stdlib/merge.c

   1 /*
   2  * Copyright (c) 1999 Apple Computer, Inc. All rights reserved.
   3  *
   4  * @APPLE_LICENSE_HEADER_START@
   5  *
   6  * Copyright (c) 1999-2003 Apple Computer, Inc.  All Rights Reserved.
   7  *
   8  * This file contains Original Code and/or Modifications of Original Code
   9  * as defined in and that are subject to the Apple Public Source License
  10  * Version 2.0 (the 'License'). You may not use this file except in
  11  * compliance with the License. Please obtain a copy of the License at
  12  * http://www.opensource.apple.com/apsl/ and read it before using this
  13  * file.
  14  *
  15  * The Original Code and all software distributed under the License are
  16  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  17  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  18  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  20  * Please see the License for the specific language governing rights and
  21  * limitations under the License.
  22  *
  23  * @APPLE_LICENSE_HEADER_END@
  24  */
  25 /*
  26  * Copyright (c) 1992, 1993
  27  *      The Regents of the University of California.  All rights reserved.
  28  *
  29  * This code is derived from software contributed to Berkeley by
  30  * Peter McIlroy.
  31  *
  32  * Redistribution and use in source and binary forms, with or without
  33  * modification, are permitted provided that the following conditions
  34  * are met:
  35  * 1. Redistributions of source code must retain the above copyright
  36  *    notice, this list of conditions and the following disclaimer.
  37  * 2. Redistributions in binary form must reproduce the above copyright
  38  *    notice, this list of conditions and the following disclaimer in the
  39  *    documentation and/or other materials provided with the distribution.
  40  * 3. All advertising materials mentioning features or use of this software
  41  *    must display the following acknowledgement:
  42  *      This product includes software developed by the University of
  43  *      California, Berkeley and its contributors.
  44  * 4. Neither the name of the University nor the names of its contributors
  45  *    may be used to endorse or promote products derived from this software
  46  *    without specific prior written permission.
  47  *
  48  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  49  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  50  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  51  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  52  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  53  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  54  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  55  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  56  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  57  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  58  * SUCH DAMAGE.
  59  */
  60
  61
  62 /*
  63  * Hybrid exponential search/linear search merge sort with hybrid
  64  * natural/pairwise first pass.  Requires about .3% more comparisons
  65  * for random data than LSMS with pairwise first pass alone.
  66  * It works for objects as small as two bytes.
  67  */
  68
  69 #define NATURAL
  70 #define THRESHOLD 16    /* Best choice for natural merge cut-off. */
  71
  72 /* #define NATURAL to get hybrid natural merge.
  73  * (The default is pairwise merging.)
  74  */
  75
  76 #include <sys/types.h>
  77
  78 #include <errno.h>
  79 #include <stdlib.h>
  80 #include <string.h>
  81
  82 static void setup __P((u_char *, u_char *, size_t, size_t, int (*)()));
  83 static void insertionsort __P((u_char *, size_t, size_t, int (*)()));
  84
  85 #define ISIZE sizeof(int)
  86 #define PSIZE sizeof(u_char *)
  87 #define ICOPY_LIST(src, dst, last)                              \
  88         do                                                      \
  89         *(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE;    \
  90         while(src < last)
  91 #define ICOPY_ELT(src, dst, i)                                  \
  92         do                                                      \
  93         *(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE;  \
  94         while (i -= ISIZE)
  95
  96 #define CCOPY_LIST(src, dst, last)              \
  97         do                                      \
  98                 *dst++ = *src++;                \
  99         while (src < last)
 100 #define CCOPY_ELT(src, dst, i)                  \
 101         do                                      \
 102                 *dst++ = *src++;                \
 103         while (i -= 1)
 104
 105 /*
 106  * Find the next possible pointer head.  (Trickery for forcing an array
 107  * to do double duty as a linked list when objects do not align with word
 108  * boundaries.
 109  */
 110 /* Assumption: PSIZE is a power of 2. */
 111 #define EVAL(p) (u_char **)                                             \
 112         ((u_char *)0 +                                                  \
 113             (((u_char *)p + PSIZE - 1 - (u_char *) 0) & ~(PSIZE - 1)))
 114
 115 /*
 116  * Arguments are as for qsort.
 117  */
 118 int
 119 mergesort(base, nmemb, size, cmp)
 120         void *base;
 121         size_t nmemb;
 122         register size_t size;
 123         int (*cmp) __P((const void *, const void *));
 124 {
 125         register int i, sense;
 126         int big, iflag;
 127         register u_char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2;
 128         u_char *list2, *list1, *p2, *p, *last, **p1;
 129
 130         if (size < PSIZE / 2) {         /* Pointers must fit into 2 * size. */
 131                 errno = EINVAL;
 132                 return (-1);
 133         }
 134
 135         /*
 136          * XXX
 137          * Stupid subtraction for the Cray.
 138          */
 139         iflag = 0;
 140         if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE))
 141                 iflag = 1;
 142
 143         if ((list2 = malloc(nmemb * size + PSIZE)) == NULL)
 144                 return (-1);
 145
 146         list1 = base;
 147         setup(list1, list2, nmemb, size, cmp);
 148         last = list2 + nmemb * size;
 149         i = big = 0;
 150         while (*EVAL(list2) != last) {
 151             l2 = list1;
 152             p1 = EVAL(list1);
 153             for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) {
 154                 p2 = *EVAL(p2);
 155                 f1 = l2;
 156                 f2 = l1 = list1 + (p2 - list2);
 157                 if (p2 != last)
 158                         p2 = *EVAL(p2);
 159                 l2 = list1 + (p2 - list2);
 160                 while (f1 < l1 && f2 < l2) {
 161                         if ((*cmp)(f1, f2) <= 0) {
 162                                 q = f2;
 163                                 b = f1, t = l1;
 164                                 sense = -1;
 165                         } else {
 166                                 q = f1;
 167                                 b = f2, t = l2;
 168                                 sense = 0;
 169                         }
 170                         if (!big) {     /* here i = 0 */
 171 LINEAR:                         while ((b += size) < t && cmp(q, b) >sense)
 172                                         if (++i == 6) {
 173                                                 big = 1;
 174                                                 goto EXPONENTIAL;
 175                                         }
 176                         } else {
 177 EXPONENTIAL:                    for (i = size; ; i <<= 1)
 178                                         if ((p = (b + i)) >= t) {
 179                                                 if ((p = t - size) > b &&
 180                                                     (*cmp)(q, p) <= sense)
 181                                                         t = p;
 182                                                 else
 183                                                         b = p;
 184                                                 break;
 185                                         } else if ((*cmp)(q, p) <= sense) {
 186                                                 t = p;
 187                                                 if (i == size)
 188                                                         big = 0;
 189                                                 goto FASTCASE;
 190                                         } else
 191                                                 b = p;
 192 SLOWCASE:                       while (t > b+size) {
 193                                         i = (((t - b) / size) >> 1) * size;
 194                                         if ((*cmp)(q, p = b + i) <= sense)
 195                                                 t = p;
 196                                         else
 197                                                 b = p;
 198                                 }
 199                                 goto COPY;
 200 FASTCASE:                       while (i > size)
 201                                         if ((*cmp)(q,
 202                                                 p = b + (i >>= 1)) <= sense)
 203                                                 t = p;
 204                                         else
 205                                                 b = p;
 206 COPY:                           b = t;
 207                         }
 208                         i = size;
 209                         if (q == f1) {
 210                                 if (iflag) {
 211                                         ICOPY_LIST(f2, tp2, b);
 212                                         ICOPY_ELT(f1, tp2, i);
 213                                 } else {
 214                                         CCOPY_LIST(f2, tp2, b);
 215                                         CCOPY_ELT(f1, tp2, i);
 216                                 }
 217                         } else {
 218                                 if (iflag) {
 219                                         ICOPY_LIST(f1, tp2, b);
 220                                         ICOPY_ELT(f2, tp2, i);
 221                                 } else {
 222                                         CCOPY_LIST(f1, tp2, b);
 223                                         CCOPY_ELT(f2, tp2, i);
 224                                 }
 225                         }
 226                 }
 227                 if (f2 < l2) {
 228                         if (iflag)
 229                                 ICOPY_LIST(f2, tp2, l2);
 230                         else
 231                                 CCOPY_LIST(f2, tp2, l2);
 232                 } else if (f1 < l1) {
 233                         if (iflag)
 234                                 ICOPY_LIST(f1, tp2, l1);
 235                         else
 236                                 CCOPY_LIST(f1, tp2, l1);
 237                 }
 238                 *p1 = l2;
 239             }
 240             tp2 = list1;        /* swap list1, list2 */
 241             list1 = list2;
 242             list2 = tp2;
 243             last = list2 + nmemb*size;
 244         }
 245         if (base == list2) {
 246                 memmove(list2, list1, nmemb*size);
 247                 list2 = list1;
 248         }
 249         free(list2);
 250         return (0);
 251 }
 252
 253 #define swap(a, b) {                                    \
 254                 s = b;                                  \
 255                 i = size;                               \
 256                 do {                                    \
 257                         tmp = *a; *a++ = *s; *s++ = tmp; \
 258                 } while (--i);                          \
 259                 a -= size;                              \
 260         }
 261 #define reverse(bot, top) {                             \
 262         s = top;                                        \
 263         do {                                            \
 264                 i = size;                               \
 265                 do {                                    \
 266                         tmp = *bot; *bot++ = *s; *s++ = tmp; \
 267                 } while (--i);                          \
 268                 s -= size2;                             \
 269         } while(bot < s);                               \
 270 }
 271
 272 /*
 273  * Optional hybrid natural/pairwise first pass.  Eats up list1 in runs of
 274  * increasing order, list2 in a corresponding linked list.  Checks for runs
 275  * when THRESHOLD/2 pairs compare with same sense.  (Only used when NATURAL
 276  * is defined.  Otherwise simple pairwise merging is used.)
 277  */
 278 void
 279 setup(list1, list2, n, size, cmp)
 280         size_t n, size;
 281         int (*cmp) __P((const void *, const void *));
 282         u_char *list1, *list2;
 283 {
 284         int i, length, size2, tmp, sense;
 285         u_char *f1, *f2, *s, *l2, *last, *p2;
 286
 287         size2 = size*2;
 288         if (n <= 5) {
 289                 insertionsort(list1, n, size, cmp);
 290                 *EVAL(list2) = (u_char*) list2 + n*size;
 291                 return;
 292         }
 293         /*
 294          * Avoid running pointers out of bounds; limit n to evens
 295          * for simplicity.
 296          */
 297         i = 4 + (n & 1);
 298         insertionsort(list1 + (n - i) * size, i, size, cmp);
 299         last = list1 + size * (n - i);
 300         *EVAL(list2 + (last - list1)) = list2 + n * size;
 301
 302 #ifdef NATURAL
 303         p2 = list2;
 304         f1 = list1;
 305         sense = (cmp(f1, f1 + size) > 0);
 306         for (; f1 < last; sense = !sense) {
 307                 length = 2;
 308                                         /* Find pairs with same sense. */
 309                 for (f2 = f1 + size2; f2 < last; f2 += size2) {
 310                         if ((cmp(f2, f2+ size) > 0) != sense)
 311                                 break;
 312                         length += 2;
 313                 }
 314                 if (length < THRESHOLD) {               /* Pairwise merge */
 315                         do {
 316                                 p2 = *EVAL(p2) = f1 + size2 - list1 + list2;
 317                                 if (sense > 0)
 318                                         swap (f1, f1 + size);
 319                         } while ((f1 += size2) < f2);
 320                 } else {                                /* Natural merge */
 321                         l2 = f2;
 322                         for (f2 = f1 + size2; f2 < l2; f2 += size2) {
 323                                 if ((cmp(f2-size, f2) > 0) != sense) {
 324                                         p2 = *EVAL(p2) = f2 - list1 + list2;
 325                                         if (sense > 0)
 326                                                 reverse(f1, f2-size);
 327                                         f1 = f2;
 328                                 }
 329                         }
 330                         if (sense > 0)
 331                                 reverse (f1, f2-size);
 332                         f1 = f2;
 333                         if (f2 < last || cmp(f2 - size, f2) > 0)
 334                                 p2 = *EVAL(p2) = f2 - list1 + list2;
 335                         else
 336                                 p2 = *EVAL(p2) = list2 + n*size;
 337                 }
 338         }
 339 #else           /* pairwise merge only. */
 340         for (f1 = list1, p2 = list2; f1 < last; f1 += size2) {
 341                 p2 = *EVAL(p2) = p2 + size2;
 342                 if (cmp (f1, f1 + size) > 0)
 343                         swap(f1, f1 + size);
 344         }
 345 #endif /* NATURAL */
 346 }
 347
 348 /*
 349  * This is to avoid out-of-bounds addresses in sorting the
 350  * last 4 elements.
 351  */
 352 static void
 353 insertionsort(a, n, size, cmp)
 354         u_char *a;
 355         size_t n, size;
 356         int (*cmp) __P((const void *, const void *));
 357 {
 358         u_char *ai, *s, *t, *u, tmp;
 359         int i;
 360
 361         for (ai = a+size; --n >= 1; ai += size)
 362                 for (t = ai; t > a; t -= size) {
 363                         u = t - size;
 364                         if (cmp(u, t) <= 0)
 365                                 break;
 366                         swap(u, t);
 367                 }
 368 }