merge.c revision 92889 
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 92889 2002-03-21 18:49:23Z obrien $");
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 __P((u_char *, u_char *, size_t, size_t, int (*)()));
64static void insertionsort __P((u_char *, size_t, size_t, int (*)()));
65
66#define ISIZE sizeof(int)
67#define PSIZE sizeof(u_char *)
68#define ICOPY_LIST(src, dst, last)				\
69	do							\
70	*(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE;	\
71	while(src < last)
72#define ICOPY_ELT(src, dst, i)					\
73	do							\
74	*(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE;	\
75	while (i -= ISIZE)
76
77#define CCOPY_LIST(src, dst, last)		\
78	do					\
79		*dst++ = *src++;		\
80	while (src < last)
81#define CCOPY_ELT(src, dst, i)			\
82	do					\
83		*dst++ = *src++;		\
84	while (i -= 1)
85
86/*
87 * Find the next possible pointer head.  (Trickery for forcing an array
88 * to do double duty as a linked list when objects do not align with word
89 * boundaries.
90 */
91/* Assumption: PSIZE is a power of 2. */
92#define EVAL(p) (u_char **)						\
93	((u_char *)0 +							\
94	    (((u_char *)p + PSIZE - 1 - (u_char *) 0) & ~(PSIZE - 1)))
95
96/*
97 * Arguments are as for qsort.
98 */
99int
100mergesort(base, nmemb, size, cmp)
101	void *base;
102	size_t nmemb;
103	size_t size;
104	int (*cmp) __P((const void *, const void *));
105{
106	int i, sense;
107	int big, iflag;
108	u_char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2;
109	u_char *list2, *list1, *p2, *p, *last, **p1;
110
111	if (size < PSIZE / 2) {		/* Pointers must fit into 2 * size. */
112		errno = EINVAL;
113		return (-1);
114	}
115
116	if (nmemb == 0)
117		return (0);
118
119	/*
120	 * XXX
121	 * Stupid subtraction for the Cray.
122	 */
123	iflag = 0;
124	if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE))
125		iflag = 1;
126
127	if ((list2 = malloc(nmemb * size + PSIZE)) == NULL)
128		return (-1);
129
130	list1 = base;
131	setup(list1, list2, nmemb, size, cmp);
132	last = list2 + nmemb * size;
133	i = big = 0;
134	while (*EVAL(list2) != last) {
135	    l2 = list1;
136	    p1 = EVAL(list1);
137	    for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) {
138	    	p2 = *EVAL(p2);
139	    	f1 = l2;
140	    	f2 = l1 = list1 + (p2 - list2);
141	    	if (p2 != last)
142	    		p2 = *EVAL(p2);
143	    	l2 = list1 + (p2 - list2);
144	    	while (f1 < l1 && f2 < l2) {
145	    		if ((*cmp)(f1, f2) <= 0) {
146	    			q = f2;
147	    			b = f1, t = l1;
148	    			sense = -1;
149	    		} else {
150	    			q = f1;
151	    			b = f2, t = l2;
152	    			sense = 0;
153	    		}
154	    		if (!big) {	/* here i = 0 */
155				while ((b += size) < t && cmp(q, b) >sense)
156	    				if (++i == 6) {
157	    					big = 1;
158	    					goto EXPONENTIAL;
159	    				}
160	    		} else {
161EXPONENTIAL:	    		for (i = size; ; i <<= 1)
162	    				if ((p = (b + i)) >= t) {
163	    					if ((p = t - size) > b &&
164						    (*cmp)(q, p) <= sense)
165	    						t = p;
166	    					else
167	    						b = p;
168	    					break;
169	    				} else if ((*cmp)(q, p) <= sense) {
170	    					t = p;
171	    					if (i == size)
172	    						big = 0;
173	    					goto FASTCASE;
174	    				} else
175	    					b = p;
176				while (t > b+size) {
177	    				i = (((t - b) / size) >> 1) * size;
178	    				if ((*cmp)(q, p = b + i) <= sense)
179	    					t = p;
180	    				else
181	    					b = p;
182	    			}
183	    			goto COPY;
184FASTCASE:	    		while (i > size)
185	    				if ((*cmp)(q,
186	    					p = b + (i >>= 1)) <= sense)
187	    					t = p;
188	    				else
189	    					b = p;
190COPY:	    			b = t;
191	    		}
192	    		i = size;
193	    		if (q == f1) {
194	    			if (iflag) {
195	    				ICOPY_LIST(f2, tp2, b);
196	    				ICOPY_ELT(f1, tp2, i);
197	    			} else {
198	    				CCOPY_LIST(f2, tp2, b);
199	    				CCOPY_ELT(f1, tp2, i);
200	    			}
201	    		} else {
202	    			if (iflag) {
203	    				ICOPY_LIST(f1, tp2, b);
204	    				ICOPY_ELT(f2, tp2, i);
205	    			} else {
206	    				CCOPY_LIST(f1, tp2, b);
207	    				CCOPY_ELT(f2, tp2, i);
208	    			}
209	    		}
210	    	}
211	    	if (f2 < l2) {
212	    		if (iflag)
213	    			ICOPY_LIST(f2, tp2, l2);
214	    		else
215	    			CCOPY_LIST(f2, tp2, l2);
216	    	} else if (f1 < l1) {
217	    		if (iflag)
218	    			ICOPY_LIST(f1, tp2, l1);
219	    		else
220	    			CCOPY_LIST(f1, tp2, l1);
221	    	}
222	    	*p1 = l2;
223	    }
224	    tp2 = list1;	/* swap list1, list2 */
225	    list1 = list2;
226	    list2 = tp2;
227	    last = list2 + nmemb*size;
228	}
229	if (base == list2) {
230		memmove(list2, list1, nmemb*size);
231		list2 = list1;
232	}
233	free(list2);
234	return (0);
235}
236
237#define	swap(a, b) {					\
238		s = b;					\
239		i = size;				\
240		do {					\
241			tmp = *a; *a++ = *s; *s++ = tmp; \
242		} while (--i);				\
243		a -= size;				\
244	}
245#define reverse(bot, top) {				\
246	s = top;					\
247	do {						\
248		i = size;				\
249		do {					\
250			tmp = *bot; *bot++ = *s; *s++ = tmp; \
251		} while (--i);				\
252		s -= size2;				\
253	} while(bot < s);				\
254}
255
256/*
257 * Optional hybrid natural/pairwise first pass.  Eats up list1 in runs of
258 * increasing order, list2 in a corresponding linked list.  Checks for runs
259 * when THRESHOLD/2 pairs compare with same sense.  (Only used when NATURAL
260 * is defined.  Otherwise simple pairwise merging is used.)
261 */
262void
263setup(list1, list2, n, size, cmp)
264	size_t n, size;
265	int (*cmp) __P((const void *, const void *));
266	u_char *list1, *list2;
267{
268	int i, length, size2, tmp, sense;
269	u_char *f1, *f2, *s, *l2, *last, *p2;
270
271	size2 = size*2;
272	if (n <= 5) {
273		insertionsort(list1, n, size, cmp);
274		*EVAL(list2) = (u_char*) list2 + n*size;
275		return;
276	}
277	/*
278	 * Avoid running pointers out of bounds; limit n to evens
279	 * for simplicity.
280	 */
281	i = 4 + (n & 1);
282	insertionsort(list1 + (n - i) * size, i, size, cmp);
283	last = list1 + size * (n - i);
284	*EVAL(list2 + (last - list1)) = list2 + n * size;
285
286#ifdef NATURAL
287	p2 = list2;
288	f1 = list1;
289	sense = (cmp(f1, f1 + size) > 0);
290	for (; f1 < last; sense = !sense) {
291		length = 2;
292					/* Find pairs with same sense. */
293		for (f2 = f1 + size2; f2 < last; f2 += size2) {
294			if ((cmp(f2, f2+ size) > 0) != sense)
295				break;
296			length += 2;
297		}
298		if (length < THRESHOLD) {		/* Pairwise merge */
299			do {
300				p2 = *EVAL(p2) = f1 + size2 - list1 + list2;
301				if (sense > 0)
302					swap (f1, f1 + size);
303			} while ((f1 += size2) < f2);
304		} else {				/* Natural merge */
305			l2 = f2;
306			for (f2 = f1 + size2; f2 < l2; f2 += size2) {
307				if ((cmp(f2-size, f2) > 0) != sense) {
308					p2 = *EVAL(p2) = f2 - list1 + list2;
309					if (sense > 0)
310						reverse(f1, f2-size);
311					f1 = f2;
312				}
313			}
314			if (sense > 0)
315				reverse (f1, f2-size);
316			f1 = f2;
317			if (f2 < last || cmp(f2 - size, f2) > 0)
318				p2 = *EVAL(p2) = f2 - list1 + list2;
319			else
320				p2 = *EVAL(p2) = list2 + n*size;
321		}
322	}
323#else		/* pairwise merge only. */
324	for (f1 = list1, p2 = list2; f1 < last; f1 += size2) {
325		p2 = *EVAL(p2) = p2 + size2;
326		if (cmp (f1, f1 + size) > 0)
327			swap(f1, f1 + size);
328	}
329#endif /* NATURAL */
330}
331
332/*
333 * This is to avoid out-of-bounds addresses in sorting the
334 * last 4 elements.
335 */
336static void
337insertionsort(a, n, size, cmp)
338	u_char *a;
339	size_t n, size;
340	int (*cmp) __P((const void *, const void *));
341{
342	u_char *ai, *s, *t, *u, tmp;
343	int i;
344
345	for (ai = a+size; --n >= 1; ai += size)
346		for (t = ai; t > a; t -= size) {
347			u = t - size;
348			if (cmp(u, t) <= 0)
349				break;
350			swap(u, t);
351		}
352}
353