RFR: Reference processors might use non-forwarded alive checks

[Roman Kennke]

Am 29.05.2017 um 17:14 schrieb Aleksey Shipilev:
> Hi,
>
> Current RPs are using ShenandoahIsForwardedAliveClosure, which always go via
> fwdptr, assuming there could be a forwarded copy. However, with update-refs
> phase, this is excessive.
>
> Therefore, it makes sense to specialize for this case:
>   http://cr.openjdk.java.net/~shade/shenandoah/refproc-isalive/webrev.01/
>
> Improves concmark times, class unloading and weak refs a bit:
>   http://cr.openjdk.java.net/~shade/shenandoah/refproc-isalive/before.txt
>   http://cr.openjdk.java.net/~shade/shenandoah/refproc-isalive/after.txt
>
> Testing: hotspot_gc_shenandoah
>
> -Aleksey
>
>
Ok