Update of /cvsroot/monetdb/pathfinder/compiler/Tests
In directory sc8-pr-cvs16:/tmp/cvs-serv26652/Tests
Modified Files:
usageH.stable.out
Log Message:
Thetajoin operator:
Introduced a thetajoin operator in the logical algebra (and the physical
algebra and the MIL generation). This new thetajoin can handle a list of
conjunctive predicates where each predicate represents an (in-)equality
condition.
Thetajoin introduction:
A thetajoin operator is introduced either by a small pattern
(select-comparison-cross) in opt_general.brg or by a new introduction
phase in intro_thetajoin.c. The latter introduction phase uses a
selection as a basis to find the *correct* equi-join and transforms
this equi-join into a thetajoin (for more details please look into
the code).
Thetajoin optimization:
Similar to the MVD optimization phase that pushes cross product
operators up in the DAG structure this checkin provides a new
optimization phase that pushes thetajoin operators up in the DAG
(opt_thetajoin.c). These rewrites result in thetajoin operators
that contain multiple predicates. (E.g. the transformed equi-joins
eventually hit the selection that triggered the rewrite and thus form
a new scope-dependent value-based join -- see XMark Q3 and Q4.)
Additional changes:
opt_general.brg
* Extended thetajoin pattern to cope with inequality comparisons
* Rewrite the pattern: semijoin (distinct (Rel), Rel) into
distinct (semijoin (Rel, Rel) as the semijoin hopefully reduces
the cardinality more than the distinct operator. (This speeds
up XMark Q12 by a factor of 2.5.)
intro_proxy.c
* Added a preceding phase that removes all semijoin operators.
(Using only simply checks more proxies are now detected.)
* Removed unnest optimization phase (thetajoin optimization
does the same in a more general scenario).
physical.c
* Toyed around with the cost model
* Fixed 'PFord_order_dir_at' bug
planner.c
* Generated more ordered alternatives (even if an ordering is
not required) after an operator is mapped to physical algebra.
In some situations this avoids in a later step planning the same
sort operator multiple times. (We only introduce orderings for
iter-like columns and sort at most two columns. Otherwise the
plan space explodes for some sample queries.)
* Extended list of possible orderings for the rownum operator
Index: usageH.stable.out
===================================================================
RCS file: /cvsroot/monetdb/pathfinder/compiler/Tests/usageH.stable.out,v
retrieving revision 1.26
retrieving revision 1.27
diff -u -d -r1.26 -r1.27
--- usageH.stable.out 21 Mar 2007 10:06:46 -0000 1.26
+++ usageH.stable.out 7 May 2007 10:17:03 -0000 1.27
@@ -79,7 +79,7 @@
P infer all properties
(used for debug output and physical algebra
_ does nothing (used for structuring the options)
- (default is: '-o
OIKDCG_VGO_[J]OKVCG}IM_{_[J]OKVCG}IM_{_[J]OKVCGCG}IM_{_[J]OKVCG}IMS{_[J]OKVCGCG}IMS{_[J]OKVCG}IMS{_[J]OKVCGCGP')
+ (default is: '-o
OIKDCG_VGO_[J]OKVCG}IM__{_[J]OKVCG}IM__{_[J]OKVCGCG}IM__{_[J]OKVCG}IMTS{_[J]OKVCGCG}IMTS{_[J]OKVCG}IMTS{_[J]OKVCGCGP')
= -e[0|1], --dead-code-elimination=[0|1]: dead code elimination:
0 disable dead code elimination
1 enable dead code elimination (default)
-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/
_______________________________________________
Monetdb-pf-checkins mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/monetdb-pf-checkins
