sysdeps/alpha/memchr.c

   1 /* Copyright (C) 2010 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, write to the Free
  16    Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
  17    02111-1307 USA.  */
  18
  19 #include <string.h>
  20 #include <bp-sym.h>
  21
  22 typedef unsigned long word;
  23
  24 static inline word
  25 ldq_u(const void *s)
  26 {
  27   return *(const word *)((word)s & -8);
  28 }
  29
  30 #define unlikely(X)     __builtin_expect ((X), 0)
  31 #define prefetch(X)     __builtin_prefetch ((void *)(X), 0)
  32
  33 #define cmpbeq0(X)      __builtin_alpha_cmpbge(0, (X))
  34 #define find(X, Y)      cmpbeq0 ((X) ^ (Y))
  35
  36 /* Search no more than N bytes of S for C.  */
  37
  38 void *
  39 __memchr (const void *s, int xc, size_t n)
  40 {
  41   const word *s_align;
  42   word t, current, found, mask, offset;
  43
  44   if (unlikely (n == 0))
  45     return 0;
  46
  47   current = ldq_u (s);
  48
  49   /* Replicate low byte of XC into all bytes of C.  */
  50   t = xc & 0xff;                        /* 0000000c */
  51   t = (t << 8) | t;                     /* 000000cc */
  52   t = (t << 16) | t;                    /* 0000cccc */
  53   const word c = (t << 32) | t;         /* cccccccc */
  54
  55   /* Align the source, and decrement the count by the number
  56      of bytes searched in the first word.  */
  57   s_align = (const word *)((word)s & -8);
  58   n += ((word)s & 7);
  59
  60   /* Deal with misalignment in the first word for the comparison.  */
  61   mask = (1ul << ((word)s & 7)) - 1;
  62
  63   /* If the entire string fits within one word, we may need masking
  64      at both the front and the back of the string.  */
  65   if (unlikely (n <= 8))
  66     {
  67       mask |= -1ul << n;
  68       goto last_quad;
  69     }
  70
  71   found = find (current, c) & ~mask;
  72   if (unlikely (found))
  73     goto found_it;
  74
  75   s_align++;
  76   n -= 8;
  77
  78   /* If the block is sufficiently large, align to cacheline and prefetch.  */
  79   if (unlikely (n >= 256))
  80     {
  81       /* Prefetch 3 cache lines beyond the one we're working on.  */
  82       prefetch (s_align + 8);
  83       prefetch (s_align + 16);
  84       prefetch (s_align + 24);
  85
  86       while ((word)s_align & 63)
  87         {
  88           current = *s_align;
  89           found = find (current, c);
  90           if (found)
  91             goto found_it;
  92           s_align++;
  93           n -= 8;
  94         }
  95
  96         /* Within each cacheline, advance the load for the next word
  97            before the test for the previous word is complete.  This
  98            allows us to hide the 3 cycle L1 cache load latency.  We
  99            only perform this advance load within a cacheline to prevent
 100            reading across page boundary.  */
 101 #define CACHELINE_LOOP                          \
 102         do {                                    \
 103           word i, next = s_align[0];            \
 104           for (i = 0; i < 7; ++i)               \
 105             {                                   \
 106               current = next;                   \
 107               next = s_align[1];                \
 108               found = find (current, c);        \
 109               if (unlikely (found))             \
 110                 goto found_it;                  \
 111               s_align++;                        \
 112             }                                   \
 113           current = next;                       \
 114           found = find (current, c);            \
 115           if (unlikely (found))                 \
 116             goto found_it;                      \
 117           s_align++;                            \
 118           n -= 64;                              \
 119         } while (0)
 120
 121       /* While there's still lots more data to potentially be read,
 122          continue issuing prefetches for the 4th cacheline out.  */
 123       while (n >= 256)
 124         {
 125           prefetch (s_align + 24);
 126           CACHELINE_LOOP;
 127         }
 128
 129       /* Up to 3 cache lines remaining.  Continue issuing advanced
 130          loads, but stop prefetching.  */
 131       while (n >= 64)
 132         CACHELINE_LOOP;
 133
 134       /* We may have exhausted the buffer.  */
 135       if (n == 0)
 136         return NULL;
 137     }
 138
 139   /* Quadword aligned loop.  */
 140   current = *s_align;
 141   while (n > 8)
 142     {
 143       found = find (current, c);
 144       if (unlikely (found))
 145         goto found_it;
 146       current = *++s_align;
 147       n -= 8;
 148     }
 149
 150   /* The last word may need masking at the tail of the compare.  */
 151   mask = -1ul << n;
 152  last_quad:
 153   found = find (current, c) & ~mask;
 154   if (found == 0)
 155     return NULL;
 156
 157  found_it:
 158 #ifdef __alpha_cix__
 159   offset = __builtin_alpha_cttz (found);
 160 #else
 161   /* Extract LSB.  */
 162   found &= -found;
 163
 164   /* Binary search for the LSB.  */
 165   offset  = (found & 0x0f ? 0 : 4);
 166   offset += (found & 0x33 ? 0 : 2);
 167   offset += (found & 0x55 ? 0 : 1);
 168 #endif
 169
 170   return (void *)((word)s_align + offset);
 171 }
 172
 173 #ifdef weak_alias
 174 weak_alias (__memchr, BP_SYM (memchr))
 175 #endif
 176 libc_hidden_builtin_def (memchr)