stdlib/FreeBSD/heapsort.c

   1 /*-
   2  * Copyright (c) 1991, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * This code is derived from software contributed to Berkeley by
   6  * Ronnie Kon at Mindcraft Inc., Kevin Lew and Elmer Yglesias.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. All advertising materials mentioning features or use of this software
  17  *    must display the following acknowledgement:
  18  *      This product includes software developed by the University of
  19  *      California, Berkeley and its contributors.
  20  * 4. Neither the name of the University nor the names of its contributors
  21  *    may be used to endorse or promote products derived from this software
  22  *    without specific prior written permission.
  23  *
  24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  34  * SUCH DAMAGE.
  35  */
  36
  37 #if defined(LIBC_SCCS) && !defined(lint)
  38 static char sccsid[] = "@(#)heapsort.c  8.1 (Berkeley) 6/4/93";
  39 #endif /* LIBC_SCCS and not lint */
  40 #include <sys/cdefs.h>
  41 __FBSDID("$FreeBSD: src/lib/libc/stdlib/heapsort.c,v 1.4 2002/03/21 22:48:41 obrien Exp $");
  42
  43 #include <errno.h>
  44 #include <stddef.h>
  45 #include <stdlib.h>
  46
  47 /*
  48  * Swap two areas of size number of bytes.  Although qsort(3) permits random
  49  * blocks of memory to be sorted, sorting pointers is almost certainly the
  50  * common case (and, were it not, could easily be made so).  Regardless, it
  51  * isn't worth optimizing; the SWAP's get sped up by the cache, and pointer
  52  * arithmetic gets lost in the time required for comparison function calls.
  53  */
  54 #define SWAP(a, b, count, size, tmp) { \
  55         count = size; \
  56         do { \
  57                 tmp = *a; \
  58                 *a++ = *b; \
  59                 *b++ = tmp; \
  60         } while (--count); \
  61 }
  62
  63 /* Copy one block of size size to another. */
  64 #define COPY(a, b, count, size, tmp1, tmp2) { \
  65         count = size; \
  66         tmp1 = a; \
  67         tmp2 = b; \
  68         do { \
  69                 *tmp1++ = *tmp2++; \
  70         } while (--count); \
  71 }
  72
  73 /*
  74  * Build the list into a heap, where a heap is defined such that for
  75  * the records K1 ... KN, Kj/2 >= Kj for 1 <= j/2 <= j <= N.
  76  *
  77  * There two cases.  If j == nmemb, select largest of Ki and Kj.  If
  78  * j < nmemb, select largest of Ki, Kj and Kj+1.
  79  */
  80 #define CREATE(initval, nmemb, par_i, child_i, par, child, size, count, tmp) { \
  81         for (par_i = initval; (child_i = par_i * 2) <= nmemb; \
  82             par_i = child_i) { \
  83                 child = base + child_i * size; \
  84                 if (child_i < nmemb && compar(child, child + size) < 0) { \
  85                         child += size; \
  86                         ++child_i; \
  87                 } \
  88                 par = base + par_i * size; \
  89                 if (compar(child, par) <= 0) \
  90                         break; \
  91                 SWAP(par, child, count, size, tmp); \
  92         } \
  93 }
  94
  95 /*
  96  * Select the top of the heap and 'heapify'.  Since by far the most expensive
  97  * action is the call to the compar function, a considerable optimization
  98  * in the average case can be achieved due to the fact that k, the displaced
  99  * elememt, is ususally quite small, so it would be preferable to first
 100  * heapify, always maintaining the invariant that the larger child is copied
 101  * over its parent's record.
 102  *
 103  * Then, starting from the *bottom* of the heap, finding k's correct place,
 104  * again maintianing the invariant.  As a result of the invariant no element
 105  * is 'lost' when k is assigned its correct place in the heap.
 106  *
 107  * The time savings from this optimization are on the order of 15-20% for the
 108  * average case. See Knuth, Vol. 3, page 158, problem 18.
 109  *
 110  * XXX Don't break the #define SELECT line, below.  Reiser cpp gets upset.
 111  */
 112 #define SELECT(par_i, child_i, nmemb, par, child, size, k, count, tmp1, tmp2) { \
 113         for (par_i = 1; (child_i = par_i * 2) <= nmemb; par_i = child_i) { \
 114                 child = base + child_i * size; \
 115                 if (child_i < nmemb && compar(child, child + size) < 0) { \
 116                         child += size; \
 117                         ++child_i; \
 118                 } \
 119                 par = base + par_i * size; \
 120                 COPY(par, child, count, size, tmp1, tmp2); \
 121         } \
 122         for (;;) { \
 123                 child_i = par_i; \
 124                 par_i = child_i / 2; \
 125                 child = base + child_i * size; \
 126                 par = base + par_i * size; \
 127                 if (child_i == 1 || compar(k, par) < 0) { \
 128                         COPY(child, k, count, size, tmp1, tmp2); \
 129                         break; \
 130                 } \
 131                 COPY(child, par, count, size, tmp1, tmp2); \
 132         } \
 133 }
 134
 135 /*
 136  * Heapsort -- Knuth, Vol. 3, page 145.  Runs in O (N lg N), both average
 137  * and worst.  While heapsort is faster than the worst case of quicksort,
 138  * the BSD quicksort does median selection so that the chance of finding
 139  * a data set that will trigger the worst case is nonexistent.  Heapsort's
 140  * only advantage over quicksort is that it requires little additional memory.
 141  */
 142 int
 143 heapsort(vbase, nmemb, size, compar)
 144         void *vbase;
 145         size_t nmemb, size;
 146         int (*compar)(const void *, const void *);
 147 {
 148         int cnt, i, j, l;
 149         char tmp, *tmp1, *tmp2;
 150         char *base, *k, *p, *t;
 151
 152         if (nmemb <= 1)
 153                 return (0);
 154
 155         if (!size) {
 156                 errno = EINVAL;
 157                 return (-1);
 158         }
 159
 160         if ((k = malloc(size)) == NULL)
 161                 return (-1);
 162
 163         /*
 164          * Items are numbered from 1 to nmemb, so offset from size bytes
 165          * below the starting address.
 166          */
 167         base = (char *)vbase - size;
 168
 169         for (l = nmemb / 2 + 1; --l;)
 170                 CREATE(l, nmemb, i, j, t, p, size, cnt, tmp);
 171
 172         /*
 173          * For each element of the heap, save the largest element into its
 174          * final slot, save the displaced element (k), then recreate the
 175          * heap.
 176          */
 177         while (nmemb > 1) {
 178                 COPY(k, base + nmemb * size, cnt, size, tmp1, tmp2);
 179                 COPY(base + nmemb * size, base + size, cnt, size, tmp1, tmp2);
 180                 --nmemb;
 181                 SELECT(i, j, nmemb, t, p, size, k, cnt, tmp1, tmp2);
 182         }
 183         free(k);
 184         return (0);
 185 }