string/memrchr.c

   1 /* memrchr -- find the last occurrence of a byte in a memory block
   2    Copyright (C) 1991, 93, 96, 97, 99, 2000 Free Software Foundation, Inc.
   3    This file is part of the GNU C Library.
   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
  10    The GNU C Library is free software; you can redistribute it and/or
  11    modify it under the terms of the GNU Lesser General Public
  12    License as published by the Free Software Foundation; either
  13    version 2.1 of the License, or (at your option) any later version.
  14
  15    The GNU C Library is distributed in the hope that it will be useful,
  16    but WITHOUT ANY WARRANTY; without even the implied warranty of
  17    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18    Lesser General Public License for more details.
  19
  20    You should have received a copy of the GNU Lesser General Public
  21    License along with the GNU C Library; if not, write to the Free
  22    Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
  23    02111-1307 USA.  */
  24
  25 #include <stdlib.h>
  26
  27 #ifdef HAVE_CONFIG_H
  28 # include <config.h>
  29 #endif
  30
  31 #undef __ptr_t
  32 #if defined __cplusplus || (defined __STDC__ && __STDC__)
  33 # define __ptr_t void *
  34 #else /* Not C++ or ANSI C.  */
  35 # define __ptr_t char *
  36 #endif /* C++ or ANSI C.  */
  37
  38 #if defined _LIBC
  39 # include <string.h>
  40 # include <memcopy.h>
  41 #else
  42 # define reg_char char
  43 #endif
  44
  45 #if defined HAVE_LIMITS_H || defined _LIBC
  46 # include <limits.h>
  47 #endif
  48
  49 #define LONG_MAX_32_BITS 2147483647
  50
  51 #ifndef LONG_MAX
  52 # define LONG_MAX LONG_MAX_32_BITS
  53 #endif
  54
  55 #include <sys/types.h>
  56
  57 #undef __memrchr
  58 #undef memrchr
  59
  60 #ifndef weak_alias
  61 # define __memrchr memrchr
  62 #endif
  63
  64 /* Search no more than N bytes of S for C.  */
  65 __ptr_t
  66 #ifndef MEMRCHR
  67 __memrchr
  68 #else
  69 MEMRCHR
  70 #endif
  71      (s, c_in, n)
  72      const __ptr_t s;
  73      int c_in;
  74      size_t n;
  75 {
  76   const unsigned char *char_ptr;
  77   const unsigned long int *longword_ptr;
  78   unsigned long int longword, magic_bits, charmask;
  79   unsigned reg_char c;
  80
  81   c = (unsigned char) c_in;
  82
  83   /* Handle the last few characters by reading one character at a time.
  84      Do this until CHAR_PTR is aligned on a longword boundary.  */
  85   for (char_ptr = (const unsigned char *) s + n;
  86        n > 0 && ((unsigned long int) char_ptr
  87                  & (sizeof (longword) - 1)) != 0;
  88        --n)
  89     if (*--char_ptr == c)
  90       return (__ptr_t) char_ptr;
  91
  92   /* All these elucidatory comments refer to 4-byte longwords,
  93      but the theory applies equally well to 8-byte longwords.  */
  94
  95   longword_ptr = (const unsigned long int *) char_ptr;
  96
  97   /* Bits 31, 24, 16, and 8 of this number are zero.  Call these bits
  98      the "holes."  Note that there is a hole just to the left of
  99      each byte, with an extra at the end:
 100
 101      bits:  01111110 11111110 11111110 11111111
 102      bytes: AAAAAAAA BBBBBBBB CCCCCCCC DDDDDDDD
 103
 104      The 1-bits make sure that carries propagate to the next 0-bit.
 105      The 0-bits provide holes for carries to fall into.  */
 106
 107   if (sizeof (longword) != 4 && sizeof (longword) != 8)
 108     abort ();
 109
 110 #if LONG_MAX <= LONG_MAX_32_BITS
 111   magic_bits = 0x7efefeff;
 112 #else
 113   magic_bits = ((unsigned long int) 0x7efefefe << 32) | 0xfefefeff;
 114 #endif
 115
 116   /* Set up a longword, each of whose bytes is C.  */
 117   charmask = c | (c << 8);
 118   charmask |= charmask << 16;
 119 #if LONG_MAX > LONG_MAX_32_BITS
 120   charmask |= charmask << 32;
 121 #endif
 122
 123   /* Instead of the traditional loop which tests each character,
 124      we will test a longword at a time.  The tricky part is testing
 125      if *any of the four* bytes in the longword in question are zero.  */
 126   while (n >= sizeof (longword))
 127     {
 128       /* We tentatively exit the loop if adding MAGIC_BITS to
 129          LONGWORD fails to change any of the hole bits of LONGWORD.
 130
 131          1) Is this safe?  Will it catch all the zero bytes?
 132          Suppose there is a byte with all zeros.  Any carry bits
 133          propagating from its left will fall into the hole at its
 134          least significant bit and stop.  Since there will be no
 135          carry from its most significant bit, the LSB of the
 136          byte to the left will be unchanged, and the zero will be
 137          detected.
 138
 139          2) Is this worthwhile?  Will it ignore everything except
 140          zero bytes?  Suppose every byte of LONGWORD has a bit set
 141          somewhere.  There will be a carry into bit 8.  If bit 8
 142          is set, this will carry into bit 16.  If bit 8 is clear,
 143          one of bits 9-15 must be set, so there will be a carry
 144          into bit 16.  Similarly, there will be a carry into bit
 145          24.  If one of bits 24-30 is set, there will be a carry
 146          into bit 31, so all of the hole bits will be changed.
 147
 148          The one misfire occurs when bits 24-30 are clear and bit
 149          31 is set; in this case, the hole at bit 31 is not
 150          changed.  If we had access to the processor carry flag,
 151          we could close this loophole by putting the fourth hole
 152          at bit 32!
 153
 154          So it ignores everything except 128's, when they're aligned
 155          properly.
 156
 157          3) But wait!  Aren't we looking for C, not zero?
 158          Good point.  So what we do is XOR LONGWORD with a longword,
 159          each of whose bytes is C.  This turns each byte that is C
 160          into a zero.  */
 161
 162       longword = *--longword_ptr ^ charmask;
 163
 164       /* Add MAGIC_BITS to LONGWORD.  */
 165       if ((((longword + magic_bits)
 166
 167             /* Set those bits that were unchanged by the addition.  */
 168             ^ ~longword)
 169
 170            /* Look at only the hole bits.  If any of the hole bits
 171               are unchanged, most likely one of the bytes was a
 172               zero.  */
 173            & ~magic_bits) != 0)
 174         {
 175           /* Which of the bytes was C?  If none of them were, it was
 176              a misfire; continue the search.  */
 177
 178           const unsigned char *cp = (const unsigned char *) longword_ptr;
 179
 180 #if LONG_MAX > 2147483647
 181           if (cp[7] == c)
 182             return (__ptr_t) &cp[7];
 183           if (cp[6] == c)
 184             return (__ptr_t) &cp[6];
 185           if (cp[5] == c)
 186             return (__ptr_t) &cp[5];
 187           if (cp[4] == c)
 188             return (__ptr_t) &cp[4];
 189 #endif
 190           if (cp[3] == c)
 191             return (__ptr_t) &cp[3];
 192           if (cp[2] == c)
 193             return (__ptr_t) &cp[2];
 194           if (cp[1] == c)
 195             return (__ptr_t) &cp[1];
 196           if (cp[0] == c)
 197             return (__ptr_t) cp;
 198         }
 199
 200       n -= sizeof (longword);
 201     }
 202
 203   char_ptr = (const unsigned char *) longword_ptr;
 204
 205   while (n-- > 0)
 206     {
 207       if (*--char_ptr == c)
 208         return (__ptr_t) char_ptr;
 209     }
 210
 211   return 0;
 212 }
 213 #ifndef MEMRCHR
 214 # ifdef weak_alias
 215 weak_alias (__memrchr, memrchr)
 216 # endif
 217 #endif