Hello Michael Ho,
I'd like you to reexamine a change. Please visit
http://gerrit.cloudera.org:8080/7081
to look at the new patch set (#2).
Change subject: IMPALA-5347: reduce codegen overhead of timestamp trunc()
......................................................................
IMPALA-5347: reduce codegen overhead of timestamp trunc()
Trunc has many implementations that are switched between based on a
string argument. Before this patch all implementations were compiled for
every call to trunc(), which added a lot of unnecessary codegen time.
This patch avoids the problem by moving the implementation out of the
cross-compiled code.
Testing:
Ran expr-test.
I ran the repro query from IMPALA-5347 and verified that codegen time
was significantly reduced from ~1.4s to ~.35s.
Perf:
I ran the following targeted benchmark:
set num_nodes=1;
set num_scanner_threads=1;
select count(*) from lineitem where trunc(l_shipdate, 'yy') >=
'1998-01-01'
The end-to-end query latency was reduced to 0.52s from 0.72s on
average. The time spent in the scanner increased slightly from
around 390ms to around 410ms. This seems like a good-tradeoff.
Change-Id: I222258f51b2093a38929df847fdb5d25bb9aafc3
---
M be/src/exprs/CMakeLists.txt
M be/src/exprs/udf-builtins-ir.cc
A be/src/exprs/udf-builtins.cc
M be/src/exprs/udf-builtins.h
4 files changed, 277 insertions(+), 229 deletions(-)
git pull ssh://gerrit.cloudera.org:29418/Impala-ASF refs/changes/81/7081/2
--
To view, visit http://gerrit.cloudera.org:8080/7081
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings
Gerrit-MessageType: newpatchset
Gerrit-Change-Id: I222258f51b2093a38929df847fdb5d25bb9aafc3
Gerrit-PatchSet: 2
Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-Owner: Tim Armstrong <tarmstr...@cloudera.com>
Gerrit-Reviewer: Michael Ho <k...@cloudera.com>
