[
https://issues.apache.org/jira/browse/IMPALA-4065?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16506691#comment-16506691
]
Tianyi Wang commented on IMPALA-4065:
-------------------------------------
I observed more than 2X speed up in the TopNNode with an "order by int_col"
extreme case, but not so much for the sort case in IMPALA-3816. It seems LLVM
doesn't optimize the template code well if there are indirect calls among them.
> Inline comparator calls into TopN::InsertBatch()
> ------------------------------------------------
>
> Key: IMPALA-4065
> URL: https://issues.apache.org/jira/browse/IMPALA-4065
> Project: IMPALA
> Issue Type: Bug
> Components: Backend
> Affects Versions: Impala 2.7.0
> Reporter: Tim Armstrong
> Assignee: Tianyi Wang
> Priority: Minor
> Labels: codegen, ramp-up
>
> This is the more interesting follow-on from IMPALA-3815. We should inline the
> Compare() calls in the codegen'd TopN code to avoid the indirect function
> pointer call.
> The tricky aspect is that the Compare() calls are called from
> std::priority_queue, and we don't have a way to force-inline those functions
> at the moment.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
