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 __memrchr (s, c_in, n)
  67      const __ptr_t s;
  68      int c_in;
  69      size_t n;
  70 {
  71   const unsigned char *char_ptr;
  72   const unsigned long int *longword_ptr;
  73   unsigned long int longword, magic_bits, charmask;
  74   unsigned reg_char c;
  75
  76   c = (unsigned char) c_in;
  77
  78   /* Handle the last few characters by reading one character at a time.
  79      Do this until CHAR_PTR is aligned on a longword boundary.  */
  80   for (char_ptr = (const unsigned char *) s + n;
  81        n > 0 && ((unsigned long int) char_ptr
  82                  & (sizeof (longword) - 1)) != 0;
  83        --n)
  84     if (*--char_ptr == c)
  85       return (__ptr_t) char_ptr;
  86
  87   /* All these elucidatory comments refer to 4-byte longwords,
  88      but the theory applies equally well to 8-byte longwords.  */
  89
  90   longword_ptr = (const unsigned long int *) char_ptr;
  91
  92   /* Bits 31, 24, 16, and 8 of this number are zero.  Call these bits
  93      the "holes."  Note that there is a hole just to the left of
  94      each byte, with an extra at the end:
  95
  96      bits:  01111110 11111110 11111110 11111111
  97      bytes: AAAAAAAA BBBBBBBB CCCCCCCC DDDDDDDD
  98
  99      The 1-bits make sure that carries propagate to the next 0-bit.
 100      The 0-bits provide holes for carries to fall into.  */
 101
 102   if (sizeof (longword) != 4 && sizeof (longword) != 8)
 103     abort ();
 104
 105 #if LONG_MAX <= LONG_MAX_32_BITS
 106   magic_bits = 0x7efefeff;
 107 #else
 108   magic_bits = ((unsigned long int) 0x7efefefe << 32) | 0xfefefeff;
 109 #endif
 110
 111   /* Set up a longword, each of whose bytes is C.  */
 112   charmask = c | (c << 8);
 113   charmask |= charmask << 16;
 114 #if LONG_MAX > LONG_MAX_32_BITS
 115   charmask |= charmask << 32;
 116 #endif
 117
 118   /* Instead of the traditional loop which tests each character,
 119      we will test a longword at a time.  The tricky part is testing
 120      if *any of the four* bytes in the longword in question are zero.  */
 121   while (n >= sizeof (longword))
 122     {
 123       /* We tentatively exit the loop if adding MAGIC_BITS to
 124          LONGWORD fails to change any of the hole bits of LONGWORD.
 125
 126          1) Is this safe?  Will it catch all the zero bytes?
 127          Suppose there is a byte with all zeros.  Any carry bits
 128          propagating from its left will fall into the hole at its
 129          least significant bit and stop.  Since there will be no
 130          carry from its most significant bit, the LSB of the
 131          byte to the left will be unchanged, and the zero will be
 132          detected.
 133
 134          2) Is this worthwhile?  Will it ignore everything except
 135          zero bytes?  Suppose every byte of LONGWORD has a bit set
 136          somewhere.  There will be a carry into bit 8.  If bit 8
 137          is set, this will carry into bit 16.  If bit 8 is clear,
 138          one of bits 9-15 must be set, so there will be a carry
 139          into bit 16.  Similarly, there will be a carry into bit
 140          24.  If one of bits 24-30 is set, there will be a carry
 141          into bit 31, so all of the hole bits will be changed.
 142
 143          The one misfire occurs when bits 24-30 are clear and bit
 144          31 is set; in this case, the hole at bit 31 is not
 145          changed.  If we had access to the processor carry flag,
 146          we could close this loophole by putting the fourth hole
 147          at bit 32!
 148
 149          So it ignores everything except 128's, when they're aligned
 150          properly.
 151
 152          3) But wait!  Aren't we looking for C, not zero?
 153          Good point.  So what we do is XOR LONGWORD with a longword,
 154          each of whose bytes is C.  This turns each byte that is C
 155          into a zero.  */
 156
 157       longword = *--longword_ptr ^ charmask;
 158
 159       /* Add MAGIC_BITS to LONGWORD.  */
 160       if ((((longword + magic_bits)
 161
 162             /* Set those bits that were unchanged by the addition.  */
 163             ^ ~longword)
 164
 165            /* Look at only the hole bits.  If any of the hole bits
 166               are unchanged, most likely one of the bytes was a
 167               zero.  */
 168            & ~magic_bits) != 0)
 169         {
 170           /* Which of the bytes was C?  If none of them were, it was
 171              a misfire; continue the search.  */
 172
 173           const unsigned char *cp = (const unsigned char *) longword_ptr;
 174
 175 #if LONG_MAX > 2147483647
 176           if (cp[7] == c)
 177             return (__ptr_t) &cp[7];
 178           if (cp[6] == c)
 179             return (__ptr_t) &cp[6];
 180           if (cp[5] == c)
 181             return (__ptr_t) &cp[5];
 182           if (cp[4] == c)
 183             return (__ptr_t) &cp[4];
 184 #endif
 185           if (cp[3] == c)
 186             return (__ptr_t) &cp[3];
 187           if (cp[2] == c)
 188             return (__ptr_t) &cp[2];
 189           if (cp[1] == c)
 190             return (__ptr_t) &cp[1];
 191           if (cp[0] == c)
 192             return (__ptr_t) cp;
 193         }
 194
 195       n -= sizeof (longword);
 196     }
 197
 198   char_ptr = (const unsigned char *) longword_ptr;
 199
 200   while (n-- > 0)
 201     {
 202       if (*--char_ptr == c)
 203         return (__ptr_t) char_ptr;
 204     }
 205
 206   return 0;
 207 }
 208 #ifdef weak_alias
 209 weak_alias (__memrchr, memrchr)
 210 #endif