1/*	$NetBSD: merge.c,v 1.12 2009/02/12 03:12:36 lukem Exp $	*/
2
3/*-
4 * Copyright (c) 1992, 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.
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
38static char sccsid[] = "from: @(#)merge.c	8.2 (Berkeley) 2/14/94";
39#else
40__RCSID("$NetBSD: merge.c,v 1.12 2009/02/12 03:12:36 lukem Exp $");
41#endif
42#endif /* LIBC_SCCS and not lint */
43
44/*
45 * Hybrid exponential search/linear search merge sort with hybrid
46 * natural/pairwise first pass.  Requires about .3% more comparisons
47 * for random data than LSMS with pairwise first pass alone.
48 * It works for objects as small as two bytes.
49 */
50
51#define NATURAL
52#define THRESHOLD 16	/* Best choice for natural merge cut-off. */
53
54/* #define NATURAL to get hybrid natural merge.
55 * (The default is pairwise merging.)
56 */
57
58#include "namespace.h"
59#include <sys/types.h>
60
61#include <assert.h>
62#include <errno.h>
63#include <stdlib.h>
64#include <string.h>
65
66#ifdef __weak_alias
67__weak_alias(mergesort,_mergesort)
68#endif
69
70static void setup(u_char *, u_char *, size_t, size_t,
71    int (*)(const void *, const void *));
72static void insertionsort(u_char *, size_t, size_t,
73    int (*)(const void *, const void *));
74
75#define ISIZE sizeof(int)
76#define PSIZE sizeof(u_char *)
77#define ICOPY_LIST(src, dst, last)				\
78	do							\
79	*(int*)(void *)dst = *(int*)(void *)src,		\
80		src += ISIZE, dst += ISIZE;			\
81	while(src < last)
82#define ICOPY_ELT(src, dst, i)					\
83	do							\
84	*(int*)(void *)dst = *(int*)(void *)src,		\
85	src += ISIZE, dst += ISIZE;				\
86	while (i -= ISIZE)
87
88#define CCOPY_LIST(src, dst, last)		\
89	do					\
90		*dst++ = *src++;		\
91	while (src < last)
92#define CCOPY_ELT(src, dst, i)			\
93	do					\
94		*dst++ = *src++;		\
95	while (i -= 1)
96
97/*
98 * Find the next possible pointer head.  (Trickery for forcing an array
99 * to do double duty as a linked list when objects do not align with word
100 * boundaries.
101 */
102/* Assumption: PSIZE is a power of 2. */
103#define EVAL(p) ((u_char **)(void *)					\
104    ((u_char *)0 +							\
105    (((u_char *)(void *)(p) + PSIZE - 1 - (u_char *) 0) & ~(PSIZE - 1))))
106
107/*
108 * Arguments are as for qsort.
109 */
110int
111mergesort(base, nmemb, size, cmp)
112	void *base;
113	size_t nmemb;
114	size_t size;
115	int (*cmp)(const void *, const void *);
116{
117	size_t i;
118	int sense;
119	int big, iflag;
120	u_char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2;
121	u_char *list2, *list1, *p2, *p, *last, **p1;
122
123	_DIAGASSERT(base != NULL);
124	_DIAGASSERT(cmp != NULL);
125
126	if (size < PSIZE / 2) {		/* Pointers must fit into 2 * size. */
127		errno = EINVAL;
128		return (-1);
129	}
130
131	/*
132	 * XXX
133	 * Stupid subtraction for the Cray.
134	 */
135	iflag = 0;
136	if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE))
137		iflag = 1;
138
139	if ((list2 = malloc(nmemb * size + PSIZE)) == NULL)
140		return (-1);
141
142	list1 = base;
143	setup(list1, list2, nmemb, size, cmp);
144	last = list2 + nmemb * size;
145	i = big = 0;
146	while (*EVAL(list2) != last) {
147	    l2 = list1;
148	    p1 = EVAL(list1);
149	    for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) {
150	    	p2 = *EVAL(p2);
151	    	f1 = l2;
152	    	f2 = l1 = list1 + (p2 - list2);
153	    	if (p2 != last)
154	    		p2 = *EVAL(p2);
155	    	l2 = list1 + (p2 - list2);
156	    	while (f1 < l1 && f2 < l2) {
157	    		if ((*cmp)(f1, f2) <= 0) {
158	    			q = f2;
159	    			b = f1, t = l1;
160	    			sense = -1;
161	    		} else {
162	    			q = f1;
163	    			b = f2, t = l2;
164	    			sense = 0;
165	    		}
166	    		if (!big) {	/* here i = 0 */
167#if 0
168LINEAR:
169#endif
170				while ((b += size) < t && cmp(q, b) >sense)
171	    				if (++i == 6) {
172	    					big = 1;
173	    					goto EXPONENTIAL;
174	    				}
175	    		} else {
176EXPONENTIAL:	    		for (i = size; ; i <<= 1)
177	    				if ((p = (b + i)) >= t) {
178	    					if ((p = t - size) > b &&
179						    (*cmp)(q, p) <= sense)
180	    						t = p;
181	    					else
182	    						b = p;
183	    					break;
184	    				} else if ((*cmp)(q, p) <= sense) {
185	    					t = p;
186	    					if (i == size)
187	    						big = 0;
188	    					goto FASTCASE;
189	    				} else
190	    					b = p;
191#if 0
192SLOWCASE:
193#endif
194				while (t > b+size) {
195	    				i = (((t - b) / size) >> 1) * size;
196	    				if ((*cmp)(q, p = b + i) <= sense)
197	    					t = p;
198	    				else
199	    					b = p;
200	    			}
201	    			goto COPY;
202FASTCASE:	    		while (i > size)
203	    				if ((*cmp)(q,
204	    					p = b +
205						    (i = (unsigned int) i >> 1)) <= sense)
206	    					t = p;
207	    				else
208	    					b = p;
209COPY:	    			b = t;
210	    		}
211	    		i = size;
212	    		if (q == f1) {
213	    			if (iflag) {
214	    				ICOPY_LIST(f2, tp2, b);
215	    				ICOPY_ELT(f1, tp2, i);
216	    			} else {
217	    				CCOPY_LIST(f2, tp2, b);
218	    				CCOPY_ELT(f1, tp2, i);
219	    			}
220	    		} else {
221	    			if (iflag) {
222	    				ICOPY_LIST(f1, tp2, b);
223	    				ICOPY_ELT(f2, tp2, i);
224	    			} else {
225	    				CCOPY_LIST(f1, tp2, b);
226	    				CCOPY_ELT(f2, tp2, i);
227	    			}
228	    		}
229	    	}
230	    	if (f2 < l2) {
231	    		if (iflag)
232	    			ICOPY_LIST(f2, tp2, l2);
233	    		else
234	    			CCOPY_LIST(f2, tp2, l2);
235	    	} else if (f1 < l1) {
236	    		if (iflag)
237	    			ICOPY_LIST(f1, tp2, l1);
238	    		else
239	    			CCOPY_LIST(f1, tp2, l1);
240	    	}
241	    	*p1 = l2;
242	    }
243	    tp2 = list1;	/* swap list1, list2 */
244	    list1 = list2;
245	    list2 = tp2;
246	    last = list2 + nmemb*size;
247	}
248	if (base == list2) {
249		memmove(list2, list1, nmemb*size);
250		list2 = list1;
251	}
252	free(list2);
253	return (0);
254}
255
256#define	swap(a, b) {					\
257		s = b;					\
258		i = size;				\
259		do {					\
260			tmp = *a; *a++ = *s; *s++ = tmp; \
261		} while (--i);				\
262		a -= size;				\
263	}
264#define reverse(bot, top) {				\
265	s = top;					\
266	do {						\
267		i = size;				\
268		do {					\
269			tmp = *bot; *bot++ = *s; *s++ = tmp; \
270		} while (--i);				\
271		s -= size2;				\
272	} while(bot < s);				\
273}
274
275/*
276 * Optional hybrid natural/pairwise first pass.  Eats up list1 in runs of
277 * increasing order, list2 in a corresponding linked list.  Checks for runs
278 * when THRESHOLD/2 pairs compare with same sense.  (Only used when NATURAL
279 * is defined.  Otherwise simple pairwise merging is used.)
280 */
281
282/* XXX: shouldn't this function be static? - lukem 990810 */
283void
284setup(list1, list2, n, size, cmp)
285	size_t n, size;
286	int (*cmp)(const void *, const void *);
287	u_char *list1, *list2;
288{
289	int i, length, size2, tmp, sense;
290	u_char *f1, *f2, *s, *l2, *last, *p2;
291
292	_DIAGASSERT(cmp != NULL);
293	_DIAGASSERT(list1 != NULL);
294	_DIAGASSERT(list2 != NULL);
295
296	size2 = size*2;
297	if (n <= 5) {
298		insertionsort(list1, n, size, cmp);
299		*EVAL(list2) = list2 + n*size;
300		return;
301	}
302	/*
303	 * Avoid running pointers out of bounds; limit n to evens
304	 * for simplicity.
305	 */
306	i = 4 + (n & 1);
307	insertionsort(list1 + (n - i) * size, (size_t)i, size, cmp);
308	last = list1 + size * (n - i);
309	*EVAL(list2 + (last - list1)) = list2 + n * size;
310
311#ifdef NATURAL
312	p2 = list2;
313	f1 = list1;
314	sense = (cmp(f1, f1 + size) > 0);
315	for (; f1 < last; sense = !sense) {
316		length = 2;
317					/* Find pairs with same sense. */
318		for (f2 = f1 + size2; f2 < last; f2 += size2) {
319			if ((cmp(f2, f2+ size) > 0) != sense)
320				break;
321			length += 2;
322		}
323		if (length < THRESHOLD) {		/* Pairwise merge */
324			do {
325				p2 = *EVAL(p2) = f1 + size2 - list1 + list2;
326				if (sense > 0)
327					swap (f1, f1 + size);
328			} while ((f1 += size2) < f2);
329		} else {				/* Natural merge */
330			l2 = f2;
331			for (f2 = f1 + size2; f2 < l2; f2 += size2) {
332				if ((cmp(f2-size, f2) > 0) != sense) {
333					p2 = *EVAL(p2) = f2 - list1 + list2;
334					if (sense > 0)
335						reverse(f1, f2-size);
336					f1 = f2;
337				}
338			}
339			if (sense > 0)
340				reverse (f1, f2-size);
341			f1 = f2;
342			if (f2 < last || cmp(f2 - size, f2) > 0)
343				p2 = *EVAL(p2) = f2 - list1 + list2;
344			else
345				p2 = *EVAL(p2) = list2 + n*size;
346		}
347	}
348#else		/* pairwise merge only. */
349	for (f1 = list1, p2 = list2; f1 < last; f1 += size2) {
350		p2 = *EVAL(p2) = p2 + size2;
351		if (cmp (f1, f1 + size) > 0)
352			swap(f1, f1 + size);
353	}
354#endif /* NATURAL */
355}
356
357/*
358 * This is to avoid out-of-bounds addresses in sorting the
359 * last 4 elements.
360 */
361static void
362insertionsort(a, n, size, cmp)
363	u_char *a;
364	size_t n, size;
365	int (*cmp)(const void *, const void *);
366{
367	u_char *ai, *s, *t, *u, tmp;
368	int i;
369
370	_DIAGASSERT(a != NULL);
371	_DIAGASSERT(cmp != NULL);
372
373	for (ai = a+size; --n >= 1; ai += size)
374		for (t = ai; t > a; t -= size) {
375			u = t - size;
376			if (cmp(u, t) <= 0)
377				break;
378			swap(u, t);
379		}
380}
381