RFR: Consistent liveness for humongous regions

[Roman Kennke]

Hi Aleksey,
thanks for improving the code. I pushed your version.

Roman


On Fri, Feb 23, 2018 at 10:43 AM, Aleksey Shipilev <shade at redhat.com> wrote:
> On 02/22/2018 10:45 PM, Roman Kennke wrote:
>> When counting liveness on humongous regions, we current give all
>> liveness to the start region, and leave the rest alone. For this
>> reason, we have put exceptions to several asserts. This patch makes
>> liveness counting more consistent for humongous regions by spreading
>> it over all involved regions, just as it is done at allocation time.
>> It also removes the exceptions in the asserts. It also removes an
>> exception for traversal-gc, which I believe is fixed by this mornings
>> liveness init patch.
>>
>> Passes hotspot_gc_shenandoah
>>
>> http://cr.openjdk.java.net/~rkennke/fix-humongous-live/webrev.00/
>
> We need to optimize for the most frequent case (normal objects) better. Right now it does two
> heap_region_index_containing calls per object. We can also use utility methods to calculate things.
>
> Better version:
>  http://cr.openjdk.java.net/~shade/shenandoah/humongous-accounting/webrev.00/
>
> This has the second changeset you can apply over yours:
>  http://cr.openjdk.java.net/~shade/shenandoah/humongous-accounting/webrev.00/shenandoah-jdk.changeset
>
> -Aleksey
>