]> git.saurik.com Git - apple/libc.git/blob - stdlib.subproj/merge.c
Libc-167.tar.gz
[apple/libc.git] / stdlib.subproj / merge.c
1 /*
2 * Copyright (c) 1999 Apple Computer, Inc. All rights reserved.
3 *
4 * @APPLE_LICENSE_HEADER_START@
5 *
6 * The contents of this file constitute Original Code as defined in and
7 * are subject to the Apple Public Source License Version 1.1 (the
8 * "License"). You may not use this file except in compliance with the
9 * License. Please obtain a copy of the License at
10 * http://www.apple.com/publicsource and read it before using this file.
11 *
12 * This Original Code and all software distributed under the License are
13 * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
14 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
15 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the
17 * License for the specific language governing rights and limitations
18 * under the License.
19 *
20 * @APPLE_LICENSE_HEADER_END@
21 */
22 /*
23 * Copyright (c) 1992, 1993
24 * The Regents of the University of California. All rights reserved.
25 *
26 * This code is derived from software contributed to Berkeley by
27 * Peter McIlroy.
28 *
29 * Redistribution and use in source and binary forms, with or without
30 * modification, are permitted provided that the following conditions
31 * are met:
32 * 1. Redistributions of source code must retain the above copyright
33 * notice, this list of conditions and the following disclaimer.
34 * 2. Redistributions in binary form must reproduce the above copyright
35 * notice, this list of conditions and the following disclaimer in the
36 * documentation and/or other materials provided with the distribution.
37 * 3. All advertising materials mentioning features or use of this software
38 * must display the following acknowledgement:
39 * This product includes software developed by the University of
40 * California, Berkeley and its contributors.
41 * 4. Neither the name of the University nor the names of its contributors
42 * may be used to endorse or promote products derived from this software
43 * without specific prior written permission.
44 *
45 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
46 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
47 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
48 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
49 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
50 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
51 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
52 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
53 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
54 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
55 * SUCH DAMAGE.
56 */
57
58
59 /*
60 * Hybrid exponential search/linear search merge sort with hybrid
61 * natural/pairwise first pass. Requires about .3% more comparisons
62 * for random data than LSMS with pairwise first pass alone.
63 * It works for objects as small as two bytes.
64 */
65
66 #define NATURAL
67 #define THRESHOLD 16 /* Best choice for natural merge cut-off. */
68
69 /* #define NATURAL to get hybrid natural merge.
70 * (The default is pairwise merging.)
71 */
72
73 #include <sys/types.h>
74
75 #include <errno.h>
76 #include <stdlib.h>
77 #include <string.h>
78
79 static void setup __P((u_char *, u_char *, size_t, size_t, int (*)()));
80 static void insertionsort __P((u_char *, size_t, size_t, int (*)()));
81
82 #define ISIZE sizeof(int)
83 #define PSIZE sizeof(u_char *)
84 #define ICOPY_LIST(src, dst, last) \
85 do \
86 *(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE; \
87 while(src < last)
88 #define ICOPY_ELT(src, dst, i) \
89 do \
90 *(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE; \
91 while (i -= ISIZE)
92
93 #define CCOPY_LIST(src, dst, last) \
94 do \
95 *dst++ = *src++; \
96 while (src < last)
97 #define CCOPY_ELT(src, dst, i) \
98 do \
99 *dst++ = *src++; \
100 while (i -= 1)
101
102 /*
103 * Find the next possible pointer head. (Trickery for forcing an array
104 * to do double duty as a linked list when objects do not align with word
105 * boundaries.
106 */
107 /* Assumption: PSIZE is a power of 2. */
108 #define EVAL(p) (u_char **) \
109 ((u_char *)0 + \
110 (((u_char *)p + PSIZE - 1 - (u_char *) 0) & ~(PSIZE - 1)))
111
112 /*
113 * Arguments are as for qsort.
114 */
115 int
116 mergesort(base, nmemb, size, cmp)
117 void *base;
118 size_t nmemb;
119 register size_t size;
120 int (*cmp) __P((const void *, const void *));
121 {
122 register int i, sense;
123 int big, iflag;
124 register u_char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2;
125 u_char *list2, *list1, *p2, *p, *last, **p1;
126
127 if (size < PSIZE / 2) { /* Pointers must fit into 2 * size. */
128 errno = EINVAL;
129 return (-1);
130 }
131
132 /*
133 * XXX
134 * Stupid subtraction for the Cray.
135 */
136 iflag = 0;
137 if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE))
138 iflag = 1;
139
140 if ((list2 = malloc(nmemb * size + PSIZE)) == NULL)
141 return (-1);
142
143 list1 = base;
144 setup(list1, list2, nmemb, size, cmp);
145 last = list2 + nmemb * size;
146 i = big = 0;
147 while (*EVAL(list2) != last) {
148 l2 = list1;
149 p1 = EVAL(list1);
150 for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) {
151 p2 = *EVAL(p2);
152 f1 = l2;
153 f2 = l1 = list1 + (p2 - list2);
154 if (p2 != last)
155 p2 = *EVAL(p2);
156 l2 = list1 + (p2 - list2);
157 while (f1 < l1 && f2 < l2) {
158 if ((*cmp)(f1, f2) <= 0) {
159 q = f2;
160 b = f1, t = l1;
161 sense = -1;
162 } else {
163 q = f1;
164 b = f2, t = l2;
165 sense = 0;
166 }
167 if (!big) { /* here i = 0 */
168 LINEAR: while ((b += size) < t && cmp(q, b) >sense)
169 if (++i == 6) {
170 big = 1;
171 goto EXPONENTIAL;
172 }
173 } else {
174 EXPONENTIAL: for (i = size; ; i <<= 1)
175 if ((p = (b + i)) >= t) {
176 if ((p = t - size) > b &&
177 (*cmp)(q, p) <= sense)
178 t = p;
179 else
180 b = p;
181 break;
182 } else if ((*cmp)(q, p) <= sense) {
183 t = p;
184 if (i == size)
185 big = 0;
186 goto FASTCASE;
187 } else
188 b = p;
189 SLOWCASE: while (t > b+size) {
190 i = (((t - b) / size) >> 1) * size;
191 if ((*cmp)(q, p = b + i) <= sense)
192 t = p;
193 else
194 b = p;
195 }
196 goto COPY;
197 FASTCASE: while (i > size)
198 if ((*cmp)(q,
199 p = b + (i >>= 1)) <= sense)
200 t = p;
201 else
202 b = p;
203 COPY: b = t;
204 }
205 i = size;
206 if (q == f1) {
207 if (iflag) {
208 ICOPY_LIST(f2, tp2, b);
209 ICOPY_ELT(f1, tp2, i);
210 } else {
211 CCOPY_LIST(f2, tp2, b);
212 CCOPY_ELT(f1, tp2, i);
213 }
214 } else {
215 if (iflag) {
216 ICOPY_LIST(f1, tp2, b);
217 ICOPY_ELT(f2, tp2, i);
218 } else {
219 CCOPY_LIST(f1, tp2, b);
220 CCOPY_ELT(f2, tp2, i);
221 }
222 }
223 }
224 if (f2 < l2) {
225 if (iflag)
226 ICOPY_LIST(f2, tp2, l2);
227 else
228 CCOPY_LIST(f2, tp2, l2);
229 } else if (f1 < l1) {
230 if (iflag)
231 ICOPY_LIST(f1, tp2, l1);
232 else
233 CCOPY_LIST(f1, tp2, l1);
234 }
235 *p1 = l2;
236 }
237 tp2 = list1; /* swap list1, list2 */
238 list1 = list2;
239 list2 = tp2;
240 last = list2 + nmemb*size;
241 }
242 if (base == list2) {
243 memmove(list2, list1, nmemb*size);
244 list2 = list1;
245 }
246 free(list2);
247 return (0);
248 }
249
250 #define swap(a, b) { \
251 s = b; \
252 i = size; \
253 do { \
254 tmp = *a; *a++ = *s; *s++ = tmp; \
255 } while (--i); \
256 a -= size; \
257 }
258 #define reverse(bot, top) { \
259 s = top; \
260 do { \
261 i = size; \
262 do { \
263 tmp = *bot; *bot++ = *s; *s++ = tmp; \
264 } while (--i); \
265 s -= size2; \
266 } while(bot < s); \
267 }
268
269 /*
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.)
274 */
275 void
276 setup(list1, list2, n, size, cmp)
277 size_t n, size;
278 int (*cmp) __P((const void *, const void *));
279 u_char *list1, *list2;
280 {
281 int i, length, size2, tmp, sense;
282 u_char *f1, *f2, *s, *l2, *last, *p2;
283
284 size2 = size*2;
285 if (n <= 5) {
286 insertionsort(list1, n, size, cmp);
287 *EVAL(list2) = (u_char*) list2 + n*size;
288 return;
289 }
290 /*
291 * Avoid running pointers out of bounds; limit n to evens
292 * for simplicity.
293 */
294 i = 4 + (n & 1);
295 insertionsort(list1 + (n - i) * size, i, size, cmp);
296 last = list1 + size * (n - i);
297 *EVAL(list2 + (last - list1)) = list2 + n * size;
298
299 #ifdef NATURAL
300 p2 = list2;
301 f1 = list1;
302 sense = (cmp(f1, f1 + size) > 0);
303 for (; f1 < last; sense = !sense) {
304 length = 2;
305 /* Find pairs with same sense. */
306 for (f2 = f1 + size2; f2 < last; f2 += size2) {
307 if ((cmp(f2, f2+ size) > 0) != sense)
308 break;
309 length += 2;
310 }
311 if (length < THRESHOLD) { /* Pairwise merge */
312 do {
313 p2 = *EVAL(p2) = f1 + size2 - list1 + list2;
314 if (sense > 0)
315 swap (f1, f1 + size);
316 } while ((f1 += size2) < f2);
317 } else { /* Natural merge */
318 l2 = f2;
319 for (f2 = f1 + size2; f2 < l2; f2 += size2) {
320 if ((cmp(f2-size, f2) > 0) != sense) {
321 p2 = *EVAL(p2) = f2 - list1 + list2;
322 if (sense > 0)
323 reverse(f1, f2-size);
324 f1 = f2;
325 }
326 }
327 if (sense > 0)
328 reverse (f1, f2-size);
329 f1 = f2;
330 if (f2 < last || cmp(f2 - size, f2) > 0)
331 p2 = *EVAL(p2) = f2 - list1 + list2;
332 else
333 p2 = *EVAL(p2) = list2 + n*size;
334 }
335 }
336 #else /* pairwise merge only. */
337 for (f1 = list1, p2 = list2; f1 < last; f1 += size2) {
338 p2 = *EVAL(p2) = p2 + size2;
339 if (cmp (f1, f1 + size) > 0)
340 swap(f1, f1 + size);
341 }
342 #endif /* NATURAL */
343 }
344
345 /*
346 * This is to avoid out-of-bounds addresses in sorting the
347 * last 4 elements.
348 */
349 static void
350 insertionsort(a, n, size, cmp)
351 u_char *a;
352 size_t n, size;
353 int (*cmp) __P((const void *, const void *));
354 {
355 u_char *ai, *s, *t, *u, tmp;
356 int i;
357
358 for (ai = a+size; --n >= 1; ai += size)
359 for (t = ai; t > a; t -= size) {
360 u = t - size;
361 if (cmp(u, t) <= 0)
362 break;
363 swap(u, t);
364 }
365 }