rpmdb/merge.c

Go to the documentation of this file.
00001 /*@-bounds -mustmod -sizeoftype @*/
00002 #ifndef __APPLE__
00003 /*-
00004  * Copyright (c) 1992, 1993
00005  *      The Regents of the University of California.  All rights reserved.
00006  *
00007  * This code is derived from software contributed to Berkeley by
00008  * Peter McIlroy.
00009  *
00010  * Redistribution and use in source and binary forms, with or without
00011  * modification, are permitted provided that the following conditions
00012  * are met:
00013  * 1. Redistributions of source code must retain the above copyright
00014  *    notice, this list of conditions and the following disclaimer.
00015  * 2. Redistributions in binary form must reproduce the above copyright
00016  *    notice, this list of conditions and the following disclaimer in the
00017  *    documentation and/or other materials provided with the distribution.
00018  * 3. All advertising materials mentioning features or use of this software
00019  *    must display the following acknowledgement:
00020  *      This product includes software developed by the University of
00021  *      California, Berkeley and its contributors.
00022  * 4. Neither the name of the University nor the names of its contributors
00023  *    may be used to endorse or promote products derived from this software
00024  *    without specific prior written permission.
00025  *
00026  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
00027  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
00028  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
00029  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
00030  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
00031  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
00032  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
00033  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
00034  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
00035  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
00036  * SUCH DAMAGE.
00037  */
00038 
00039 #if defined(LIBC_SCCS) && !defined(lint)
00040 static char sccsid[] = "@(#)merge.c     8.2 (Berkeley) 2/14/94";
00041 #endif /* LIBC_SCCS and not lint */
00042 
00043 /*
00044  * Hybrid exponential search/linear search merge sort with hybrid
00045  * natural/pairwise first pass.  Requires about .3% more comparisons
00046  * for random data than LSMS with pairwise first pass alone.
00047  * It works for objects as small as two bytes.
00048  */
00049 
00050 #define NATURAL
00051 #define THRESHOLD 16    /* Best choice for natural merge cut-off. */
00052 
00053 /* #define NATURAL to get hybrid natural merge.
00054  * (The default is pairwise merging.)
00055  */
00056 
00057 #include "system.h"
00058 
00059 #define ISIZE sizeof(int)
00060 #define PSIZE sizeof(unsigned char *)
00061 #define ICOPY_LIST(src, dst, last)                              \
00062         do                                                      \
00063         *(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE;    \
00064         while(src < last)
00065 #define ICOPY_ELT(src, dst, i)                                  \
00066         do                                                      \
00067         *(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE;  \
00068         while (i -= ISIZE)
00069 
00070 #define CCOPY_LIST(src, dst, last)              \
00071         do                                      \
00072                 *dst++ = *src++;                \
00073         while (src < last)
00074 #define CCOPY_ELT(src, dst, i)                  \
00075         do                                      \
00076                 *dst++ = *src++;                \
00077         while (i -= 1)
00078 
00079 /*
00080  * Find the next possible pointer head.  (Trickery for forcing an array
00081  * to do double duty as a linked list when objects do not align with word
00082  * boundaries.
00083  */
00084 /* Assumption: PSIZE is a power of 2. */
00085 #define EVAL(p) (unsigned char **)                                      \
00086     ((unsigned char *)0 +                                               \
00087         (((unsigned char *)p + PSIZE - 1 - (unsigned char *) 0) & ~(PSIZE - 1)))
00088 
00089 #define swap(a, b) {                                    \
00090                 s = b;                                  \
00091                 i = size;                               \
00092                 do {                                    \
00093                         tmp = *a; *a++ = *s; *s++ = tmp; \
00094                 } while (--i);                          \
00095                 a -= size;                              \
00096         }
00097 #define reverse(bot, top) {                             \
00098         s = top;                                        \
00099         do {                                            \
00100                 i = size;                               \
00101                 do {                                    \
00102                         tmp = *bot; *bot++ = *s; *s++ = tmp; \
00103                 } while (--i);                          \
00104                 s -= size2;                             \
00105         } while(bot < s);                               \
00106 }
00107 
00108 /*
00109  * This is to avoid out-of-bounds addresses in sorting the
00110  * last 4 elements.
00111  */
00112 static void
00113 insertionsort(unsigned char *a, size_t n, size_t size,
00114                 int (*cmp) (const void *, const void *))
00115         /*@modifies *a @*/
00116 {
00117         u_char *ai, *s, *t, *u, tmp;
00118         int i;
00119 
00120         for (ai = a+size; --n >= 1; ai += size)
00121                 for (t = ai; t > a; t -= size) {
00122                         u = t - size;
00123                         if (cmp(u, t) <= 0)
00124                                 /*@innerbreak@*/ break;
00125                         swap(u, t);
00126                 }
00127 }
00128 
00129 /*
00130  * Optional hybrid natural/pairwise first pass.  Eats up list1 in runs of
00131  * increasing order, list2 in a corresponding linked list.  Checks for runs
00132  * when THRESHOLD/2 pairs compare with same sense.  (Only used when NATURAL
00133  * is defined.  Otherwise simple pairwise merging is used.)
00134  */
00135 static void
00136 setup(unsigned char *list1, /*@out@*/ unsigned char *list2,
00137                 size_t n, size_t size, int (*cmp) (const void *, const void *))
00138         /*@modifies list1, list2 @*/
00139 {
00140         int i, length, size2, tmp, sense;
00141         unsigned char *f1, *f2, *s, *l2, *last, *p2;
00142 
00143         size2 = size*2;
00144         if (n <= 5) {
00145                 insertionsort(list1, n, size, cmp);
00146                 *EVAL(list2) = (unsigned char*) list2 + n*size;
00147                 return;
00148         }
00149         /*
00150          * Avoid running pointers out of bounds; limit n to evens
00151          * for simplicity.
00152          */
00153         i = 4 + (n & 1);
00154         insertionsort(list1 + (n - i) * size, i, size, cmp);
00155         last = list1 + size * (n - i);
00156         *EVAL(list2 + (last - list1)) = list2 + n * size;
00157 
00158 #ifdef NATURAL
00159         p2 = list2;
00160         f1 = list1;
00161         sense = (cmp(f1, f1 + size) > 0);
00162         for (; f1 < last; sense = !sense) {
00163                 length = 2;
00164                                         /* Find pairs with same sense. */
00165                 for (f2 = f1 + size2; f2 < last; f2 += size2) {
00166                         if ((cmp(f2, f2+ size) > 0) != sense)
00167                                 /*@innerbreak@*/ break;
00168                         length += 2;
00169                 }
00170                 if (length < THRESHOLD) {               /* Pairwise merge */
00171                         do {
00172                                 p2 = *EVAL(p2) = f1 + size2 - list1 + list2;
00173                                 if (sense > 0)
00174                                         swap (f1, f1 + size);
00175                         } while ((f1 += size2) < f2);
00176                 } else {                                /* Natural merge */
00177                         l2 = f2;
00178                         for (f2 = f1 + size2; f2 < l2; f2 += size2) {
00179                                 if ((cmp(f2-size, f2) > 0) != sense) {
00180                                         p2 = *EVAL(p2) = f2 - list1 + list2;
00181                                         if (sense > 0)
00182                                                 reverse(f1, f2-size);
00183                                         f1 = f2;
00184                                 }
00185                         }
00186                         if (sense > 0)
00187                                 reverse (f1, f2-size);
00188                         f1 = f2;
00189                         if (f2 < last || cmp(f2 - size, f2) > 0)
00190                                 p2 = *EVAL(p2) = f2 - list1 + list2;
00191                         else
00192                                 p2 = *EVAL(p2) = list2 + n*size;
00193                 }
00194         }
00195 #else           /* pairwise merge only. */
00196         for (f1 = list1, p2 = list2; f1 < last; f1 += size2) {
00197                 p2 = *EVAL(p2) = p2 + size2;
00198                 if (cmp (f1, f1 + size) > 0)
00199                         swap(f1, f1 + size);
00200         }
00201 #endif /* NATURAL */
00202 }
00203 
00204 /*
00205  * Arguments are as for qsort.
00206  */
00207 int
00208 mergesort(void *base, size_t nmemb, size_t size,
00209                 int (*cmp) (const void *, const void *))
00210 {
00211         register int i, sense;
00212         int big, iflag;
00213         register unsigned char *f1, *f2, *t, *b, *q, *l1, *l2;
00214         /*@dependent@*/
00215         register unsigned char *tp2;
00216         /*@owned@*/
00217         unsigned char *list2;
00218         /*@dependent@*/
00219         unsigned char *list1;
00220         unsigned char *p2, *p, *last, **p1;
00221 
00222         if (size < PSIZE / 2) {         /* Pointers must fit into 2 * size. */
00223                 errno = EINVAL;
00224                 return (-1);
00225         }
00226 
00227         if (nmemb == 0)
00228                 return (0);
00229 
00230         /*
00231          * XXX
00232          * Stupid subtraction for the Cray.
00233          */
00234         iflag = 0;
00235         if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE))
00236                 iflag = 1;
00237 
00238         if ((list2 = malloc(nmemb * size + PSIZE)) == NULL)
00239                 return (-1);
00240 
00241         list1 = base;
00242         setup(list1, list2, nmemb, size, cmp);
00243         last = list2 + nmemb * size;
00244         i = big = 0;
00245 /*@-branchstate@*/
00246         while (*EVAL(list2) != last) {
00247             l2 = list1;
00248             p1 = EVAL(list1);
00249             for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) {
00250                 p2 = *EVAL(p2);
00251                 f1 = l2;
00252                 f2 = l1 = list1 + (p2 - list2);
00253                 if (p2 != last)
00254                         p2 = *EVAL(p2);
00255                 l2 = list1 + (p2 - list2);
00256                 while (f1 < l1 && f2 < l2) {
00257                         if ((*cmp)(f1, f2) <= 0) {
00258                                 q = f2;
00259                                 b = f1, t = l1;
00260                                 sense = -1;
00261                         } else {
00262                                 q = f1;
00263                                 b = f2, t = l2;
00264                                 sense = 0;
00265                         }
00266                         if (!big) {     /* here i = 0 */
00267                                 while ((b += size) < t && cmp(q, b) >sense)
00268                                         if (++i == 6) {
00269                                                 big = 1;
00270                                                 goto EXPONENTIAL;
00271                                         }
00272                         } else {
00273 /*@-shiftimplementation@*/
00274 EXPONENTIAL:                    for (i = size; ; i <<= 1)
00275                                         if ((p = (b + i)) >= t) {
00276                                                 if ((p = t - size) > b &&
00277                                                     (*cmp)(q, p) <= sense)
00278                                                         t = p;
00279                                                 else
00280                                                         b = p;
00281                                                 /*@innerbreak@*/ break;
00282                                         } else if ((*cmp)(q, p) <= sense) {
00283                                                 t = p;
00284                                                 if (i == size)
00285                                                         big = 0;
00286                                                 goto FASTCASE;
00287                                         } else
00288                                                 b = p;
00289                                 /*@-infloopsuncon@*/
00290                                 while (t > b+size) {
00291                                         i = (((t - b) / size) >> 1) * size;
00292                                         if ((*cmp)(q, p = b + i) <= sense)
00293                                                 t = p;
00294                                         else
00295                                                 b = p;
00296                                 }
00297                                 goto COPY;
00298 FASTCASE:                       while (i > size)
00299                                         if ((*cmp)(q,
00300                                                 p = b + (i >>= 1)) <= sense)
00301                                                 t = p;
00302                                         else
00303                                                 b = p;
00304                                 /*@=infloopsuncon@*/
00305 /*@=shiftimplementation@*/
00306 COPY:                           b = t;
00307                         }
00308                         i = size;
00309                         if (q == f1) {
00310                                 if (iflag) {
00311                                         ICOPY_LIST(f2, tp2, b);
00312                                         ICOPY_ELT(f1, tp2, i);
00313                                 } else {
00314                                         CCOPY_LIST(f2, tp2, b);
00315                                         CCOPY_ELT(f1, tp2, i);
00316                                 }
00317                         } else {
00318                                 if (iflag) {
00319                                         ICOPY_LIST(f1, tp2, b);
00320                                         ICOPY_ELT(f2, tp2, i);
00321                                 } else {
00322                                         CCOPY_LIST(f1, tp2, b);
00323                                         CCOPY_ELT(f2, tp2, i);
00324                                 }
00325                         }
00326                 }
00327                 if (f2 < l2) {
00328                         if (iflag)
00329                                 ICOPY_LIST(f2, tp2, l2);
00330                         else
00331                                 CCOPY_LIST(f2, tp2, l2);
00332                 } else if (f1 < l1) {
00333                         if (iflag)
00334                                 ICOPY_LIST(f1, tp2, l1);
00335                         else
00336                                 CCOPY_LIST(f1, tp2, l1);
00337                 }
00338                 *p1 = l2;
00339             }
00340 /*@-dependenttrans@*/
00341             tp2 = list1;        /* swap list1, list2 */
00342             list1 = list2;
00343             list2 = tp2;
00344 /*@=dependenttrans@*/
00345             last = list2 + nmemb*size;
00346         }
00347 /*@=branchstate@*/
00348         if (base == list2) {
00349                 memmove(list2, list1, nmemb*size);
00350                 list2 = list1;
00351         }
00352 /*@-usereleased@*/
00353         free(list2);
00354 /*@=usereleased@*/
00355         return (0);
00356 }
00357 #else
00358 /* mergesort is implemented in System on Mac OS X */
00359 #endif /* __APPLE__ */
00360 /*@=bounds =mustmod =sizeoftype @*/

Generated on Fri Oct 12 08:44:54 2007 for rpm by  doxygen 1.5.2