sysdeps/generic/memcmp.c

   1 /* Copyright (C) 1991, 1993, 1995, 1997 Free Software Foundation, Inc.
   2    This file is part of the GNU C Library.
   3    Contributed by Torbjorn Granlund (tege@sics.se).
   4
   5    The GNU C Library is free software; you can redistribute it and/or
   6    modify it under the terms of the GNU Library General Public License as
   7    published by the Free Software Foundation; either version 2 of the
   8    License, or (at your option) any later version.
   9
  10    The GNU C Library is distributed in the hope that it will be useful,
  11    but WITHOUT ANY WARRANTY; without even the implied warranty of
  12    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13    Library General Public License for more details.
  14
  15    You should have received a copy of the GNU Library General Public
  16    License along with the GNU C Library; see the file COPYING.LIB.  If not,
  17    write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  18    Boston, MA 02111-1307, USA.  */
  19
  20 #ifdef HAVE_CONFIG_H
  21 #include "config.h"
  22 #endif
  23
  24 #undef  __ptr_t
  25 #if defined (__cplusplus) || (defined (__STDC__) && __STDC__)
  26 #define __ptr_t void *
  27 #else /* Not C++ or ANSI C.  */
  28 #undef  const
  29 #define const
  30 #define __ptr_t char *
  31 #endif /* C++ or ANSI C.  */
  32
  33 #if defined (HAVE_STRING_H) || defined (_LIBC)
  34 #include <string.h>
  35 #endif
  36
  37 #undef memcmp
  38
  39 #ifdef _LIBC
  40
  41 #include <memcopy.h>
  42
  43 #else   /* Not in the GNU C library.  */
  44
  45 #include <sys/types.h>
  46
  47 /* Type to use for aligned memory operations.
  48    This should normally be the biggest type supported by a single load
  49    and store.  Must be an unsigned type.  */
  50 #define op_t    unsigned long int
  51 #define OPSIZ   (sizeof(op_t))
  52
  53 /* Threshold value for when to enter the unrolled loops.  */
  54 #define OP_T_THRES      16
  55
  56 /* Type to use for unaligned operations.  */
  57 typedef unsigned char byte;
  58
  59 #ifndef WORDS_BIGENDIAN
  60 #define MERGE(w0, sh_1, w1, sh_2) (((w0) >> (sh_1)) | ((w1) << (sh_2)))
  61 #else
  62 #define MERGE(w0, sh_1, w1, sh_2) (((w0) << (sh_1)) | ((w1) >> (sh_2)))
  63 #endif
  64
  65 #endif  /* In the GNU C library.  */
  66
  67 #ifdef WORDS_BIGENDIAN
  68 #define CMP_LT_OR_GT(a, b) ((a) > (b) ? 1 : -1)
  69 #else
  70 #define CMP_LT_OR_GT(a, b) memcmp_bytes ((a), (b))
  71 #endif
  72
  73 /* BE VERY CAREFUL IF YOU CHANGE THIS CODE!  */
  74
  75 /* The strategy of this memcmp is:
  76
  77    1. Compare bytes until one of the block pointers is aligned.
  78
  79    2. Compare using memcmp_common_alignment or
  80       memcmp_not_common_alignment, regarding the alignment of the other
  81       block after the initial byte operations.  The maximum number of
  82       full words (of type op_t) are compared in this way.
  83
  84    3. Compare the few remaining bytes.  */
  85
  86 #ifndef WORDS_BIGENDIAN
  87 /* memcmp_bytes -- Compare A and B bytewise in the byte order of the machine.
  88    A and B are known to be different.
  89    This is needed only on little-endian machines.  */
  90
  91 static int memcmp_bytes __P((op_t, op_t));
  92
  93 #ifdef  __GNUC__
  94 __inline
  95 #endif
  96 static int
  97 memcmp_bytes (a, b)
  98      op_t a, b;
  99 {
 100   long int srcp1 = (long int) &a;
 101   long int srcp2 = (long int) &b;
 102   op_t a0, b0;
 103
 104   do
 105     {
 106       a0 = ((byte *) srcp1)[0];
 107       b0 = ((byte *) srcp2)[0];
 108       srcp1 += 1;
 109       srcp2 += 1;
 110     }
 111   while (a0 == b0);
 112   return a0 - b0;
 113 }
 114 #endif
 115
 116 static int memcmp_common_alignment __P((long, long, size_t));
 117
 118 /* memcmp_common_alignment -- Compare blocks at SRCP1 and SRCP2 with LEN `op_t'
 119    objects (not LEN bytes!).  Both SRCP1 and SRCP2 should be aligned for
 120    memory operations on `op_t's.  */
 121 #ifdef  __GNUC__
 122 __inline
 123 #endif
 124 static int
 125 memcmp_common_alignment (srcp1, srcp2, len)
 126      long int srcp1;
 127      long int srcp2;
 128      size_t len;
 129 {
 130   op_t a0, a1;
 131   op_t b0, b1;
 132
 133   switch (len % 4)
 134     {
 135     default: /* Avoid warning about uninitialized local variables.  */
 136     case 2:
 137       a0 = ((op_t *) srcp1)[0];
 138       b0 = ((op_t *) srcp2)[0];
 139       srcp1 -= 2 * OPSIZ;
 140       srcp2 -= 2 * OPSIZ;
 141       len += 2;
 142       goto do1;
 143     case 3:
 144       a1 = ((op_t *) srcp1)[0];
 145       b1 = ((op_t *) srcp2)[0];
 146       srcp1 -= OPSIZ;
 147       srcp2 -= OPSIZ;
 148       len += 1;
 149       goto do2;
 150     case 0:
 151       if (OP_T_THRES <= 3 * OPSIZ && len == 0)
 152         return 0;
 153       a0 = ((op_t *) srcp1)[0];
 154       b0 = ((op_t *) srcp2)[0];
 155       goto do3;
 156     case 1:
 157       a1 = ((op_t *) srcp1)[0];
 158       b1 = ((op_t *) srcp2)[0];
 159       srcp1 += OPSIZ;
 160       srcp2 += OPSIZ;
 161       len -= 1;
 162       if (OP_T_THRES <= 3 * OPSIZ && len == 0)
 163         goto do0;
 164       /* Fall through.  */
 165     }
 166
 167   do
 168     {
 169       a0 = ((op_t *) srcp1)[0];
 170       b0 = ((op_t *) srcp2)[0];
 171       if (a1 != b1)
 172         return CMP_LT_OR_GT (a1, b1);
 173
 174     do3:
 175       a1 = ((op_t *) srcp1)[1];
 176       b1 = ((op_t *) srcp2)[1];
 177       if (a0 != b0)
 178         return CMP_LT_OR_GT (a0, b0);
 179
 180     do2:
 181       a0 = ((op_t *) srcp1)[2];
 182       b0 = ((op_t *) srcp2)[2];
 183       if (a1 != b1)
 184         return CMP_LT_OR_GT (a1, b1);
 185
 186     do1:
 187       a1 = ((op_t *) srcp1)[3];
 188       b1 = ((op_t *) srcp2)[3];
 189       if (a0 != b0)
 190         return CMP_LT_OR_GT (a0, b0);
 191
 192       srcp1 += 4 * OPSIZ;
 193       srcp2 += 4 * OPSIZ;
 194       len -= 4;
 195     }
 196   while (len != 0);
 197
 198   /* This is the right position for do0.  Please don't move
 199      it into the loop.  */
 200  do0:
 201   if (a1 != b1)
 202     return CMP_LT_OR_GT (a1, b1);
 203   return 0;
 204 }
 205
 206 static int memcmp_not_common_alignment __P((long, long, size_t));
 207
 208 /* memcmp_not_common_alignment -- Compare blocks at SRCP1 and SRCP2 with LEN
 209    `op_t' objects (not LEN bytes!).  SRCP2 should be aligned for memory
 210    operations on `op_t', but SRCP1 *should be unaligned*.  */
 211 #ifdef  __GNUC__
 212 __inline
 213 #endif
 214 static int
 215 memcmp_not_common_alignment (srcp1, srcp2, len)
 216      long int srcp1;
 217      long int srcp2;
 218      size_t len;
 219 {
 220   op_t a0, a1, a2, a3;
 221   op_t b0, b1, b2, b3;
 222   op_t x;
 223   int shl, shr;
 224
 225   /* Calculate how to shift a word read at the memory operation
 226      aligned srcp1 to make it aligned for comparison.  */
 227
 228   shl = 8 * (srcp1 % OPSIZ);
 229   shr = 8 * OPSIZ - shl;
 230
 231   /* Make SRCP1 aligned by rounding it down to the beginning of the `op_t'
 232      it points in the middle of.  */
 233   srcp1 &= -OPSIZ;
 234
 235   switch (len % 4)
 236     {
 237     default: /* Avoid warning about uninitialized local variables.  */
 238     case 2:
 239       a1 = ((op_t *) srcp1)[0];
 240       a2 = ((op_t *) srcp1)[1];
 241       b2 = ((op_t *) srcp2)[0];
 242       srcp1 -= 1 * OPSIZ;
 243       srcp2 -= 2 * OPSIZ;
 244       len += 2;
 245       goto do1;
 246     case 3:
 247       a0 = ((op_t *) srcp1)[0];
 248       a1 = ((op_t *) srcp1)[1];
 249       b1 = ((op_t *) srcp2)[0];
 250       srcp2 -= 1 * OPSIZ;
 251       len += 1;
 252       goto do2;
 253     case 0:
 254       if (OP_T_THRES <= 3 * OPSIZ && len == 0)
 255         return 0;
 256       a3 = ((op_t *) srcp1)[0];
 257       a0 = ((op_t *) srcp1)[1];
 258       b0 = ((op_t *) srcp2)[0];
 259       srcp1 += 1 * OPSIZ;
 260       goto do3;
 261     case 1:
 262       a2 = ((op_t *) srcp1)[0];
 263       a3 = ((op_t *) srcp1)[1];
 264       b3 = ((op_t *) srcp2)[0];
 265       srcp1 += 2 * OPSIZ;
 266       srcp2 += 1 * OPSIZ;
 267       len -= 1;
 268       if (OP_T_THRES <= 3 * OPSIZ && len == 0)
 269         goto do0;
 270       /* Fall through.  */
 271     }
 272
 273   do
 274     {
 275       a0 = ((op_t *) srcp1)[0];
 276       b0 = ((op_t *) srcp2)[0];
 277       x = MERGE(a2, shl, a3, shr);
 278       if (x != b3)
 279         return CMP_LT_OR_GT (x, b3);
 280
 281     do3:
 282       a1 = ((op_t *) srcp1)[1];
 283       b1 = ((op_t *) srcp2)[1];
 284       x = MERGE(a3, shl, a0, shr);
 285       if (x != b0)
 286         return CMP_LT_OR_GT (x, b0);
 287
 288     do2:
 289       a2 = ((op_t *) srcp1)[2];
 290       b2 = ((op_t *) srcp2)[2];
 291       x = MERGE(a0, shl, a1, shr);
 292       if (x != b1)
 293         return CMP_LT_OR_GT (x, b1);
 294
 295     do1:
 296       a3 = ((op_t *) srcp1)[3];
 297       b3 = ((op_t *) srcp2)[3];
 298       x = MERGE(a1, shl, a2, shr);
 299       if (x != b2)
 300         return CMP_LT_OR_GT (x, b2);
 301
 302       srcp1 += 4 * OPSIZ;
 303       srcp2 += 4 * OPSIZ;
 304       len -= 4;
 305     }
 306   while (len != 0);
 307
 308   /* This is the right position for do0.  Please don't move
 309      it into the loop.  */
 310  do0:
 311   x = MERGE(a2, shl, a3, shr);
 312   if (x != b3)
 313     return CMP_LT_OR_GT (x, b3);
 314   return 0;
 315 }
 316
 317 int
 318 memcmp (s1, s2, len)
 319      const __ptr_t s1;
 320      const __ptr_t s2;
 321      size_t len;
 322 {
 323   op_t a0;
 324   op_t b0;
 325   long int srcp1 = (long int) s1;
 326   long int srcp2 = (long int) s2;
 327   op_t res;
 328
 329   if (len >= OP_T_THRES)
 330     {
 331       /* There are at least some bytes to compare.  No need to test
 332          for LEN == 0 in this alignment loop.  */
 333       while (srcp2 % OPSIZ != 0)
 334         {
 335           a0 = ((byte *) srcp1)[0];
 336           b0 = ((byte *) srcp2)[0];
 337           srcp1 += 1;
 338           srcp2 += 1;
 339           res = a0 - b0;
 340           if (res != 0)
 341             return res;
 342           len -= 1;
 343         }
 344
 345       /* SRCP2 is now aligned for memory operations on `op_t'.
 346          SRCP1 alignment determines if we can do a simple,
 347          aligned compare or need to shuffle bits.  */
 348
 349       if (srcp1 % OPSIZ == 0)
 350         res = memcmp_common_alignment (srcp1, srcp2, len / OPSIZ);
 351       else
 352         res = memcmp_not_common_alignment (srcp1, srcp2, len / OPSIZ);
 353       if (res != 0)
 354         return res;
 355
 356       /* Number of bytes remaining in the interval [0..OPSIZ-1].  */
 357       srcp1 += len & -OPSIZ;
 358       srcp2 += len & -OPSIZ;
 359       len %= OPSIZ;
 360     }
 361
 362   /* There are just a few bytes to compare.  Use byte memory operations.  */
 363   while (len != 0)
 364     {
 365       a0 = ((byte *) srcp1)[0];
 366       b0 = ((byte *) srcp2)[0];
 367       srcp1 += 1;
 368       srcp2 += 1;
 369       res = a0 - b0;
 370       if (res != 0)
 371         return res;
 372       len -= 1;
 373     }
 374
 375   return 0;
 376 }
 377
 378 #ifdef weak_alias
 379 #undef bcmp
 380 weak_alias (memcmp, bcmp)
 381 #endif