Changeset: 045c3e7b8a55 for MonetDB
URL: https://dev.monetdb.org/hg/MonetDB/rev/045c3e7b8a55
Modified Files:
gdk/gdk_bat.c
Branch: persistent_sketches
Log Message:
COLcopy2 and BUNappendmulti revisited for sketches
diffs (244 lines):
diff --git a/gdk/gdk_bat.c b/gdk/gdk_bat.c
--- a/gdk/gdk_bat.c
+++ b/gdk/gdk_bat.c
@@ -354,6 +354,12 @@ COLnew2(oid hseq, int tt, BUN cap, role_
goto bailout;
}
+ if (bn->tvheap && width == 0 && ATOMheap(tt, bn->tvheap, cap) !=
GDK_SUCCEED) {
+ HEAPfree(bn->theap, true);
+ HEAPfree(bn->tsketch, true);
+ goto bailout;
+ }
+
if (ATOMsketchable(tt) && role == PERSISTENT) {
assert(bn->tsketch);
if (HEAPalloc(bn->tsketch, SKETCH_SIZE, sizeof(uint8_t)) !=
GDK_SUCCEED) {
@@ -363,12 +369,6 @@ COLnew2(oid hseq, int tt, BUN cap, role_
memset(bn->tsketch->base, 0, SKETCH_SIZE);
}
- if (bn->tvheap && width == 0 && ATOMheap(tt, bn->tvheap, cap) !=
GDK_SUCCEED) {
- HEAPfree(bn->theap, true);
- HEAPfree(bn->tsketch, true);
- goto bailout;
- }
-
BBPcacheit(bn, true);
TRC_DEBUG(ALGO, "-> " ALGOBATFMT "\n", ALGOBATPAR(bn));
return bn;
@@ -924,8 +924,9 @@ COLcopy2(BAT *b, int tt, bool writable,
memcpy(bn->theap->base, bi.base, bi.hfree);
bn->theap->free = bi.hfree;
bn->theap->dirty = true;
- if (bn->tsketch && bi.skh) {
- memcpy(bn->theap->base, bi.skh, SKETCH_SIZE);
+ if (bn->tsketch && b->tsketch && bi.count) {
+ assert(b->tsketch->base);
+ memcpy(bn->theap->base, b->tsketch->base, SKETCH_SIZE);
bn->tsketch->dirty = true;
}
@@ -945,10 +946,16 @@ COLcopy2(BAT *b, int tt, bool writable,
if (bunfastapp_nocheck(bn, t) != GDK_SUCCEED)
goto bunins_failed;
}
- if (bn->tsketch) {
- if (bat_sketch(&bi, NULL, (uint8_t (*)[CLZ_BUCKETS])
bn->tsketch->base) != GDK_SUCCEED)
- goto bunins_failed;
- bn->tsketch->dirty = true;
+ if (bn->tsketch && bi.count) {
+ if (b->tsketch) {
+ assert(b->tsketch->base);
+ memcpy(bn->theap->base, b->tsketch->base,
SKETCH_SIZE);
+ bn->tsketch->dirty = true;
+ } else {
+ if (bat_sketch(&bi, NULL, (uint8_t
(*)[CLZ_BUCKETS]) bn->tsketch->base) != GDK_SUCCEED)
+ goto bunins_failed;
+ bn->tsketch->dirty = true;
+ }
}
TIMEOUT_CHECK(qry_ctx,
GOTO_LABEL_TIMEOUT_HANDLER(bunins_failed, qry_ctx));
bn->theap->dirty |= bi.count > 0;
@@ -980,8 +987,8 @@ COLcopy2(BAT *b, int tt, bool writable,
memcpy(Tloc(bn, 0), bi.base, bn->theap->free);
bn->theap->dirty |= bi.count > 0;
algo = "memcpy heap";
- if (bn->tsketch) {
- memcpy(bn->tsketch->base, bi.skh, SKETCH_SIZE);
+ if (bn->tsketch && bi.count) {
+ memcpy(bn->tsketch->base, b->tsketch->base,
SKETCH_SIZE);
bn->tsketch->dirty = true;
}
}
@@ -990,11 +997,10 @@ COLcopy2(BAT *b, int tt, bool writable,
/* set properties (note that types may have changed in the copy) */
if (ATOMtype(tt) == ATOMtype(bi.type)) {
- if (ATOMtype(tt) == TYPE_oid) {
+ if (ATOMtype(tt) == TYPE_oid)
BATtseqbase(bn, bi.tseq);
- } else {
+ else
BATtseqbase(bn, oid_nil);
- }
BATkey(bn, bi.key);
bn->tsorted = bi.sorted;
bn->trevsorted = bi.revsorted;
@@ -1010,8 +1016,16 @@ COLcopy2(BAT *b, int tt, bool writable,
bn->tnil = bi.nil;
bn->tminpos = bi.minpos;
bn->tmaxpos = bi.maxpos;
- if (!bi.key)
- bn->tunique_est = bi.unique_est;
+ if (!bi.key) {
+ if (bi.unique_est) {
+ bn->tunique_est = bi.unique_est;
+ } else {
+ if (bn->tsketch) {
+ assert(bn->tsketch->base);
+ bn->tunique_est =
sketch_estimate((uint8_t (*)[CLZ_BUCKETS]) bn->tsketch->base);
+ }
+ }
+ }
} else if (ATOMstorage(tt) == ATOMstorage(b->ttype) &&
ATOMcompare(tt) == ATOMcompare(b->ttype)) {
BUN h = bi.count;
@@ -1038,14 +1052,32 @@ COLcopy2(BAT *b, int tt, bool writable,
}
bn->tminpos = bi.minpos;
bn->tmaxpos = bi.maxpos;
- if (!bi.key)
- bn->tunique_est = bi.unique_est;
+ if (!bi.key) {
+ if (bi.unique_est) {
+ bn->tunique_est = bi.unique_est;
+ } else {
+ if (bn->tsketch) {
+ assert(bn->tsketch->base);
+ bn->tunique_est =
sketch_estimate((uint8_t (*)[CLZ_BUCKETS]) bn->tsketch->base);
+ }
+ }
+ }
} else {
bn->tsorted = bn->trevsorted = false; /* set based on count
later */
bn->tnonil = bn->tnil = false;
bn->tkey = false;
bn->tnosorted = bn->tnorevsorted = 0;
bn->tnokey[0] = bn->tnokey[1] = 0;
+ if (!bi.key) {
+ if (bi.unique_est) {
+ bn->tunique_est = bi.unique_est;
+ } else {
+ if (bn->tsketch) {
+ assert(bn->tsketch->base);
+ bn->tunique_est =
sketch_estimate((uint8_t (*)[CLZ_BUCKETS]) bn->tsketch->base);
+ }
+ }
+ }
}
if (BATcount(bn) <= 1) {
bn->tsorted = ATOMlinear(b->ttype);
@@ -1088,7 +1120,8 @@ BUNappendmulti(BAT *b, const void *value
if (count == 0)
return GDK_SUCCEED;
- TRC_DEBUG(ALGO, ALGOBATFMT " appending " BUNFMT " values%s\n",
ALGOBATPAR(b), count, values ? "" : " (all nil)");
+ TRC_DEBUG(ALGO, ALGOBATFMT " appending " BUNFMT " values%s\n",
+ ALGOBATPAR(b), count, values ? "" : " (all nil)");
p = BATcount(b); /* insert at end */
if (p == BUN_MAX || BATcount(b) + count >= BUN_MAX) {
@@ -1103,11 +1136,9 @@ BUNappendmulti(BAT *b, const void *value
if (b->ttype == TYPE_void && BATtdense(b)) {
const oid *ovals = values;
bool dense = b->batCount == 0 || (ovals != NULL && b->tseqbase
+ 1 == ovals[0]);
- if (ovals) {
- for (BUN i = 1; dense && i < count; i++) {
+ if (ovals)
+ for (BUN i = 1; dense && i < count; i++)
dense = ovals[i - 1] + 1 == ovals[i];
- }
- }
if (dense) {
MT_lock_set(&b->theaplock);
assert(BATtdense(b)); /* no change (coverity) */
@@ -1123,9 +1154,8 @@ BUNappendmulti(BAT *b, const void *value
}
}
- if (!b->ustr && unshare_varsized_heap(b) != GDK_SUCCEED) {
+ if (!b->ustr && unshare_varsized_heap(b) != GDK_SUCCEED)
return GDK_FAIL;
- }
if (BATcount(b) + count > BATcapacity(b)) {
/* if needed space exceeds a normal growth extend just
@@ -1262,16 +1292,15 @@ BUNappendmulti(BAT *b, const void *value
nunique = b->thash ? b->thash->nunique : 0;
}
if (b->tsketch) {
- /* printf("sketch_update_many updating dense
for bat id %d\n", b->batCacheid); */
- if (sketch_update_many(&bi, (uint8_t
(*)[CLZ_BUCKETS]) bi.skh->base,
+ printf("BUNappendmulti updating dense for bat
id %d\n", b->batCacheid);
+ MT_lock_set(&b->theaplock);
+ if (sketch_update_many(&bi, (uint8_t
(*)[CLZ_BUCKETS]) b->tsketch->base,
values, count) !=
GDK_SUCCEED) {
MT_rwlock_wrunlock(&b->thashlock);
- if (minbound)
- VALclear(&minprop);
- if (maxbound)
- VALclear(&maxprop);
+ MT_lock_unset(&b->theaplock);
return GDK_FAIL;
}
+ MT_lock_unset(&b->theaplock);
}
} else if (ATOMstorage(b->ttype) == TYPE_msk) {
bi.minpos = bi.maxpos = BUN_NONE;
@@ -1320,12 +1349,15 @@ BUNappendmulti(BAT *b, const void *value
}
nunique = b->thash ? b->thash->nunique : 0;
if (b->tsketch) {
- /* printf("sketch_update_many updating dense
for bat id %d\n", b->batCacheid); */
- if(sketch_update_many(&bi, (uint8_t
(*)[CLZ_BUCKETS]) bi.skh->base,
+ printf("BUNappendmulti updating dense for bat
id %d\n", b->batCacheid);
+ MT_lock_set(&b->theaplock);
+ if(sketch_update_many(&bi, (uint8_t
(*)[CLZ_BUCKETS]) b->tsketch->base,
values, count) !=
GDK_SUCCEED) {
MT_rwlock_wrunlock(&b->thashlock);
+ MT_lock_unset(&b->theaplock);
return GDK_FAIL;
}
+ MT_lock_unset(&b->theaplock);
}
}
} else {
@@ -1351,8 +1383,8 @@ BUNappendmulti(BAT *b, const void *value
}
b->tminpos = bi.minpos;
b->tmaxpos = bi.maxpos;
- if (count > BATcount(b) / gdk_unique_estimate_keep_fraction)
- b->tunique_est = 0;
+ /* if (count > BATcount(b) / gdk_unique_estimate_keep_fraction) */
+ /* b->tunique_est = 0; */
if (b->ttype == TYPE_oid) {
/* spend extra effort on oid (possible candidate list) */
@@ -1485,11 +1517,13 @@ BUNappendmulti(BAT *b, const void *value
b->tsorted = b->trevsorted = b->tkey = false;
}
BATsetcount(b, p);
- if (nunique != 0)
+ if (nunique != 0) {
+ if (b->tsketch && b->tsketch->base)
+ printf("BUNappendmulti would be estimation %f for
unique bat id %d\n", sketch_estimate((uint8_t (*)[CLZ_BUCKETS])
b->tsketch->base), b->batCacheid);
b->tunique_est = (double) nunique;
- else if (b->tsketch) {
- /* printf("BUNappendmulti estimating for bat id %d\n",
b->batCacheid); */
+ } else if (b->tsketch) {
b->tunique_est = sketch_estimate((uint8_t (*)[CLZ_BUCKETS])
b->tsketch->base);
+ printf("BUNappendmulti estimating %f for bat id %d\n",
b->tunique_est, b->batCacheid);
}
MT_lock_unset(&b->theaplock);
MT_rwlock_wrunlock(&b->thashlock);
_______________________________________________
checkin-list mailing list -- [email protected]
To unsubscribe send an email to [email protected]
