sysdeps/tile/memcmp.c

   1 /* Copyright (C) 1991-2015 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 #ifdef HAVE_CONFIG_H
  19 # include "config.h"
  20 #endif
  21
  22 #undef  __ptr_t
  23 #define __ptr_t void *
  24
  25 #if defined HAVE_STRING_H || defined _LIBC
  26 # include <string.h>
  27 #endif
  28
  29 #undef memcmp
  30
  31 #ifndef MEMCMP
  32 # define MEMCMP memcmp
  33 #endif
  34
  35 #ifdef _LIBC
  36
  37 # include <memcopy.h>
  38 # include <endian.h>
  39
  40 # if __BYTE_ORDER == __BIG_ENDIAN
  41 #  define WORDS_BIGENDIAN
  42 # endif
  43
  44 #else   /* Not in the GNU C library.  */
  45
  46 # include <sys/types.h>
  47
  48 /* Type to use for aligned memory operations.
  49    This should normally be the biggest type supported by a single load
  50    and store.  Must be an unsigned type.  */
  51 # define op_t   unsigned long int
  52 # define OPSIZ  (sizeof(op_t))
  53
  54 /* Threshold value for when to enter the unrolled loops.  */
  55 # define OP_T_THRES     16
  56
  57 /* Type to use for unaligned operations.  */
  58 typedef unsigned char byte;
  59
  60 #endif  /* In the GNU C library.  */
  61
  62 /* Provide the appropriate builtins to shift two registers based on
  63    the alignment of a pointer held in a third register, and to reverse
  64    the bytes in a word.  */
  65 #ifdef __tilegx__
  66 #define DBLALIGN __insn_dblalign
  67 #define REVBYTES __insn_revbytes
  68 #else
  69 #define DBLALIGN __insn_dword_align
  70 #define REVBYTES __insn_bytex
  71 #endif
  72
  73 #ifdef WORDS_BIGENDIAN
  74 # define CMP_LT_OR_GT(a, b) ((a) > (b) ? 1 : -1)
  75 #else
  76 # define CMP_LT_OR_GT(a, b) (REVBYTES(a) > REVBYTES(b) ? 1 : -1)
  77 #endif
  78
  79 /* BE VERY CAREFUL IF YOU CHANGE THIS CODE!  */
  80
  81 /* The strategy of this memcmp is:
  82
  83    1. Compare bytes until one of the block pointers is aligned.
  84
  85    2. Compare using memcmp_common_alignment or
  86       memcmp_not_common_alignment, regarding the alignment of the other
  87       block after the initial byte operations.  The maximum number of
  88       full words (of type op_t) are compared in this way.
  89
  90    3. Compare the few remaining bytes.  */
  91
  92 static int memcmp_common_alignment (long, long, size_t) __THROW;
  93
  94 /* memcmp_common_alignment -- Compare blocks at SRCP1 and SRCP2 with LEN `op_t'
  95    objects (not LEN bytes!).  Both SRCP1 and SRCP2 should be aligned for
  96    memory operations on `op_t's.  */
  97 static int
  98 memcmp_common_alignment (srcp1, srcp2, len)
  99      long int srcp1;
 100      long int srcp2;
 101      size_t len;
 102 {
 103   op_t a0, a1;
 104   op_t b0, b1;
 105
 106   switch (len % 4)
 107     {
 108     default: /* Avoid warning about uninitialized local variables.  */
 109     case 2:
 110       a0 = ((op_t *) srcp1)[0];
 111       b0 = ((op_t *) srcp2)[0];
 112       srcp1 += OPSIZ;
 113       srcp2 += OPSIZ;
 114       len += 2;
 115       goto do1;
 116     case 3:
 117       a1 = ((op_t *) srcp1)[0];
 118       b1 = ((op_t *) srcp2)[0];
 119       srcp1 += OPSIZ;
 120       srcp2 += OPSIZ;
 121       len += 1;
 122       goto do2;
 123     case 0:
 124       if (OP_T_THRES <= 3 * OPSIZ && len == 0)
 125         return 0;
 126       a0 = ((op_t *) srcp1)[0];
 127       b0 = ((op_t *) srcp2)[0];
 128       srcp1 += OPSIZ;
 129       srcp2 += OPSIZ;
 130       goto do3;
 131     case 1:
 132       a1 = ((op_t *) srcp1)[0];
 133       b1 = ((op_t *) srcp2)[0];
 134       srcp1 += OPSIZ;
 135       srcp2 += OPSIZ;
 136       len -= 1;
 137       if (OP_T_THRES <= 3 * OPSIZ && len == 0)
 138         goto do0;
 139       /* Fall through.  */
 140     }
 141
 142   do
 143     {
 144       a0 = ((op_t *) srcp1)[0];
 145       b0 = ((op_t *) srcp2)[0];
 146       srcp1 += OPSIZ;
 147       srcp2 += OPSIZ;
 148       if (__glibc_likely (a1 != b1))
 149         return CMP_LT_OR_GT (a1, b1);
 150
 151     do3:
 152       a1 = ((op_t *) srcp1)[0];
 153       b1 = ((op_t *) srcp2)[0];
 154       srcp1 += OPSIZ;
 155       srcp2 += OPSIZ;
 156       if (__glibc_likely (a0 != b0))
 157         return CMP_LT_OR_GT (a0, b0);
 158
 159     do2:
 160       a0 = ((op_t *) srcp1)[0];
 161       b0 = ((op_t *) srcp2)[0];
 162       srcp1 += OPSIZ;
 163       srcp2 += OPSIZ;
 164       if (__glibc_likely (a1 != b1))
 165         return CMP_LT_OR_GT (a1, b1);
 166
 167     do1:
 168       a1 = ((op_t *) srcp1)[0];
 169       b1 = ((op_t *) srcp2)[0];
 170       srcp1 += OPSIZ;
 171       srcp2 += OPSIZ;
 172       if (__glibc_likely (a0 != b0))
 173         return CMP_LT_OR_GT (a0, b0);
 174
 175       len -= 4;
 176     }
 177   while (len != 0);
 178
 179   /* This is the right position for do0.  Please don't move
 180      it into the loop.  */
 181  do0:
 182   if (__glibc_likely (a1 != b1))
 183     return CMP_LT_OR_GT (a1, b1);
 184   return 0;
 185 }
 186
 187 static int memcmp_not_common_alignment (long, long, size_t) __THROW;
 188
 189 /* memcmp_not_common_alignment -- Compare blocks at SRCP1 and SRCP2 with LEN
 190    `op_t' objects (not LEN bytes!).  SRCP2 should be aligned for memory
 191    operations on `op_t', but SRCP1 *should be unaligned*.  */
 192 static int
 193 memcmp_not_common_alignment (srcp1, srcp2, len)
 194      long int srcp1;
 195      long int srcp2;
 196      size_t len;
 197 {
 198   void * srcp1i;
 199   op_t a0, a1, a2, a3;
 200   op_t b0, b1, b2, b3;
 201   op_t x;
 202
 203   /* Calculate how to shift a word read at the memory operation
 204      aligned srcp1 to make it aligned for comparison.  */
 205
 206   srcp1i = (void *) srcp1;
 207
 208   /* Make SRCP1 aligned by rounding it down to the beginning of the `op_t'
 209      it points in the middle of.  */
 210   srcp1 &= -OPSIZ;
 211
 212   switch (len % 4)
 213     {
 214     default: /* Avoid warning about uninitialized local variables.  */
 215     case 2:
 216       a1 = ((op_t *) srcp1)[0];
 217       a2 = ((op_t *) srcp1)[1];
 218       b2 = ((op_t *) srcp2)[0];
 219       srcp1 += 2 * OPSIZ;
 220       srcp2 += 1 * OPSIZ;
 221       len += 2;
 222       goto do1;
 223     case 3:
 224       a0 = ((op_t *) srcp1)[0];
 225       a1 = ((op_t *) srcp1)[1];
 226       b1 = ((op_t *) srcp2)[0];
 227       srcp1 += 2 * OPSIZ;
 228       srcp2 += 1 * OPSIZ;
 229       len += 1;
 230       goto do2;
 231     case 0:
 232       if (OP_T_THRES <= 3 * OPSIZ && len == 0)
 233         return 0;
 234       a3 = ((op_t *) srcp1)[0];
 235       a0 = ((op_t *) srcp1)[1];
 236       b0 = ((op_t *) srcp2)[0];
 237       srcp1 += 2 * OPSIZ;
 238       srcp2 += 1 * OPSIZ;
 239       goto do3;
 240     case 1:
 241       a2 = ((op_t *) srcp1)[0];
 242       a3 = ((op_t *) srcp1)[1];
 243       b3 = ((op_t *) srcp2)[0];
 244       srcp1 += 2 * OPSIZ;
 245       srcp2 += 1 * OPSIZ;
 246       len -= 1;
 247       if (OP_T_THRES <= 3 * OPSIZ && len == 0)
 248         goto do0;
 249       /* Fall through.  */
 250     }
 251
 252   do
 253     {
 254       a0 = ((op_t *) srcp1)[0];
 255       b0 = ((op_t *) srcp2)[0];
 256       x = DBLALIGN (a2, a3, srcp1i);
 257       srcp1 += OPSIZ;
 258       srcp2 += OPSIZ;
 259       if (__glibc_likely (x != b3))
 260         return CMP_LT_OR_GT (x, b3);
 261
 262     do3:
 263       a1 = ((op_t *) srcp1)[0];
 264       b1 = ((op_t *) srcp2)[0];
 265       x = DBLALIGN (a3, a0, srcp1i);
 266       srcp1 += OPSIZ;
 267       srcp2 += OPSIZ;
 268       if (__glibc_likely (x != b0))
 269         return CMP_LT_OR_GT (x, b0);
 270
 271     do2:
 272       a2 = ((op_t *) srcp1)[0];
 273       b2 = ((op_t *) srcp2)[0];
 274       x = DBLALIGN (a0, a1, srcp1i);
 275       srcp1 += OPSIZ;
 276       srcp2 += OPSIZ;
 277       if (__glibc_likely (x != b1))
 278         return CMP_LT_OR_GT (x, b1);
 279
 280     do1:
 281       a3 = ((op_t *) srcp1)[0];
 282       b3 = ((op_t *) srcp2)[0];
 283       x = DBLALIGN (a1, a2, srcp1i);
 284       srcp1 += OPSIZ;
 285       srcp2 += OPSIZ;
 286       if (__glibc_likely (x != b2))
 287         return CMP_LT_OR_GT (x, b2);
 288
 289       len -= 4;
 290     }
 291   while (len != 0);
 292
 293   /* This is the right position for do0.  Please don't move
 294      it into the loop.  */
 295  do0:
 296   x = DBLALIGN (a2, a3, srcp1i);
 297   if (__glibc_likely (x != b3))
 298     return CMP_LT_OR_GT (x, b3);
 299   return 0;
 300 }
 301
 302 int
 303 MEMCMP (s1, s2, len)
 304      const __ptr_t s1;
 305      const __ptr_t s2;
 306      size_t len;
 307 {
 308   op_t a0;
 309   op_t b0;
 310   long int srcp1 = (long int) s1;
 311   long int srcp2 = (long int) s2;
 312   int res;
 313
 314   if (len >= OP_T_THRES)
 315     {
 316       /* There are at least some bytes to compare.  No need to test
 317          for LEN == 0 in this alignment loop.  */
 318       while (srcp2 % OPSIZ != 0)
 319         {
 320           a0 = ((byte *) srcp1)[0];
 321           b0 = ((byte *) srcp2)[0];
 322           srcp1 += 1;
 323           srcp2 += 1;
 324           res = a0 - b0;
 325           if (__glibc_likely (res != 0))
 326             return res;
 327           len -= 1;
 328         }
 329
 330       /* SRCP2 is now aligned for memory operations on `op_t'.
 331          SRCP1 alignment determines if we can do a simple,
 332          aligned compare or need to shuffle bits.  */
 333
 334       if (srcp1 % OPSIZ == 0)
 335         res = memcmp_common_alignment (srcp1, srcp2, len / OPSIZ);
 336       else
 337         res = memcmp_not_common_alignment (srcp1, srcp2, len / OPSIZ);
 338       if (res != 0)
 339         return res;
 340
 341       /* Number of bytes remaining in the interval [0..OPSIZ-1].  */
 342       srcp1 += len & -OPSIZ;
 343       srcp2 += len & -OPSIZ;
 344       len %= OPSIZ;
 345     }
 346
 347   /* There are just a few bytes to compare.  Use byte memory operations.  */
 348   while (len != 0)
 349     {
 350       a0 = ((byte *) srcp1)[0];
 351       b0 = ((byte *) srcp2)[0];
 352       srcp1 += 1;
 353       srcp2 += 1;
 354       res = a0 - b0;
 355       if (__glibc_likely (res != 0))
 356         return res;
 357       len -= 1;
 358     }
 359
 360   return 0;
 361 }
 362 libc_hidden_builtin_def(memcmp)
 363 #ifdef weak_alias
 364 # undef bcmp
 365 weak_alias (memcmp, bcmp)
 366 #endif