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. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 #if defined(LIBC_SCCS) && !defined(lint)
34 static char sccsid[] = "@(#)merge.c 8.2 (Berkeley) 2/14/94";
35 #endif /* LIBC_SCCS and not lint */
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
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.
47 #define THRESHOLD 16 /* Best choice for natural merge cut-off. */
49 /* #define NATURAL to get hybrid natural merge.
50 * (The default is pairwise merging.)
53 #include <sys/types.h>
59 #ifdef I_AM_MERGESORT_B
60 #include "block_abi.h"
61 #define DECLARE_CMP DECLARE_BLOCK(int, cmp, const void *, const void *)
62 typedef DECLARE_BLOCK(int, cmp_t, const void *, const void *);
63 #define CMP(x, y) CALL_BLOCK(cmp, x, y)
65 typedef int (*cmp_t)(const void *, const void *);
66 #define CMP(x, y) cmp(x, y)
69 static void setup(u_char *, u_char *, size_t, size_t, cmp_t);
70 static void insertionsort(u_char *, size_t, size_t, cmp_t);
72 #define ISIZE sizeof(int)
73 #define PSIZE sizeof(u_char *)
74 #define ICOPY_LIST(src, dst, last) \
76 *(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE; \
78 #define ICOPY_ELT(src, dst, i) \
80 *(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE; \
83 #define CCOPY_LIST(src, dst, last) \
87 #define CCOPY_ELT(src, dst, i) \
93 * Find the next possible pointer head. (Trickery for forcing an array
94 * to do double duty as a linked list when objects do not align with word
97 /* Assumption: PSIZE is a power of 2. */
98 #define EVAL(p) (u_char **) \
100 (((u_char *)p + PSIZE - 1 - (u_char *) 0) & ~(PSIZE - 1)))
102 #ifdef I_AM_MERGESORT_B
103 int mergesort_b(void *, size_t, size_t, cmp_t);
105 int mergesort(void *, size_t, size_t, cmp_t);
109 * Arguments are as for qsort.
112 #ifdef I_AM_MERGESORT_B
113 mergesort_b(void *base, size_t nmemb, size_t size, cmp_t cmp)
115 mergesort(void *base, size_t nmemb, size_t size, cmp_t cmp)
121 u_char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2;
122 u_char *list2, *list1, *p2, *p, *last, **p1;
124 if (size < PSIZE / 2) { /* Pointers must fit into 2 * size. */
134 * Stupid subtraction for the Cray.
137 if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE))
140 if ((list2 = malloc(nmemb * size + PSIZE)) == NULL)
144 setup(list1, list2, nmemb, size, cmp);
145 last = list2 + nmemb * size;
147 while (*EVAL(list2) != last) {
150 for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) {
153 f2 = l1 = list1 + (p2 - list2);
156 l2 = list1 + (p2 - list2);
157 while (f1 < l1 && f2 < l2) {
158 if (CMP(f1, f2) <= 0) {
167 if (!big) { /* here i = 0 */
168 while ((b += size) < t && CMP(q, b) >sense)
174 EXPONENTIAL: for (i = size; ; i <<= 1)
175 if ((p = (b + i)) >= t) {
176 if ((p = t - size) > b &&
182 } else if (CMP(q, p) <= sense) {
190 i = (((t - b) / size) >> 1) * size;
191 if (CMP(q, p = b + i) <= sense)
197 FASTCASE: while (i > size)
199 p = b + (i >>= 1)) <= sense)
208 ICOPY_LIST(f2, tp2, b);
209 ICOPY_ELT(f1, tp2, i);
211 CCOPY_LIST(f2, tp2, b);
212 CCOPY_ELT(f1, tp2, i);
216 ICOPY_LIST(f1, tp2, b);
217 ICOPY_ELT(f2, tp2, i);
219 CCOPY_LIST(f1, tp2, b);
220 CCOPY_ELT(f2, tp2, i);
226 ICOPY_LIST(f2, tp2, l2);
228 CCOPY_LIST(f2, tp2, l2);
229 } else if (f1 < l1) {
231 ICOPY_LIST(f1, tp2, l1);
233 CCOPY_LIST(f1, tp2, l1);
237 tp2 = list1; /* swap list1, list2 */
240 last = list2 + nmemb*size;
243 memmove(list2, list1, nmemb*size);
250 #define swap(a, b) { \
254 tmp = *a; *a++ = *s; *s++ = tmp; \
258 #define reverse(bot, top) { \
263 tmp = *bot; *bot++ = *s; *s++ = tmp; \
270 * Optional hybrid natural/pairwise first pass. Eats up list1 in runs of
271 * increasing order, list2 in a corresponding linked list. Checks for runs
272 * when THRESHOLD/2 pairs compare with same sense. (Only used when NATURAL
273 * is defined. Otherwise simple pairwise merging is used.)
276 setup(u_char *list1, u_char *list2, size_t n, size_t size, cmp_t cmp)
278 int i, length, size2, tmp, sense;
279 u_char *f1, *f2, *s, *l2, *last, *p2;
283 insertionsort(list1, n, size, cmp);
284 *EVAL(list2) = (u_char*) list2 + n*size;
288 * Avoid running pointers out of bounds; limit n to evens
292 insertionsort(list1 + (n - i) * size, i, size, cmp);
293 last = list1 + size * (n - i);
294 *EVAL(list2 + (last - list1)) = list2 + n * size;
299 sense = (CMP(f1, f1 + size) > 0);
300 for (; f1 < last; sense = !sense) {
302 /* Find pairs with same sense. */
303 for (f2 = f1 + size2; f2 < last; f2 += size2) {
304 if ((CMP(f2, f2+ size) > 0) != sense)
308 if (length < THRESHOLD) { /* Pairwise merge */
310 p2 = *EVAL(p2) = f1 + size2 - list1 + list2;
312 swap (f1, f1 + size);
313 } while ((f1 += size2) < f2);
314 } else { /* Natural merge */
316 for (f2 = f1 + size2; f2 < l2; f2 += size2) {
317 if ((CMP(f2-size, f2) > 0) != sense) {
318 p2 = *EVAL(p2) = f2 - list1 + list2;
320 reverse(f1, f2-size);
325 reverse (f1, f2-size);
327 if (f2 < last || CMP(f2 - size, f2) > 0)
328 p2 = *EVAL(p2) = f2 - list1 + list2;
330 p2 = *EVAL(p2) = list2 + n*size;
333 #else /* pairwise merge only. */
334 for (f1 = list1, p2 = list2; f1 < last; f1 += size2) {
335 p2 = *EVAL(p2) = p2 + size2;
336 if (CMP (f1, f1 + size) > 0)
343 * This is to avoid out-of-bounds addresses in sorting the
347 insertionsort(u_char *a, size_t n, size_t size, cmp_t cmp)
349 u_char *ai, *s, *t, *u, tmp;
352 for (ai = a+size; --n >= 1; ai += size)
353 for (t = ai; t > a; t -= size) {