1/*-
2 * SPDX-License-Identifier: BSD-3-Clause
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#if defined(LIBC_SCCS) && !defined(lint)
36static char sccsid[] = "@(#)merge.c	8.2 (Berkeley) 2/14/94";
37#endif /* LIBC_SCCS and not lint */
38
39/*
40 * Hybrid exponential search/linear search merge sort with hybrid
41 * natural/pairwise first pass.  Requires about .3% more comparisons
42 * for random data than LSMS with pairwise first pass alone.
43 * It works for objects as small as two bytes.
44 */
45
46#define NATURAL
47#define THRESHOLD 16	/* Best choice for natural merge cut-off. */
48
49/* #define NATURAL to get hybrid natural merge.
50 * (The default is pairwise merging.)
51 */
52
53#include <sys/types.h>
54#include <stdint.h>
55
56#include <errno.h>
57#include <stdlib.h>
58#include <string.h>
59
60typedef int (*cmp_t)(const void *, const void *);
61#define	CMP(x, y)	cmp(x, y)
62
63static void setup(u_char *, u_char *, size_t, size_t, cmp_t);
64static void insertionsort(u_char *, size_t, size_t, cmp_t);
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 roundup2(x, y)	(((x) + ((y) - 1)) & (~((y) - 1)))
93#define EVAL(p) (u_char **)roundup2((uintptr_t)p, PSIZE)
94
95/*
96 * Arguments are as for qsort.
97 */
98int
99mergesort(void *base, size_t nmemb, size_t size, cmp_t cmp)
100{
101	size_t i;
102	int sense;
103	int big, iflag;
104	u_char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2;
105	u_char *list2, *list1, *p2, *p, *last, **p1;
106
107	if (size < PSIZE / 2) {		/* Pointers must fit into 2 * size. */
108		errno = EINVAL;
109		return (-1);
110	}
111
112	if (nmemb == 0)
113		return (0);
114
115	iflag = 0;
116	if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE))
117		iflag = 1;
118
119	if ((list2 = malloc(nmemb * size + PSIZE)) == NULL)
120		return (-1);
121
122	list1 = base;
123	setup(list1, list2, nmemb, size, cmp);
124	last = list2 + nmemb * size;
125	i = big = 0;
126	while (*EVAL(list2) != last) {
127		l2 = list1;
128		p1 = EVAL(list1);
129		for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) {
130			p2 = *EVAL(p2);
131			f1 = l2;
132			f2 = l1 = list1 + (p2 - list2);
133			if (p2 != last)
134				p2 = *EVAL(p2);
135			l2 = list1 + (p2 - list2);
136			while (f1 < l1 && f2 < l2) {
137				if (CMP(f1, f2) <= 0) {
138					q = f2;
139					b = f1, t = l1;
140					sense = -1;
141				} else {
142					q = f1;
143					b = f2, t = l2;
144					sense = 0;
145				}
146				if (!big) {	/* here i = 0 */
147				while ((b += size) < t && CMP(q, b) >sense)
148						if (++i == 6) {
149							big = 1;
150							goto EXPONENTIAL;
151						}
152				} else {
153EXPONENTIAL:	    		for (i = size; ; i <<= 1)
154						if ((p = (b + i)) >= t) {
155							if ((p = t - size) > b &&
156							CMP(q, p) <= sense)
157								t = p;
158							else
159								b = p;
160							break;
161						} else if (CMP(q, p) <= sense) {
162							t = p;
163							if (i == size)
164								big = 0;
165							goto FASTCASE;
166						} else
167							b = p;
168				while (t > b+size) {
169						i = (((t - b) / size) >> 1) * size;
170						if (CMP(q, p = b + i) <= sense)
171							t = p;
172						else
173							b = p;
174					}
175					goto COPY;
176FASTCASE:	    		while (i > size)
177						if (CMP(q,
178							p = b + (i >>= 1)) <= sense)
179							t = p;
180						else
181							b = p;
182COPY:	    			b = t;
183				}
184				i = size;
185				if (q == f1) {
186					if (iflag) {
187						ICOPY_LIST(f2, tp2, b);
188						ICOPY_ELT(f1, tp2, i);
189					} else {
190						CCOPY_LIST(f2, tp2, b);
191						CCOPY_ELT(f1, tp2, i);
192					}
193				} else {
194					if (iflag) {
195						ICOPY_LIST(f1, tp2, b);
196						ICOPY_ELT(f2, tp2, i);
197					} else {
198						CCOPY_LIST(f1, tp2, b);
199						CCOPY_ELT(f2, tp2, i);
200					}
201				}
202			}
203			if (f2 < l2) {
204				if (iflag)
205					ICOPY_LIST(f2, tp2, l2);
206				else
207					CCOPY_LIST(f2, tp2, l2);
208			} else if (f1 < l1) {
209				if (iflag)
210					ICOPY_LIST(f1, tp2, l1);
211				else
212					CCOPY_LIST(f1, tp2, l1);
213			}
214			*p1 = l2;
215		}
216		tp2 = list1;	/* swap list1, list2 */
217		list1 = list2;
218		list2 = tp2;
219		last = list2 + nmemb*size;
220	}
221	if (base == list2) {
222		memmove(list2, list1, nmemb*size);
223		list2 = list1;
224	}
225	free(list2);
226	return (0);
227}
228
229#define	swap(a, b) {					\
230		s = b;					\
231		i = size;				\
232		do {					\
233			tmp = *a; *a++ = *s; *s++ = tmp; \
234		} while (--i);				\
235		a -= size;				\
236	}
237#define reverse(bot, top) {				\
238	s = top;					\
239	do {						\
240		i = size;				\
241		do {					\
242			tmp = *bot; *bot++ = *s; *s++ = tmp; \
243		} while (--i);				\
244		s -= size2;				\
245	} while(bot < s);				\
246}
247
248/*
249 * Optional hybrid natural/pairwise first pass.  Eats up list1 in runs of
250 * increasing order, list2 in a corresponding linked list.  Checks for runs
251 * when THRESHOLD/2 pairs compare with same sense.  (Only used when NATURAL
252 * is defined.  Otherwise simple pairwise merging is used.)
253 */
254static
255void
256setup(u_char *list1, u_char *list2, size_t n, size_t size, cmp_t cmp)
257{
258	int i, length, size2, tmp, sense;
259	u_char *f1, *f2, *s, *l2, *last, *p2;
260
261	size2 = size*2;
262	if (n <= 5) {
263		insertionsort(list1, n, size, cmp);
264		*EVAL(list2) = (u_char*) list2 + n*size;
265		return;
266	}
267	/*
268	 * Avoid running pointers out of bounds; limit n to evens
269	 * for simplicity.
270	 */
271	i = 4 + (n & 1);
272	insertionsort(list1 + (n - i) * size, i, size, cmp);
273	last = list1 + size * (n - i);
274	*EVAL(list2 + (last - list1)) = list2 + n * size;
275
276#ifdef NATURAL
277	p2 = list2;
278	f1 = list1;
279	sense = (CMP(f1, f1 + size) > 0);
280	for (; f1 < last; sense = !sense) {
281		length = 2;
282					/* Find pairs with same sense. */
283		for (f2 = f1 + size2; f2 < last; f2 += size2) {
284			if ((CMP(f2, f2+ size) > 0) != sense)
285				break;
286			length += 2;
287		}
288		if (length < THRESHOLD) {		/* Pairwise merge */
289			do {
290				p2 = *EVAL(p2) = f1 + size2 - list1 + list2;
291				if (sense > 0)
292					swap (f1, f1 + size);
293			} while ((f1 += size2) < f2);
294		} else {				/* Natural merge */
295			l2 = f2;
296			for (f2 = f1 + size2; f2 < l2; f2 += size2) {
297				if ((CMP(f2-size, f2) > 0) != sense) {
298					p2 = *EVAL(p2) = f2 - list1 + list2;
299					if (sense > 0)
300						reverse(f1, f2-size);
301					f1 = f2;
302				}
303			}
304			if (sense > 0)
305				reverse (f1, f2-size);
306			f1 = f2;
307			if (f2 < last || CMP(f2 - size, f2) > 0)
308				p2 = *EVAL(p2) = f2 - list1 + list2;
309			else
310				p2 = *EVAL(p2) = list2 + n*size;
311		}
312	}
313#else		/* pairwise merge only. */
314	for (f1 = list1, p2 = list2; f1 < last; f1 += size2) {
315		p2 = *EVAL(p2) = p2 + size2;
316		if (CMP (f1, f1 + size) > 0)
317			swap(f1, f1 + size);
318	}
319#endif /* NATURAL */
320}
321
322/*
323 * This is to avoid out-of-bounds addresses in sorting the
324 * last 4 elements.
325 */
326static void
327insertionsort(u_char *a, size_t n, size_t size, cmp_t cmp)
328{
329	u_char *ai, *s, *t, *u, tmp;
330	int i;
331
332	for (ai = a+size; --n >= 1; ai += size)
333		for (t = ai; t > a; t -= size) {
334			u = t - size;
335			if (CMP(u, t) <= 0)
336				break;
337			swap(u, t);
338		}
339}
340