ptr_t) &cp[7];
 180 #endif
 181         }
 182     }
 183 }
 184 libc_hidden_def (__rawmemchr)
 185 weak_alias (__rawmemchr, rawmemchr)