1 /* $OpenBSD: merge.c,v 1.10 2015/06/21 03:20:56 millert Exp $ */
3 * Copyright (c) 1992, 1993
4 * The Regents of the University of California. All rights reserved.
6 * This code is derived from software contributed to Berkeley by
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * Hybrid exponential search/linear search merge sort with hybrid
36 * natural/pairwise first pass. Requires about .3% more comparisons
37 * for random data than LSMS with pairwise first pass alone.
38 * It works for objects as small as two bytes.
42 #define THRESHOLD 16 /* Best choice for natural merge cut-off. */
44 /* #define NATURAL to get hybrid natural merge.
45 * (The default is pairwise merging.)
48 #include <sys/types.h>
54 static void setup(u_char *list1, u_char *list2, size_t n, size_t size,
55 int (*cmp)(const void *, const void *));
56 static void insertionsort(u_char *a, size_t n, size_t size,
57 int (*cmp)(const void *, const void *));
59 #define ISIZE sizeof(int)
60 #define PSIZE sizeof(u_char *)
61 #define ICOPY_LIST(src, dst, last) \
63 *(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE; \
65 #define ICOPY_ELT(src, dst, i) \
67 *(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE; \
70 #define CCOPY_LIST(src, dst, last) \
74 #define CCOPY_ELT(src, dst, i) \
80 * Find the next possible pointer head. (Trickery for forcing an array
81 * to do double duty as a linked list when objects do not align with word
84 /* Assumption: PSIZE is a power of 2. */
85 #define EVAL(p) (u_char **) \
87 (((u_char *)p + PSIZE - 1 - (u_char *) 0) & ~(PSIZE - 1)))
90 * Arguments are as for qsort.
93 mergesort(void *base, size_t nmemb, size_t size,
94 int (*cmp)(const void *, const void *))
98 u_char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2;
99 u_char *list2, *list1, *p2, *p, *last, **p1;
101 if (size < PSIZE / 2) { /* Pointers must fit into 2 * size. */
111 * Stupid subtraction for the Cray.
114 if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE))
117 if ((list2 = malloc(nmemb * size + PSIZE)) == NULL)
121 setup(list1, list2, nmemb, size, cmp);
122 last = list2 + nmemb * size;
124 while (*EVAL(list2) != last) {
127 for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) {
130 f2 = l1 = list1 + (p2 - list2);
133 l2 = list1 + (p2 - list2);
134 while (f1 < l1 && f2 < l2) {
135 if ((*cmp)(f1, f2) <= 0) {
144 if (!big) { /* here i = 0 */
145 while ((b += size) < t && cmp(q, b) >sense)
151 EXPONENTIAL: for (i = size; ; i <<= 1)
152 if ((p = (b + i)) >= t) {
153 if ((p = t - size) > b &&
154 (*cmp)(q, p) <= sense)
159 } else if ((*cmp)(q, p) <= sense) {
167 i = (((t - b) / size) >> 1) * size;
168 if ((*cmp)(q, p = b + i) <= sense)
174 FASTCASE: while (i > size)
176 p = b + (i >>= 1)) <= sense)
185 ICOPY_LIST(f2, tp2, b);
186 ICOPY_ELT(f1, tp2, i);
188 CCOPY_LIST(f2, tp2, b);
189 CCOPY_ELT(f1, tp2, i);
193 ICOPY_LIST(f1, tp2, b);
194 ICOPY_ELT(f2, tp2, i);
196 CCOPY_LIST(f1, tp2, b);
197 CCOPY_ELT(f2, tp2, i);
203 ICOPY_LIST(f2, tp2, l2);
205 CCOPY_LIST(f2, tp2, l2);
206 } else if (f1 < l1) {
208 ICOPY_LIST(f1, tp2, l1);
210 CCOPY_LIST(f1, tp2, l1);
214 tp2 = list1; /* swap list1, list2 */
217 last = list2 + nmemb*size;
220 memmove(list2, list1, nmemb*size);
227 #define swap(a, b) { \
231 tmp = *a; *a++ = *s; *s++ = tmp; \
235 #define reverse(bot, top) { \
240 tmp = *bot; *bot++ = *s; *s++ = tmp; \
247 * Optional hybrid natural/pairwise first pass. Eats up list1 in runs of
248 * increasing order, list2 in a corresponding linked list. Checks for runs
249 * when THRESHOLD/2 pairs compare with same sense. (Only used when NATURAL
250 * is defined. Otherwise simple pairwise merging is used.)
253 setup(u_char *list1, u_char *list2, size_t n, size_t size,
254 int (*cmp)(const void *, const void *))
256 int i, length, size2, sense;
257 u_char tmp, *f1, *f2, *s, *l2, *last, *p2;
261 insertionsort(list1, n, size, cmp);
262 *EVAL(list2) = (u_char*) list2 + n*size;
266 * Avoid running pointers out of bounds; limit n to evens
270 insertionsort(list1 + (n - i) * size, i, size, cmp);
271 last = list1 + size * (n - i);
272 *EVAL(list2 + (last - list1)) = list2 + n * size;
277 sense = (cmp(f1, f1 + size) > 0);
278 for (; f1 < last; sense = !sense) {
280 /* Find pairs with same sense. */
281 for (f2 = f1 + size2; f2 < last; f2 += size2) {
282 if ((cmp(f2, f2+ size) > 0) != sense)
286 if (length < THRESHOLD) { /* Pairwise merge */
288 p2 = *EVAL(p2) = f1 + size2 - list1 + list2;
290 swap (f1, f1 + size);
291 } while ((f1 += size2) < f2);
292 } else { /* Natural merge */
294 for (f2 = f1 + size2; f2 < l2; f2 += size2) {
295 if ((cmp(f2-size, f2) > 0) != sense) {
296 p2 = *EVAL(p2) = f2 - list1 + list2;
298 reverse(f1, f2-size);
303 reverse (f1, f2-size);
305 if (f2 < last || cmp(f2 - size, f2) > 0)
306 p2 = *EVAL(p2) = f2 - list1 + list2;
308 p2 = *EVAL(p2) = list2 + n*size;
311 #else /* pairwise merge only. */
312 for (f1 = list1, p2 = list2; f1 < last; f1 += size2) {
313 p2 = *EVAL(p2) = p2 + size2;
314 if (cmp (f1, f1 + size) > 0)
321 * This is to avoid out-of-bounds addresses in sorting the
325 insertionsort(u_char *a, size_t n, size_t size,
326 int (*cmp)(const void *, const void *))
328 u_char *ai, *s, *t, *u, tmp;
331 for (ai = a+size; --n >= 1; ai += size)
332 for (t = ai; t > a; t -= size) {