1/***************************************************************************
2 * LPRng - An Extended Print Spooler System
3 *
4 * Copyright 1988-2003, Patrick Powell, San Diego, CA
5 *     papowell@lprng.com
6 * See LICENSE for conditions of use.
7 *
8 ***************************************************************************/
9
10 static char *const _id =
11"$Id: merge.c,v 1.1.1.1 2008/10/15 03:28:27 james26_jang Exp $";
12
13
14/*-
15 * copyright (c) 1992, 1993
16 *	The Regents of the University of California.  All rights reserved.
17 *
18 * This code is derived from software contributed to Berkeley by
19 * Peter McIlroy.
20 *
21 * Redistribution and use in source and binary forms, with or without
22 * modification, are permitted provided that the following conditions
23 * are met:
24 * 1. Redistributions of source code must retain the above copyright
25 *    notice, this list of conditions and the following disclaimer.
26 * 2. Redistributions in binary form must reproduce the above copyright
27 *    notice, this list of conditions and the following disclaimer in the
28 *    documentation and/or other materials provided with the distribution.
29 * 3. All advertising materials mentioning features or use of this software
30 *    must display the following acknowledgement:
31 *	This product includes software developed by the University of
32 *	California, Berkeley and its contributors.
33 * 4. Neither the name of the University nor the names of its contributors
34 *    may be used to endorse or promote products derived from this software
35 *    without specific prior written permission.
36 *
37 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
38 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
39 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
40 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
41 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
42 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
43 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
45 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
46 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
47 * SUCH DAMAGE.
48 */
49
50#if defined(LIBC_SCCS) && !defined(lint)
51 static char sccsid[] = "@(#)merge.c	8.2 (Berkeley) 2/14/94";
52#endif /* LIBC_SCCS and not lint */
53
54/*
55 * Hybrid exponential search/linear search merge sort with hybrid
56 * natural/pairwise first pass.  Requires about .3% more comparisons
57 * for random data than LSMS with pairwise first pass alone.
58 * It works for objects as small as two bytes.
59 */
60
61#define NATURAL
62#define THRESHOLD 16	/* Best choice for natural merge cut-off. */
63
64/* #define NATURAL to get hybrid natural merge.
65 * (The default is pairwise merging.)
66 */
67
68#include "lp.h"
69#include "merge.h"
70
71 static void
72 setup(unsigned char *list1, unsigned char *list2, size_t n, size_t size,
73	int (*cmp)(const void *, const void *, const void *), const void * arg);
74 static void
75 insertionsort(unsigned char *a, size_t n, size_t size,
76	int (*cmp)(const void *, const void *, const void *), const void * arg);
77
78#define ISIZE sizeof(int)
79#define PSIZE sizeof(unsigned char *)
80#define ICOPY_LIST(src, dst, last)				\
81	do							\
82	*(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE;	\
83	while(src < last)
84#define ICOPY_ELT(src, dst, i)					\
85	do							\
86	*(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE;	\
87	while (i -= ISIZE)
88
89#define CCOPY_LIST(src, dst, last)		\
90	do					\
91		*dst++ = *src++;		\
92	while (src < last)
93#define CCOPY_ELT(src, dst, i)			\
94	do					\
95		*dst++ = *src++;		\
96	while (i -= 1)
97
98/*
99 * Find the next possible pointer head.  (Trickery for forcing an array
100 * to do double duty as a linked list when objects do not align with word
101 * boundaries.
102 */
103/* Assumption: PSIZE is a power of 2. */
104#define EVAL(p) (unsigned char **)						\
105	((unsigned char *)0 +							\
106	    (((unsigned char *)p + PSIZE - 1 - (unsigned char *) 0) & ~(PSIZE - 1)))
107
108/*
109 * We pass an additional arguement to the comparison routines.
110 * This lack of a third argument is a real defect in the qsort, etc.
111 * code
112 */
113int
114Mergesort(void *base, size_t nmemb, size_t size,
115	int (*cmp)(const void *, const void *, const void *), const void * arg)
116{
117	int i, sense;
118	int big, iflag;
119	unsigned char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2;
120	unsigned char *list2, *list1, *p2, *p, *last, **p1;
121
122	if( nmemb < 2 ){
123		return(0);
124	}
125	if (size < PSIZE / 2) {		/* Pointers must fit into 2 * size. */
126		errno = EINVAL;
127		return (-1);
128	}
129
130	/*
131	 * XXX
132	 * Stupid subtraction for the Cray.
133	 */
134	iflag = 0;
135	if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE))
136		iflag = 1;
137
138	list2 = malloc_or_die( (nmemb * size + PSIZE),__FILE__,__LINE__);
139
140	list1 = base;
141	setup(list1, list2, nmemb, size, cmp, arg);
142	last = list2 + nmemb * size;
143	i = big = 0;
144	while (*EVAL(list2) != last) {
145	    l2 = list1;
146	    p1 = EVAL(list1);
147	    for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) {
148	    	p2 = *EVAL(p2);
149	    	f1 = l2;
150	    	f2 = l1 = list1 + (p2 - list2);
151	    	if (p2 != last)
152	    		p2 = *EVAL(p2);
153	    	l2 = list1 + (p2 - list2);
154	    	while (f1 < l1 && f2 < l2) {
155	    		if ((*cmp)(f1, f2, arg) <= 0) {
156	    			q = f2;
157	    			b = f1, t = l1;
158	    			sense = -1;
159	    		} else {
160	    			q = f1;
161	    			b = f2, t = l2;
162	    			sense = 0;
163	    		}
164	    		if (!big) {	/* here i = 0 */
165/*LINEAR:*/			while ((b += size) < t && cmp(q, b, arg) >sense)
166	    				if (++i == 6) {
167	    					big = 1;
168	    					goto EXPONENTIAL;
169	    				}
170	    		} else {
171 EXPONENTIAL:	    		for (i = size; ; i <<= 1)
172	    				if ((p = (b + i)) >= t) {
173	    					if ((p = t - size) > b &&
174						    (*cmp)(q, p, arg) <= sense)
175	    						t = p;
176	    					else
177	    						b = p;
178	    					break;
179	    				} else if ((*cmp)(q, p, arg) <= sense) {
180	    					t = p;
181	    					if (i == (int)size)
182	    						big = 0;
183	    					goto FASTCASE;
184	    				} else
185	    					b = p;
186/*SLOWCASE:*/    		while (t > b+size) {
187	    				i = (((t - b) / size) >> 1) * size;
188	    				if ((*cmp)(q, p = b + i, arg) <= sense)
189	    					t = p;
190	    				else
191	    					b = p;
192	    			}
193	    			goto COPY;
194 FASTCASE:	    		while (i > (int)size)
195	    				if ((*cmp)(q,
196	    					p = b + (i >>= 1), arg) <= sense)
197	    					t = p;
198	    				else
199	    					b = p;
200 COPY:	    			b = t;
201	    		}
202	    		i = size;
203	    		if (q == f1) {
204	    			if (iflag) {
205	    				ICOPY_LIST(f2, tp2, b);
206	    				ICOPY_ELT(f1, tp2, i);
207	    			} else {
208	    				CCOPY_LIST(f2, tp2, b);
209	    				CCOPY_ELT(f1, tp2, i);
210	    			}
211	    		} else {
212	    			if (iflag) {
213	    				ICOPY_LIST(f1, tp2, b);
214	    				ICOPY_ELT(f2, tp2, i);
215	    			} else {
216	    				CCOPY_LIST(f1, tp2, b);
217	    				CCOPY_ELT(f2, tp2, i);
218	    			}
219	    		}
220	    	}
221	    	if (f2 < l2) {
222	    		if (iflag)
223	    			ICOPY_LIST(f2, tp2, l2);
224	    		else
225	    			CCOPY_LIST(f2, tp2, l2);
226	    	} else if (f1 < l1) {
227	    		if (iflag)
228	    			ICOPY_LIST(f1, tp2, l1);
229	    		else
230	    			CCOPY_LIST(f1, tp2, l1);
231	    	}
232	    	*p1 = l2;
233	    }
234	    tp2 = list1;	/* swap list1, list2 */
235	    list1 = list2;
236	    list2 = tp2;
237	    last = list2 + nmemb*size;
238	}
239	if (base == list2) {
240		memmove(list2, list1, nmemb*size);
241		list2 = list1;
242	}
243	free(list2);
244	return (0);
245}
246
247#define	swap(a, b) {					\
248		s = b;					\
249		i = size;				\
250		do {					\
251			tmp = *a; *a++ = *s; *s++ = tmp; \
252		} while (--i);				\
253		a -= size;				\
254	}
255#define reverse(bot, top) {				\
256	s = top;					\
257	do {						\
258		i = size;				\
259		do {					\
260			tmp = *bot; *bot++ = *s; *s++ = tmp; \
261		} while (--i);				\
262		s -= size2;				\
263	} while(bot < s);				\
264}
265
266/*
267 * Optional hybrid natural/pairwise first pass.  Eats up list1 in runs of
268 * increasing order, list2 in a corresponding linked list.  Checks for runs
269 * when THRESHOLD/2 pairs compare with same sense.  (Only used when NATURAL
270 * is defined.  Otherwise simple pairwise merging is used.)
271 */
272 static void
273 setup(unsigned char *list1, unsigned char *list2, size_t n, size_t size,
274	int (*cmp)(const void *, const void *, const void *), const void * arg)
275{
276	int i, length, size2, tmp, sense;
277	unsigned char *f1, *f2, *s, *l2, *last, *p2;
278
279	size2 = size*2;
280	if (n <= 5) {
281		insertionsort(list1, n, size, cmp, arg);
282		*EVAL(list2) = (unsigned char*) list2 + n*size;
283		return;
284	}
285	/*
286	 * Avoid running pointers out of bounds; limit n to evens
287	 * for simplicity.
288	 */
289	i = 4 + (n & 1);
290	insertionsort(list1 + (n - i) * size, i, size, cmp, arg);
291	last = list1 + size * (n - i);
292	*EVAL(list2 + (last - list1)) = list2 + n * size;
293
294#ifdef NATURAL
295	p2 = list2;
296	f1 = list1;
297	sense = (cmp(f1, f1 + size, arg) > 0);
298	for (; f1 < last; sense = !sense) {
299		length = 2;
300					/* Find pairs with same sense. */
301		for (f2 = f1 + size2; f2 < last; f2 += size2) {
302			if ((cmp(f2, f2+ size, arg) > 0) != sense)
303				break;
304			length += 2;
305		}
306		if (length < THRESHOLD) {		/* Pairwise merge */
307			do {
308				p2 = *EVAL(p2) = f1 + size2 - list1 + list2;
309				if (sense > 0)
310					swap (f1, f1 + size);
311			} while ((f1 += size2) < f2);
312		} else {				/* Natural merge */
313			l2 = f2;
314			for (f2 = f1 + size2; f2 < l2; f2 += size2) {
315				if ((cmp(f2-size, f2, arg) > 0) != sense) {
316					p2 = *EVAL(p2) = f2 - list1 + list2;
317					if (sense > 0)
318						reverse(f1, f2-size);
319					f1 = f2;
320				}
321			}
322			if (sense > 0)
323				reverse (f1, f2-size);
324			f1 = f2;
325			if (f2 < last || cmp(f2 - size, f2, arg) > 0)
326				p2 = *EVAL(p2) = f2 - list1 + list2;
327			else
328				p2 = *EVAL(p2) = list2 + n*size;
329		}
330	}
331#else		/* pairwise merge only. */
332	for (f1 = list1, p2 = list2; f1 < last; f1 += size2) {
333		p2 = *EVAL(p2) = p2 + size2;
334		if (cmp (f1, f1 + size, arg) > 0)
335			swap(f1, f1 + size);
336	}
337#endif /* NATURAL */
338}
339
340/*
341 * This is to avoid out-of-bounds addresses in sorting the
342 * last 4 elements.
343 */
344 static void
345 insertionsort(unsigned char *a, size_t n, size_t size,
346	int (*cmp)(const void *, const void *, const void *), const void * arg)
347{
348	unsigned char *ai, *s, *t, *u, tmp;
349	int i;
350
351	for (ai = a+size; --n >= 1; ai += size)
352		for (t = ai; t > a; t -= size) {
353			u = t - size;
354			if (cmp(u, t, arg) <= 0)
355				break;
356			swap(u, t);
357		}
358}
359