merge.c revision 40896 
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
41/*
42 * Hybrid exponential search/linear search merge sort with hybrid
43 * natural/pairwise first pass.  Requires about .3% more comparisons
44 * for random data than LSMS with pairwise first pass alone.
45 * It works for objects as small as two bytes.
46 */
47
48#define NATURAL
49#define THRESHOLD 16	/* Best choice for natural merge cut-off. */
50
51/* #define NATURAL to get hybrid natural merge.
52 * (The default is pairwise merging.)
53 */
54
55#include <sys/types.h>
56
57#include <errno.h>
58#include <stdlib.h>
59#include <string.h>
60
61static void setup __P((u_char *, u_char *, size_t, size_t, int (*)()));
62static void insertionsort __P((u_char *, size_t, size_t, int (*)()));
63
64#define ISIZE sizeof(int)
65#define PSIZE sizeof(u_char *)
66#define ICOPY_LIST(src, dst, last)				\
67	do							\
68	*(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE;	\
69	while(src < last)
70#define ICOPY_ELT(src, dst, i)					\
71	do							\
72	*(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE;	\
73	while (i -= ISIZE)
74
75#define CCOPY_LIST(src, dst, last)		\
76	do					\
77		*dst++ = *src++;		\
78	while (src < last)
79#define CCOPY_ELT(src, dst, i)			\
80	do					\
81		*dst++ = *src++;		\
82	while (i -= 1)
83
84/*
85 * Find the next possible pointer head.  (Trickery for forcing an array
86 * to do double duty as a linked list when objects do not align with word
87 * boundaries.
88 */
89/* Assumption: PSIZE is a power of 2. */
90#define EVAL(p) (u_char **)						\
91	((u_char *)0 +							\
92	    (((u_char *)p + PSIZE - 1 - (u_char *) 0) & ~(PSIZE - 1)))
93
94/*
95 * Arguments are as for qsort.
96 */
97int
98mergesort(base, nmemb, size, cmp)
99	void *base;
100	size_t nmemb;
101	register size_t size;
102	int (*cmp) __P((const void *, const void *));
103{
104	register int i, sense;
105	int big, iflag;
106	register u_char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2;
107	u_char *list2, *list1, *p2, *p, *last, **p1;
108
109	if (size < PSIZE / 2) {		/* Pointers must fit into 2 * size. */
110		errno = EINVAL;
111		return (-1);
112	}
113
114	if (nmemb == 0)
115		return (0);
116
117	/*
118	 * XXX
119	 * Stupid subtraction for the Cray.
120	 */
121	iflag = 0;
122	if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE))
123		iflag = 1;
124
125	if ((list2 = malloc(nmemb * size + PSIZE)) == NULL)
126		return (-1);
127
128	list1 = base;
129	setup(list1, list2, nmemb, size, cmp);
130	last = list2 + nmemb * size;
131	i = big = 0;
132	while (*EVAL(list2) != last) {
133	    l2 = list1;
134	    p1 = EVAL(list1);
135	    for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) {
136	    	p2 = *EVAL(p2);
137	    	f1 = l2;
138	    	f2 = l1 = list1 + (p2 - list2);
139	    	if (p2 != last)
140	    		p2 = *EVAL(p2);
141	    	l2 = list1 + (p2 - list2);
142	    	while (f1 < l1 && f2 < l2) {
143	    		if ((*cmp)(f1, f2) <= 0) {
144	    			q = f2;
145	    			b = f1, t = l1;
146	    			sense = -1;
147	    		} else {
148	    			q = f1;
149	    			b = f2, t = l2;
150	    			sense = 0;
151	    		}
152	    		if (!big) {	/* here i = 0 */
153LINEAR:	    			while ((b += size) < t && cmp(q, b) >sense)
154	    				if (++i == 6) {
155	    					big = 1;
156	    					goto EXPONENTIAL;
157	    				}
158	    		} else {
159EXPONENTIAL:	    		for (i = size; ; i <<= 1)
160	    				if ((p = (b + i)) >= t) {
161	    					if ((p = t - size) > b &&
162						    (*cmp)(q, p) <= sense)
163	    						t = p;
164	    					else
165	    						b = p;
166	    					break;
167	    				} else if ((*cmp)(q, p) <= sense) {
168	    					t = p;
169	    					if (i == size)
170	    						big = 0;
171	    					goto FASTCASE;
172	    				} else
173	    					b = p;
174SLOWCASE:	    		while (t > b+size) {
175	    				i = (((t - b) / size) >> 1) * size;
176	    				if ((*cmp)(q, p = b + i) <= sense)
177	    					t = p;
178	    				else
179	    					b = p;
180	    			}
181	    			goto COPY;
182FASTCASE:	    		while (i > size)
183	    				if ((*cmp)(q,
184	    					p = b + (i >>= 1)) <= sense)
185	    					t = p;
186	    				else
187	    					b = p;
188COPY:	    			b = t;
189	    		}
190	    		i = size;
191	    		if (q == f1) {
192	    			if (iflag) {
193	    				ICOPY_LIST(f2, tp2, b);
194	    				ICOPY_ELT(f1, tp2, i);
195	    			} else {
196	    				CCOPY_LIST(f2, tp2, b);
197	    				CCOPY_ELT(f1, tp2, i);
198	    			}
199	    		} else {
200	    			if (iflag) {
201	    				ICOPY_LIST(f1, tp2, b);
202	    				ICOPY_ELT(f2, tp2, i);
203	    			} else {
204	    				CCOPY_LIST(f1, tp2, b);
205	    				CCOPY_ELT(f2, tp2, i);
206	    			}
207	    		}
208	    	}
209	    	if (f2 < l2) {
210	    		if (iflag)
211	    			ICOPY_LIST(f2, tp2, l2);
212	    		else
213	    			CCOPY_LIST(f2, tp2, l2);
214	    	} else if (f1 < l1) {
215	    		if (iflag)
216	    			ICOPY_LIST(f1, tp2, l1);
217	    		else
218	    			CCOPY_LIST(f1, tp2, l1);
219	    	}
220	    	*p1 = l2;
221	    }
222	    tp2 = list1;	/* swap list1, list2 */
223	    list1 = list2;
224	    list2 = tp2;
225	    last = list2 + nmemb*size;
226	}
227	if (base == list2) {
228		memmove(list2, list1, nmemb*size);
229		list2 = list1;
230	}
231	free(list2);
232	return (0);
233}
234
235#define	swap(a, b) {					\
236		s = b;					\
237		i = size;				\
238		do {					\
239			tmp = *a; *a++ = *s; *s++ = tmp; \
240		} while (--i);				\
241		a -= size;				\
242	}
243#define reverse(bot, top) {				\
244	s = top;					\
245	do {						\
246		i = size;				\
247		do {					\
248			tmp = *bot; *bot++ = *s; *s++ = tmp; \
249		} while (--i);				\
250		s -= size2;				\
251	} while(bot < s);				\
252}
253
254/*
255 * Optional hybrid natural/pairwise first pass.  Eats up list1 in runs of
256 * increasing order, list2 in a corresponding linked list.  Checks for runs
257 * when THRESHOLD/2 pairs compare with same sense.  (Only used when NATURAL
258 * is defined.  Otherwise simple pairwise merging is used.)
259 */
260void
261setup(list1, list2, n, size, cmp)
262	size_t n, size;
263	int (*cmp) __P((const void *, const void *));
264	u_char *list1, *list2;
265{
266	int i, length, size2, tmp, sense;
267	u_char *f1, *f2, *s, *l2, *last, *p2;
268
269	size2 = size*2;
270	if (n <= 5) {
271		insertionsort(list1, n, size, cmp);
272		*EVAL(list2) = (u_char*) list2 + n*size;
273		return;
274	}
275	/*
276	 * Avoid running pointers out of bounds; limit n to evens
277	 * for simplicity.
278	 */
279	i = 4 + (n & 1);
280	insertionsort(list1 + (n - i) * size, i, size, cmp);
281	last = list1 + size * (n - i);
282	*EVAL(list2 + (last - list1)) = list2 + n * size;
283
284#ifdef NATURAL
285	p2 = list2;
286	f1 = list1;
287	sense = (cmp(f1, f1 + size) > 0);
288	for (; f1 < last; sense = !sense) {
289		length = 2;
290					/* Find pairs with same sense. */
291		for (f2 = f1 + size2; f2 < last; f2 += size2) {
292			if ((cmp(f2, f2+ size) > 0) != sense)
293				break;
294			length += 2;
295		}
296		if (length < THRESHOLD) {		/* Pairwise merge */
297			do {
298				p2 = *EVAL(p2) = f1 + size2 - list1 + list2;
299				if (sense > 0)
300					swap (f1, f1 + size);
301			} while ((f1 += size2) < f2);
302		} else {				/* Natural merge */
303			l2 = f2;
304			for (f2 = f1 + size2; f2 < l2; f2 += size2) {
305				if ((cmp(f2-size, f2) > 0) != sense) {
306					p2 = *EVAL(p2) = f2 - list1 + list2;
307					if (sense > 0)
308						reverse(f1, f2-size);
309					f1 = f2;
310				}
311			}
312			if (sense > 0)
313				reverse (f1, f2-size);
314			f1 = f2;
315			if (f2 < last || cmp(f2 - size, f2) > 0)
316				p2 = *EVAL(p2) = f2 - list1 + list2;
317			else
318				p2 = *EVAL(p2) = list2 + n*size;
319		}
320	}
321#else		/* pairwise merge only. */
322	for (f1 = list1, p2 = list2; f1 < last; f1 += size2) {
323		p2 = *EVAL(p2) = p2 + size2;
324		if (cmp (f1, f1 + size) > 0)
325			swap(f1, f1 + size);
326	}
327#endif /* NATURAL */
328}
329
330/*
331 * This is to avoid out-of-bounds addresses in sorting the
332 * last 4 elements.
333 */
334static void
335insertionsort(a, n, size, cmp)
336	u_char *a;
337	size_t n, size;
338	int (*cmp) __P((const void *, const void *));
339{
340	u_char *ai, *s, *t, *u, tmp;
341	int i;
342
343	for (ai = a+size; --n >= 1; ai += size)
344		for (t = ai; t > a; t -= size) {
345			u = t - size;
346			if (cmp(u, t) <= 0)
347				break;
348			swap(u, t);
349		}
350}
351