xref: /netbsd-src/lib/libc/stdlib/radixsort.c (revision 53b02e147d4ed531c0d2a5ca9b3e8026ba3e99b5)
1 /*	$NetBSD: radixsort.c,v 1.21 2021/10/29 11:03:46 nia Exp $	*/
2 
3 /*-
4  * Copyright (c) 1990, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Peter McIlroy and by Dan Bernstein at New York University,
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 
35 #include <sys/cdefs.h>
36 #if defined(LIBC_SCCS) && !defined(lint)
37 #if 0
38 static char sccsid[] = "@(#)radixsort.c	8.2 (Berkeley) 4/28/95";
39 #else
40 __RCSID("$NetBSD: radixsort.c,v 1.21 2021/10/29 11:03:46 nia Exp $");
41 #endif
42 #endif /* LIBC_SCCS and not lint */
43 
44 /*
45  * Radixsort routines.
46  *
47  * Program r_sort_a() is unstable but uses O(logN) extra memory for a stack.
48  * Use radixsort(a, n, trace, endchar) for this case.
49  *
50  * For stable sorting (using N extra pointers) use sradixsort(), which calls
51  * r_sort_b().
52  *
53  * For a description of this code, see D. McIlroy, P. McIlroy, K. Bostic,
54  * "Engineering Radix Sort".
55  */
56 
57 #include "namespace.h"
58 #include <sys/types.h>
59 
60 #include <assert.h>
61 #include <errno.h>
62 #include <stdlib.h>
63 
64 #ifdef __weak_alias
65 __weak_alias(radixsort,_radixsort)
66 __weak_alias(sradixsort,_sradixsort)
67 #endif
68 
69 typedef struct {
70 	const u_char **sa;
71 	int sn, si;
72 } stack;
73 
74 static inline void simplesort(const u_char **, int, int, const u_char *, u_int);
75 static void r_sort_a(const u_char **, int, int, const u_char *, u_int);
76 static void r_sort_b(const u_char **,
77 	    const u_char **, int, int, const u_char *, u_int);
78 
79 #define	THRESHOLD	20		/* Divert to simplesort(). */
80 #define	SIZE		512		/* Default stack size. */
81 
82 #define SETUP {								\
83 	if (tab == NULL) {						\
84 		tr = tr0;						\
85 		for (c = 0; c < endch; c++)				\
86 			tr0[c] = c + 1;					\
87 		tr0[c] = 0;						\
88 		for (c++; c < 256; c++)					\
89 			tr0[c] = c;					\
90 		endch = 0;						\
91 	} else {							\
92 		endch = tab[endch];					\
93 		tr = tab;						\
94 		if (endch != 0 && endch != 255) {			\
95 			errno = EINVAL;					\
96 			return (-1);					\
97 		}							\
98 	}								\
99 }
100 
101 int
102 radixsort(const u_char **a, int n, const u_char *tab, u_int endch)
103 {
104 	const u_char *tr;
105 	u_int c;
106 	u_char tr0[256];
107 
108 	_DIAGASSERT(a != NULL);
109 
110 	SETUP;
111 	r_sort_a(a, n, 0, tr, endch);
112 	return (0);
113 }
114 
115 int
116 sradixsort(const u_char **a, int n, const u_char *tab, u_int endch)
117 {
118 	const u_char *tr, **ta;
119 	u_int c;
120 	u_char tr0[256];
121 
122 	_DIAGASSERT(a != NULL);
123 	if (a == NULL) {
124 		errno = EFAULT;
125 		return (-1);
126 	}
127 
128 	SETUP;
129 	if (n < THRESHOLD)
130 		simplesort(a, n, 0, tr, endch);
131 	else {
132 		ta = NULL;
133 		if (reallocarr(&ta, n, sizeof(a)) != 0) {
134 			errno = ENOMEM;
135 			return (-1);
136 		}
137 		r_sort_b(a, ta, n, 0, tr, endch);
138 		free(ta);
139 	}
140 	return (0);
141 }
142 
143 #define empty(s)	(s >= sp)
144 #define pop(a, n, i)	a = (--sp)->sa, n = sp->sn, i = sp->si
145 #define push(a, n, i)	sp->sa = a, sp->sn = n, (sp++)->si = i
146 #define swap(a, b, t)	t = a, a = b, b = t
147 
148 /* Unstable, in-place sort. */
149 static void
150 r_sort_a(const u_char **a, int n, int i, const u_char *tr, u_int endch)
151 {
152 	static u_int count[256], nc, bmin;
153 	u_int c;
154 	const u_char **ak, *r;
155 	stack s[SIZE], *sp, *sp0, *sp1, temp;
156 	u_int *cp, bigc;
157 	const u_char **an, *t, **aj, **top[256];
158 
159 	_DIAGASSERT(a != NULL);
160 	_DIAGASSERT(tr != NULL);
161 
162 	/* Set up stack. */
163 	sp = s;
164 	push(a, n, i);
165 	while (!empty(s)) {
166 		pop(a, n, i);
167 		if (n < THRESHOLD) {
168 			simplesort(a, n, i, tr, endch);
169 			continue;
170 		}
171 		an = a + n;
172 
173 		/* Make character histogram. */
174 		if (nc == 0) {
175 			bmin = 255;	/* First occupied bin, excluding eos. */
176 			for (ak = a; ak < an;) {
177 				c = tr[(*ak++)[i]];
178 				if (++count[c] == 1 && c != endch) {
179 					if (c < bmin)
180 						bmin = c;
181 					nc++;
182 				}
183 			}
184 			if (sp + nc > s + SIZE) {	/* Get more stack. */
185 				r_sort_a(a, n, i, tr, endch);
186 				continue;
187 			}
188 		}
189 
190 		/*
191 		 * Set top[]; push incompletely sorted bins onto stack.
192 		 * top[] = pointers to last out-of-place element in bins.
193 		 * count[] = counts of elements in bins.
194 		 * Before permuting: top[c-1] + count[c] = top[c];
195 		 * during deal: top[c] counts down to top[c-1].
196 		 */
197 		sp0 = sp1 = sp;		/* Stack position of biggest bin. */
198 		bigc = 2;		/* Size of biggest bin. */
199 		if (endch == 0)		/* Special case: set top[eos]. */
200 			top[0] = ak = a + count[0];
201 		else {
202 			ak = a;
203 			top[255] = an;
204 		}
205 		for (cp = count + bmin; nc > 0; cp++) {
206 			while (*cp == 0)	/* Find next non-empty pile. */
207 				cp++;
208 			if (*cp > 1) {
209 				if (*cp > bigc) {
210 					bigc = *cp;
211 					sp1 = sp;
212 				}
213 				push(ak, *cp, i+1);
214 			}
215 			top[cp-count] = ak += *cp;
216 			nc--;
217 		}
218 		swap(*sp0, *sp1, temp);	/* Play it safe -- biggest bin last. */
219 
220 		/*
221 		 * Permute misplacements home.  Already home: everything
222 		 * before aj, and in bin[c], items from top[c] on.
223 		 * Inner loop:
224 		 *	r = next element to put in place;
225 		 *	ak = top[r[i]] = location to put the next element.
226 		 *	aj = bottom of 1st disordered bin.
227 		 * Outer loop:
228 		 *	Once the 1st disordered bin is done, ie. aj >= ak,
229 		 *	aj<-aj + count[c] connects the bins in a linked list;
230 		 *	reset count[c].
231 		 */
232 		for (aj = a; aj < an;  *aj = r, aj += count[c], count[c] = 0)
233 			for (r = *aj;  aj < (ak = --top[c = tr[r[i]]]);)
234 				swap(*ak, r, t);
235 	}
236 }
237 
238 /* Stable sort, requiring additional memory. */
239 static void
240 r_sort_b(const u_char **a, const u_char **ta, int n, int i, const u_char *tr,
241     u_int endch)
242 {
243 	static u_int count[256], nc, bmin;
244 	u_int c;
245 	const u_char **ak, **ai;
246 	stack s[512], *sp, *sp0, *sp1, temp;
247 	const u_char **top[256];
248 	u_int *cp, bigc;
249 
250 	_DIAGASSERT(a != NULL);
251 	_DIAGASSERT(ta != NULL);
252 	_DIAGASSERT(tr != NULL);
253 
254 	sp = s;
255 	push(a, n, i);
256 	while (!empty(s)) {
257 		pop(a, n, i);
258 		if (n < THRESHOLD) {
259 			simplesort(a, n, i, tr, endch);
260 			continue;
261 		}
262 
263 		if (nc == 0) {
264 			bmin = 255;
265 			for (ak = a + n; --ak >= a;) {
266 				c = tr[(*ak)[i]];
267 				if (++count[c] == 1 && c != endch) {
268 					if (c < bmin)
269 						bmin = c;
270 					nc++;
271 				}
272 			}
273 			if (sp + nc > s + SIZE) {
274 				r_sort_b(a, ta, n, i, tr, endch);
275 				continue;
276 			}
277 		}
278 
279 		sp0 = sp1 = sp;
280 		bigc = 2;
281 		if (endch == 0) {
282 			top[0] = ak = a + count[0];
283 			count[0] = 0;
284 		} else {
285 			ak = a;
286 			top[255] = a + n;
287 			count[255] = 0;
288 		}
289 		for (cp = count + bmin; nc > 0; cp++) {
290 			while (*cp == 0)
291 				cp++;
292 			if ((c = *cp) > 1) {
293 				if (c > bigc) {
294 					bigc = c;
295 					sp1 = sp;
296 				}
297 				push(ak, c, i+1);
298 			}
299 			top[cp-count] = ak += c;
300 			*cp = 0;			/* Reset count[]. */
301 			nc--;
302 		}
303 		swap(*sp0, *sp1, temp);
304 
305 		for (ak = ta + n, ai = a+n; ak > ta;)	/* Copy to temp. */
306 			*--ak = *--ai;
307 		for (ak = ta+n; --ak >= ta;)		/* Deal to piles. */
308 			*--top[tr[(*ak)[i]]] = *ak;
309 	}
310 }
311 
312 /* insertion sort */
313 static inline void
314 simplesort(const u_char **a, int n, int b, const u_char *tr, u_int endch)
315 {
316 	u_char ch;
317 	const u_char  **ak, **ai, *s, *t;
318 
319 	_DIAGASSERT(a != NULL);
320 	_DIAGASSERT(tr != NULL);
321 
322 	for (ak = a+1; --n >= 1; ak++)
323 		for (ai = ak; ai > a; ai--) {
324 			for (s = ai[0] + b, t = ai[-1] + b;
325 			    (ch = tr[*s]) != endch; s++, t++)
326 				if (ch != tr[*t])
327 					break;
328 			if (ch >= tr[*t])
329 				break;
330 			swap(ai[0], ai[-1], s);
331 		}
332 }
333