RFR 8146991: Introduce per-worker preserved mark stacks in ParallelGC

Tony Printezis tprintezis at twitter.com
Fri Mar 4 21:42:20 UTC 2016


Hi Thomas,

Cool, thanks. So, what else should we do on this front? Do you want me to also refactor G1 to use the new abstractions? Or are you guys happy with the way G1 is right now (last time I looked at the code I thought there was some opportunity for re-use). Also, there’s an additional change to cut down on the number of preserved marks that are pushed on he stacks (given that, as we discussed on a separate e-mail thread, when biased locking is enabled most objects do not have 0x1 as their mark word). Interested in both?

Tony

On March 4, 2016 at 11:52:34 AM, Thomas Schatzl (thomas.schatzl at oracle.com) wrote:

Hi,  

On Fri, 2016-03-04 at 11:14 -0500, Tony Printezis wrote:  
> Thomas,  
>  
> I confirmed that the code below was incorrect and doing the addition  
> before calling restore() fixes the issue. Latest webrev here:  
>  
> http://cr.openjdk.java.net/~tonyp/8146991/webrev.2/  
>  

looks good to me. JPRT has been fine, although I messed up running  
the other tests - I forgot to specify ParallelGC as collector to use.  

I will redo that on Monday. I do not expect any issues.  

Thanks,  
Thomas  

-----

Tony Printezis | JVM/GC Engineer / VM Team | Twitter

@TonyPrintezis
tprintezis at twitter.com

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openjdk.java.net/pipermail/hotspot-gc-dev/attachments/20160304/203f487b/attachment.html>


More information about the hotspot-gc-dev mailing list