harmony-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Xiao-Feng Li" <xiaofeng...@gmail.com>
Subject Re: [DRLVM][GCv5] (H-2945) scalable parallel generational or non-generational collection
Date Tue, 09 Jan 2007 08:07:04 GMT
Egor, thanks for the help. GenArrayCopy and GenArrayCopyRepMove are
two possible sources. I will try to identify if some other opts also
have impact on write barrier.

Thanks,
xiaofeng

On 08 Jan 2007 12:54:35 +0600, Egor Pasko <egor.pasko@gmail.com> wrote:
> On the 0x256 day of Apache Harmony Xiao-Feng Li wrote:
> > Hi, I've submitted a patch for GCv5 at URL:
> > https://issues.apache.org/jira/browse/HARMONY-2945, which achieves
> > good parallelization scalability in real SMP machines. It can works in
> > generational or nongenerational mode. I have tested it on a Intel
> > Tulsa platform which has four Pentium-D dual-core processors (totally
> > 8 cores). It performs pretty well. With more processors and larger
> > heap size, this parallel GC can demonstrate more benefits.
> >
> > There are two dependences:
> > 1. To fully exploit this GC's adavantage, helper inlining should be
> > enabled for it. It depends on Harmony-2652 which inlines write
> > barrier.
> > 2. Some JIT optimizations do not implement write barrer for reference
> > field update. These opts have to be disabled to make the generational
> > mode work.
>
> Xiao-Feng, what specific optimizations did you have to disable? I
> believe, we can re-enable them in GCv5 mode pretty fast.
>
> > At the same time before the two dependences are resolved, people still
> > can play with it in non-generational mode. But the performance will be
> > impacted seriously without helper inlinings.
> >
> > Dear commiter, please help to integrate it into Harmony SVN. Please
> > apply the patch at working_vm directory.
> >
> > Thanks,
> > xiaofeng
> >
>
> --
> Egor Pasko
>
>

Mime
View raw message