RFR: Scan remembered [v3]

[Kelvin Nilsen]

On Fri, 15 Jan 2021 12:46:30 GMT, Roman Kennke <rkennke at openjdk.org> wrote:

>> As discussed in email messages exchanged on shenandoah-dev at openjdk.java.net, there are potential synergies between the SATB-based remembered set implementation and the existing SATB pre-write barrier.  Especially if the default Shenandoah configuration chooses not to use SATB barrier, that increases the appeal of borrowing the G1 barrier implementation.
>
> If we were to choose the I-U barrier as default, that would still be possible/interesting to piggy-back the card-marking on. The I-U barrier is very similar to the SATB-barrier: it is also employed at reference-stores, and it also uses a reference-queue, pretty much like SATB. The only difference (barrier-side) is that, instead of enqueuing the previous value of a field before the store, it enqueues the *new* value, about to be stored (or even already-stored when done as post-barrier, it doesn't really matter).

Thanks for the email clarifying how the I-U barrier works.  I hadn't yet studied that.  I agree that we could also piggy-back card marking on that barrier.  I will explore further.

-------------

PR: https://git.openjdk.java.net/shenandoah/pull/12