string/memcmp.c

   1 /* Copyright (C) 1991-2015 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 Lesser General Public
   7    License as published by the Free Software Foundation; either
   8    version 2.1 of the 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    Lesser General Public License for more details.
  14
  15    You should have received a copy of the GNU Lesser General Public
  16    License along with the GNU C Library; if not, see
  17    <http://www.gnu.org/licenses/>.  */
  18
  19 #ifdef HAVE_CONFIG_H
  20 # include "config.h"
  21 #endif
  22
  23 #undef  __ptr_t
  24 #define __ptr_t void *
  25
  26 #if defined HAVE_STRING_H || defined _LIBC
  27 # include <string.h>
  28 #endif
  29
  30 #undef memcmp
  31
  32 #ifndef MEMCMP
  33 # define MEMCMP memcmp
  34 #endif
  35
  36 #ifdef _LIBC
  37
  38 # include <memcopy.h>
  39 # include <endian.h>
  40
  41 # if __BYTE_ORDER == __BIG_ENDIAN
  42 #  define WORDS_BIGENDIAN
  43 # endif
  44
  45 #else   /* Not in the GNU C library.  */
  46
  47 # include <sys/types.h>
  48
  49 /* Type to use for aligned memory operations.
  50    This should normally be the biggest type supported by a single load
  51    and store.  Must be an unsigned type.  */
  52 # define op_t   unsigned long int
  53 # define OPSIZ  (sizeof(op_t))
  54
  55 /* Threshold value for when to enter the unrolled loops.  */
  56 # define OP_T_THRES     16
  57
  58 /* Type to use for unaligned operations.  */
  59 typedef unsigned char byte;
  60
  61 # ifndef WORDS_BIGENDIAN
  62 #  define MERGE(w0, sh_1, w1, sh_2) (((w0) >> (sh_1)) | ((w1) << (sh_2)))
  63 # else
  64 #  define MERGE(w0, sh_1, w1, sh_2) (((w0) << (sh_1)) | ((w1) >> (sh_2)))
  65 # endif
  66
  67 #endif  /* In the GNU C library.  */
  68
  69 #ifdef WORDS_BIGENDIAN
  70 # define CMP_LT_OR_GT(a, b) ((a) > (b) ? 1 : -1)
  71 #else
  72 # define CMP_LT_OR_GT(a, b) memcmp_bytes ((a), (b))
  73 #endif
  74
  75 /* BE VERY CAREFUL IF YOU CHANGE THIS CODE!  */
  76
  77 /* The strategy of this memcmp is:
  78
  79    1. Compare bytes until one of the block pointers is aligned.
  80
  81    2. Compare using memcmp_common_alignment or
  82       memcmp_not_common_alignment, regarding the alignment of the other
  83       block after the initial byte operations.  The maximum number of
  84       full words (of type op_t) are compared in this way.
  85
  86    3. Compare the few remaining bytes.  */
  87
  88 #ifndef WORDS_BIGENDIAN
  89 /* memcmp_bytes -- Compare A and B bytewise in the byte order of the machine.
  90    A and B are known to be different.
  91    This is needed only on little-endian machines.  */
  92
  93 static int memcmp_bytes (op_t, op_t) __THROW;
  94
  95 static int
  96 memcmp_bytes (a, b)
  97      op_t a, b;
  98 {
  99   long int srcp1 = (long int) &a;
 100   long int srcp2 = (long int) &b;
 101   op_t a0, b0;
 102
 103   do
 104     {
 105       a0 = ((byte *) srcp1)[0];
 106       b0 = ((byte *) srcp2)[0];
 107       srcp1 += 1;
 108       srcp2 += 1;
 109     }
 110   while (a0 == b0);
 111   return a0 - b0;
 112 }
 113 #endif
 114
 115 static int memcmp_common_alignment (long, long, size_t) __THROW;
 116
 117 /* memcmp_common_alignment -- Compare blocks at SRCP1 and SRCP2 with LEN `op_t'
 118    objects (not LEN bytes!).  Both SRCP1 and SRCP2 should be aligned for
 119    memory operations on `op_t's.  */
 120 static int
 121 memcmp_common_alignment (long int srcp1, long int srcp2, size_t len)
 122 {
 123   op_t a0, a1;
 124   op_t b0, b1;
 125
 126   switch (len % 4)
 127     {
 128     default: /* Avoid warning about uninitialized local variables.  */
 129     case 2:
 130       a0 = ((op_t *) srcp1)[0];
 131       b0 = ((op_t *) srcp2)[0];
 132       srcp1 -= 2 * OPSIZ;
 133       srcp2 -= 2 * OPSIZ;
 134       len += 2;
 135       goto do1;
 136     case 3:
 137       a1 = ((op_t *) srcp1)[0];
 138       b1 = ((op_t *) srcp2)[0];
 139       srcp1 -= OPSIZ;
 140       srcp2 -= OPSIZ;
 141       len += 1;
 142       goto do2;
 143     case 0:
 144       if (OP_T_THRES <= 3 * OPSIZ && len == 0)
 145         return 0;
 146       a0 = ((op_t *) srcp1)[0];
 147       b0 = ((op_t *) srcp2)[0];
 148       goto do3;
 149     case 1:
 150       a1 = ((op_t *) srcp1)[0];
 151       b1 = ((op_t *) srcp2)[0];
 152       srcp1 += OPSIZ;
 153       srcp2 += OPSIZ;
 154       len -= 1;
 155       if (OP_T_THRES <= 3 * OPSIZ && len == 0)
 156         goto do0;
 157       /* Fall through.  */
 158     }
 159
 160   do
 161     {
 162       a0 = ((op_t *) srcp1)[0];
 163       b0 = ((op_t *) srcp2)[0];
 164       if (a1 != b1)
 165         return CMP_LT_OR_GT (a1, b1);
 166
 167     do3:
 168       a1 = ((op_t *) srcp1)[1];
 169       b1 = ((op_t *) srcp2)[1];
 170       if (a0 != b0)
 171         return CMP_LT_OR_GT (a0, b0);
 172
 173     do2:
 174       a0 = ((op_t *) srcp1)[2];
 175       b0 = ((op_t *) srcp2)[2];
 176       if (a1 != b1)
 177         return CMP_LT_OR_GT (a1, b1);
 178
 179     do1:
 180       a1 = ((op_t *) srcp1)[3];
 181       b1 = ((op_t *) srcp2)[3];
 182       if (a0 != b0)
 183         return CMP_LT_OR_GT (a0, b0);
 184
 185       srcp1 += 4 * OPSIZ;
 186       srcp2 += 4 * OPSIZ;
 187       len -= 4;
 188     }
 189   while (len != 0);
 190
 191   /* This is the right position for do0.  Please don't move
 192      it into the loop.  */
 193  do0:
 194   if (a1 != b1)
 195     return CMP_LT_OR_GT (a1, b1);
 196   return 0;
 197 }
 198
 199 static int memcmp_not_common_alignment (long, long, size_t) __THROW;
 200
 201 /* memcmp_not_common_alignment -- Compare blocks at SRCP1 and SRCP2 with LEN
 202    `op_t' objects (not LEN bytes!).  SRCP2 should be aligned for memory
 203    operations on `op_t', but SRCP1 *should be unaligned*.  */
 204 static int
 205 memcmp_not_common_alignment (long int srcp1, long int srcp2, size_t len)
 206 {
 207   op_t a0, a1, a2, a3;
 208   op_t b0, b1, b2, b3;
 209   op_t x;
 210   int shl, shr;
 211
 212   /* Calculate how to shift a word read at the memory operation
 213      aligned srcp1 to make it aligned for comparison.  */
 214
 215   shl = 8 * (srcp1 % OPSIZ);
 216   shr = 8 * OPSIZ - shl;
 217
 218   /* Make SRCP1 aligned by rounding it down to the beginning of the `op_t'
 219      it points in the middle of.  */
 220   srcp1 &= -OPSIZ;
 221
 222   switch (len % 4)
 223     {
 224     default: /* Avoid warning about uninitialized local variables.  */
 225     case 2:
 226       a1 = ((op_t *) srcp1)[0];
 227       a2 = ((op_t *) srcp1)[1];
 228       b2 = ((op_t *) srcp2)[0];
 229       srcp1 -= 1 * OPSIZ;
 230       srcp2 -= 2 * OPSIZ;
 231       len += 2;
 232       goto do1;
 233     case 3:
 234       a0 = ((op_t *) srcp1)[0];
 235       a1 = ((op_t *) srcp1)[1];
 236       b1 = ((op_t *) srcp2)[0];
 237       srcp2 -= 1 * OPSIZ;
 238       len += 1;
 239       goto do2;
 240     case 0:
 241       if (OP_T_THRES <= 3 * OPSIZ && len == 0)
 242         return 0;
 243       a3 = ((op_t *) srcp1)[0];
 244       a0 = ((op_t *) srcp1)[1];
 245       b0 = ((op_t *) srcp2)[0];
 246       srcp1 += 1 * OPSIZ;
 247       goto do3;
 248     case 1:
 249       a2 = ((op_t *) srcp1)[0];
 250       a3 = ((op_t *) srcp1)[1];
 251       b3 = ((op_t *) srcp2)[0];
 252       srcp1 += 2 * OPSIZ;
 253       srcp2 += 1 * OPSIZ;
 254       len -= 1;
 255       if (OP_T_THRES <= 3 * OPSIZ && len == 0)
 256         goto do0;
 257       /* Fall through.  */
 258     }
 259
 260   do
 261     {
 262       a0 = ((op_t *) srcp1)[0];
 263       b0 = ((op_t *) srcp2)[0];
 264       x = MERGE(a2, shl, a3, shr);
 265       if (x != b3)
 266         return CMP_LT_OR_GT (x, b3);
 267
 268     do3:
 269       a1 = ((op_t *) srcp1)[1];
 270       b1 = ((op_t *) srcp2)[1];
 271       x = MERGE(a3, shl, a0, shr);
 272       if (x != b0)
 273         return CMP_LT_OR_GT (x, b0);
 274
 275     do2:
 276       a2 = ((op_t *) srcp1)[2];
 277       b2 = ((op_t *) srcp2)[2];
 278       x = MERGE(a0, shl, a1, shr);
 279       if (x != b1)
 280         return CMP_LT_OR_GT (x, b1);
 281
 282     do1:
 283       a3 = ((op_t *) srcp1)[3];
 284       b3 = ((op_t *) srcp2)[3];
 285       x = MERGE(a1, shl, a2, shr);
 286       if (x != b2)
 287         return CMP_LT_OR_GT (x, b2);
 288
 289       srcp1 += 4 * OPSIZ;
 290       srcp2 += 4 * OPSIZ;
 291       len -= 4;
 292     }
 293   while (len != 0);
 294
 295   /* This is the right position for do0.  Please don't move
 296      it into the loop.  */
 297  do0:
 298   x = MERGE(a2, shl, a3, shr);
 299   if (x != b3)
 300     return CMP_LT_OR_GT (x, b3);
 301   return 0;
 302 }
 303
 304 int
 305 MEMCMP (const __ptr_t s1, const __ptr_t s2, size_t len)
 306 {
 307   op_t a0;
 308   op_t b0;
 309   long int srcp1 = (long int) s1;
 310   long int srcp2 = (long int) s2;
 311   op_t res;
 312
 313   if (len >= OP_T_THRES)
 314     {
 315       /* There are at least some bytes to compare.  No need to test
 316          for LEN == 0 in this alignment loop.  */
 317       while (srcp2 % OPSIZ != 0)
 318         {
 319           a0 = ((byte *) srcp1)[0];
 320           b0 = ((byte *) srcp2)[0];
 321           srcp1 += 1;
 322           srcp2 += 1;
 323           res = a0 - b0;
 324           if (res != 0)
 325             return res;
 326           len -= 1;
 327         }
 328
 329       /* SRCP2 is now aligned for memory operations on `op_t'.
 330          SRCP1 alignment determines if we can do a simple,
 331          aligned compare or need to shuffle bits.  */
 332
 333       if (srcp1 % OPSIZ == 0)
 334         res = memcmp_common_alignment (srcp1, srcp2, len / OPSIZ);
 335       else
 336         res = memcmp_not_common_alignment (srcp1, srcp2, len / OPSIZ);
 337       if (res != 0)
 338         return res;
 339
 340       /* Number of bytes remaining in the interval [0..OPSIZ-1].  */
 341       srcp1 += len & -OPSIZ;
 342       srcp2 += len & -OPSIZ;
 343       len %= OPSIZ;
 344     }
 345
 346   /* There are just a few bytes to compare.  Use byte memory operations.  */
 347   while (len != 0)
 348     {
 349       a0 = ((byte *) srcp1)[0];
 350       b0 = ((byte *) srcp2)[0];
 351       srcp1 += 1;
 352       srcp2 += 1;
 353       res = a0 - b0;
 354       if (res != 0)
 355         return res;
 356       len -= 1;
 357     }
 358
 359   return 0;
 360 }
 361 libc_hidden_builtin_def(memcmp)
 362 #ifdef weak_alias
 363 # undef bcmp
 364 weak_alias (memcmp, bcmp)
 365 #endif