xref: /dragonfly/lib/libc/stdlib/heapsort.c (revision dc71b7ab81c4f5270d3668e1625d94a58895fa7a)
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. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  * @(#)heapsort.c   8.1 (Berkeley) 6/4/93
33  * $FreeBSD: src/lib/libc/stdlib/heapsort.c,v 1.6 2008/01/13 02:11:10 das Exp $
34  * $DragonFly: src/lib/libc/stdlib/heapsort.c,v 1.5 2005/11/20 12:37:48 swildner Exp $
35  */
36 
37 #include <errno.h>
38 #include <stddef.h>
39 #include <stdlib.h>
40 
41 /*
42  * Swap two areas of size number of bytes.  Although qsort(3) permits random
43  * blocks of memory to be sorted, sorting pointers is almost certainly the
44  * common case (and, were it not, could easily be made so).  Regardless, it
45  * isn't worth optimizing; the SWAP's get sped up by the cache, and pointer
46  * arithmetic gets lost in the time required for comparison function calls.
47  */
48 #define   SWAP(a, b, count, size, tmp) { \
49           count = size; \
50           do { \
51                     tmp = *a; \
52                     *a++ = *b; \
53                     *b++ = tmp; \
54           } while (--count); \
55 }
56 
57 /* Copy one block of size size to another. */
58 #define COPY(a, b, count, size, tmp1, tmp2) { \
59           count = size; \
60           tmp1 = a; \
61           tmp2 = b; \
62           do { \
63                     *tmp1++ = *tmp2++; \
64           } while (--count); \
65 }
66 
67 /*
68  * Build the list into a heap, where a heap is defined such that for
69  * the records K1 ... KN, Kj/2 >= Kj for 1 <= j/2 <= j <= N.
70  *
71  * There two cases.  If j == nmemb, select largest of Ki and Kj.  If
72  * j < nmemb, select largest of Ki, Kj and Kj+1.
73  */
74 #define CREATE(initval, nmemb, par_i, child_i, par, child, size, count, tmp) { \
75           for (par_i = initval; (child_i = par_i * 2) <= nmemb; \
76               par_i = child_i) { \
77                     child = base + child_i * size; \
78                     if (child_i < nmemb && compar(child, child + size) < 0) { \
79                               child += size; \
80                               ++child_i; \
81                     } \
82                     par = base + par_i * size; \
83                     if (compar(child, par) <= 0) \
84                               break; \
85                     SWAP(par, child, count, size, tmp); \
86           } \
87 }
88 
89 /*
90  * Select the top of the heap and 'heapify'.  Since by far the most expensive
91  * action is the call to the compar function, a considerable optimization
92  * in the average case can be achieved due to the fact that k, the displaced
93  * elememt, is ususally quite small, so it would be preferable to first
94  * heapify, always maintaining the invariant that the larger child is copied
95  * over its parent's record.
96  *
97  * Then, starting from the *bottom* of the heap, finding k's correct place,
98  * again maintianing the invariant.  As a result of the invariant no element
99  * is 'lost' when k is assigned its correct place in the heap.
100  *
101  * The time savings from this optimization are on the order of 15-20% for the
102  * average case. See Knuth, Vol. 3, page 158, problem 18.
103  *
104  * XXX Don't break the #define SELECT line, below.  Reiser cpp gets upset.
105  */
106 #define SELECT(par_i, child_i, nmemb, par, child, size, k, count, tmp1, tmp2) { \
107           for (par_i = 1; (child_i = par_i * 2) <= nmemb; par_i = child_i) { \
108                     child = base + child_i * size; \
109                     if (child_i < nmemb && compar(child, child + size) < 0) { \
110                               child += size; \
111                               ++child_i; \
112                     } \
113                     par = base + par_i * size; \
114                     COPY(par, child, count, size, tmp1, tmp2); \
115           } \
116           for (;;) { \
117                     child_i = par_i; \
118                     par_i = child_i / 2; \
119                     child = base + child_i * size; \
120                     par = base + par_i * size; \
121                     if (child_i == 1 || compar(k, par) < 0) { \
122                               COPY(child, k, count, size, tmp1, tmp2); \
123                               break; \
124                     } \
125                     COPY(child, par, count, size, tmp1, tmp2); \
126           } \
127 }
128 
129 /*
130  * Heapsort -- Knuth, Vol. 3, page 145.  Runs in O (N lg N), both average
131  * and worst.  While heapsort is faster than the worst case of quicksort,
132  * the BSD quicksort does median selection so that the chance of finding
133  * a data set that will trigger the worst case is nonexistent.  Heapsort's
134  * only advantage over quicksort is that it requires little additional memory.
135  */
136 int
heapsort(void * vbase,size_t nmemb,size_t size,int (* compar)(const void *,const void *))137 heapsort(void *vbase, size_t nmemb, size_t size,
138            int (*compar)(const void *, const void *))
139 {
140           size_t cnt, i, j, l;
141           char tmp, *tmp1, *tmp2;
142           char *base, *k, *p, *t;
143 
144           if (nmemb <= 1)
145                     return (0);
146 
147           if (!size) {
148                     errno = EINVAL;
149                     return (-1);
150           }
151 
152           if ((k = malloc(size)) == NULL)
153                     return (-1);
154 
155           /*
156            * Items are numbered from 1 to nmemb, so offset from size bytes
157            * below the starting address.
158            */
159           base = (char *)vbase - size;
160 
161           for (l = nmemb / 2 + 1; --l;)
162                     CREATE(l, nmemb, i, j, t, p, size, cnt, tmp);
163 
164           /*
165            * For each element of the heap, save the largest element into its
166            * final slot, save the displaced element (k), then recreate the
167            * heap.
168            */
169           while (nmemb > 1) {
170                     COPY(k, base + nmemb * size, cnt, size, tmp1, tmp2);
171                     COPY(base + nmemb * size, base + size, cnt, size, tmp1, tmp2);
172                     --nmemb;
173                     SELECT(i, j, nmemb, t, p, size, k, cnt, tmp1, tmp2);
174           }
175           free(k);
176           return (0);
177 }
178