xref: /haiku/src/system/libroot/posix/stdlib/merge.c (revision 1294543de9ac0eff000eaea1b18368c36435d08e)
1 /*-
2  * Copyright (c) 1992, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Peter McIlroy.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
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.
23  *
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
34  * SUCH DAMAGE.
35  */
36 
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 
41 /*
42  * Hybrid exponential search/linear search merge sort with hybrid
43  * natural/pairwise first pass.  Requires about .3% more comparisons
44  * for random data than LSMS with pairwise first pass alone.
45  * It works for objects as small as two bytes.
46  */
47 
48 #define NATURAL
49 #define THRESHOLD 16	/* Best choice for natural merge cut-off. */
50 
51 /* #define NATURAL to get hybrid natural merge.
52  * (The default is pairwise merging.)
53  */
54 
55 #include <sys/types.h>
56 
57 #include <errno.h>
58 #include <stdlib.h>
59 #include <string.h>
60 
61 static void setup(u_char *, u_char *, size_t, size_t, int (*)());
62 static void insertionsort(u_char *, size_t, size_t, int (*)());
63 
64 #define ISIZE sizeof(int)
65 #define PSIZE sizeof(u_char *)
66 #define ICOPY_LIST(src, dst, last)				\
67 	do							\
68 	*(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE;	\
69 	while(src < last)
70 #define ICOPY_ELT(src, dst, i)					\
71 	do							\
72 	*(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE;	\
73 	while (i -= ISIZE)
74 
75 #define CCOPY_LIST(src, dst, last)		\
76 	do					\
77 		*dst++ = *src++;		\
78 	while (src < last)
79 #define CCOPY_ELT(src, dst, i)			\
80 	do					\
81 		*dst++ = *src++;		\
82 	while (i -= 1)
83 
84 /*
85  * Find the next possible pointer head.  (Trickery for forcing an array
86  * to do double duty as a linked list when objects do not align with word
87  * boundaries.
88  */
89 /* Assumption: PSIZE is a power of 2. */
90 #define EVAL(p) (u_char **)						\
91 	((u_char *)0 +							\
92 	    (((u_char *)p + PSIZE - 1 - (u_char *) 0) & ~(PSIZE - 1)))
93 
94 /*
95  * Arguments are as for qsort.
96  */
97 int
98 mergesort(void *base, size_t nmemb, size_t size, int (*cmp)(void const *, void const *))
99 {
100 	size_t i;
101 	int sense;
102 	int big, iflag;
103 	u_char *f1;
104 	u_char *f2;
105 	u_char *t;
106 	u_char *b;
107 	u_char *tp2;
108 	u_char *q;
109 	u_char *l1;
110 	u_char *l2;
111 	u_char *list2;
112 	u_char *list1;
113 	u_char *p2;
114 	u_char *p;
115 	u_char *last;
116 	u_char **p1;
117 
118 	if (size < PSIZE / 2) {		/* Pointers must fit into 2 * size. */
119 //		errno = EINVAL;
120 		return (-1);
121 	}
122 
123 	if (nmemb == 0) {
124 		return (0);
125 	}
126 
127 	/*
128 	 * XXX
129 	 * Stupid subtraction for the Cray.
130 	 */
131 	iflag = 0;
132 	if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE))
133 		iflag = 1;
134 
135 	if ((list2 = malloc(nmemb * size + PSIZE)) == NULL)
136 		return (-1);
137 
138 	list1 = base;
139 	setup(list1, list2, nmemb, size, cmp);
140 	last = list2 + nmemb * size;
141 	i = big = 0;
142 	while (*EVAL(list2) != last) {
143 	    l2 = list1;
144 	    p1 = EVAL(list1);
145 	    for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) {
146 	    	p2 = *EVAL(p2);
147 	    	f1 = l2;
148 	    	f2 = l1 = list1 + (p2 - list2);
149 	    	if (p2 != last)
150 	    		p2 = *EVAL(p2);
151 	    	l2 = list1 + (p2 - list2);
152 	    	while (f1 < l1 && f2 < l2) {
153 	    		if ((*cmp)(f1, f2) <= 0) {
154 	    			q = f2;
155 	    			b = f1, t = l1;
156 	    			sense = -1;
157 	    		} else {
158 	    			q = f1;
159 	    			b = f2, t = l2;
160 	    			sense = 0;
161 	    		}
162 	    		if (!big) {	/* here i = 0 */
163 				while ((b += size) < t && cmp(q, b) >sense)
164 	    				if (++i == 6) {
165 	    					big = 1;
166 	    					goto EXPONENTIAL;
167 	    				}
168 	    		} else {
169 EXPONENTIAL:	    		for (i = size; ; i <<= 1)
170 	    				if ((p = (b + i)) >= t) {
171 	    					if ((p = t - size) > b &&
172 						    (*cmp)(q, p) <= sense)
173 	    						t = p;
174 	    					else
175 	    						b = p;
176 	    					break;
177 	    				} else if ((*cmp)(q, p) <= sense) {
178 	    					t = p;
179 	    					if (i == size)
180 	    						big = 0;
181 	    					goto FASTCASE;
182 	    				} else
183 	    					b = p;
184 				while (t > b+size) {
185 	    				i = (((t - b) / size) >> 1) * size;
186 	    				if ((*cmp)(q, p = b + i) <= sense)
187 	    					t = p;
188 	    				else
189 	    					b = p;
190 	    			}
191 	    			goto COPY;
192 FASTCASE:	    		while (i > size)
193 	    				if ((*cmp)(q,
194 	    					p = b + (i >>= 1)) <= sense)
195 	    					t = p;
196 	    				else
197 	    					b = p;
198 COPY:	    			b = t;
199 	    		}
200 	    		i = size;
201 	    		if (q == f1) {
202 	    			if (iflag) {
203 	    				ICOPY_LIST(f2, tp2, b);
204 	    				ICOPY_ELT(f1, tp2, i);
205 	    			} else {
206 	    				CCOPY_LIST(f2, tp2, b);
207 	    				CCOPY_ELT(f1, tp2, i);
208 	    			}
209 	    		} else {
210 	    			if (iflag) {
211 	    				ICOPY_LIST(f1, tp2, b);
212 	    				ICOPY_ELT(f2, tp2, i);
213 	    			} else {
214 	    				CCOPY_LIST(f1, tp2, b);
215 	    				CCOPY_ELT(f2, tp2, i);
216 	    			}
217 	    		}
218 	    	}
219 	    	if (f2 < l2) {
220 	    		if (iflag)
221 	    			ICOPY_LIST(f2, tp2, l2);
222 	    		else
223 	    			CCOPY_LIST(f2, tp2, l2);
224 	    	} else if (f1 < l1) {
225 	    		if (iflag)
226 	    			ICOPY_LIST(f1, tp2, l1);
227 	    		else
228 	    			CCOPY_LIST(f1, tp2, l1);
229 	    	}
230 	    	*p1 = l2;
231 	    }
232 	    tp2 = list1;	/* swap list1, list2 */
233 	    list1 = list2;
234 	    list2 = tp2;
235 	    last = list2 + nmemb*size;
236 	}
237 	if (base == list2) {
238 		memmove(list2, list1, nmemb*size);
239 		list2 = list1;
240 	}
241 	free(list2);
242 	return (0);
243 }
244 
245 #define	swap(a, b) {					\
246 		s = b;					\
247 		i = size;				\
248 		do {					\
249 			tmp = *a; *a++ = *s; *s++ = tmp; \
250 		} while (--i);				\
251 		a -= size;				\
252 	}
253 #define reverse(bot, top) {				\
254 	s = top;					\
255 	do {						\
256 		i = size;				\
257 		do {					\
258 			tmp = *bot; *bot++ = *s; *s++ = tmp; \
259 		} while (--i);				\
260 		s -= size2;				\
261 	} while(bot < s);				\
262 }
263 
264 /*
265  * Optional hybrid natural/pairwise first pass.  Eats up list1 in runs of
266  * increasing order, list2 in a corresponding linked list.  Checks for runs
267  * when THRESHOLD/2 pairs compare with same sense.  (Only used when NATURAL
268  * is defined.  Otherwise simple pairwise merging is used.)
269  */
270 static
271 void
272 setup(u_char *list1, u_char *list2, size_t n, size_t size, int (*cmp)(void const *, void const *))
273 {
274 	int i, length, size2, tmp, sense;
275 	u_char *f1, *f2, *s, *l2, *last, *p2;
276 
277 	size2 = size*2;
278 	if (n <= 5) {
279 		insertionsort(list1, n, size, cmp);
280 		*EVAL(list2) = (u_char*) list2 + n*size;
281 		return;
282 	}
283 	/*
284 	 * Avoid running pointers out of bounds; limit n to evens
285 	 * for simplicity.
286 	 */
287 	i = 4 + (n & 1);
288 	insertionsort(list1 + (n - i) * size, i, size, cmp);
289 	last = list1 + size * (n - i);
290 	*EVAL(list2 + (last - list1)) = list2 + n * size;
291 
292 #ifdef NATURAL
293 	p2 = list2;
294 	f1 = list1;
295 	sense = (cmp(f1, f1 + size) > 0);
296 	for (; f1 < last; sense = !sense) {
297 		length = 2;
298 					/* Find pairs with same sense. */
299 		for (f2 = f1 + size2; f2 < last; f2 += size2) {
300 			if ((cmp(f2, f2+ size) > 0) != sense)
301 				break;
302 			length += 2;
303 		}
304 		if (length < THRESHOLD) {		/* Pairwise merge */
305 			do {
306 				p2 = *EVAL(p2) = f1 + size2 - list1 + list2;
307 				if (sense > 0)
308 					swap (f1, f1 + size);
309 			} while ((f1 += size2) < f2);
310 		} else {				/* Natural merge */
311 			l2 = f2;
312 			for (f2 = f1 + size2; f2 < l2; f2 += size2) {
313 				if ((cmp(f2-size, f2) > 0) != sense) {
314 					p2 = *EVAL(p2) = f2 - list1 + list2;
315 					if (sense > 0)
316 						reverse(f1, f2-size);
317 					f1 = f2;
318 				}
319 			}
320 			if (sense > 0)
321 				reverse (f1, f2-size);
322 			f1 = f2;
323 			if (f2 < last || cmp(f2 - size, f2) > 0)
324 				p2 = *EVAL(p2) = f2 - list1 + list2;
325 			else
326 				p2 = *EVAL(p2) = list2 + n*size;
327 		}
328 	}
329 #else		/* pairwise merge only. */
330 	for (f1 = list1, p2 = list2; f1 < last; f1 += size2) {
331 		p2 = *EVAL(p2) = p2 + size2;
332 		if (cmp (f1, f1 + size) > 0)
333 			swap(f1, f1 + size);
334 	}
335 #endif /* NATURAL */
336 }
337 
338 /*
339  * This is to avoid out-of-bounds addresses in sorting the
340  * last 4 elements.
341  */
342 static
343 void
344 insertionsort(u_char *a, size_t n,size_t  size, int (*cmp)(const void *, const void *))
345 {
346 	u_char *ai;
347 	u_char *s;
348 	u_char *t;
349 	u_char *u;
350 	u_char tmp;
351 	int i;
352 
353 	for (ai = a+size; --n >= 1; ai += size) {
354 		for (t = ai; t > a; t -= size) {
355 			u = t - size;
356 			if (cmp(u, t) <= 0)
357 				break;
358 			swap(u, t);
359 		}
360 	}
361 }
362