xref: /freebsd/lib/libc/stdlib/merge.c (revision 3c4ba5f55438f7afd4f4b0b56f88f2bb505fd6a6)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1992, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Peter McIlroy.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 
35 #if defined(LIBC_SCCS) && !defined(lint)
36 static char sccsid[] = "@(#)merge.c	8.2 (Berkeley) 2/14/94";
37 #endif /* LIBC_SCCS and not lint */
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
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/param.h>
56 
57 #include <errno.h>
58 #include <stdlib.h>
59 #include <string.h>
60 
61 #ifdef I_AM_MERGESORT_B
62 #include "block_abi.h"
63 #define	DECLARE_CMP	DECLARE_BLOCK(int, cmp, const void *, const void *)
64 typedef DECLARE_BLOCK(int, cmp_t, const void *, const void *);
65 #define	CMP(x, y)	CALL_BLOCK(cmp, x, y)
66 #else
67 typedef int (*cmp_t)(const void *, const void *);
68 #define	CMP(x, y)	cmp(x, y)
69 #endif
70 
71 static void setup(u_char *, u_char *, size_t, size_t, cmp_t);
72 static void insertionsort(u_char *, size_t, size_t, cmp_t);
73 
74 #define ISIZE sizeof(int)
75 #define PSIZE sizeof(u_char *)
76 #define ICOPY_LIST(src, dst, last)				\
77 	do							\
78 	*(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE;	\
79 	while(src < last)
80 #define ICOPY_ELT(src, dst, i)					\
81 	do							\
82 	*(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE;	\
83 	while (i -= ISIZE)
84 
85 #define CCOPY_LIST(src, dst, last)		\
86 	do					\
87 		*dst++ = *src++;		\
88 	while (src < last)
89 #define CCOPY_ELT(src, dst, i)			\
90 	do					\
91 		*dst++ = *src++;		\
92 	while (i -= 1)
93 
94 /*
95  * Find the next possible pointer head.  (Trickery for forcing an array
96  * to do double duty as a linked list when objects do not align with word
97  * boundaries.
98  */
99 /* Assumption: PSIZE is a power of 2. */
100 #define EVAL(p) (u_char **)roundup2((uintptr_t)p, PSIZE)
101 
102 #ifdef I_AM_MERGESORT_B
103 int mergesort_b(void *, size_t, size_t, cmp_t);
104 #else
105 int mergesort(void *, size_t, size_t, cmp_t);
106 #endif
107 
108 /*
109  * Arguments are as for qsort.
110  */
111 int
112 #ifdef I_AM_MERGESORT_B
113 mergesort_b(void *base, size_t nmemb, size_t size, cmp_t cmp)
114 #else
115 mergesort(void *base, size_t nmemb, size_t size, cmp_t cmp)
116 #endif
117 {
118 	size_t i;
119 	int sense;
120 	int big, iflag;
121 	u_char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2;
122 	u_char *list2, *list1, *p2, *p, *last, **p1;
123 
124 	if (size < PSIZE / 2) {		/* Pointers must fit into 2 * size. */
125 		errno = EINVAL;
126 		return (-1);
127 	}
128 
129 	if (nmemb == 0)
130 		return (0);
131 
132 	iflag = 0;
133 	if (__is_aligned(size, ISIZE) && __is_aligned(base, ISIZE))
134 		iflag = 1;
135 
136 	if ((list2 = malloc(nmemb * size + PSIZE)) == NULL)
137 		return (-1);
138 
139 	list1 = base;
140 	setup(list1, list2, nmemb, size, cmp);
141 	last = list2 + nmemb * size;
142 	i = big = 0;
143 	while (*EVAL(list2) != last) {
144 	    l2 = list1;
145 	    p1 = EVAL(list1);
146 	    for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) {
147 	    	p2 = *EVAL(p2);
148 	    	f1 = l2;
149 	    	f2 = l1 = list1 + (p2 - list2);
150 	    	if (p2 != last)
151 	    		p2 = *EVAL(p2);
152 	    	l2 = list1 + (p2 - list2);
153 	    	while (f1 < l1 && f2 < l2) {
154 	    		if (CMP(f1, f2) <= 0) {
155 	    			q = f2;
156 	    			b = f1, t = l1;
157 	    			sense = -1;
158 	    		} else {
159 	    			q = f1;
160 	    			b = f2, t = l2;
161 	    			sense = 0;
162 	    		}
163 	    		if (!big) {	/* here i = 0 */
164 				while ((b += size) < t && CMP(q, b) >sense)
165 	    				if (++i == 6) {
166 	    					big = 1;
167 	    					goto EXPONENTIAL;
168 	    				}
169 	    		} else {
170 EXPONENTIAL:	    		for (i = size; ; i <<= 1)
171 	    				if ((p = (b + i)) >= t) {
172 	    					if ((p = t - size) > b &&
173 						    CMP(q, p) <= sense)
174 	    						t = p;
175 	    					else
176 	    						b = p;
177 	    					break;
178 	    				} else if (CMP(q, p) <= sense) {
179 	    					t = p;
180 	    					if (i == size)
181 	    						big = 0;
182 	    					goto FASTCASE;
183 	    				} else
184 	    					b = p;
185 				while (t > b+size) {
186 	    				i = (((t - b) / size) >> 1) * size;
187 	    				if (CMP(q, p = b + i) <= sense)
188 	    					t = p;
189 	    				else
190 	    					b = p;
191 	    			}
192 	    			goto COPY;
193 FASTCASE:	    		while (i > size)
194 	    				if (CMP(q,
195 	    					p = b + (i >>= 1)) <= sense)
196 	    					t = p;
197 	    				else
198 	    					b = p;
199 COPY:	    			b = t;
200 	    		}
201 	    		i = size;
202 	    		if (q == f1) {
203 	    			if (iflag) {
204 	    				ICOPY_LIST(f2, tp2, b);
205 	    				ICOPY_ELT(f1, tp2, i);
206 	    			} else {
207 	    				CCOPY_LIST(f2, tp2, b);
208 	    				CCOPY_ELT(f1, tp2, i);
209 	    			}
210 	    		} else {
211 	    			if (iflag) {
212 	    				ICOPY_LIST(f1, tp2, b);
213 	    				ICOPY_ELT(f2, tp2, i);
214 	    			} else {
215 	    				CCOPY_LIST(f1, tp2, b);
216 	    				CCOPY_ELT(f2, tp2, i);
217 	    			}
218 	    		}
219 	    	}
220 	    	if (f2 < l2) {
221 	    		if (iflag)
222 	    			ICOPY_LIST(f2, tp2, l2);
223 	    		else
224 	    			CCOPY_LIST(f2, tp2, l2);
225 	    	} else if (f1 < l1) {
226 	    		if (iflag)
227 	    			ICOPY_LIST(f1, tp2, l1);
228 	    		else
229 	    			CCOPY_LIST(f1, tp2, l1);
230 	    	}
231 	    	*p1 = l2;
232 	    }
233 	    tp2 = list1;	/* swap list1, list2 */
234 	    list1 = list2;
235 	    list2 = tp2;
236 	    last = list2 + nmemb*size;
237 	}
238 	if (base == list2) {
239 		memmove(list2, list1, nmemb*size);
240 		list2 = list1;
241 	}
242 	free(list2);
243 	return (0);
244 }
245 
246 #define	swap(a, b) {					\
247 		s = b;					\
248 		i = size;				\
249 		do {					\
250 			tmp = *a; *a++ = *s; *s++ = tmp; \
251 		} while (--i);				\
252 		a -= size;				\
253 	}
254 #define reverse(bot, top) {				\
255 	s = top;					\
256 	do {						\
257 		i = size;				\
258 		do {					\
259 			tmp = *bot; *bot++ = *s; *s++ = tmp; \
260 		} while (--i);				\
261 		s -= size2;				\
262 	} while(bot < s);				\
263 }
264 
265 /*
266  * Optional hybrid natural/pairwise first pass.  Eats up list1 in runs of
267  * increasing order, list2 in a corresponding linked list.  Checks for runs
268  * when THRESHOLD/2 pairs compare with same sense.  (Only used when NATURAL
269  * is defined.  Otherwise simple pairwise merging is used.)
270  */
271 void
272 setup(u_char *list1, u_char *list2, size_t n, size_t size, cmp_t cmp)
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 void
343 insertionsort(u_char *a, size_t n, size_t size, cmp_t cmp)
344 {
345 	u_char *ai, *s, *t, *u, tmp;
346 	int i;
347 
348 	for (ai = a+size; --n >= 1; ai += size)
349 		for (t = ai; t > a; t -= size) {
350 			u = t - size;
351 			if (CMP(u, t) <= 0)
352 				break;
353 			swap(u, t);
354 		}
355 }
356