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  * 4. 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.
19  *
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
30  * SUCH DAMAGE.
31  */
32 
33 #include <sys/cdefs.h>
34 
35 /*
36  * Hybrid exponential search/linear search merge sort with hybrid
37  * natural/pairwise first pass.  Requires about .3% more comparisons
38  * for random data than LSMS with pairwise first pass alone.
39  * It works for objects as small as two bytes.
40  */
41 
42 #define NATURAL
43 #define THRESHOLD 16	/* Best choice for natural merge cut-off. */
44 
45 /* #define NATURAL to get hybrid natural merge.
46  * (The default is pairwise merging.)
47  */
48 
49 #include <sys/types.h>
50 
51 #include <errno.h>
52 #include <stdlib.h>
53 #include <string.h>
54 
55 #include "got_compat.h"
56 
57 static void setup(unsigned char *, unsigned char *, size_t, size_t,
58     int (*)(const void *, const void *));
59 static void insertionsort(unsigned char *, size_t, size_t,
60     int (*)(const void *, const void *));
61 
62 #define ISIZE sizeof(int)
63 #define PSIZE sizeof(unsigned char *)
64 #define ICOPY_LIST(src, dst, last)				\
65 	do							\
66 	*(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE;	\
67 	while(src < last)
68 #define ICOPY_ELT(src, dst, i)					\
69 	do							\
70 	*(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE;	\
71 	while (i -= ISIZE)
72 
73 #define CCOPY_LIST(src, dst, last)		\
74 	do					\
75 		*dst++ = *src++;		\
76 	while (src < last)
77 #define CCOPY_ELT(src, dst, i)			\
78 	do					\
79 		*dst++ = *src++;		\
80 	while (i -= 1)
81 
82 /*
83  * Find the next possible pointer head.  (Trickery for forcing an array
84  * to do double duty as a linked list when objects do not align with word
85  * boundaries.
86  */
87 /* Assumption: PSIZE is a power of 2. */
88 #define EVAL(p) (unsigned char **)					\
89 	((unsigned char *)0 +						\
90 	 (((unsigned char *)p + PSIZE - 1 -				\
91 	   (unsigned char *)0) & ~(PSIZE - 1)))
92 
93 /*
94  * Arguments are as for qsort.
95  */
96 int
97 mergesort(void *base, size_t nmemb, size_t size,
98 	int (*cmp)(const void *, const void *))
99 {
100 	size_t i;
101 	int sense;
102 	int big, iflag;
103 	unsigned char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2;
104 	unsigned char *list2, *list1, *p2, *p, *last, **p1;
105 
106 	if (size < PSIZE / 2) {		/* Pointers must fit into 2 * size. */
107 		errno = EINVAL;
108 		return (-1);
109 	}
110 
111 	if (nmemb == 0)
112 		return (0);
113 
114 	/*
115 	 * XXX
116 	 * Stupid subtraction for the Cray.
117 	 */
118 	iflag = 0;
119 	if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE))
120 		iflag = 1;
121 
122 	if ((list2 = malloc(nmemb * size + PSIZE)) == NULL)
123 		return (-1);
124 
125 	list1 = base;
126 	setup(list1, list2, nmemb, size, cmp);
127 	last = list2 + nmemb * size;
128 	i = big = 0;
129 	while (*EVAL(list2) != last) {
130 	    l2 = list1;
131 	    p1 = EVAL(list1);
132 	    for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) {
133 	    	p2 = *EVAL(p2);
134 	    	f1 = l2;
135 	    	f2 = l1 = list1 + (p2 - list2);
136 	    	if (p2 != last)
137 	    		p2 = *EVAL(p2);
138 	    	l2 = list1 + (p2 - list2);
139 	    	while (f1 < l1 && f2 < l2) {
140 	    		if ((*cmp)(f1, f2) <= 0) {
141 	    			q = f2;
142 	    			b = f1, t = l1;
143 	    			sense = -1;
144 	    		} else {
145 	    			q = f1;
146 	    			b = f2, t = l2;
147 	    			sense = 0;
148 	    		}
149 	    		if (!big) {	/* here i = 0 */
150 				while ((b += size) < t && cmp(q, b) >sense)
151 	    				if (++i == 6) {
152 	    					big = 1;
153 	    					goto EXPONENTIAL;
154 	    				}
155 	    		} else {
156 EXPONENTIAL:	    		for (i = size; ; i <<= 1)
157 	    				if ((p = (b + i)) >= t) {
158 	    					if ((p = t - size) > b &&
159 						    (*cmp)(q, p) <= sense)
160 	    						t = p;
161 	    					else
162 	    						b = p;
163 	    					break;
164 	    				} else if ((*cmp)(q, p) <= sense) {
165 	    					t = p;
166 	    					if (i == size)
167 	    						big = 0;
168 	    					goto FASTCASE;
169 	    				} else
170 	    					b = p;
171 				while (t > b+size) {
172 	    				i = (((t - b) / size) >> 1) * size;
173 	    				if ((*cmp)(q, p = b + i) <= sense)
174 	    					t = p;
175 	    				else
176 	    					b = p;
177 	    			}
178 	    			goto COPY;
179 FASTCASE:	    		while (i > size)
180 	    				if ((*cmp)(q,
181 	    					p = b + (i >>= 1)) <= sense)
182 	    					t = p;
183 	    				else
184 	    					b = p;
185 COPY:	    			b = t;
186 	    		}
187 	    		i = size;
188 	    		if (q == f1) {
189 	    			if (iflag) {
190 	    				ICOPY_LIST(f2, tp2, b);
191 	    				ICOPY_ELT(f1, tp2, i);
192 	    			} else {
193 	    				CCOPY_LIST(f2, tp2, b);
194 	    				CCOPY_ELT(f1, tp2, i);
195 	    			}
196 	    		} else {
197 	    			if (iflag) {
198 	    				ICOPY_LIST(f1, tp2, b);
199 	    				ICOPY_ELT(f2, tp2, i);
200 	    			} else {
201 	    				CCOPY_LIST(f1, tp2, b);
202 	    				CCOPY_ELT(f2, tp2, i);
203 	    			}
204 	    		}
205 	    	}
206 	    	if (f2 < l2) {
207 	    		if (iflag)
208 	    			ICOPY_LIST(f2, tp2, l2);
209 	    		else
210 	    			CCOPY_LIST(f2, tp2, l2);
211 	    	} else if (f1 < l1) {
212 	    		if (iflag)
213 	    			ICOPY_LIST(f1, tp2, l1);
214 	    		else
215 	    			CCOPY_LIST(f1, tp2, l1);
216 	    	}
217 	    	*p1 = l2;
218 	    }
219 	    tp2 = list1;	/* swap list1, list2 */
220 	    list1 = list2;
221 	    list2 = tp2;
222 	    last = list2 + nmemb*size;
223 	}
224 	if (base == list2) {
225 		memmove(list2, list1, nmemb*size);
226 		list2 = list1;
227 	}
228 	free(list2);
229 	return (0);
230 }
231 
232 #define	swap(a, b) {					\
233 		s = b;					\
234 		i = size;				\
235 		do {					\
236 			tmp = *a; *a++ = *s; *s++ = tmp; \
237 		} while (--i);				\
238 		a -= size;				\
239 	}
240 #define reverse(bot, top) {				\
241 	s = top;					\
242 	do {						\
243 		i = size;				\
244 		do {					\
245 			tmp = *bot; *bot++ = *s; *s++ = tmp; \
246 		} while (--i);				\
247 		s -= size2;				\
248 	} while(bot < s);				\
249 }
250 
251 /*
252  * Optional hybrid natural/pairwise first pass.  Eats up list1 in runs of
253  * increasing order, list2 in a corresponding linked list.  Checks for runs
254  * when THRESHOLD/2 pairs compare with same sense.  (Only used when NATURAL
255  * is defined.  Otherwise simple pairwise merging is used.)
256  */
257 static void
258 setup(unsigned char *list1, unsigned char *list2, size_t n, size_t size,
259 	int (*cmp)(const void *, const void *))
260 {
261 	int i, length, size2, tmp, sense;
262 	unsigned char *f1, *f2, *s, *l2, *last, *p2;
263 
264 	size2 = size*2;
265 	if (n <= 5) {
266 		insertionsort(list1, n, size, cmp);
267 		*EVAL(list2) = (unsigned char*) list2 + n*size;
268 		return;
269 	}
270 	/*
271 	 * Avoid running pointers out of bounds; limit n to evens
272 	 * for simplicity.
273 	 */
274 	i = 4 + (n & 1);
275 	insertionsort(list1 + (n - i) * size, i, size, cmp);
276 	last = list1 + size * (n - i);
277 	*EVAL(list2 + (last - list1)) = list2 + n * size;
278 
279 #ifdef NATURAL
280 	p2 = list2;
281 	f1 = list1;
282 	sense = (cmp(f1, f1 + size) > 0);
283 	for (; f1 < last; sense = !sense) {
284 		length = 2;
285 					/* Find pairs with same sense. */
286 		for (f2 = f1 + size2; f2 < last; f2 += size2) {
287 			if ((cmp(f2, f2+ size) > 0) != sense)
288 				break;
289 			length += 2;
290 		}
291 		if (length < THRESHOLD) {		/* Pairwise merge */
292 			do {
293 				p2 = *EVAL(p2) = f1 + size2 - list1 + list2;
294 				if (sense > 0)
295 					swap (f1, f1 + size);
296 			} while ((f1 += size2) < f2);
297 		} else {				/* Natural merge */
298 			l2 = f2;
299 			for (f2 = f1 + size2; f2 < l2; f2 += size2) {
300 				if ((cmp(f2-size, f2) > 0) != sense) {
301 					p2 = *EVAL(p2) = f2 - list1 + list2;
302 					if (sense > 0)
303 						reverse(f1, f2-size);
304 					f1 = f2;
305 				}
306 			}
307 			if (sense > 0)
308 				reverse (f1, f2-size);
309 			f1 = f2;
310 			if (f2 < last || cmp(f2 - size, f2) > 0)
311 				p2 = *EVAL(p2) = f2 - list1 + list2;
312 			else
313 				p2 = *EVAL(p2) = list2 + n*size;
314 		}
315 	}
316 #else		/* pairwise merge only. */
317 	for (f1 = list1, p2 = list2; f1 < last; f1 += size2) {
318 		p2 = *EVAL(p2) = p2 + size2;
319 		if (cmp (f1, f1 + size) > 0)
320 			swap(f1, f1 + size);
321 	}
322 #endif /* NATURAL */
323 }
324 
325 /*
326  * This is to avoid out-of-bounds addresses in sorting the
327  * last 4 elements.
328  */
329 static void
330 insertionsort(unsigned char *a, size_t n, size_t size,
331 	int (*cmp)(const void *, const void *))
332 {
333 	unsigned char *ai, *s, *t, *u, tmp;
334 	int i;
335 
336 	for (ai = a+size; --n >= 1; ai += size)
337 		for (t = ai; t > a; t -= size) {
338 			u = t - size;
339 			if (cmp(u, t) <= 0)
340 				break;
341 			swap(u, t);
342 		}
343 }