RFR 8080232: Eagerly promote objects in nmethod to reduce scanning in ParNew

Carsten Varming varming at gmail.com
Wed Feb 17 02:24:36 UTC 2016


Dear Jon,

Oops, I don't know what I was thinking when I found that ticket. I'll try
to put it back in its old state. I filed a new ticket: JDK-8150013.

I folded the pruning of the scavengable nmethod list into
CodeCache::scavenge_root_nmethods_do, thus removing the need for a separate
pruning step after the young collection.

You can find the new webrev here:
http://cr.openjdk.java.net/~cvarming/1/webrev/

Carsten



On Tue, Feb 16, 2016 at 6:53 PM, Jon Masamitsu <jon.masamitsu at oracle.com>
wrote:

> Carsten,
>
> This looks like a good change, but 8080232 describes a different problem.
> 8080232 proposed early promotion for objects in nmethods.  This change
> removes nmethods from a list when it is determined that they don't refer to
> scavengable objects.  Can you create a new CR for this?
>
> Jon
>
>
> On 02/15/2016 12:53 PM, Carsten Varming wrote:
>
> Dear GC members,
>
> I would like to contribute a fix for JDK-8080232. The bug is marked as
> resolved because it was withdrawn. The fix is simple, so I would like to
> reopen the ticket. At Twitter we have seen the time to fix relocations
> after a ParNew cycle drop by three to four orders of magnitude with this
> patch.
>
> I have put a webrev here:
> http://cr.openjdk.java.net/~cvarming/prune_scavengable_nmethods/0/webrev/
>
> I will need a sponsor.
>
> Carsten
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openjdk.java.net/pipermail/hotspot-gc-dev/attachments/20160216/363a33e3/attachment.html>


More information about the hotspot-gc-dev mailing list