xref: /netbsd-src/lib/libc/stdlib/radixsort.c (revision 0b9f50897e9a9c6709320fafb4c3787fddcc0a45)
1 /*-
2  * Copyright (c) 1990 The Regents of the University of California.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33 
34 #if defined(LIBC_SCCS) && !defined(lint)
35 /*static char *sccsid = "from: @(#)radixsort.c	5.7 (Berkeley) 2/23/91";*/
36 static char *rcsid = "$Id: radixsort.c,v 1.3 1993/08/26 00:48:07 jtc Exp $";
37 #endif /* LIBC_SCCS and not lint */
38 
39 #include <sys/types.h>
40 #include <limits.h>
41 #include <stdlib.h>
42 #include <stddef.h>
43 #include <string.h>
44 
45 /*
46  * __rspartition is the cutoff point for a further partitioning instead
47  * of a shellsort.  If it changes check __rsshell_increments.  Both of
48  * these are exported, as the best values are data dependent.
49  */
50 #define	NPARTITION	40
51 int __rspartition = NPARTITION;
52 int __rsshell_increments[] = { 4, 1, 0, 0, 0, 0, 0, 0 };
53 
54 /*
55  * Stackp points to context structures, where each structure schedules a
56  * partitioning.  Radixsort exits when the stack is empty.
57  *
58  * If the buckets are placed on the stack randomly, the worst case is when
59  * all the buckets but one contain (npartitions + 1) elements and the bucket
60  * pushed on the stack last contains the rest of the elements.  In this case,
61  * stack growth is bounded by:
62  *
63  *	limit = (nelements / (npartitions + 1)) - 1;
64  *
65  * This is a very large number, 52,377,648 for the maximum 32-bit signed int.
66  *
67  * By forcing the largest bucket to be pushed on the stack first, the worst
68  * case is when all but two buckets each contain (npartitions + 1) elements,
69  * with the remaining elements split equally between the first and last
70  * buckets pushed on the stack.  In this case, stack growth is bounded when:
71  *
72  *	for (partition_cnt = 0; nelements > npartitions; ++partition_cnt)
73  *		nelements =
74  *		    (nelements - (npartitions + 1) * (nbuckets - 2)) / 2;
75  * The bound is:
76  *
77  *	limit = partition_cnt * (nbuckets - 1);
78  *
79  * This is a much smaller number, 4590 for the maximum 32-bit signed int.
80  */
81 #define	NBUCKETS	(UCHAR_MAX + 1)
82 
83 typedef struct _stack {
84 	const u_char **bot;
85 	int indx, nmemb;
86 } CONTEXT;
87 
88 #define	STACKPUSH { \
89 	stackp->bot = p; \
90 	stackp->nmemb = nmemb; \
91 	stackp->indx = indx; \
92 	++stackp; \
93 }
94 #define	STACKPOP { \
95 	if (stackp == stack) \
96 		break; \
97 	--stackp; \
98 	bot = stackp->bot; \
99 	nmemb = stackp->nmemb; \
100 	indx = stackp->indx; \
101 }
102 
103 /*
104  * A variant of MSD radix sorting; see Knuth Vol. 3, page 177, and 5.2.5,
105  * Ex. 10 and 12.  Also, "Three Partition Refinement Algorithms, Paige
106  * and Tarjan, SIAM J. Comput. Vol. 16, No. 6, December 1987.
107  *
108  * This uses a simple sort as soon as a bucket crosses a cutoff point,
109  * rather than sorting the entire list after partitioning is finished.
110  * This should be an advantage.
111  *
112  * This is pure MSD instead of LSD of some number of MSD, switching to
113  * the simple sort as soon as possible.  Takes linear time relative to
114  * the number of bytes in the strings.
115  */
116 int
117 #if __STDC__
118 radixsort(const u_char **l1, int nmemb, const u_char *tab, u_char endbyte)
119 #else
120 radixsort(l1, nmemb, tab, endbyte)
121 	const u_char **l1;
122 	register int nmemb;
123 	const u_char *tab;
124 	u_char endbyte;
125 #endif
126 {
127 	register int i, indx, t1, t2;
128 	register const u_char **l2;
129 	register const u_char **p;
130 	register const u_char **bot;
131 	register const u_char *tr;
132 	CONTEXT *stack, *stackp;
133 	int c[NBUCKETS + 1], max;
134 	u_char ltab[NBUCKETS];
135 	static void shellsort();
136 
137 	if (nmemb <= 1)
138 		return(0);
139 
140 	/*
141 	 * T1 is the constant part of the equation, the number of elements
142 	 * represented on the stack between the top and bottom entries.
143 	 * It doesn't get rounded as the divide by 2 rounds down (correct
144 	 * for a value being subtracted).  T2, the nelem value, has to be
145 	 * rounded up before each divide because we want an upper bound;
146 	 * this could overflow if nmemb is the maximum int.
147 	 */
148 	t1 = ((__rspartition + 1) * (NBUCKETS - 2)) >> 1;
149 	for (i = 0, t2 = nmemb; t2 > __rspartition; i += NBUCKETS - 1)
150 		t2 = ((t2 + 1) >> 1) - t1;
151 	if (i) {
152 		if (!(stack = stackp = (CONTEXT *)malloc(i * sizeof(CONTEXT))))
153 			return(-1);
154 	} else
155 		stack = stackp = NULL;
156 
157 	/*
158 	 * There are two arrays, one provided by the user (l1), and the
159 	 * temporary one (l2).  The data is sorted to the temporary stack,
160 	 * and then copied back.  The speedup of using index to determine
161 	 * which stack the data is on and simply swapping stacks back and
162 	 * forth, thus avoiding the copy every iteration, turns out to not
163 	 * be any faster than the current implementation.
164 	 */
165 	if (!(l2 = (const u_char **)malloc(sizeof(u_char *) * nmemb)))
166 		return(-1);
167 
168 	/*
169 	 * Tr references a table of sort weights; multiple entries may
170 	 * map to the same weight; EOS char must have the lowest weight.
171 	 */
172 	if (tab)
173 		tr = tab;
174 	else {
175 		for (t1 = 0, t2 = endbyte; t1 < t2; ++t1)
176 			ltab[t1] = t1 + 1;
177 		ltab[t2] = 0;
178 		for (t1 = endbyte + 1; t1 < NBUCKETS; ++t1)
179 			ltab[t1] = t1;
180 		tr = ltab;
181 	}
182 
183 	/* First sort is entire stack */
184 	bot = l1;
185 	indx = 0;
186 
187 	for (;;) {
188 		/* Clear bucket count array */
189 		bzero((char *)c, sizeof(c));
190 
191 		/*
192 		 * Compute number of items that sort to the same bucket
193 		 * for this index.
194 		 */
195 		for (p = bot, i = nmemb; --i >= 0;)
196 			++c[tr[(*p++)[indx]]];
197 
198 		/*
199 		 * Sum the number of characters into c, dividing the temp
200 		 * stack into the right number of buckets for this bucket,
201 		 * this index.  C contains the cumulative total of keys
202 		 * before and included in this bucket, and will later be
203 		 * used as an index to the bucket.  c[NBUCKETS] contains
204 		 * the total number of elements, for determining how many
205 		 * elements the last bucket contains.  At the same time
206 		 * find the largest bucket so it gets pushed first.
207 		 */
208 		for (i = max = t1 = 0, t2 = __rspartition; i <= NBUCKETS; ++i) {
209 			if (c[i] > t2) {
210 				t2 = c[i];
211 				max = i;
212 			}
213 			t1 = c[i] += t1;
214 		}
215 
216 		/*
217 		 * Partition the elements into buckets; c decrements through
218 		 * the bucket, and ends up pointing to the first element of
219 		 * the bucket.
220 		 */
221 		for (i = nmemb; --i >= 0;) {
222 			--p;
223 			l2[--c[tr[(*p)[indx]]]] = *p;
224 		}
225 
226 		/* Copy the partitioned elements back to user stack */
227 		bcopy(l2, bot, nmemb * sizeof(u_char *));
228 
229 		++indx;
230 		/*
231 		 * Sort buckets as necessary; don't sort c[0], it's the
232 		 * EOS character bucket, and nothing can follow EOS.
233 		 */
234 		for (i = max; i; --i) {
235 			if ((nmemb = c[i + 1] - (t1 = c[i])) < 2)
236 				continue;
237 			p = bot + t1;
238 			if (nmemb > __rspartition)
239 				STACKPUSH
240 			else
241 				shellsort(p, indx, nmemb, tr);
242 		}
243 		for (i = max + 1; i < NBUCKETS; ++i) {
244 			if ((nmemb = c[i + 1] - (t1 = c[i])) < 2)
245 				continue;
246 			p = bot + t1;
247 			if (nmemb > __rspartition)
248 				STACKPUSH
249 			else
250 				shellsort(p, indx, nmemb, tr);
251 		}
252 		/* Break out when stack is empty */
253 		STACKPOP
254 	}
255 
256 	free((char *)l2);
257 	free((char *)stack);
258 	return(0);
259 }
260 
261 /*
262  * Shellsort (diminishing increment sort) from Data Structures and
263  * Algorithms, Aho, Hopcraft and Ullman, 1983 Edition, page 290;
264  * see also Knuth Vol. 3, page 84.  The increments are selected from
265  * formula (8), page 95.  Roughly O(N^3/2).
266  */
267 static void
268 shellsort(p, indx, nmemb, tr)
269 	register u_char **p, *tr;
270 	register int indx, nmemb;
271 {
272 	register u_char ch, *s1, *s2;
273 	register int incr, *incrp, t1, t2;
274 
275 	for (incrp = __rsshell_increments; incr = *incrp++;)
276 		for (t1 = incr; t1 < nmemb; ++t1)
277 			for (t2 = t1 - incr; t2 >= 0;) {
278 				s1 = p[t2] + indx;
279 				s2 = p[t2 + incr] + indx;
280 				while ((ch = tr[*s1++]) == tr[*s2] && ch)
281 					++s2;
282 				if (ch > tr[*s2]) {
283 					s1 = p[t2];
284 					p[t2] = p[t2 + incr];
285 					p[t2 + incr] = s1;
286 					t2 -= incr;
287 				} else
288 					break;
289 			}
290 }
291