sysdeps/alpha/memchr.c

   1 /* Copyright (C) 2010-2015 Free Software Foundation, Inc.
   2    This file is part of the GNU C Library.
   3
   4    The GNU C Library is free software; you can redistribute it and/or
   5    modify it under the terms of the GNU Lesser General Public
   6    License as published by the Free Software Foundation; either
   7    version 2.1 of the License, or (at your option) any later version.
   8
   9    The GNU C Library is distributed in the hope that it will be useful,
  10    but WITHOUT ANY WARRANTY; without even the implied warranty of
  11    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  12    Lesser General Public License for more details.
  13
  14    You should have received a copy of the GNU Lesser General Public
  15    License along with the GNU C Library.  If not, see
  16    <http://www.gnu.org/licenses/>.  */
  17
  18 #include <string.h>
  19
  20 typedef unsigned long word;
  21
  22 static inline word
  23 ldq_u(const void *s)
  24 {
  25   return *(const word *)((word)s & -8);
  26 }
  27
  28 #define unlikely(X)     __builtin_expect ((X), 0)
  29 #define prefetch(X)     __builtin_prefetch ((void *)(X), 0)
  30
  31 #define cmpbeq0(X)      __builtin_alpha_cmpbge(0, (X))
  32 #define find(X, Y)      cmpbeq0 ((X) ^ (Y))
  33
  34 /* Search no more than N bytes of S for C.  */
  35
  36 void *
  37 __memchr (const void *s, int xc, size_t n)
  38 {
  39   const word *s_align;
  40   word t, current, found, mask, offset;
  41
  42   if (unlikely (n == 0))
  43     return 0;
  44
  45   current = ldq_u (s);
  46
  47   /* Replicate low byte of XC into all bytes of C.  */
  48   t = xc & 0xff;                        /* 0000000c */
  49   t = (t << 8) | t;                     /* 000000cc */
  50   t = (t << 16) | t;                    /* 0000cccc */
  51   const word c = (t << 32) | t;         /* cccccccc */
  52
  53   /* Align the source, and decrement the count by the number
  54      of bytes searched in the first word.  */
  55   s_align = (const word *)((word)s & -8);
  56   n += ((word)s & 7);
  57
  58   /* Deal with misalignment in the first word for the comparison.  */
  59   mask = (1ul << ((word)s & 7)) - 1;
  60
  61   /* If the entire string fits within one word, we may need masking
  62      at both the front and the back of the string.  */
  63   if (unlikely (n <= 8))
  64     {
  65       mask |= -1ul << n;
  66       goto last_quad;
  67     }
  68
  69   found = find (current, c) & ~mask;
  70   if (unlikely (found))
  71     goto found_it;
  72
  73   s_align++;
  74   n -= 8;
  75
  76   /* If the block is sufficiently large, align to cacheline and prefetch.  */
  77   if (unlikely (n >= 256))
  78     {
  79       /* Prefetch 3 cache lines beyond the one we're working on.  */
  80       prefetch (s_align + 8);
  81       prefetch (s_align + 16);
  82       prefetch (s_align + 24);
  83
  84       while ((word)s_align & 63)
  85         {
  86           current = *s_align;
  87           found = find (current, c);
  88           if (found)
  89             goto found_it;
  90           s_align++;
  91           n -= 8;
  92         }
  93
  94         /* Within each cacheline, advance the load for the next word
  95            before the test for the previous word is complete.  This
  96            allows us to hide the 3 cycle L1 cache load latency.  We
  97            only perform this advance load within a cacheline to prevent
  98            reading across page boundary.  */
  99 #define CACHELINE_LOOP                          \
 100         do {                                    \
 101           word i, next = s_align[0];            \
 102           for (i = 0; i < 7; ++i)               \
 103             {                                   \
 104               current = next;                   \
 105               next = s_align[1];                \
 106               found = find (current, c);        \
 107               if (unlikely (found))             \
 108                 goto found_it;                  \
 109               s_align++;                        \
 110             }                                   \
 111           current = next;                       \
 112           found = find (current, c);            \
 113           if (unlikely (found))                 \
 114             goto found_it;                      \
 115           s_align++;                            \
 116           n -= 64;                              \
 117         } while (0)
 118
 119       /* While there's still lots more data to potentially be read,
 120          continue issuing prefetches for the 4th cacheline out.  */
 121       while (n >= 256)
 122         {
 123           prefetch (s_align + 24);
 124           CACHELINE_LOOP;
 125         }
 126
 127       /* Up to 3 cache lines remaining.  Continue issuing advanced
 128          loads, but stop prefetching.  */
 129       while (n >= 64)
 130         CACHELINE_LOOP;
 131
 132       /* We may have exhausted the buffer.  */
 133       if (n == 0)
 134         return NULL;
 135     }
 136
 137   /* Quadword aligned loop.  */
 138   current = *s_align;
 139   while (n > 8)
 140     {
 141       found = find (current, c);
 142       if (unlikely (found))
 143         goto found_it;
 144       current = *++s_align;
 145       n -= 8;
 146     }
 147
 148   /* The last word may need masking at the tail of the compare.  */
 149   mask = -1ul << n;
 150  last_quad:
 151   found = find (current, c) & ~mask;
 152   if (found == 0)
 153     return NULL;
 154
 155  found_it:
 156 #ifdef __alpha_cix__
 157   offset = __builtin_alpha_cttz (found);
 158 #else
 159   /* Extract LSB.  */
 160   found &= -found;
 161
 162   /* Binary search for the LSB.  */
 163   offset  = (found & 0x0f ? 0 : 4);
 164   offset += (found & 0x33 ? 0 : 2);
 165   offset += (found & 0x55 ? 0 : 1);
 166 #endif
 167
 168   return (void *)((word)s_align + offset);
 169 }
 170
 171 #ifdef weak_alias
 172 weak_alias (__memchr, memchr)
 173 #endif
 174 libc_hidden_builtin_def (memchr)