xref: /netbsd-src/lib/libc/stdlib/radixsort.c (revision ce63d6c20fc4ec8ddc95c84bb229e3c4ecf82b69)
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[] = "@(#)radixsort.c	5.7 (Berkeley) 2/23/91";
36 #endif /* LIBC_SCCS and not lint */
37 
38 #include <sys/types.h>
39 #include <limits.h>
40 #include <stdlib.h>
41 #include <stddef.h>
42 #include <string.h>
43 
44 /*
45  * __rspartition is the cutoff point for a further partitioning instead
46  * of a shellsort.  If it changes check __rsshell_increments.  Both of
47  * these are exported, as the best values are data dependent.
48  */
49 #define	NPARTITION	40
50 int __rspartition = NPARTITION;
51 int __rsshell_increments[] = { 4, 1, 0, 0, 0, 0, 0, 0 };
52 
53 /*
54  * Stackp points to context structures, where each structure schedules a
55  * partitioning.  Radixsort exits when the stack is empty.
56  *
57  * If the buckets are placed on the stack randomly, the worst case is when
58  * all the buckets but one contain (npartitions + 1) elements and the bucket
59  * pushed on the stack last contains the rest of the elements.  In this case,
60  * stack growth is bounded by:
61  *
62  *	limit = (nelements / (npartitions + 1)) - 1;
63  *
64  * This is a very large number, 52,377,648 for the maximum 32-bit signed int.
65  *
66  * By forcing the largest bucket to be pushed on the stack first, the worst
67  * case is when all but two buckets each contain (npartitions + 1) elements,
68  * with the remaining elements split equally between the first and last
69  * buckets pushed on the stack.  In this case, stack growth is bounded when:
70  *
71  *	for (partition_cnt = 0; nelements > npartitions; ++partition_cnt)
72  *		nelements =
73  *		    (nelements - (npartitions + 1) * (nbuckets - 2)) / 2;
74  * The bound is:
75  *
76  *	limit = partition_cnt * (nbuckets - 1);
77  *
78  * This is a much smaller number, 4590 for the maximum 32-bit signed int.
79  */
80 #define	NBUCKETS	(UCHAR_MAX + 1)
81 
82 typedef struct _stack {
83 	const u_char **bot;
84 	int indx, nmemb;
85 } CONTEXT;
86 
87 #define	STACKPUSH { \
88 	stackp->bot = p; \
89 	stackp->nmemb = nmemb; \
90 	stackp->indx = indx; \
91 	++stackp; \
92 }
93 #define	STACKPOP { \
94 	if (stackp == stack) \
95 		break; \
96 	--stackp; \
97 	bot = stackp->bot; \
98 	nmemb = stackp->nmemb; \
99 	indx = stackp->indx; \
100 }
101 
102 /*
103  * A variant of MSD radix sorting; see Knuth Vol. 3, page 177, and 5.2.5,
104  * Ex. 10 and 12.  Also, "Three Partition Refinement Algorithms, Paige
105  * and Tarjan, SIAM J. Comput. Vol. 16, No. 6, December 1987.
106  *
107  * This uses a simple sort as soon as a bucket crosses a cutoff point,
108  * rather than sorting the entire list after partitioning is finished.
109  * This should be an advantage.
110  *
111  * This is pure MSD instead of LSD of some number of MSD, switching to
112  * the simple sort as soon as possible.  Takes linear time relative to
113  * the number of bytes in the strings.
114  */
115 int
116 #if __STDC__
117 radixsort(const u_char **l1, int nmemb, const u_char *tab, u_char endbyte)
118 #else
119 radixsort(l1, nmemb, tab, endbyte)
120 	const u_char **l1;
121 	register int nmemb;
122 	const u_char *tab;
123 	u_char endbyte;
124 #endif
125 {
126 	register int i, indx, t1, t2;
127 	register const u_char **l2;
128 	register const u_char **p;
129 	register const u_char **bot;
130 	register const u_char *tr;
131 	CONTEXT *stack, *stackp;
132 	int c[NBUCKETS + 1], max;
133 	u_char ltab[NBUCKETS];
134 	static void shellsort();
135 
136 	if (nmemb <= 1)
137 		return(0);
138 
139 	/*
140 	 * T1 is the constant part of the equation, the number of elements
141 	 * represented on the stack between the top and bottom entries.
142 	 * It doesn't get rounded as the divide by 2 rounds down (correct
143 	 * for a value being subtracted).  T2, the nelem value, has to be
144 	 * rounded up before each divide because we want an upper bound;
145 	 * this could overflow if nmemb is the maximum int.
146 	 */
147 	t1 = ((__rspartition + 1) * (NBUCKETS - 2)) >> 1;
148 	for (i = 0, t2 = nmemb; t2 > __rspartition; i += NBUCKETS - 1)
149 		t2 = ((t2 + 1) >> 1) - t1;
150 	if (i) {
151 		if (!(stack = stackp = (CONTEXT *)malloc(i * sizeof(CONTEXT))))
152 			return(-1);
153 	} else
154 		stack = stackp = NULL;
155 
156 	/*
157 	 * There are two arrays, one provided by the user (l1), and the
158 	 * temporary one (l2).  The data is sorted to the temporary stack,
159 	 * and then copied back.  The speedup of using index to determine
160 	 * which stack the data is on and simply swapping stacks back and
161 	 * forth, thus avoiding the copy every iteration, turns out to not
162 	 * be any faster than the current implementation.
163 	 */
164 	if (!(l2 = (const u_char **)malloc(sizeof(u_char *) * nmemb)))
165 		return(-1);
166 
167 	/*
168 	 * Tr references a table of sort weights; multiple entries may
169 	 * map to the same weight; EOS char must have the lowest weight.
170 	 */
171 	if (tab)
172 		tr = tab;
173 	else {
174 		for (t1 = 0, t2 = endbyte; t1 < t2; ++t1)
175 			ltab[t1] = t1 + 1;
176 		ltab[t2] = 0;
177 		for (t1 = endbyte + 1; t1 < NBUCKETS; ++t1)
178 			ltab[t1] = t1;
179 		tr = ltab;
180 	}
181 
182 	/* First sort is entire stack */
183 	bot = l1;
184 	indx = 0;
185 
186 	for (;;) {
187 		/* Clear bucket count array */
188 		bzero((char *)c, sizeof(c));
189 
190 		/*
191 		 * Compute number of items that sort to the same bucket
192 		 * for this index.
193 		 */
194 		for (p = bot, i = nmemb; --i >= 0;)
195 			++c[tr[(*p++)[indx]]];
196 
197 		/*
198 		 * Sum the number of characters into c, dividing the temp
199 		 * stack into the right number of buckets for this bucket,
200 		 * this index.  C contains the cumulative total of keys
201 		 * before and included in this bucket, and will later be
202 		 * used as an index to the bucket.  c[NBUCKETS] contains
203 		 * the total number of elements, for determining how many
204 		 * elements the last bucket contains.  At the same time
205 		 * find the largest bucket so it gets pushed first.
206 		 */
207 		for (i = max = t1 = 0, t2 = __rspartition; i <= NBUCKETS; ++i) {
208 			if (c[i] > t2) {
209 				t2 = c[i];
210 				max = i;
211 			}
212 			t1 = c[i] += t1;
213 		}
214 
215 		/*
216 		 * Partition the elements into buckets; c decrements through
217 		 * the bucket, and ends up pointing to the first element of
218 		 * the bucket.
219 		 */
220 		for (i = nmemb; --i >= 0;) {
221 			--p;
222 			l2[--c[tr[(*p)[indx]]]] = *p;
223 		}
224 
225 		/* Copy the partitioned elements back to user stack */
226 		bcopy(l2, bot, nmemb * sizeof(u_char *));
227 
228 		++indx;
229 		/*
230 		 * Sort buckets as necessary; don't sort c[0], it's the
231 		 * EOS character bucket, and nothing can follow EOS.
232 		 */
233 		for (i = max; i; --i) {
234 			if ((nmemb = c[i + 1] - (t1 = c[i])) < 2)
235 				continue;
236 			p = bot + t1;
237 			if (nmemb > __rspartition)
238 				STACKPUSH
239 			else
240 				shellsort(p, indx, nmemb, tr);
241 		}
242 		for (i = max + 1; i < NBUCKETS; ++i) {
243 			if ((nmemb = c[i + 1] - (t1 = c[i])) < 2)
244 				continue;
245 			p = bot + t1;
246 			if (nmemb > __rspartition)
247 				STACKPUSH
248 			else
249 				shellsort(p, indx, nmemb, tr);
250 		}
251 		/* Break out when stack is empty */
252 		STACKPOP
253 	}
254 
255 	free((char *)l2);
256 	free((char *)stack);
257 	return(0);
258 }
259 
260 /*
261  * Shellsort (diminishing increment sort) from Data Structures and
262  * Algorithms, Aho, Hopcraft and Ullman, 1983 Edition, page 290;
263  * see also Knuth Vol. 3, page 84.  The increments are selected from
264  * formula (8), page 95.  Roughly O(N^3/2).
265  */
266 static void
267 shellsort(p, indx, nmemb, tr)
268 	register u_char **p, *tr;
269 	register int indx, nmemb;
270 {
271 	register u_char ch, *s1, *s2;
272 	register int incr, *incrp, t1, t2;
273 
274 	for (incrp = __rsshell_increments; incr = *incrp++;)
275 		for (t1 = incr; t1 < nmemb; ++t1)
276 			for (t2 = t1 - incr; t2 >= 0;) {
277 				s1 = p[t2] + indx;
278 				s2 = p[t2 + incr] + indx;
279 				while ((ch = tr[*s1++]) == tr[*s2] && ch)
280 					++s2;
281 				if (ch > tr[*s2]) {
282 					s1 = p[t2];
283 					p[t2] = p[t2 + incr];
284 					p[t2 + incr] = s1;
285 					t2 -= incr;
286 				} else
287 					break;
288 			}
289 }
290