harmony-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Aleksey Shipilev (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HARMONY-5718) [classlib][luni][performance] j.u.IdentityHashMap ariphmetic improvements
Date Thu, 17 Apr 2008 21:53:21 GMT

    [ https://issues.apache.org/jira/browse/HARMONY-5718?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12590213#action_12590213
] 

Aleksey Shipilev commented on HARMONY-5718:
-------------------------------------------

I have more precise idea which would require much less changes: if we need to compute the
mask then it's just (elementData.length - 1), if elementData.length is power of 2. Clean patch
will be later.

> [classlib][luni][performance] j.u.IdentityHashMap ariphmetic improvements
> -------------------------------------------------------------------------
>
>                 Key: HARMONY-5718
>                 URL: https://issues.apache.org/jira/browse/HARMONY-5718
>             Project: Harmony
>          Issue Type: Sub-task
>          Components: Classlib
>            Reporter: Aleksey Shipilev
>         Attachments: cl-identityhashmap-rc1.patch, cl-identityhashmap-rc2.patch, cl-identityhashmap-rc3.patch
>
>
> Profiling and disassembly shows that much of the time on MT/ThreadLocalBench is spent
on % and / operations in java.util.IdentityHashMap.(findIndex|getModuloHash). The dividers
though are 2^k, so we can replace them with mask and shift operations.

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


Mime
View raw message