/*-
* Copyright (c) 1980, 1983 The Regents of the University of California.
* All rights reserved.
*
* Redistribution and use in source and binary forms are permitted
* provided that: (1) source distributions retain this entire copyright
* notice and comment, and (2) distributions including binaries display
* the following acknowledgement: ``This product includes software
* developed by the University of California, Berkeley and its contributors''
* in the documentation or other materials provided with the distribution
* and in all advertising materials mentioning features or use of this
* software. Neither the name of the University nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
* THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
* IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
* WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
*/
/* Note: I have modified this file as follows:
* Liam
*
* $Id: lqsort.c,v 1.8 1996/06/11 22:07:00 lee Exp $
*
* $Log: lqsort.c,v $
* Revision 1.8 96/06/11 22:07:00 lee
* BIF is a long now.
*
* Revision 1.7 1996/05/15 22:59:02 lee
* api change, plus ? bugfix?
*
* Revision 1.6 94/03/04 11:53:31 lee
* fix core dump...
*
* Revision 1.5 94/02/26 14:53:47 lee
* API change
*
* Revision 1.4 92/02/15 06:43:40 lee
* declared qst() as void.
*
* Revision 1.3 92/02/15 05:34:45 lee
* Added a macro that doesn't work...
*
* Revision 1.2 92/01/31 00:14:55 lee
* gcc warnings; added an INLINE.
*
* Revision 1.1 91/06/16 18:08:39 lee
* Initial revision
*
*/
#if defined(LIBC_SCCS) && !defined(lint)
static char sccsid[] = "@(#)qsort.c 5.7 (Berkeley) 5/17/90";
#endif /* LIBC_SCCS and not lint */
#include "error.h"
#include <stdio.h>
#include <sys/types.h>
#include "globals.h"
#include "liblqtext.h"
#include "fileinfo.h"
#include "wordplace.h"
/*
* qsort.c:
* Our own version of the system qsort routine which is faster by an average
* of 25%, with lows and highs of 10% and 50%.
* The THRESHold below is the insertion sort threshold, and has been adjusted
* for records of size 48 bytes.
* The MTHREShold is where we stop finding a better median.
*/
#define THRESH 4 /* threshold for insertion */
#define MTHRESH 6 /* threshold for median */
#if 0
#define qcmp(F1,F2) wpqcmp((t_WordPlace *)F1, (t_WordPlace *)F2)
/* this is commented out because it dumps core sometimes... */
/* There's a function for this below, if you have problems. */
#define wpqcmp(F1,F2) \
(((F1)->FID == (F2)->FID) ? ( \
((F1)->BlockInFile == (F2)->BlockInFile) ? \
((F1)->WordInBlock - (F2)->WordInBlock) : \
((F1)->BlockInFile - (F2)->BlockInFile) ) : \
((F1)->FID - (F2)->FID))
#else
/* I'll leave this in case the macro version above makes some compilers
* explode...
*/
static int
qcmp(F1, F2) /* the comparison routine */
register t_WordPlace *F1, *F2;
{
if (F1->FID != F2->FID) {
return F1->FID - F2->FID;
} else {
long diff;
if (F1->BlockInFile != F2->BlockInFile) {
diff = F1->BlockInFile - F2->BlockInFile;
} else {
diff = F1->WordInBlock - F2->WordInBlock;
}
if (diff < 0) return -1;
if (diff > 0) return 1;
return 0;
}
}
#endif
static int qsz; /* size of each record */
static int thresh; /* THRESHold in chars */
static int mthresh; /* MTHRESHold in chars */
/*
* qsort:
* First, set up some global parameters for qst to share. Then, quicksort
* with qst(), and then a cleanup insertion sort ourselves. Sound simple?
* It's not...
*/
PRIVATE void qst();
/* <Function>
* <Name>LQT_SortWordPlaces
* <Class>Database/Update, Database/Words
* <Purpose>
* Sorts the given WordPlace array using Quicker Sort
* to the in-memory stop list, to be ignored by LQT_ReadWord.
* A WordPlace array must be sorted in ascending order by FID, then
* by Block In File, then by Word Within Block, in order to be
* written to the database. Since this is exactly the order generated
* by reading files one at a time from beginning to end, this
* routine is not currently used.
* <Notes>
* Buggy, I think.
* </Function>
*/
API void
LQT_SortWordPlaces(db, NumberOfWordPlaces, WordPlaces)
t_LQTEXT_Database *db;
unsigned long NumberOfWordPlaces;
t_WordPlace *WordPlaces;
{
char *base = (char *) WordPlaces;
register char c, *i, *j, *lo, *hi;
char *min, *max;
if (NumberOfWordPlaces <= 1) return;
#if 0 /* untested */
if (NumberOfWordPlaces == 2) {
if (qcmp(&WordPlaces[0], &WordPlaces[1]) > 0) {
t_WordPlace tmp = WordPlaces[0]; /* struct copy */
WprdPlaces[0] = WordPlaces[1];
WordPlaces[1] = tmp;
}
return;
}
#endif
qsz = sizeof(t_WordPlace);
thresh = qsz * THRESH;
mthresh = qsz * MTHRESH;
max = base + NumberOfWordPlaces * qsz;
if (NumberOfWordPlaces >= THRESH) {
qst(base, max);
hi = base + thresh;
} else {
hi = max;
}
/*
* First put smallest element, which must be in the first THRESH, in
* the first position as a sentinel. This is done just by searching
* the first THRESH elements (or the first n if n < THRESH), finding
* the min, and swapping it into the first position.
*/
for (j = lo = base; (lo += qsz) < hi; )
if (qcmp(j, lo) > 0)
j = lo;
if (j != base) {
/* swap j into place */
for (i = base, hi = base + qsz; i < hi; ) {
c = *j;
*j++ = *i;
*i++ = c;
}
}
/*
* With our sentinel in place, we now run the following hyper-fast
* insertion sort. For each remaining element, min, from [1] to [n-1],
* set hi to the index of the element AFTER which this one goes.
* Then, do the standard insertion sort shift on a character at a time
* basis for each element in the frob.
*/
for (min = base; (hi = min += qsz) < max; ) {
while (qcmp(hi -= qsz, min) > 0)
/* void */;
if ((hi += qsz) != min) {
for (lo = min + qsz; --lo >= min; ) {
c = *lo;
for (i = j = lo; (j -= qsz) >= hi; i = j)
*i = *j;
*i = c;
}
}
}
}
/*
* qst:
* Do a quicksort
* First, find the median element, and put that one in the first place as the
* discriminator. (This "median" is just the median of the first, last and
* middle elements). (Using this median instead of the first element is a big
* win). Then, the usual partitioning/swapping, followed by moving the
* discriminator into the right place. Then, figure out the sizes of the two
* partions, do the smaller one recursively and the larger one via a repeat of
* this code. Stopping when there are less than THRESH elements in a partition
* and cleaning up with an insertion sort (in our caller) is a huge win.
* All data swaps are done in-line, which is space-losing but time-saving.
* (And there are only three places where this is done).
*/
static void
qst(base, max)
char *base, *max;
{
register char c, *i, *j, *jj;
register int ii;
char *mid, *tmp;
int lo, hi;
/*
* At the top here, lo is the number of characters of elements in the
* current partition. (Which should be max - base).
* Find the median of the first, last, and middle element and make
* that the middle element. Set j to largest of first and middle.
* If max is larger than that guy, then it's that guy, else compare
* max with loser of first and take larger. Things are set up to
* prefer the middle, then the first in case of ties.
*/
lo = max - base; /* number of elements as chars */
do {
mid = i = base + qsz * ((lo / qsz) >> 1);
if (lo >= mthresh) {
j = (qcmp((jj = base), i) > 0 ? jj : i);
if (qcmp(j, (tmp = max - qsz)) > 0) {
/* switch to first loser */
j = (j == jj ? i : jj);
if (qcmp(j, tmp) < 0) {
j = tmp;
}
}
if (j != i) {
ii = qsz;
do {
c = *i;
*i++ = *j;
*j++ = c;
} while (--ii);
}
}
/*
* Semi-standard quicksort partitioning/swapping
*/
for (i = base, j = max - qsz; ; ) {
while (i < mid && qcmp(i, mid) <= 0) {
i += qsz;
}
while (j > mid) {
if (qcmp(mid, j) <= 0) {
j -= qsz;
continue;
}
tmp = i + qsz; /* value of i after swap */
if (i == mid) {
/* j <-> mid, new mid is j */
mid = jj = j;
} else {
/* i <-> j */
jj = j;
j -= qsz;
}
goto swap;
}
if (i == mid) {
break;
} else {
/* i <-> mid, new mid is i */
jj = mid;
tmp = mid = i; /* value of i after swap */
j -= qsz;
}
swap:
ii = qsz;
do {
c = *i;
*i++ = *jj;
*jj++ = c;
} while (--ii);
i = tmp;
}
/*
* Look at sizes of the two partitions, do the smaller
* one first by recursion, then do the larger one by
* making sure lo is its size, base and max are update
* correctly, and branching back. But only repeat
* (recursively or by branching) if the partition is
* of at least size THRESH.
*/
i = (j = mid) + qsz;
if ((lo = j - base) <= (hi = max - i)) {
if (lo >= thresh) {
qst(base, j);
}
base = i;
lo = hi;
} else {
if (hi >= thresh) {
qst(i, max);
}
max = j;
}
} while (lo >= thresh);
}