RFR (S): Parallelize partial GC cleanup

[Zhengyu Gu]

ShenandoahCollectionSet already has claim_next() to claim regions for 
parallel processing.

Thanks,

-Zhengyu

On 05/03/2017 10:12 AM, Aleksey Shipilev wrote:
> Hi,
>
> Matrix and bitmap cleanups take significant time in partial GC. The obvious
> solution is to parallelize it at this point, since we have mined out the
> sequential improvements.
>
> Fix:
>   http://cr.openjdk.java.net/~shade/shenandoah/partial-recycle-parallel/webrev.01/
>
> Improves partial GC times significantly under heavy allocation pressure:
>   http://cr.openjdk.java.net/~shade/shenandoah/partial-recycle-parallel/perf.txt
>
> Testing: hotspot_gc_shenandoah
>
> Thanks,
> -Aleksey
>