xref: /freebsd/lib/libc/stdlib/heapsort.c (revision 5f4c09dd85bff675e0ca63c55ea3c517e0fddfcc)
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 <errno.h>
41 #include <stddef.h>
42 #include <stdlib.h>
43 
44 #ifdef I_AM_HEAPSORT_B
45 #include "block_abi.h"
46 #define COMPAR(x, y) CALL_BLOCK(compar, x, y)
47 typedef DECLARE_BLOCK(int, heapsort_block, const void *, const void *);
48 #else
49 #define COMPAR(x, y) compar(x, y)
50 #endif
51 
52 /*
53  * Swap two areas of size number of bytes.  Although qsort(3) permits random
54  * blocks of memory to be sorted, sorting pointers is almost certainly the
55  * common case (and, were it not, could easily be made so).  Regardless, it
56  * isn't worth optimizing; the SWAP's get sped up by the cache, and pointer
57  * arithmetic gets lost in the time required for comparison function calls.
58  */
59 #define	SWAP(a, b, count, size, tmp) { \
60 	count = size; \
61 	do { \
62 		tmp = *a; \
63 		*a++ = *b; \
64 		*b++ = tmp; \
65 	} while (--count); \
66 }
67 
68 /* Copy one block of size size to another. */
69 #define COPY(a, b, count, size, tmp1, tmp2) { \
70 	count = size; \
71 	tmp1 = a; \
72 	tmp2 = b; \
73 	do { \
74 		*tmp1++ = *tmp2++; \
75 	} while (--count); \
76 }
77 
78 /*
79  * Build the list into a heap, where a heap is defined such that for
80  * the records K1 ... KN, Kj/2 >= Kj for 1 <= j/2 <= j <= N.
81  *
82  * There two cases.  If j == nmemb, select largest of Ki and Kj.  If
83  * j < nmemb, select largest of Ki, Kj and Kj+1.
84  */
85 #define CREATE(initval, nmemb, par_i, child_i, par, child, size, count, tmp) { \
86 	for (par_i = initval; (child_i = par_i * 2) <= nmemb; \
87 	    par_i = child_i) { \
88 		child = base + child_i * size; \
89 		if (child_i < nmemb && COMPAR(child, child + size) < 0) { \
90 			child += size; \
91 			++child_i; \
92 		} \
93 		par = base + par_i * size; \
94 		if (COMPAR(child, par) <= 0) \
95 			break; \
96 		SWAP(par, child, count, size, tmp); \
97 	} \
98 }
99 
100 /*
101  * Select the top of the heap and 'heapify'.  Since by far the most expensive
102  * action is the call to the compar function, a considerable optimization
103  * in the average case can be achieved due to the fact that k, the displaced
104  * elememt, is usually quite small, so it would be preferable to first
105  * heapify, always maintaining the invariant that the larger child is copied
106  * over its parent's record.
107  *
108  * Then, starting from the *bottom* of the heap, finding k's correct place,
109  * again maintianing the invariant.  As a result of the invariant no element
110  * is 'lost' when k is assigned its correct place in the heap.
111  *
112  * The time savings from this optimization are on the order of 15-20% for the
113  * average case. See Knuth, Vol. 3, page 158, problem 18.
114  *
115  * XXX Don't break the #define SELECT line, below.  Reiser cpp gets upset.
116  */
117 #define SELECT(par_i, child_i, nmemb, par, child, size, k, count, tmp1, tmp2) { \
118 	for (par_i = 1; (child_i = par_i * 2) <= nmemb; par_i = child_i) { \
119 		child = base + child_i * size; \
120 		if (child_i < nmemb && COMPAR(child, child + size) < 0) { \
121 			child += size; \
122 			++child_i; \
123 		} \
124 		par = base + par_i * size; \
125 		COPY(par, child, count, size, tmp1, tmp2); \
126 	} \
127 	for (;;) { \
128 		child_i = par_i; \
129 		par_i = child_i / 2; \
130 		child = base + child_i * size; \
131 		par = base + par_i * size; \
132 		if (child_i == 1 || COMPAR(k, par) < 0) { \
133 			COPY(child, k, count, size, tmp1, tmp2); \
134 			break; \
135 		} \
136 		COPY(child, par, count, size, tmp1, tmp2); \
137 	} \
138 }
139 
140 #ifdef I_AM_HEAPSORT_B
141 int heapsort_b(void *, size_t, size_t, heapsort_block);
142 #else
143 int heapsort(void *, size_t, size_t,
144     int (*)(const void *, const void *));
145 #endif
146 /*
147  * Heapsort -- Knuth, Vol. 3, page 145.  Runs in O (N lg N), both average
148  * and worst.  While heapsort is faster than the worst case of quicksort,
149  * the BSD quicksort does median selection so that the chance of finding
150  * a data set that will trigger the worst case is nonexistent.  Heapsort's
151  * only advantage over quicksort is that it requires little additional memory.
152  */
153 #ifdef I_AM_HEAPSORT_B
154 int
155 heapsort_b(void *vbase, size_t nmemb, size_t size, heapsort_block compar)
156 #else
157 int
158 heapsort(void *vbase, size_t nmemb, size_t size,
159     int (*compar)(const void *, const void *))
160 #endif
161 {
162 	size_t cnt, i, j, l;
163 	char tmp, *tmp1, *tmp2;
164 	char *base, *k, *p, *t;
165 
166 	if (nmemb <= 1)
167 		return (0);
168 
169 	if (!size) {
170 		errno = EINVAL;
171 		return (-1);
172 	}
173 
174 	if ((k = malloc(size)) == NULL)
175 		return (-1);
176 
177 	/*
178 	 * Items are numbered from 1 to nmemb, so offset from size bytes
179 	 * below the starting address.
180 	 */
181 	base = (char *)vbase - size;
182 
183 	for (l = nmemb / 2 + 1; --l;)
184 		CREATE(l, nmemb, i, j, t, p, size, cnt, tmp);
185 
186 	/*
187 	 * For each element of the heap, save the largest element into its
188 	 * final slot, save the displaced element (k), then recreate the
189 	 * heap.
190 	 */
191 	while (nmemb > 1) {
192 		COPY(k, base + nmemb * size, cnt, size, tmp1, tmp2);
193 		COPY(base + nmemb * size, base + size, cnt, size, tmp1, tmp2);
194 		--nmemb;
195 		SELECT(i, j, nmemb, t, p, size, k, cnt, tmp1, tmp2);
196 	}
197 	free(k);
198 	return (0);
199 }
200