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
36 * @(#)merge.c 8.2 (Berkeley) 2/14/94
37 * $DragonFly: src/lib/libc/stdlib/merge.c,v 1.5 2005/03/09 02:44:42 drhodus Exp $
41 * Hybrid exponential search/linear search merge sort with hybrid
42 * natural/pairwise first pass. Requires about .3% more comparisons
43 * for random data than LSMS with pairwise first pass alone.
44 * It works for objects as small as two bytes.
48 #define THRESHOLD 16 /* Best choice for natural merge cut-off. */
50 /* #define NATURAL to get hybrid natural merge.
51 * (The default is pairwise merging.)
54 #include <sys/types.h>
60 static void setup(u_char *, u_char *, size_t, size_t,
61 int (*)(const void *, const void *));
62 static void insertionsort(u_char *, size_t, size_t,
63 int (*)(const void *, const void *));
65 #define ISIZE sizeof(int)
66 #define PSIZE sizeof(u_char *)
67 #define ICOPY_LIST(src, dst, last) \
69 *(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE; \
71 #define ICOPY_ELT(src, dst, i) \
73 *(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE; \
76 #define CCOPY_LIST(src, dst, last) \
80 #define CCOPY_ELT(src, dst, i) \
86 * Find the next possible pointer head. (Trickery for forcing an array
87 * to do double duty as a linked list when objects do not align with word
90 /* Assumption: PSIZE is a power of 2. */
91 #define EVAL(p) (u_char **) \
93 (((u_char *)p + PSIZE - 1 - (u_char *) 0) & ~(PSIZE - 1)))
96 * Arguments are as for qsort.
99 mergesort(base, nmemb, size, cmp)
103 int (*cmp) (const void *, const void *);
108 u_char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2;
109 u_char *list2, *list1, *p2, *p, *last, **p1;
111 if (size < PSIZE / 2) { /* Pointers must fit into 2 * size. */
121 * Stupid subtraction for the Cray.
124 if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE))
127 if ((list2 = malloc(nmemb * size + PSIZE)) == NULL)
131 setup(list1, list2, nmemb, size, cmp);
132 last = list2 + nmemb * size;
134 while (*EVAL(list2) != last) {
137 for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) {
140 f2 = l1 = list1 + (p2 - list2);
143 l2 = list1 + (p2 - list2);
144 while (f1 < l1 && f2 < l2) {
145 if ((*cmp)(f1, f2) <= 0) {
154 if (!big) { /* here i = 0 */
155 while ((b += size) < t && cmp(q, b) >sense)
161 EXPONENTIAL: for (i = size; ; i <<= 1)
162 if ((p = (b + i)) >= t) {
163 if ((p = t - size) > b &&
164 (*cmp)(q, p) <= sense)
169 } else if ((*cmp)(q, p) <= sense) {
177 i = (((t - b) / size) >> 1) * size;
178 if ((*cmp)(q, p = b + i) <= sense)
184 FASTCASE: while (i > size)
186 p = b + (i >>= 1)) <= sense)
195 ICOPY_LIST(f2, tp2, b);
196 ICOPY_ELT(f1, tp2, i);
198 CCOPY_LIST(f2, tp2, b);
199 CCOPY_ELT(f1, tp2, i);
203 ICOPY_LIST(f1, tp2, b);
204 ICOPY_ELT(f2, tp2, i);
206 CCOPY_LIST(f1, tp2, b);
207 CCOPY_ELT(f2, tp2, i);
213 ICOPY_LIST(f2, tp2, l2);
215 CCOPY_LIST(f2, tp2, l2);
216 } else if (f1 < l1) {
218 ICOPY_LIST(f1, tp2, l1);
220 CCOPY_LIST(f1, tp2, l1);
224 tp2 = list1; /* swap list1, list2 */
227 last = list2 + nmemb*size;
230 memmove(list2, list1, nmemb*size);
237 #define swap(a, b) { \
241 tmp = *a; *a++ = *s; *s++ = tmp; \
245 #define reverse(bot, top) { \
250 tmp = *bot; *bot++ = *s; *s++ = tmp; \
257 * Optional hybrid natural/pairwise first pass. Eats up list1 in runs of
258 * increasing order, list2 in a corresponding linked list. Checks for runs
259 * when THRESHOLD/2 pairs compare with same sense. (Only used when NATURAL
260 * is defined. Otherwise simple pairwise merging is used.)
263 setup(list1, list2, n, size, cmp)
265 int (*cmp) (const void *, const void *);
266 u_char *list1, *list2;
268 int i, length, size2, tmp, sense;
269 u_char *f1, *f2, *s, *l2, *last, *p2;
273 insertionsort(list1, n, size, cmp);
274 *EVAL(list2) = (u_char*) list2 + n*size;
278 * Avoid running pointers out of bounds; limit n to evens
282 insertionsort(list1 + (n - i) * size, i, size, cmp);
283 last = list1 + size * (n - i);
284 *EVAL(list2 + (last - list1)) = list2 + n * size;
289 sense = (cmp(f1, f1 + size) > 0);
290 for (; f1 < last; sense = !sense) {
292 /* Find pairs with same sense. */
293 for (f2 = f1 + size2; f2 < last; f2 += size2) {
294 if ((cmp(f2, f2+ size) > 0) != sense)
298 if (length < THRESHOLD) { /* Pairwise merge */
300 p2 = *EVAL(p2) = f1 + size2 - list1 + list2;
302 swap (f1, f1 + size);
303 } while ((f1 += size2) < f2);
304 } else { /* Natural merge */
306 for (f2 = f1 + size2; f2 < l2; f2 += size2) {
307 if ((cmp(f2-size, f2) > 0) != sense) {
308 p2 = *EVAL(p2) = f2 - list1 + list2;
310 reverse(f1, f2-size);
315 reverse (f1, f2-size);
317 if (f2 < last || cmp(f2 - size, f2) > 0)
318 p2 = *EVAL(p2) = f2 - list1 + list2;
320 p2 = *EVAL(p2) = list2 + n*size;
323 #else /* pairwise merge only. */
324 for (f1 = list1, p2 = list2; f1 < last; f1 += size2) {
325 p2 = *EVAL(p2) = p2 + size2;
326 if (cmp (f1, f1 + size) > 0)
333 * This is to avoid out-of-bounds addresses in sorting the
337 insertionsort(a, n, size, cmp)
340 int (*cmp) (const void *, const void *);
342 u_char *ai, *s, *t, *u, tmp;
345 for (ai = a+size; --n >= 1; ai += size)
346 for (t = ai; t > a; t -= size) {