Changeset: 9cbea7bc661c for MonetDB
URL: http://dev.monetdb.org/hg/MonetDB?cmd=changeset;node=9cbea7bc661c
Modified Files:
gdk/gdk_join.c
Branch: default
Log Message:
Implemented specialized versions of mergejoin for int and lng.
The most common set of options are fixed in these implementations.
diffs (truncated from 632 to 300 lines):
diff --git a/gdk/gdk_join.c b/gdk/gdk_join.c
--- a/gdk/gdk_join.c
+++ b/gdk/gdk_join.c
@@ -1001,6 +1001,608 @@ mergejoin_void(BAT *r1, BAT *r2, BAT *l,
* r; otherwise all matches are returned.
*/
static gdk_return
+mergejoin_int(BAT *r1, BAT *r2, BAT *l, BAT *r,
+ int nil_matches, BUN maxsize, lng t0, int swapped)
+{
+ BUN lstart, lend;
+ BUN rstart, rend;
+ BUN lscan, rscan; /* opportunistic scan window */
+ const int *lvals, *rvals;
+ int v;
+ BUN nl, nr;
+ oid lv;
+ BUN i;
+
+ ALGODEBUG fprintf(stderr, "#mergejoin_int(l=%s#" BUNFMT "[%s]%s%s%s,"
+ "r=%s#" BUNFMT "[%s]%s%s%s)%s\n",
+ BATgetId(l), BATcount(l), ATOMname(l->ttype),
+ l->tsorted ? "-sorted" : "",
+ l->trevsorted ? "-revsorted" : "",
+ l->tkey & 1 ? "-key" : "",
+ BATgetId(r), BATcount(r), ATOMname(r->ttype),
+ r->tsorted ? "-sorted" : "",
+ r->trevsorted ? "-revsorted" : "",
+ r->tkey & 1 ? "-key" : "",
+ swapped ? " swapped" : "");
+
+ assert(BAThdense(l));
+ assert(BAThdense(r));
+ assert(ATOMtype(l->ttype) == ATOMtype(r->ttype));
+ assert(r->tsorted || r->trevsorted);
+
+ lstart = rstart = 0;
+ lend = BATcount(l);
+ rend = BATcount(r);
+ lvals = (const int *) Tloc(l, BUNfirst(l));
+ rvals = (const int *) Tloc(r, BUNfirst(r));
+ assert(!r->tvarsized || !r->ttype);
+
+ /* basic properties will be adjusted if necessary later on,
+ * they were initially set by joininitresults() */
+
+ if (lend == 0 || rend == 0) {
+ /* there are no matches */
+ return nomatch(r1, r2, l, r, lstart, lend, NULL, NULL,
+ 0, 0, "mergejoin_int", t0);
+ }
+
+ /* determine opportunistic scan window for l and r */
+ for (nl = lend - lstart, lscan = 4; nl > 0; lscan++)
+ nl >>= 1;
+ for (nr = rend - rstart, rscan = 4; nr > 0; rscan++)
+ nr >>= 1;
+
+ if (!nil_matches) {
+ /* skip over nils at the start of the columns */
+ if (lscan < lend - lstart && lvals[lstart + lscan] == int_nil) {
+ lstart = binsearch_int(NULL, 0, lvals, lstart + lscan,
+ lend - 1, int_nil, 1, 1);
+ } else {
+ while (lvals[lstart] == int_nil)
+ lstart++;
+ }
+ if (rscan < rend - rstart && rvals[rstart + rscan] == int_nil) {
+ rstart = binsearch_int(NULL, 0, rvals, rstart + rscan,
+ rend - 1, int_nil, 1, 1);
+ } else {
+ while (rvals[rstart] == int_nil)
+ rstart++;
+ }
+ }
+ /* from here on we don't have to worry about nil values */
+
+ while (lstart < lend && rstart < rend) {
+ v = rvals[rstart];
+
+ if (lscan < lend - lstart && lvals[lstart + lscan] < v) {
+ lstart = binsearch_int(NULL, 0, lvals, lstart + lscan,
+ lend - 1, v, 1, 0);
+ } else {
+ /* scan l for v */
+ while (lstart < lend && lvals[lstart] < v)
+ lstart++;
+ }
+ if (lstart >= lend) {
+ /* nothing found */
+ break;
+ }
+
+ /* Here we determine the next value in l that we are
+ * going to try to match in r. We will also count the
+ * number of occurrences in l of that value.
+ * Afterwards, v points to the value and nl is the
+ * number of times it occurs. Also, lstart will
+ * point to the next value to be considered (ready for
+ * the next iteration).
+ * If there are many equal values in l (more than
+ * lscan), we will use binary search to find the end
+ * of the sequence. Obviously, we can do this only if
+ * l is actually sorted (lscan > 0). */
+ nl = 1; /* we'll match (at least) one in l */
+ nr = 0; /* maybe we won't match anything in r */
+ v = lvals[lstart];
+ if (l->tkey) {
+ /* if l is key, there is a single value */
+ lstart++;
+ } else if (lscan < lend - lstart &&
+ v == lvals[lstart + lscan]) {
+ /* lots of equal values: use binary search to
+ * find end */
+ nl = binsearch_int(NULL, 0, lvals, lstart + lscan,
+ lend - 1, v, 1, 1);
+ nl -= lstart;
+ lstart += nl;
+ } else {
+ /* just scan */
+ while (++lstart < lend && v == lvals[lstart])
+ nl++;
+ }
+ /* lstart points one beyond the value we're
+ * going to match: ready for the next iteration. */
+
+ /* First we find the first value in r that is at
+ * least as large as v, then we find the first
+ * value in r that is larger than v. The difference
+ * is the number of values equal to v and is stored in
+ * nr.
+ * We will use binary search on r to find both ends of
+ * the sequence of values that are equal to v in case
+ * the position is "too far" (more than rscan
+ * away). */
+
+ /* first find the location of the first value in r
+ * that is >= v, then find the location of the first
+ * value in r that is > v; the difference is the
+ * number of values equal to v */
+
+ /* look ahead a little (rscan) in r to see whether
+ * we're better off doing a binary search */
+ if (rscan < rend - rstart && rvals[rstart + rscan] < v) {
+ /* value too far away in r: use binary
+ * search */
+ rstart = binsearch_int(NULL, 0, rvals, rstart + rscan,
+ rend - 1, v, 1, 0);
+ } else {
+ /* scan r for v */
+ while (rstart < rend && rvals[rstart] < v)
+ rstart++;
+ }
+ if (rstart == rend) {
+ /* nothing found */
+ break;
+ }
+
+ /* now find the end of the sequence of equal values v */
+
+ /* if r is key, there is zero or one match, otherwise
+ * look ahead a little (rscan) in r to see whether
+ * we're better off doing a binary search */
+ if (r->tkey) {
+ if (rstart < rend && v == rvals[rstart]) {
+ nr = 1;
+ rstart++;
+ }
+ } else if (rscan < rend - rstart &&
+ v == rvals[rstart + rscan]) {
+ /* range too large: use binary search */
+ nr = binsearch_int(NULL, 0, rvals, rstart + rscan,
+ rend - 1, v, 1, 1);
+ nr -= rstart;
+ rstart += nr;
+ } else {
+ /* scan r for end of range */
+ while (rstart < rend && v == rvals[rstart]) {
+ nr++;
+ rstart++;
+ }
+ }
+ /* rstart points to first value > v or end of
+ * r, and nr is the number of values in r that
+ * are equal to v */
+ if (nr == 0) {
+ /* no entries in r found */
+ continue;
+ }
+ /* make space: nl values in l match nr values in r, so
+ * we need to add nl * nr values in the results */
+ if (BATcount(r1) + nl * nr > BATcapacity(r1)) {
+ /* make some extra space by extrapolating how
+ * much more we need (fraction of l we've seen
+ * so far is used as the fraction of the
+ * expected result size we've produced so
+ * far) */
+ BUN newcap = (BUN) ((double) BATcount(l) / (BATcount(l)
- (lend - lstart)) * (BATcount(r1) + nl * nr) * 1.1);
+ if (newcap < nl * nr + BATcount(r1))
+ newcap = nl * nr + BATcount(r1) + 1024;
+ if (newcap > maxsize)
+ newcap = maxsize;
+ /* make sure heap.free is set properly before
+ * extending */
+ BATsetcount(r1, BATcount(r1));
+ if (BATextend(r1, newcap) != GDK_SUCCEED)
+ goto bailout;
+ BATsetcount(r2, BATcount(r2));
+ if (BATextend(r2, newcap) != GDK_SUCCEED)
+ goto bailout;
+ assert(BATcapacity(r1) == BATcapacity(r2));
+ }
+
+ /* maintain properties */
+ if (nl > 1) {
+ /* value occurs multiple times in l, so entry
+ * in r will be repeated multiple times: hence
+ * r2 is not key and not dense */
+ r2->tkey = 0;
+ r2->tdense = 0;
+ /* multiple different values will be inserted
+ * in r1 (always in order), so not reverse
+ * ordered anymore */
+ r1->trevsorted = 0;
+ }
+ if (nr > 1) {
+ /* value occurs multiple times in r, so entry
+ * in l will be repeated multiple times: hence
+ * r1 is not key and not dense */
+ r1->tkey = 0;
+ r1->tdense = 0;
+ /* multiple different values will be inserted
+ * in r2 (in order), so not reverse ordered
+ * anymore */
+ r2->trevsorted = 0;
+ if (nl > 1) {
+ /* multiple values in l match multiple
+ * values in r, so an ordered sequence
+ * will be inserted multiple times in
+ * r2, so r2 is not ordered anymore */
+ r2->tsorted = 0;
+ }
+ }
+ if (BATcount(r1) > 0) {
+ /* a new, higher value will be inserted into
+ * r1, so r1 is not reverse ordered anymore */
+ r1->trevsorted = 0;
+ /* a new higher value will be added to r2 */
+ r2->trevsorted = 0;
+ if (r1->tdense &&
+ ((oid *) r1->T->heap.base)[r1->batFirst +
r1->batCount - 1] + 1 != l->hseqbase + lstart - nl)
+ r1->tdense = 0;
+ }
+
+ if (BATcount(r2) > 0 &&
+ r2->tdense &&
+ ((oid *) r2->T->heap.base)[r2->batFirst + r2->batCount - 1]
+ 1 != r->hseqbase + rstart - nr)
+ r2->tdense = 0;
+
+ /* insert values */
+ lv = l->hseqbase + lstart - nl;
+ for (i = 0; i < nl; i++) {
+ BUN j;
+ oid rv;
+
+ rv = r->hseqbase + rstart - nr;
+ for (j = 0; j < nr; j++) {
+ APPEND(r1, lv);
+ APPEND(r2, rv);
+ rv++;
+ }
+ lv++;
+ }
+ }
+ /* also set other bits of heap to correct value to indicate size */
+ BATsetcount(r1, BATcount(r1));
+ BATsetcount(r2, BATcount(r2));
+ assert(BATcount(r1) == BATcount(r2));
+ if (BATcount(r1) > 0) {
+ if (r1->tdense)
+ r1->tseqbase = ((oid *) r1->T->heap.base)[r1->batFirst];
+ if (r2->tdense)
+ r2->tseqbase = ((oid *) r2->T->heap.base)[r2->batFirst];
+ }
+ BATseqbase(r1, 0);
+ BATseqbase(r2, 0);
+ ALGODEBUG fprintf(stderr,
"#mergejoin_int(l=%s,r=%s)=(%s#"BUNFMT"%s%s%s%s,%s#"BUNFMT"%s%s%s%s) " LLFMT
"us\n",
+ BATgetId(l), BATgetId(r),
+ BATgetId(r1), BATcount(r1),
+ r1->tsorted ? "-sorted" : "",
+ r1->trevsorted ? "-revsorted" : "",
+ r1->tdense ? "-dense" : "",
+ r1->tkey & 1 ? "-key" : "",
+ BATgetId(r2), BATcount(r2),
+ r2->tsorted ? "-sorted" : "",
+ r2->trevsorted ? "-revsorted" : "",
+ r2->tdense ? "-dense" : "",
+ r2->tkey & 1 ? "-key" : "",
+ GDKusec() - t0);
+ return GDK_SUCCEED;
_______________________________________________
checkin-list mailing list
checkin-list@monetdb.org
https://www.monetdb.org/mailman/listinfo/checkin-list
