harmony-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Dmitry Pronichkin (JIRA)" <j...@apache.org>
Subject [jira] Updated: (HARMONY-3584) [drlvm][jit][opt] arraycopy folding implementation in guarded manner
Date Thu, 03 Apr 2008 16:00:26 GMT

     [ https://issues.apache.org/jira/browse/HARMONY-3584?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel

Dmitry Pronichkin updated HARMONY-3584:

    Attachment: arraycopy_upto03042008.patch


Here's a up-to-date patch carrying arraycopy optimization out of translator.
I've tested arraycopy-intensive benchmarks and unfortunately they do not benefit from new
code generation (due to small sizes of copied arrays) despite micro-benches do very well.
And I haven't tried to find any other solution. But with patch it would be easier to check
new ideas.
So I leave an old code for arraycopy generation.

Some moments:
- genArraCopy option is removed with the patch. Do we need to keep it?
- Write barriers work when copying primitive type arrays.


> [drlvm][jit][opt] arraycopy folding implementation in guarded manner
> --------------------------------------------------------------------
>                 Key: HARMONY-3584
>                 URL: https://issues.apache.org/jira/browse/HARMONY-3584
>             Project: Harmony
>          Issue Type: Sub-task
>          Components: DRLVM
>            Reporter: George Timoshenko
>            Assignee: Mikhail Fursov
>            Priority: Minor
>         Attachments: arraycopy.patch, arraycopy_pass.patch, arraycopy_upto03042008.patch,
HARMONY-3584_arraycopy_up_to_date.patch, HARMONY-3584_arraycopy_up_to_date2.patch, intrinsic_patch.patch
> the feature (arraycopy call replacement) can be done in the guarded manner. When the
classes of the arrays are being checked at runtime. 
> (It is not obvious if it gives a performance effect or not). 
> when this approach (guarded arraycopy call "inlining") is being implemented this optimisation
should be moved out from the translator and inserted (before inliner) into the HLO path as
a separate optimisation. 

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

View raw message