merge.c revision 140098 
1/*-
2 * Copyright (c) 1992, 1993
3 *	The Regents of the University of California.  All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * Peter McIlroy.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 *    notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 *    notice, this list of conditions and the following disclaimer in the
15 *    documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 *    must display the following acknowledgement:
18 *	This product includes software developed by the University of
19 *	California, Berkeley and its contributors.
20 * 4. 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)
38static char sccsid[] = "@(#)merge.c	8.2 (Berkeley) 2/14/94";
39#endif /* LIBC_SCCS and not lint */
40#include <sys/cdefs.h>
41__FBSDID("$FreeBSD: head/lib/libc/stdlib/merge.c 140098 2005-01-12 03:39:34Z brian $");
42
43/*
44 * Hybrid exponential search/linear search merge sort with hybrid
45 * natural/pairwise first pass.  Requires about .3% more comparisons
46 * for random data than LSMS with pairwise first pass alone.
47 * It works for objects as small as two bytes.
48 */
49
50#define NATURAL
51#define THRESHOLD 16	/* Best choice for natural merge cut-off. */
52
53/* #define NATURAL to get hybrid natural merge.
54 * (The default is pairwise merging.)
55 */
56
57#include <sys/types.h>
58
59#include <errno.h>
60#include <stdlib.h>
61#include <string.h>
62
63static void setup(u_char *, u_char *, size_t, size_t,
64    int (*)(const void *, const void *));
65static void insertionsort(u_char *, size_t, size_t,
66    int (*)(const void *, const void *));
67
68#define ISIZE sizeof(int)
69#define PSIZE sizeof(u_char *)
70#define ICOPY_LIST(src, dst, last)				\
71	do							\
72	*(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE;	\
73	while(src < last)
74#define ICOPY_ELT(src, dst, i)					\
75	do							\
76	*(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE;	\
77	while (i -= ISIZE)
78
79#define CCOPY_LIST(src, dst, last)		\
80	do					\
81		*dst++ = *src++;		\
82	while (src < last)
83#define CCOPY_ELT(src, dst, i)			\
84	do					\
85		*dst++ = *src++;		\
86	while (i -= 1)
87
88/*
89 * Find the next possible pointer head.  (Trickery for forcing an array
90 * to do double duty as a linked list when objects do not align with word
91 * boundaries.
92 */
93/* Assumption: PSIZE is a power of 2. */
94#define EVAL(p) (u_char **)						\
95	((u_char *)0 +							\
96	    (((u_char *)p + PSIZE - 1 - (u_char *) 0) & ~(PSIZE - 1)))
97
98/*
99 * Arguments are as for qsort.
100 */
101int
102mergesort(base, nmemb, size, cmp)
103	void *base;
104	size_t nmemb;
105	size_t size;
106	int (*cmp)(const void *, const void *);
107{
108	size_t i;
109	int sense;
110	int big, iflag;
111	u_char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2;
112	u_char *list2, *list1, *p2, *p, *last, **p1;
113
114	if (size < PSIZE / 2) {		/* Pointers must fit into 2 * size. */
115		errno = EINVAL;
116		return (-1);
117	}
118
119	if (nmemb == 0)
120		return (0);
121
122	/*
123	 * XXX
124	 * Stupid subtraction for the Cray.
125	 */
126	iflag = 0;
127	if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE))
128		iflag = 1;
129
130	if ((list2 = malloc(nmemb * size + PSIZE)) == NULL)
131		return (-1);
132
133	list1 = base;
134	setup(list1, list2, nmemb, size, cmp);
135	last = list2 + nmemb * size;
136	i = big = 0;
137	while (*EVAL(list2) != last) {
138	    l2 = list1;
139	    p1 = EVAL(list1);
140	    for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) {
141	    	p2 = *EVAL(p2);
142	    	f1 = l2;
143	    	f2 = l1 = list1 + (p2 - list2);
144	    	if (p2 != last)
145	    		p2 = *EVAL(p2);
146	    	l2 = list1 + (p2 - list2);
147	    	while (f1 < l1 && f2 < l2) {
148	    		if ((*cmp)(f1, f2) <= 0) {
149	    			q = f2;
150	    			b = f1, t = l1;
151	    			sense = -1;
152	    		} else {
153	    			q = f1;
154	    			b = f2, t = l2;
155	    			sense = 0;
156	    		}
157	    		if (!big) {	/* here i = 0 */
158				while ((b += size) < t && cmp(q, b) >sense)
159	    				if (++i == 6) {
160	    					big = 1;
161	    					goto EXPONENTIAL;
162	    				}
163	    		} else {
164EXPONENTIAL:	    		for (i = size; ; i <<= 1)
165	    				if ((p = (b + i)) >= t) {
166	    					if ((p = t - size) > b &&
167						    (*cmp)(q, p) <= sense)
168	    						t = p;
169	    					else
170	    						b = p;
171	    					break;
172	    				} else if ((*cmp)(q, p) <= sense) {
173	    					t = p;
174	    					if (i == size)
175	    						big = 0;
176	    					goto FASTCASE;
177	    				} else
178	    					b = p;
179				while (t > b+size) {
180	    				i = (((t - b) / size) >> 1) * size;
181	    				if ((*cmp)(q, p = b + i) <= sense)
182	    					t = p;
183	    				else
184	    					b = p;
185	    			}
186	    			goto COPY;
187FASTCASE:	    		while (i > size)
188	    				if ((*cmp)(q,
189	    					p = b + (i >>= 1)) <= sense)
190	    					t = p;
191	    				else
192	    					b = p;
193COPY:	    			b = t;
194	    		}
195	    		i = size;
196	    		if (q == f1) {
197	    			if (iflag) {
198	    				ICOPY_LIST(f2, tp2, b);
199	    				ICOPY_ELT(f1, tp2, i);
200	    			} else {
201	    				CCOPY_LIST(f2, tp2, b);
202	    				CCOPY_ELT(f1, tp2, i);
203	    			}
204	    		} else {
205	    			if (iflag) {
206	    				ICOPY_LIST(f1, tp2, b);
207	    				ICOPY_ELT(f2, tp2, i);
208	    			} else {
209	    				CCOPY_LIST(f1, tp2, b);
210	    				CCOPY_ELT(f2, tp2, i);
211	    			}
212	    		}
213	    	}
214	    	if (f2 < l2) {
215	    		if (iflag)
216	    			ICOPY_LIST(f2, tp2, l2);
217	    		else
218	    			CCOPY_LIST(f2, tp2, l2);
219	    	} else if (f1 < l1) {
220	    		if (iflag)
221	    			ICOPY_LIST(f1, tp2, l1);
222	    		else
223	    			CCOPY_LIST(f1, tp2, l1);
224	    	}
225	    	*p1 = l2;
226	    }
227	    tp2 = list1;	/* swap list1, list2 */
228	    list1 = list2;
229	    list2 = tp2;
230	    last = list2 + nmemb*size;
231	}
232	if (base == list2) {
233		memmove(list2, list1, nmemb*size);
234		list2 = list1;
235	}
236	free(list2);
237	return (0);
238}
239
240#define	swap(a, b) {					\
241		s = b;					\
242		i = size;				\
243		do {					\
244			tmp = *a; *a++ = *s; *s++ = tmp; \
245		} while (--i);				\
246		a -= size;				\
247	}
248#define reverse(bot, top) {				\
249	s = top;					\
250	do {						\
251		i = size;				\
252		do {					\
253			tmp = *bot; *bot++ = *s; *s++ = tmp; \
254		} while (--i);				\
255		s -= size2;				\
256	} while(bot < s);				\
257}
258
259/*
260 * Optional hybrid natural/pairwise first pass.  Eats up list1 in runs of
261 * increasing order, list2 in a corresponding linked list.  Checks for runs
262 * when THRESHOLD/2 pairs compare with same sense.  (Only used when NATURAL
263 * is defined.  Otherwise simple pairwise merging is used.)
264 */
265void
266setup(list1, list2, n, size, cmp)
267	size_t n, size;
268	int (*cmp)(const void *, const void *);
269	u_char *list1, *list2;
270{
271	int i, length, size2, tmp, sense;
272	u_char *f1, *f2, *s, *l2, *last, *p2;
273
274	size2 = size*2;
275	if (n <= 5) {
276		insertionsort(list1, n, size, cmp);
277		*EVAL(list2) = (u_char*) list2 + n*size;
278		return;
279	}
280	/*
281	 * Avoid running pointers out of bounds; limit n to evens
282	 * for simplicity.
283	 */
284	i = 4 + (n & 1);
285	insertionsort(list1 + (n - i) * size, i, size, cmp);
286	last = list1 + size * (n - i);
287	*EVAL(list2 + (last - list1)) = list2 + n * size;
288
289#ifdef NATURAL
290	p2 = list2;
291	f1 = list1;
292	sense = (cmp(f1, f1 + size) > 0);
293	for (; f1 < last; sense = !sense) {
294		length = 2;
295					/* Find pairs with same sense. */
296		for (f2 = f1 + size2; f2 < last; f2 += size2) {
297			if ((cmp(f2, f2+ size) > 0) != sense)
298				break;
299			length += 2;
300		}
301		if (length < THRESHOLD) {		/* Pairwise merge */
302			do {
303				p2 = *EVAL(p2) = f1 + size2 - list1 + list2;
304				if (sense > 0)
305					swap (f1, f1 + size);
306			} while ((f1 += size2) < f2);
307		} else {				/* Natural merge */
308			l2 = f2;
309			for (f2 = f1 + size2; f2 < l2; f2 += size2) {
310				if ((cmp(f2-size, f2) > 0) != sense) {
311					p2 = *EVAL(p2) = f2 - list1 + list2;
312					if (sense > 0)
313						reverse(f1, f2-size);
314					f1 = f2;
315				}
316			}
317			if (sense > 0)
318				reverse (f1, f2-size);
319			f1 = f2;
320			if (f2 < last || cmp(f2 - size, f2) > 0)
321				p2 = *EVAL(p2) = f2 - list1 + list2;
322			else
323				p2 = *EVAL(p2) = list2 + n*size;
324		}
325	}
326#else		/* pairwise merge only. */
327	for (f1 = list1, p2 = list2; f1 < last; f1 += size2) {
328		p2 = *EVAL(p2) = p2 + size2;
329		if (cmp (f1, f1 + size) > 0)
330			swap(f1, f1 + size);
331	}
332#endif /* NATURAL */
333}
334
335/*
336 * This is to avoid out-of-bounds addresses in sorting the
337 * last 4 elements.
338 */
339static void
340insertionsort(a, n, size, cmp)
341	u_char *a;
342	size_t n, size;
343	int (*cmp)(const void *, const void *);
344{
345	u_char *ai, *s, *t, *u, tmp;
346	int i;
347
348	for (ai = a+size; --n >= 1; ai += size)
349		for (t = ai; t > a; t -= size) {
350			u = t - size;
351			if (cmp(u, t) <= 0)
352				break;
353			swap(u, t);
354		}
355}
356