You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@drill.apache.org by "Jinfeng Ni (JIRA)" <ji...@apache.org> on 2014/02/04 20:20:10 UTC
[jira] [Updated] (DRILL-346) Runtime generated code should evaluate
constant expression once in doSetup method
[ https://issues.apache.org/jira/browse/DRILL-346?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Jinfeng Ni updated DRILL-346:
-----------------------------
Attachment: (was: DRILL_346.1.patch.txt)
> Runtime generated code should evaluate constant expression once in doSetup method
> ---------------------------------------------------------------------------------
>
> Key: DRILL-346
> URL: https://issues.apache.org/jira/browse/DRILL-346
> Project: Apache Drill
> Issue Type: Bug
> Reporter: Jinfeng Ni
> Assignee: Jinfeng Ni
> Attachments: DRILL_346.2.patch.txt
>
>
> Currently, when runtime CodeGenerator logic generates class to evaluate an expression, it would put the code in doEval() method, regardless of whether the expression is a constant, or part of the expression is a constant. This means the generated code for the constant expression will be executed once for every record, in stead of once for each record batch.
> For a constant expression, we should put the code in doSetup() method, so that it will be executed once per record batch. Doing so would improve runtime code efficiency.
> In addition, for some function call expression, if the arguments are constant, we may want to refer to the constant arguments in doSetup(). For instance, given function : regexp_replae(input_string, regex, replace), if regex is a constant (eg : regexp_replace(varcharcol, "a[bc]", "XYZ") ), the implementation of regexp_replace would use the constant to build a java.Pattern in doSetup(), and repeatedly use that java.Pattern object in doEval() for each record. This requires CodeGenerator to make the constant arguments accessible to doSetup() method for function call.
>
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)