xref: /freebsd/lib/libc/stdlib/heapsort.c (revision 641a6cfb86023499caafe26a4d821a0b885cf00b)
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  * 4. 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 
33 #if defined(LIBC_SCCS) && !defined(lint)
34 static char sccsid[] = "@(#)heapsort.c	8.1 (Berkeley) 6/4/93";
35 #endif /* LIBC_SCCS and not lint */
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 #include <errno.h>
40 #include <stddef.h>
41 #include <stdlib.h>
42 
43 /*
44  * Swap two areas of size number of bytes.  Although qsort(3) permits random
45  * blocks of memory to be sorted, sorting pointers is almost certainly the
46  * common case (and, were it not, could easily be made so).  Regardless, it
47  * isn't worth optimizing; the SWAP's get sped up by the cache, and pointer
48  * arithmetic gets lost in the time required for comparison function calls.
49  */
50 #define	SWAP(a, b, count, size, tmp) { \
51 	count = size; \
52 	do { \
53 		tmp = *a; \
54 		*a++ = *b; \
55 		*b++ = tmp; \
56 	} while (--count); \
57 }
58 
59 /* Copy one block of size size to another. */
60 #define COPY(a, b, count, size, tmp1, tmp2) { \
61 	count = size; \
62 	tmp1 = a; \
63 	tmp2 = b; \
64 	do { \
65 		*tmp1++ = *tmp2++; \
66 	} while (--count); \
67 }
68 
69 /*
70  * Build the list into a heap, where a heap is defined such that for
71  * the records K1 ... KN, Kj/2 >= Kj for 1 <= j/2 <= j <= N.
72  *
73  * There two cases.  If j == nmemb, select largest of Ki and Kj.  If
74  * j < nmemb, select largest of Ki, Kj and Kj+1.
75  */
76 #define CREATE(initval, nmemb, par_i, child_i, par, child, size, count, tmp) { \
77 	for (par_i = initval; (child_i = par_i * 2) <= nmemb; \
78 	    par_i = child_i) { \
79 		child = base + child_i * size; \
80 		if (child_i < nmemb && compar(child, child + size) < 0) { \
81 			child += size; \
82 			++child_i; \
83 		} \
84 		par = base + par_i * size; \
85 		if (compar(child, par) <= 0) \
86 			break; \
87 		SWAP(par, child, count, size, tmp); \
88 	} \
89 }
90 
91 /*
92  * Select the top of the heap and 'heapify'.  Since by far the most expensive
93  * action is the call to the compar function, a considerable optimization
94  * in the average case can be achieved due to the fact that k, the displaced
95  * elememt, is ususally quite small, so it would be preferable to first
96  * heapify, always maintaining the invariant that the larger child is copied
97  * over its parent's record.
98  *
99  * Then, starting from the *bottom* of the heap, finding k's correct place,
100  * again maintianing the invariant.  As a result of the invariant no element
101  * is 'lost' when k is assigned its correct place in the heap.
102  *
103  * The time savings from this optimization are on the order of 15-20% for the
104  * average case. See Knuth, Vol. 3, page 158, problem 18.
105  *
106  * XXX Don't break the #define SELECT line, below.  Reiser cpp gets upset.
107  */
108 #define SELECT(par_i, child_i, nmemb, par, child, size, k, count, tmp1, tmp2) { \
109 	for (par_i = 1; (child_i = par_i * 2) <= nmemb; par_i = child_i) { \
110 		child = base + child_i * size; \
111 		if (child_i < nmemb && compar(child, child + size) < 0) { \
112 			child += size; \
113 			++child_i; \
114 		} \
115 		par = base + par_i * size; \
116 		COPY(par, child, count, size, tmp1, tmp2); \
117 	} \
118 	for (;;) { \
119 		child_i = par_i; \
120 		par_i = child_i / 2; \
121 		child = base + child_i * size; \
122 		par = base + par_i * size; \
123 		if (child_i == 1 || compar(k, par) < 0) { \
124 			COPY(child, k, count, size, tmp1, tmp2); \
125 			break; \
126 		} \
127 		COPY(child, par, count, size, tmp1, tmp2); \
128 	} \
129 }
130 
131 /*
132  * Heapsort -- Knuth, Vol. 3, page 145.  Runs in O (N lg N), both average
133  * and worst.  While heapsort is faster than the worst case of quicksort,
134  * the BSD quicksort does median selection so that the chance of finding
135  * a data set that will trigger the worst case is nonexistent.  Heapsort's
136  * only advantage over quicksort is that it requires little additional memory.
137  */
138 int
139 heapsort(vbase, nmemb, size, compar)
140 	void *vbase;
141 	size_t nmemb, size;
142 	int (*compar)(const void *, const void *);
143 {
144 	size_t cnt, i, j, l;
145 	char tmp, *tmp1, *tmp2;
146 	char *base, *k, *p, *t;
147 
148 	if (nmemb <= 1)
149 		return (0);
150 
151 	if (!size) {
152 		errno = EINVAL;
153 		return (-1);
154 	}
155 
156 	if ((k = malloc(size)) == NULL)
157 		return (-1);
158 
159 	/*
160 	 * Items are numbered from 1 to nmemb, so offset from size bytes
161 	 * below the starting address.
162 	 */
163 	base = (char *)vbase - size;
164 
165 	for (l = nmemb / 2 + 1; --l;)
166 		CREATE(l, nmemb, i, j, t, p, size, cnt, tmp);
167 
168 	/*
169 	 * For each element of the heap, save the largest element into its
170 	 * final slot, save the displaced element (k), then recreate the
171 	 * heap.
172 	 */
173 	while (nmemb > 1) {
174 		COPY(k, base + nmemb * size, cnt, size, tmp1, tmp2);
175 		COPY(base + nmemb * size, base + size, cnt, size, tmp1, tmp2);
176 		--nmemb;
177 		SELECT(i, j, nmemb, t, p, size, k, cnt, tmp1, tmp2);
178 	}
179 	free(k);
180 	return (0);
181 }
182