xref: /freebsd/lib/libc/stdlib/heapsort.c (revision faf25f48d601ae39f5752602f3020e2e92605625)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * Copyright (c) 2014 David T. Chisnall
7  * All rights reserved.
8  *
9  * This code is derived from software contributed to Berkeley by
10  * Ronnie Kon at Mindcraft Inc., Kevin Lew and Elmer Yglesias.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. 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("$FreeBSD$");
42 
43 #include <errno.h>
44 #include <stddef.h>
45 #include <stdlib.h>
46 
47 #ifdef I_AM_HEAPSORT_B
48 #include "block_abi.h"
49 #define COMPAR(x, y) CALL_BLOCK(compar, x, y)
50 typedef DECLARE_BLOCK(int, heapsort_block, const void *, const void *);
51 #else
52 #define COMPAR(x, y) compar(x, y)
53 #endif
54 
55 /*
56  * Swap two areas of size number of bytes.  Although qsort(3) permits random
57  * blocks of memory to be sorted, sorting pointers is almost certainly the
58  * common case (and, were it not, could easily be made so).  Regardless, it
59  * isn't worth optimizing; the SWAP's get sped up by the cache, and pointer
60  * arithmetic gets lost in the time required for comparison function calls.
61  */
62 #define	SWAP(a, b, count, size, tmp) { \
63 	count = size; \
64 	do { \
65 		tmp = *a; \
66 		*a++ = *b; \
67 		*b++ = tmp; \
68 	} while (--count); \
69 }
70 
71 /* Copy one block of size size to another. */
72 #define COPY(a, b, count, size, tmp1, tmp2) { \
73 	count = size; \
74 	tmp1 = a; \
75 	tmp2 = b; \
76 	do { \
77 		*tmp1++ = *tmp2++; \
78 	} while (--count); \
79 }
80 
81 /*
82  * Build the list into a heap, where a heap is defined such that for
83  * the records K1 ... KN, Kj/2 >= Kj for 1 <= j/2 <= j <= N.
84  *
85  * There two cases.  If j == nmemb, select largest of Ki and Kj.  If
86  * j < nmemb, select largest of Ki, Kj and Kj+1.
87  */
88 #define CREATE(initval, nmemb, par_i, child_i, par, child, size, count, tmp) { \
89 	for (par_i = initval; (child_i = par_i * 2) <= nmemb; \
90 	    par_i = child_i) { \
91 		child = base + child_i * size; \
92 		if (child_i < nmemb && COMPAR(child, child + size) < 0) { \
93 			child += size; \
94 			++child_i; \
95 		} \
96 		par = base + par_i * size; \
97 		if (COMPAR(child, par) <= 0) \
98 			break; \
99 		SWAP(par, child, count, size, tmp); \
100 	} \
101 }
102 
103 /*
104  * Select the top of the heap and 'heapify'.  Since by far the most expensive
105  * action is the call to the compar function, a considerable optimization
106  * in the average case can be achieved due to the fact that k, the displaced
107  * elememt, is usually quite small, so it would be preferable to first
108  * heapify, always maintaining the invariant that the larger child is copied
109  * over its parent's record.
110  *
111  * Then, starting from the *bottom* of the heap, finding k's correct place,
112  * again maintianing the invariant.  As a result of the invariant no element
113  * is 'lost' when k is assigned its correct place in the heap.
114  *
115  * The time savings from this optimization are on the order of 15-20% for the
116  * average case. See Knuth, Vol. 3, page 158, problem 18.
117  *
118  * XXX Don't break the #define SELECT line, below.  Reiser cpp gets upset.
119  */
120 #define SELECT(par_i, child_i, nmemb, par, child, size, k, count, tmp1, tmp2) { \
121 	for (par_i = 1; (child_i = par_i * 2) <= nmemb; par_i = child_i) { \
122 		child = base + child_i * size; \
123 		if (child_i < nmemb && COMPAR(child, child + size) < 0) { \
124 			child += size; \
125 			++child_i; \
126 		} \
127 		par = base + par_i * size; \
128 		COPY(par, child, count, size, tmp1, tmp2); \
129 	} \
130 	for (;;) { \
131 		child_i = par_i; \
132 		par_i = child_i / 2; \
133 		child = base + child_i * size; \
134 		par = base + par_i * size; \
135 		if (child_i == 1 || COMPAR(k, par) < 0) { \
136 			COPY(child, k, count, size, tmp1, tmp2); \
137 			break; \
138 		} \
139 		COPY(child, par, count, size, tmp1, tmp2); \
140 	} \
141 }
142 
143 #ifdef I_AM_HEAPSORT_B
144 int heapsort_b(void *, size_t, size_t, heapsort_block);
145 #else
146 int heapsort(void *, size_t, size_t,
147     int (*)(const void *, const void *));
148 #endif
149 /*
150  * Heapsort -- Knuth, Vol. 3, page 145.  Runs in O (N lg N), both average
151  * and worst.  While heapsort is faster than the worst case of quicksort,
152  * the BSD quicksort does median selection so that the chance of finding
153  * a data set that will trigger the worst case is nonexistent.  Heapsort's
154  * only advantage over quicksort is that it requires little additional memory.
155  */
156 #ifdef I_AM_HEAPSORT_B
157 int
158 heapsort_b(void *vbase, size_t nmemb, size_t size, heapsort_block compar)
159 #else
160 int
161 heapsort(void *vbase, size_t nmemb, size_t size,
162     int (*compar)(const void *, const void *))
163 #endif
164 {
165 	size_t cnt, i, j, l;
166 	char tmp, *tmp1, *tmp2;
167 	char *base, *k, *p, *t;
168 
169 	if (nmemb <= 1)
170 		return (0);
171 
172 	if (!size) {
173 		errno = EINVAL;
174 		return (-1);
175 	}
176 
177 	if ((k = malloc(size)) == NULL)
178 		return (-1);
179 
180 	/*
181 	 * Items are numbered from 1 to nmemb, so offset from size bytes
182 	 * below the starting address.
183 	 */
184 	base = (char *)vbase - size;
185 
186 	for (l = nmemb / 2 + 1; --l;)
187 		CREATE(l, nmemb, i, j, t, p, size, cnt, tmp);
188 
189 	/*
190 	 * For each element of the heap, save the largest element into its
191 	 * final slot, save the displaced element (k), then recreate the
192 	 * heap.
193 	 */
194 	while (nmemb > 1) {
195 		COPY(k, base + nmemb * size, cnt, size, tmp1, tmp2);
196 		COPY(base + nmemb * size, base + size, cnt, size, tmp1, tmp2);
197 		--nmemb;
198 		SELECT(i, j, nmemb, t, p, size, k, cnt, tmp1, tmp2);
199 	}
200 	free(k);
201 	return (0);
202 }
203