Home | History | Annotate | Line # | Download | only in stdlib
radixsort.c revision 1.1
      1  1.1  cgd /*-
      2  1.1  cgd  * Copyright (c) 1990 The Regents of the University of California.
      3  1.1  cgd  * All rights reserved.
      4  1.1  cgd  *
      5  1.1  cgd  * Redistribution and use in source and binary forms, with or without
      6  1.1  cgd  * modification, are permitted provided that the following conditions
      7  1.1  cgd  * are met:
      8  1.1  cgd  * 1. Redistributions of source code must retain the above copyright
      9  1.1  cgd  *    notice, this list of conditions and the following disclaimer.
     10  1.1  cgd  * 2. Redistributions in binary form must reproduce the above copyright
     11  1.1  cgd  *    notice, this list of conditions and the following disclaimer in the
     12  1.1  cgd  *    documentation and/or other materials provided with the distribution.
     13  1.1  cgd  * 3. All advertising materials mentioning features or use of this software
     14  1.1  cgd  *    must display the following acknowledgement:
     15  1.1  cgd  *	This product includes software developed by the University of
     16  1.1  cgd  *	California, Berkeley and its contributors.
     17  1.1  cgd  * 4. Neither the name of the University nor the names of its contributors
     18  1.1  cgd  *    may be used to endorse or promote products derived from this software
     19  1.1  cgd  *    without specific prior written permission.
     20  1.1  cgd  *
     21  1.1  cgd  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     22  1.1  cgd  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     23  1.1  cgd  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     24  1.1  cgd  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     25  1.1  cgd  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     26  1.1  cgd  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     27  1.1  cgd  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     28  1.1  cgd  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     29  1.1  cgd  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     30  1.1  cgd  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     31  1.1  cgd  * SUCH DAMAGE.
     32  1.1  cgd  */
     33  1.1  cgd 
     34  1.1  cgd #if defined(LIBC_SCCS) && !defined(lint)
     35  1.1  cgd static char sccsid[] = "@(#)radixsort.c	5.7 (Berkeley) 2/23/91";
     36  1.1  cgd #endif /* LIBC_SCCS and not lint */
     37  1.1  cgd 
     38  1.1  cgd #include <sys/types.h>
     39  1.1  cgd #include <limits.h>
     40  1.1  cgd #include <stdlib.h>
     41  1.1  cgd #include <stddef.h>
     42  1.1  cgd #include <string.h>
     43  1.1  cgd 
     44  1.1  cgd /*
     45  1.1  cgd  * __rspartition is the cutoff point for a further partitioning instead
     46  1.1  cgd  * of a shellsort.  If it changes check __rsshell_increments.  Both of
     47  1.1  cgd  * these are exported, as the best values are data dependent.
     48  1.1  cgd  */
     49  1.1  cgd #define	NPARTITION	40
     50  1.1  cgd int __rspartition = NPARTITION;
     51  1.1  cgd int __rsshell_increments[] = { 4, 1, 0, 0, 0, 0, 0, 0 };
     52  1.1  cgd 
     53  1.1  cgd /*
     54  1.1  cgd  * Stackp points to context structures, where each structure schedules a
     55  1.1  cgd  * partitioning.  Radixsort exits when the stack is empty.
     56  1.1  cgd  *
     57  1.1  cgd  * If the buckets are placed on the stack randomly, the worst case is when
     58  1.1  cgd  * all the buckets but one contain (npartitions + 1) elements and the bucket
     59  1.1  cgd  * pushed on the stack last contains the rest of the elements.  In this case,
     60  1.1  cgd  * stack growth is bounded by:
     61  1.1  cgd  *
     62  1.1  cgd  *	limit = (nelements / (npartitions + 1)) - 1;
     63  1.1  cgd  *
     64  1.1  cgd  * This is a very large number, 52,377,648 for the maximum 32-bit signed int.
     65  1.1  cgd  *
     66  1.1  cgd  * By forcing the largest bucket to be pushed on the stack first, the worst
     67  1.1  cgd  * case is when all but two buckets each contain (npartitions + 1) elements,
     68  1.1  cgd  * with the remaining elements split equally between the first and last
     69  1.1  cgd  * buckets pushed on the stack.  In this case, stack growth is bounded when:
     70  1.1  cgd  *
     71  1.1  cgd  *	for (partition_cnt = 0; nelements > npartitions; ++partition_cnt)
     72  1.1  cgd  *		nelements =
     73  1.1  cgd  *		    (nelements - (npartitions + 1) * (nbuckets - 2)) / 2;
     74  1.1  cgd  * The bound is:
     75  1.1  cgd  *
     76  1.1  cgd  *	limit = partition_cnt * (nbuckets - 1);
     77  1.1  cgd  *
     78  1.1  cgd  * This is a much smaller number, 4590 for the maximum 32-bit signed int.
     79  1.1  cgd  */
     80  1.1  cgd #define	NBUCKETS	(UCHAR_MAX + 1)
     81  1.1  cgd 
     82  1.1  cgd typedef struct _stack {
     83  1.1  cgd 	const u_char **bot;
     84  1.1  cgd 	int indx, nmemb;
     85  1.1  cgd } CONTEXT;
     86  1.1  cgd 
     87  1.1  cgd #define	STACKPUSH { \
     88  1.1  cgd 	stackp->bot = p; \
     89  1.1  cgd 	stackp->nmemb = nmemb; \
     90  1.1  cgd 	stackp->indx = indx; \
     91  1.1  cgd 	++stackp; \
     92  1.1  cgd }
     93  1.1  cgd #define	STACKPOP { \
     94  1.1  cgd 	if (stackp == stack) \
     95  1.1  cgd 		break; \
     96  1.1  cgd 	--stackp; \
     97  1.1  cgd 	bot = stackp->bot; \
     98  1.1  cgd 	nmemb = stackp->nmemb; \
     99  1.1  cgd 	indx = stackp->indx; \
    100  1.1  cgd }
    101  1.1  cgd 
    102  1.1  cgd /*
    103  1.1  cgd  * A variant of MSD radix sorting; see Knuth Vol. 3, page 177, and 5.2.5,
    104  1.1  cgd  * Ex. 10 and 12.  Also, "Three Partition Refinement Algorithms, Paige
    105  1.1  cgd  * and Tarjan, SIAM J. Comput. Vol. 16, No. 6, December 1987.
    106  1.1  cgd  *
    107  1.1  cgd  * This uses a simple sort as soon as a bucket crosses a cutoff point,
    108  1.1  cgd  * rather than sorting the entire list after partitioning is finished.
    109  1.1  cgd  * This should be an advantage.
    110  1.1  cgd  *
    111  1.1  cgd  * This is pure MSD instead of LSD of some number of MSD, switching to
    112  1.1  cgd  * the simple sort as soon as possible.  Takes linear time relative to
    113  1.1  cgd  * the number of bytes in the strings.
    114  1.1  cgd  */
    115  1.1  cgd int
    116  1.1  cgd #if __STDC__
    117  1.1  cgd radixsort(const u_char **l1, int nmemb, const u_char *tab, u_char endbyte)
    118  1.1  cgd #else
    119  1.1  cgd radixsort(l1, nmemb, tab, endbyte)
    120  1.1  cgd 	const u_char **l1;
    121  1.1  cgd 	register int nmemb;
    122  1.1  cgd 	const u_char *tab;
    123  1.1  cgd 	u_char endbyte;
    124  1.1  cgd #endif
    125  1.1  cgd {
    126  1.1  cgd 	register int i, indx, t1, t2;
    127  1.1  cgd 	register const u_char **l2;
    128  1.1  cgd 	register const u_char **p;
    129  1.1  cgd 	register const u_char **bot;
    130  1.1  cgd 	register const u_char *tr;
    131  1.1  cgd 	CONTEXT *stack, *stackp;
    132  1.1  cgd 	int c[NBUCKETS + 1], max;
    133  1.1  cgd 	u_char ltab[NBUCKETS];
    134  1.1  cgd 	static void shellsort();
    135  1.1  cgd 
    136  1.1  cgd 	if (nmemb <= 1)
    137  1.1  cgd 		return(0);
    138  1.1  cgd 
    139  1.1  cgd 	/*
    140  1.1  cgd 	 * T1 is the constant part of the equation, the number of elements
    141  1.1  cgd 	 * represented on the stack between the top and bottom entries.
    142  1.1  cgd 	 * It doesn't get rounded as the divide by 2 rounds down (correct
    143  1.1  cgd 	 * for a value being subtracted).  T2, the nelem value, has to be
    144  1.1  cgd 	 * rounded up before each divide because we want an upper bound;
    145  1.1  cgd 	 * this could overflow if nmemb is the maximum int.
    146  1.1  cgd 	 */
    147  1.1  cgd 	t1 = ((__rspartition + 1) * (NBUCKETS - 2)) >> 1;
    148  1.1  cgd 	for (i = 0, t2 = nmemb; t2 > __rspartition; i += NBUCKETS - 1)
    149  1.1  cgd 		t2 = ((t2 + 1) >> 1) - t1;
    150  1.1  cgd 	if (i) {
    151  1.1  cgd 		if (!(stack = stackp = (CONTEXT *)malloc(i * sizeof(CONTEXT))))
    152  1.1  cgd 			return(-1);
    153  1.1  cgd 	} else
    154  1.1  cgd 		stack = stackp = NULL;
    155  1.1  cgd 
    156  1.1  cgd 	/*
    157  1.1  cgd 	 * There are two arrays, one provided by the user (l1), and the
    158  1.1  cgd 	 * temporary one (l2).  The data is sorted to the temporary stack,
    159  1.1  cgd 	 * and then copied back.  The speedup of using index to determine
    160  1.1  cgd 	 * which stack the data is on and simply swapping stacks back and
    161  1.1  cgd 	 * forth, thus avoiding the copy every iteration, turns out to not
    162  1.1  cgd 	 * be any faster than the current implementation.
    163  1.1  cgd 	 */
    164  1.1  cgd 	if (!(l2 = (const u_char **)malloc(sizeof(u_char *) * nmemb)))
    165  1.1  cgd 		return(-1);
    166  1.1  cgd 
    167  1.1  cgd 	/*
    168  1.1  cgd 	 * Tr references a table of sort weights; multiple entries may
    169  1.1  cgd 	 * map to the same weight; EOS char must have the lowest weight.
    170  1.1  cgd 	 */
    171  1.1  cgd 	if (tab)
    172  1.1  cgd 		tr = tab;
    173  1.1  cgd 	else {
    174  1.1  cgd 		for (t1 = 0, t2 = endbyte; t1 < t2; ++t1)
    175  1.1  cgd 			ltab[t1] = t1 + 1;
    176  1.1  cgd 		ltab[t2] = 0;
    177  1.1  cgd 		for (t1 = endbyte + 1; t1 < NBUCKETS; ++t1)
    178  1.1  cgd 			ltab[t1] = t1;
    179  1.1  cgd 		tr = ltab;
    180  1.1  cgd 	}
    181  1.1  cgd 
    182  1.1  cgd 	/* First sort is entire stack */
    183  1.1  cgd 	bot = l1;
    184  1.1  cgd 	indx = 0;
    185  1.1  cgd 
    186  1.1  cgd 	for (;;) {
    187  1.1  cgd 		/* Clear bucket count array */
    188  1.1  cgd 		bzero((char *)c, sizeof(c));
    189  1.1  cgd 
    190  1.1  cgd 		/*
    191  1.1  cgd 		 * Compute number of items that sort to the same bucket
    192  1.1  cgd 		 * for this index.
    193  1.1  cgd 		 */
    194  1.1  cgd 		for (p = bot, i = nmemb; --i >= 0;)
    195  1.1  cgd 			++c[tr[(*p++)[indx]]];
    196  1.1  cgd 
    197  1.1  cgd 		/*
    198  1.1  cgd 		 * Sum the number of characters into c, dividing the temp
    199  1.1  cgd 		 * stack into the right number of buckets for this bucket,
    200  1.1  cgd 		 * this index.  C contains the cumulative total of keys
    201  1.1  cgd 		 * before and included in this bucket, and will later be
    202  1.1  cgd 		 * used as an index to the bucket.  c[NBUCKETS] contains
    203  1.1  cgd 		 * the total number of elements, for determining how many
    204  1.1  cgd 		 * elements the last bucket contains.  At the same time
    205  1.1  cgd 		 * find the largest bucket so it gets pushed first.
    206  1.1  cgd 		 */
    207  1.1  cgd 		for (i = max = t1 = 0, t2 = __rspartition; i <= NBUCKETS; ++i) {
    208  1.1  cgd 			if (c[i] > t2) {
    209  1.1  cgd 				t2 = c[i];
    210  1.1  cgd 				max = i;
    211  1.1  cgd 			}
    212  1.1  cgd 			t1 = c[i] += t1;
    213  1.1  cgd 		}
    214  1.1  cgd 
    215  1.1  cgd 		/*
    216  1.1  cgd 		 * Partition the elements into buckets; c decrements through
    217  1.1  cgd 		 * the bucket, and ends up pointing to the first element of
    218  1.1  cgd 		 * the bucket.
    219  1.1  cgd 		 */
    220  1.1  cgd 		for (i = nmemb; --i >= 0;) {
    221  1.1  cgd 			--p;
    222  1.1  cgd 			l2[--c[tr[(*p)[indx]]]] = *p;
    223  1.1  cgd 		}
    224  1.1  cgd 
    225  1.1  cgd 		/* Copy the partitioned elements back to user stack */
    226  1.1  cgd 		bcopy(l2, bot, nmemb * sizeof(u_char *));
    227  1.1  cgd 
    228  1.1  cgd 		++indx;
    229  1.1  cgd 		/*
    230  1.1  cgd 		 * Sort buckets as necessary; don't sort c[0], it's the
    231  1.1  cgd 		 * EOS character bucket, and nothing can follow EOS.
    232  1.1  cgd 		 */
    233  1.1  cgd 		for (i = max; i; --i) {
    234  1.1  cgd 			if ((nmemb = c[i + 1] - (t1 = c[i])) < 2)
    235  1.1  cgd 				continue;
    236  1.1  cgd 			p = bot + t1;
    237  1.1  cgd 			if (nmemb > __rspartition)
    238  1.1  cgd 				STACKPUSH
    239  1.1  cgd 			else
    240  1.1  cgd 				shellsort(p, indx, nmemb, tr);
    241  1.1  cgd 		}
    242  1.1  cgd 		for (i = max + 1; i < NBUCKETS; ++i) {
    243  1.1  cgd 			if ((nmemb = c[i + 1] - (t1 = c[i])) < 2)
    244  1.1  cgd 				continue;
    245  1.1  cgd 			p = bot + t1;
    246  1.1  cgd 			if (nmemb > __rspartition)
    247  1.1  cgd 				STACKPUSH
    248  1.1  cgd 			else
    249  1.1  cgd 				shellsort(p, indx, nmemb, tr);
    250  1.1  cgd 		}
    251  1.1  cgd 		/* Break out when stack is empty */
    252  1.1  cgd 		STACKPOP
    253  1.1  cgd 	}
    254  1.1  cgd 
    255  1.1  cgd 	free((char *)l2);
    256  1.1  cgd 	free((char *)stack);
    257  1.1  cgd 	return(0);
    258  1.1  cgd }
    259  1.1  cgd 
    260  1.1  cgd /*
    261  1.1  cgd  * Shellsort (diminishing increment sort) from Data Structures and
    262  1.1  cgd  * Algorithms, Aho, Hopcraft and Ullman, 1983 Edition, page 290;
    263  1.1  cgd  * see also Knuth Vol. 3, page 84.  The increments are selected from
    264  1.1  cgd  * formula (8), page 95.  Roughly O(N^3/2).
    265  1.1  cgd  */
    266  1.1  cgd static void
    267  1.1  cgd shellsort(p, indx, nmemb, tr)
    268  1.1  cgd 	register u_char **p, *tr;
    269  1.1  cgd 	register int indx, nmemb;
    270  1.1  cgd {
    271  1.1  cgd 	register u_char ch, *s1, *s2;
    272  1.1  cgd 	register int incr, *incrp, t1, t2;
    273  1.1  cgd 
    274  1.1  cgd 	for (incrp = __rsshell_increments; incr = *incrp++;)
    275  1.1  cgd 		for (t1 = incr; t1 < nmemb; ++t1)
    276  1.1  cgd 			for (t2 = t1 - incr; t2 >= 0;) {
    277  1.1  cgd 				s1 = p[t2] + indx;
    278  1.1  cgd 				s2 = p[t2 + incr] + indx;
    279  1.1  cgd 				while ((ch = tr[*s1++]) == tr[*s2] && ch)
    280  1.1  cgd 					++s2;
    281  1.1  cgd 				if (ch > tr[*s2]) {
    282  1.1  cgd 					s1 = p[t2];
    283  1.1  cgd 					p[t2] = p[t2 + incr];
    284  1.1  cgd 					p[t2 + incr] = s1;
    285  1.1  cgd 					t2 -= incr;
    286  1.1  cgd 				} else
    287  1.1  cgd 					break;
    288  1.1  cgd 			}
    289  1.1  cgd }
    290