RFR: Traversal: Initialize extra liveness arrays when conc-threads > par-threads

[Roman Kennke]

Hmm right.

http://cr.openjdk.java.net/~rkennke/init-extra-liveness/webrev.01/

We could allocate liveness in one chunk and initialize it in one
batch, but it probably doesn't matter much.

Ok?

On Thu, Feb 22, 2018 at 11:26 AM, Aleksey Shipilev <shade at redhat.com> wrote:
> On 02/22/2018 11:07 AM, Roman Kennke wrote:
>> Yesterday, Zhengyu discovered why sometimes we get liveness counters
>> off when running with conc-threads > par-threads. We only initialize
>> liveness arrays in init-traversal, and we do this in the workers, so
>> only the active workers get their liveness array initialized. Duh.
>>
>> This fix checks in conc-traversal worker if the worker_id is >= the
>> last number of stw workers used, and if so, initializes its own
>> liveness array.
>>
>> Test: hotspot_gc_shenandoah
>>
>> http://cr.openjdk.java.net/~rkennke/init-extra-liveness/webrev.00/
>
> While this looks moderately okay, maybe we shouldn't cut through like that, and instead initialize
> liveness for max_workers somewhere during init-traversal? You can fill the entire _liveness_local in
> bulk then too.
>
> -Aleksey
>