2 * Copyright (c) 1992, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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.
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
37 #if defined(LIBC_SCCS) && !defined(lint)
38 static char sccsid[] = "@(#)merge.c 8.2 (Berkeley) 2/14/94";
39 #endif /* LIBC_SCCS and not lint */
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
44 * Hybrid exponential search/linear search merge sort with hybrid
45 * natural/pairwise first pass. Requires about .3% more comparisons
46 * for random data than LSMS with pairwise first pass alone.
47 * It works for objects as small as two bytes.
51 #define THRESHOLD 16 /* Best choice for natural merge cut-off. */
53 /* #define NATURAL to get hybrid natural merge.
54 * (The default is pairwise merging.)
57 #include <sys/types.h>
63 static void setup(u_char *, u_char *, size_t, size_t,
64 int (*)(const void *, const void *));
65 static void insertionsort(u_char *, size_t, size_t,
66 int (*)(const void *, const void *));
68 #define ISIZE sizeof(int)
69 #define PSIZE sizeof(u_char *)
70 #define ICOPY_LIST(src, dst, last) \
72 *(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE; \
74 #define ICOPY_ELT(src, dst, i) \
76 *(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE; \
79 #define CCOPY_LIST(src, dst, last) \
83 #define CCOPY_ELT(src, dst, i) \
89 * Find the next possible pointer head. (Trickery for forcing an array
90 * to do double duty as a linked list when objects do not align with word
93 /* Assumption: PSIZE is a power of 2. */
94 #define EVAL(p) (u_char **) \
96 (((u_char *)p + PSIZE - 1 - (u_char *) 0) & ~(PSIZE - 1)))
99 * Arguments are as for qsort.
102 mergesort(base, nmemb, size, cmp)
106 int (*cmp)(const void *, const void *);
111 u_char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2;
112 u_char *list2, *list1, *p2, *p, *last, **p1;
114 if (size < PSIZE / 2) { /* Pointers must fit into 2 * size. */
124 * Stupid subtraction for the Cray.
127 if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE))
130 if ((list2 = malloc(nmemb * size + PSIZE)) == NULL)
134 setup(list1, list2, nmemb, size, cmp);
135 last = list2 + nmemb * size;
137 while (*EVAL(list2) != last) {
140 for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) {
143 f2 = l1 = list1 + (p2 - list2);
146 l2 = list1 + (p2 - list2);
147 while (f1 < l1 && f2 < l2) {
148 if ((*cmp)(f1, f2) <= 0) {
157 if (!big) { /* here i = 0 */
158 while ((b += size) < t && cmp(q, b) >sense)
164 EXPONENTIAL: for (i = size; ; i <<= 1)
165 if ((p = (b + i)) >= t) {
166 if ((p = t - size) > b &&
167 (*cmp)(q, p) <= sense)
172 } else if ((*cmp)(q, p) <= sense) {
180 i = (((t - b) / size) >> 1) * size;
181 if ((*cmp)(q, p = b + i) <= sense)
187 FASTCASE: while (i > size)
189 p = b + (i >>= 1)) <= sense)
198 ICOPY_LIST(f2, tp2, b);
199 ICOPY_ELT(f1, tp2, i);
201 CCOPY_LIST(f2, tp2, b);
202 CCOPY_ELT(f1, tp2, i);
206 ICOPY_LIST(f1, tp2, b);
207 ICOPY_ELT(f2, tp2, i);
209 CCOPY_LIST(f1, tp2, b);
210 CCOPY_ELT(f2, tp2, i);
216 ICOPY_LIST(f2, tp2, l2);
218 CCOPY_LIST(f2, tp2, l2);
219 } else if (f1 < l1) {
221 ICOPY_LIST(f1, tp2, l1);
223 CCOPY_LIST(f1, tp2, l1);
227 tp2 = list1; /* swap list1, list2 */
230 last = list2 + nmemb*size;
233 memmove(list2, list1, nmemb*size);
240 #define swap(a, b) { \
244 tmp = *a; *a++ = *s; *s++ = tmp; \
248 #define reverse(bot, top) { \
253 tmp = *bot; *bot++ = *s; *s++ = tmp; \
260 * Optional hybrid natural/pairwise first pass. Eats up list1 in runs of
261 * increasing order, list2 in a corresponding linked list. Checks for runs
262 * when THRESHOLD/2 pairs compare with same sense. (Only used when NATURAL
263 * is defined. Otherwise simple pairwise merging is used.)
266 setup(list1, list2, n, size, cmp)
268 int (*cmp)(const void *, const void *);
269 u_char *list1, *list2;
271 int i, length, size2, tmp, sense;
272 u_char *f1, *f2, *s, *l2, *last, *p2;
276 insertionsort(list1, n, size, cmp);
277 *EVAL(list2) = (u_char*) list2 + n*size;
281 * Avoid running pointers out of bounds; limit n to evens
285 insertionsort(list1 + (n - i) * size, i, size, cmp);
286 last = list1 + size * (n - i);
287 *EVAL(list2 + (last - list1)) = list2 + n * size;
292 sense = (cmp(f1, f1 + size) > 0);
293 for (; f1 < last; sense = !sense) {
295 /* Find pairs with same sense. */
296 for (f2 = f1 + size2; f2 < last; f2 += size2) {
297 if ((cmp(f2, f2+ size) > 0) != sense)
301 if (length < THRESHOLD) { /* Pairwise merge */
303 p2 = *EVAL(p2) = f1 + size2 - list1 + list2;
305 swap (f1, f1 + size);
306 } while ((f1 += size2) < f2);
307 } else { /* Natural merge */
309 for (f2 = f1 + size2; f2 < l2; f2 += size2) {
310 if ((cmp(f2-size, f2) > 0) != sense) {
311 p2 = *EVAL(p2) = f2 - list1 + list2;
313 reverse(f1, f2-size);
318 reverse (f1, f2-size);
320 if (f2 < last || cmp(f2 - size, f2) > 0)
321 p2 = *EVAL(p2) = f2 - list1 + list2;
323 p2 = *EVAL(p2) = list2 + n*size;
326 #else /* pairwise merge only. */
327 for (f1 = list1, p2 = list2; f1 < last; f1 += size2) {
328 p2 = *EVAL(p2) = p2 + size2;
329 if (cmp (f1, f1 + size) > 0)
336 * This is to avoid out-of-bounds addresses in sorting the
340 insertionsort(a, n, size, cmp)
343 int (*cmp)(const void *, const void *);
345 u_char *ai, *s, *t, *u, tmp;
348 for (ai = a+size; --n >= 1; ai += size)
349 for (t = ai; t > a; t -= size) {