sysdeps/tile/memcmp.c

   1 /* Copyright (C) 1991-2017 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 #if defined HAVE_STRING_H || defined _LIBC
  23 # include <string.h>
  24 #endif
  25
  26 #undef memcmp
  27
  28 #ifndef MEMCMP
  29 # define MEMCMP memcmp
  30 #endif
  31
  32 #ifdef _LIBC
  33
  34 # include <memcopy.h>
  35 # include <endian.h>
  36
  37 # if __BYTE_ORDER == __BIG_ENDIAN
  38 #  define WORDS_BIGENDIAN
  39 # endif
  40
  41 #else   /* Not in the GNU C library.  */
  42
  43 # include <sys/types.h>
  44
  45 /* Type to use for aligned memory operations.
  46    This should normally be the biggest type supported by a single load
  47    and store.  Must be an unsigned type.  */
  48 # define op_t   unsigned long int
  49 # define OPSIZ  (sizeof(op_t))
  50
  51 /* Threshold value for when to enter the unrolled loops.  */
  52 # define OP_T_THRES     16
  53
  54 /* Type to use for unaligned operations.  */
  55 typedef unsigned char byte;
  56
  57 #endif  /* In the GNU C library.  */
  58
  59 /* Provide the appropriate builtins to shift two registers based on
  60    the alignment of a pointer held in a third register, and to reverse
  61    the bytes in a word.  */
  62 #ifdef __tilegx__
  63 #define DBLALIGN __insn_dblalign
  64 #define REVBYTES __insn_revbytes
  65 #else
  66 #define DBLALIGN __insn_dword_align
  67 #define REVBYTES __insn_bytex
  68 #endif
  69
  70 #ifdef WORDS_BIGENDIAN
  71 # define CMP_LT_OR_GT(a, b) ((a) > (b) ? 1 : -1)
  72 #else
  73 # define CMP_LT_OR_GT(a, b) (REVBYTES(a) > REVBYTES(b) ? 1 : -1)
  74 #endif
  75
  76 /* BE VERY CAREFUL IF YOU CHANGE THIS CODE!  */
  77
  78 /* The strategy of this memcmp is:
  79
  80    1. Compare bytes until one of the block pointers is aligned.
  81
  82    2. Compare using memcmp_common_alignment or
  83       memcmp_not_common_alignment, regarding the alignment of the other
  84       block after the initial byte operations.  The maximum number of
  85       full words (of type op_t) are compared in this way.
  86
  87    3. Compare the few remaining bytes.  */
  88
  89 static int memcmp_common_alignment (long, long, size_t) __THROW;
  90
  91 /* memcmp_common_alignment -- Compare blocks at SRCP1 and SRCP2 with LEN `op_t'
  92    objects (not LEN bytes!).  Both SRCP1 and SRCP2 should be aligned for
  93    memory operations on `op_t's.  */
  94 static int
  95 memcmp_common_alignment (long int srcp1, long int srcp2, size_t len)
  96 {
  97   op_t a0, a1;
  98   op_t b0, b1;
  99
 100   switch (len % 4)
 101     {
 102     default: /* Avoid warning about uninitialized local variables.  */
 103     case 2:
 104       a0 = ((op_t *) srcp1)[0];
 105       b0 = ((op_t *) srcp2)[0];
 106       srcp1 += OPSIZ;
 107       srcp2 += OPSIZ;
 108       len += 2;
 109       goto do1;
 110     case 3:
 111       a1 = ((op_t *) srcp1)[0];
 112       b1 = ((op_t *) srcp2)[0];
 113       srcp1 += OPSIZ;
 114       srcp2 += OPSIZ;
 115       len += 1;
 116       goto do2;
 117     case 0:
 118       if (OP_T_THRES <= 3 * OPSIZ && len == 0)
 119         return 0;
 120       a0 = ((op_t *) srcp1)[0];
 121       b0 = ((op_t *) srcp2)[0];
 122       srcp1 += OPSIZ;
 123       srcp2 += OPSIZ;
 124       goto do3;
 125     case 1:
 126       a1 = ((op_t *) srcp1)[0];
 127       b1 = ((op_t *) srcp2)[0];
 128       srcp1 += OPSIZ;
 129       srcp2 += OPSIZ;
 130       len -= 1;
 131       if (OP_T_THRES <= 3 * OPSIZ && len == 0)
 132         goto do0;
 133       /* Fall through.  */
 134     }
 135
 136   do
 137     {
 138       a0 = ((op_t *) srcp1)[0];
 139       b0 = ((op_t *) srcp2)[0];
 140       srcp1 += OPSIZ;
 141       srcp2 += OPSIZ;
 142       if (__glibc_likely (a1 != b1))
 143         return CMP_LT_OR_GT (a1, b1);
 144
 145     do3:
 146       a1 = ((op_t *) srcp1)[0];
 147       b1 = ((op_t *) srcp2)[0];
 148       srcp1 += OPSIZ;
 149       srcp2 += OPSIZ;
 150       if (__glibc_likely (a0 != b0))
 151         return CMP_LT_OR_GT (a0, b0);
 152
 153     do2:
 154       a0 = ((op_t *) srcp1)[0];
 155       b0 = ((op_t *) srcp2)[0];
 156       srcp1 += OPSIZ;
 157       srcp2 += OPSIZ;
 158       if (__glibc_likely (a1 != b1))
 159         return CMP_LT_OR_GT (a1, b1);
 160
 161     do1:
 162       a1 = ((op_t *) srcp1)[0];
 163       b1 = ((op_t *) srcp2)[0];
 164       srcp1 += OPSIZ;
 165       srcp2 += OPSIZ;
 166       if (__glibc_likely (a0 != b0))
 167         return CMP_LT_OR_GT (a0, b0);
 168
 169       len -= 4;
 170     }
 171   while (len != 0);
 172
 173   /* This is the right position for do0.  Please don't move
 174      it into the loop.  */
 175  do0:
 176   if (__glibc_likely (a1 != b1))
 177     return CMP_LT_OR_GT (a1, b1);
 178   return 0;
 179 }
 180
 181 static int memcmp_not_common_alignment (long, long, size_t) __THROW;
 182
 183 /* memcmp_not_common_alignment -- Compare blocks at SRCP1 and SRCP2 with LEN
 184    `op_t' objects (not LEN bytes!).  SRCP2 should be aligned for memory
 185    operations on `op_t', but SRCP1 *should be unaligned*.  */
 186 static int
 187 memcmp_not_common_alignment (long int srcp1, long int srcp2, size_t len)
 188 {
 189   void * srcp1i;
 190   op_t a0, a1, a2, a3;
 191   op_t b0, b1, b2, b3;
 192   op_t x;
 193
 194   /* Calculate how to shift a word read at the memory operation
 195      aligned srcp1 to make it aligned for comparison.  */
 196
 197   srcp1i = (void *) srcp1;
 198
 199   /* Make SRCP1 aligned by rounding it down to the beginning of the `op_t'
 200      it points in the middle of.  */
 201   srcp1 &= -OPSIZ;
 202
 203   switch (len % 4)
 204     {
 205     default: /* Avoid warning about uninitialized local variables.  */
 206     case 2:
 207       a1 = ((op_t *) srcp1)[0];
 208       a2 = ((op_t *) srcp1)[1];
 209       b2 = ((op_t *) srcp2)[0];
 210       srcp1 += 2 * OPSIZ;
 211       srcp2 += 1 * OPSIZ;
 212       len += 2;
 213       goto do1;
 214     case 3:
 215       a0 = ((op_t *) srcp1)[0];
 216       a1 = ((op_t *) srcp1)[1];
 217       b1 = ((op_t *) srcp2)[0];
 218       srcp1 += 2 * OPSIZ;
 219       srcp2 += 1 * OPSIZ;
 220       len += 1;
 221       goto do2;
 222     case 0:
 223       if (OP_T_THRES <= 3 * OPSIZ && len == 0)
 224         return 0;
 225       a3 = ((op_t *) srcp1)[0];
 226       a0 = ((op_t *) srcp1)[1];
 227       b0 = ((op_t *) srcp2)[0];
 228       srcp1 += 2 * OPSIZ;
 229       srcp2 += 1 * OPSIZ;
 230       goto do3;
 231     case 1:
 232       a2 = ((op_t *) srcp1)[0];
 233       a3 = ((op_t *) srcp1)[1];
 234       b3 = ((op_t *) srcp2)[0];
 235       srcp1 += 2 * OPSIZ;
 236       srcp2 += 1 * OPSIZ;
 237       len -= 1;
 238       if (OP_T_THRES <= 3 * OPSIZ && len == 0)
 239         goto do0;
 240       /* Fall through.  */
 241     }
 242
 243   do
 244     {
 245       a0 = ((op_t *) srcp1)[0];
 246       b0 = ((op_t *) srcp2)[0];
 247       x = DBLALIGN (a2, a3, srcp1i);
 248       srcp1 += OPSIZ;
 249       srcp2 += OPSIZ;
 250       if (__glibc_likely (x != b3))
 251         return CMP_LT_OR_GT (x, b3);
 252
 253     do3:
 254       a1 = ((op_t *) srcp1)[0];
 255       b1 = ((op_t *) srcp2)[0];
 256       x = DBLALIGN (a3, a0, srcp1i);
 257       srcp1 += OPSIZ;
 258       srcp2 += OPSIZ;
 259       if (__glibc_likely (x != b0))
 260         return CMP_LT_OR_GT (x, b0);
 261
 262     do2:
 263       a2 = ((op_t *) srcp1)[0];
 264       b2 = ((op_t *) srcp2)[0];
 265       x = DBLALIGN (a0, a1, srcp1i);
 266       srcp1 += OPSIZ;
 267       srcp2 += OPSIZ;
 268       if (__glibc_likely (x != b1))
 269         return CMP_LT_OR_GT (x, b1);
 270
 271     do1:
 272       a3 = ((op_t *) srcp1)[0];
 273       b3 = ((op_t *) srcp2)[0];
 274       x = DBLALIGN (a1, a2, srcp1i);
 275       srcp1 += OPSIZ;
 276       srcp2 += OPSIZ;
 277       if (__glibc_likely (x != b2))
 278         return CMP_LT_OR_GT (x, b2);
 279
 280       len -= 4;
 281     }
 282   while (len != 0);
 283
 284   /* This is the right position for do0.  Please don't move
 285      it into the loop.  */
 286  do0:
 287   x = DBLALIGN (a2, a3, srcp1i);
 288   if (__glibc_likely (x != b3))
 289     return CMP_LT_OR_GT (x, b3);
 290   return 0;
 291 }
 292
 293 int
 294 MEMCMP (const void *s1, const void *s2, size_t len)
 295 {
 296   op_t a0;
 297   op_t b0;
 298   long int srcp1 = (long int) s1;
 299   long int srcp2 = (long int) s2;
 300   int res;
 301
 302   if (len >= OP_T_THRES)
 303     {
 304       /* There are at least some bytes to compare.  No need to test
 305          for LEN == 0 in this alignment loop.  */
 306       while (srcp2 % OPSIZ != 0)
 307         {
 308           a0 = ((byte *) srcp1)[0];
 309           b0 = ((byte *) srcp2)[0];
 310           srcp1 += 1;
 311           srcp2 += 1;
 312           res = a0 - b0;
 313           if (__glibc_likely (res != 0))
 314             return res;
 315           len -= 1;
 316         }
 317
 318       /* SRCP2 is now aligned for memory operations on `op_t'.
 319          SRCP1 alignment determines if we can do a simple,
 320          aligned compare or need to shuffle bits.  */
 321
 322       if (srcp1 % OPSIZ == 0)
 323         res = memcmp_common_alignment (srcp1, srcp2, len / OPSIZ);
 324       else
 325         res = memcmp_not_common_alignment (srcp1, srcp2, len / OPSIZ);
 326       if (res != 0)
 327         return res;
 328
 329       /* Number of bytes remaining in the interval [0..OPSIZ-1].  */
 330       srcp1 += len & -OPSIZ;
 331       srcp2 += len & -OPSIZ;
 332       len %= OPSIZ;
 333     }
 334
 335   /* There are just a few bytes to compare.  Use byte memory operations.  */
 336   while (len != 0)
 337     {
 338       a0 = ((byte *) srcp1)[0];
 339       b0 = ((byte *) srcp2)[0];
 340       srcp1 += 1;
 341       srcp2 += 1;
 342       res = a0 - b0;
 343       if (__glibc_likely (res != 0))
 344         return res;
 345       len -= 1;
 346     }
 347
 348   return 0;
 349 }
 350 libc_hidden_builtin_def(memcmp)
 351 #ifdef weak_alias
 352 # undef bcmp
 353 weak_alias (memcmp, bcmp)
 354 #endif