1/*	$NetBSD: merge.c,v 1.16 2018/05/23 21:21:27 joerg 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 1.16 2018/05/23 21:21:27 joerg 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 *)(void *)(p) + PSIZE - 1 - (u_char *) 0) & ~(PSIZE - 1)))
105
106/*
107 * Arguments are as for qsort.
108 */
109int
110mergesort(void *base, size_t nmemb, size_t size,
111    int (*cmp)(const void *, const void *))
112{
113	size_t i;
114	int sense;
115	int big, iflag;
116	u_char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2;
117	u_char *list2, *list1, *p2, *p, *last, **p1;
118
119	_DIAGASSERT(base != NULL);
120	_DIAGASSERT(cmp != NULL);
121
122	if (size < PSIZE / 2) {		/* Pointers must fit into 2 * size. */
123		errno = EINVAL;
124		return (-1);
125	}
126
127	if (nmemb == 0)
128		return (0);
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	if ((list2 = malloc(nmemb * size + PSIZE)) == NULL)
139		return (-1);
140
141	list1 = base;
142	setup(list1, list2, nmemb, size, cmp);
143	last = list2 + nmemb * size;
144	i = big = 0;
145	while (*EVAL(list2) != last) {
146	    l2 = list1;
147	    p1 = EVAL(list1);
148	    for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) {
149	    	p2 = *EVAL(p2);
150	    	f1 = l2;
151	    	f2 = l1 = list1 + (p2 - list2);
152	    	if (p2 != last)
153	    		p2 = *EVAL(p2);
154	    	l2 = list1 + (p2 - list2);
155	    	while (f1 < l1 && f2 < l2) {
156	    		if ((*cmp)(f1, f2) <= 0) {
157	    			q = f2;
158	    			b = f1, t = l1;
159	    			sense = -1;
160	    		} else {
161	    			q = f1;
162	    			b = f2, t = l2;
163	    			sense = 0;
164	    		}
165	    		if (!big) {	/* here i = 0 */
166#if 0
167LINEAR:
168#endif
169				while ((b += size) < t && cmp(q, b) >sense)
170	    				if (++i == 6) {
171	    					big = 1;
172	    					goto EXPONENTIAL;
173	    				}
174	    		} else {
175EXPONENTIAL:	    		for (i = size; ; i <<= 1)
176	    				if ((p = (b + i)) >= t) {
177	    					if ((p = t - size) > b &&
178						    (*cmp)(q, p) <= sense)
179	    						t = p;
180	    					else
181	    						b = p;
182	    					break;
183	    				} else if ((*cmp)(q, p) <= sense) {
184	    					t = p;
185	    					if (i == size)
186	    						big = 0;
187	    					goto FASTCASE;
188	    				} else
189	    					b = p;
190#if 0
191SLOWCASE:
192#endif
193				while (t > b+size) {
194	    				i = (((t - b) / size) >> 1) * size;
195	    				if ((*cmp)(q, p = b + i) <= sense)
196	    					t = p;
197	    				else
198	    					b = p;
199	    			}
200	    			goto COPY;
201FASTCASE:	    		while (i > size)
202	    				if ((*cmp)(q,
203	    					p = b +
204						    (i = (unsigned int) i >> 1)) <= sense)
205	    					t = p;
206	    				else
207	    					b = p;
208COPY:	    			b = t;
209	    		}
210	    		i = size;
211	    		if (q == f1) {
212	    			if (iflag) {
213	    				ICOPY_LIST(f2, tp2, b);
214	    				ICOPY_ELT(f1, tp2, i);
215	    			} else {
216	    				CCOPY_LIST(f2, tp2, b);
217	    				CCOPY_ELT(f1, tp2, i);
218	    			}
219	    		} else {
220	    			if (iflag) {
221	    				ICOPY_LIST(f1, tp2, b);
222	    				ICOPY_ELT(f2, tp2, i);
223	    			} else {
224	    				CCOPY_LIST(f1, tp2, b);
225	    				CCOPY_ELT(f2, tp2, i);
226	    			}
227	    		}
228	    	}
229	    	if (f2 < l2) {
230	    		if (iflag)
231	    			ICOPY_LIST(f2, tp2, l2);
232	    		else
233	    			CCOPY_LIST(f2, tp2, l2);
234	    	} else if (f1 < l1) {
235	    		if (iflag)
236	    			ICOPY_LIST(f1, tp2, l1);
237	    		else
238	    			CCOPY_LIST(f1, tp2, l1);
239	    	}
240	    	*p1 = l2;
241	    }
242	    tp2 = list1;	/* swap list1, list2 */
243	    list1 = list2;
244	    list2 = tp2;
245	    last = list2 + nmemb*size;
246	}
247	if (base == list2) {
248		memmove(list2, list1, nmemb*size);
249		list2 = list1;
250	}
251	free(list2);
252	return (0);
253}
254
255#define	swap(a, b) {					\
256		s = b;					\
257		i = size;				\
258		do {					\
259			tmp = *a; *a++ = *s; *s++ = tmp; \
260		} while (--i);				\
261		a -= size;				\
262	}
263#define reverse(bot, top) {				\
264	s = top;					\
265	do {						\
266		i = size;				\
267		do {					\
268			tmp = *bot; *bot++ = *s; *s++ = tmp; \
269		} while (--i);				\
270		s -= size2;				\
271	} while(bot < s);				\
272}
273
274/*
275 * Optional hybrid natural/pairwise first pass.  Eats up list1 in runs of
276 * increasing order, list2 in a corresponding linked list.  Checks for runs
277 * when THRESHOLD/2 pairs compare with same sense.  (Only used when NATURAL
278 * is defined.  Otherwise simple pairwise merging is used.)
279 */
280
281/* XXX: shouldn't this function be static? - lukem 990810 */
282void
283setup(u_char *list1, u_char *list2, size_t n, size_t size,
284    int (*cmp)(const void *, const void *))
285{
286	int length, tmp, sense;
287	u_char *f1, *f2, *s, *l2, *last, *p2;
288	size_t size2, i;
289
290	_DIAGASSERT(cmp != NULL);
291	_DIAGASSERT(list1 != NULL);
292	_DIAGASSERT(list2 != NULL);
293
294	size2 = size*2;
295	if (n <= 5) {
296		insertionsort(list1, n, size, cmp);
297		*EVAL(list2) = list2 + n*size;
298		return;
299	}
300	/*
301	 * Avoid running pointers out of bounds; limit n to evens
302	 * for simplicity.
303	 */
304	i = 4 + (n & 1);
305	insertionsort(list1 + (n - i) * size, i, size, cmp);
306	last = list1 + size * (n - i);
307	*EVAL(list2 + (last - list1)) = list2 + n * size;
308
309#ifdef NATURAL
310	p2 = list2;
311	f1 = list1;
312	sense = (cmp(f1, f1 + size) > 0);
313	for (; f1 < last; sense = !sense) {
314		length = 2;
315					/* Find pairs with same sense. */
316		for (f2 = f1 + size2; f2 < last; f2 += size2) {
317			if ((cmp(f2, f2+ size) > 0) != sense)
318				break;
319			length += 2;
320		}
321		if (length < THRESHOLD) {		/* Pairwise merge */
322			do {
323				p2 = *EVAL(p2) = f1 + size2 - list1 + list2;
324				if (sense > 0)
325					swap (f1, f1 + size);
326			} while ((f1 += size2) < f2);
327		} else {				/* Natural merge */
328			l2 = f2;
329			for (f2 = f1 + size2; f2 < l2; f2 += size2) {
330				if ((cmp(f2-size, f2) > 0) != sense) {
331					p2 = *EVAL(p2) = f2 - list1 + list2;
332					if (sense > 0)
333						reverse(f1, f2 - size);
334					f1 = f2;
335				}
336			}
337			if (sense > 0)
338				reverse(f1, f2 - size);
339			f1 = f2;
340			if (f2 < last || cmp(f2 - size, f2) > 0)
341				p2 = *EVAL(p2) = f2 - list1 + list2;
342			else
343				p2 = *EVAL(p2) = list2 + n*size;
344		}
345	}
346#else		/* pairwise merge only. */
347	for (f1 = list1, p2 = list2; f1 < last; f1 += size2) {
348		p2 = *EVAL(p2) = p2 + size2;
349		if (cmp (f1, f1 + size) > 0)
350			swap(f1, f1 + size);
351	}
352#endif /* NATURAL */
353}
354
355/*
356 * This is to avoid out-of-bounds addresses in sorting the
357 * last 4 elements.
358 */
359static void
360insertionsort(u_char *a, size_t n, size_t size,
361    int (*cmp)(const void *, const void *))
362{
363	u_char *ai, *s, *t, *u, tmp;
364	size_t i;
365
366	_DIAGASSERT(a != NULL);
367	_DIAGASSERT(cmp != NULL);
368
369	for (ai = a+size; --n >= 1; ai += size)
370		for (t = ai; t > a; t -= size) {
371			u = t - size;
372			if (cmp(u, t) <= 0)
373				break;
374			swap(u, t);
375		}
376}
377